the quality of surgical care for radical cystectomy … · 2010-02-08 · the quality of surgical...
TRANSCRIPT
THE QUALITY OF SURGICAL CARE FOR RADICAL CYSTECTOMY IN
ONTARIO FROM 1992 TO 2004
by
Girish Satish Kulkarni
A thesis submitted in conformity with the requirements
for the degree of Doctor of Philosophy in Clinical Epidemiology,
Graduate Department of Health Policy, Management, and Evaluation,
in the University of Toronto
© Copyright by Girish Satish Kulkarni, 2008
ii
THESIS ABSTRACT
Thesis Title: The Quality of Surgical Care for Radical Cystectomy in Ontario from 1992
to 2004
Degree: Doctor of Philosophy (PhD) in Clinical Epidemiology
Year of Convocation: 2008
Student: Girish Satish Kulkarni
Graduate Department: Health Policy, Management and Evaluation
University: University of Toronto
Background: This thesis is composed of three studies pertaining to the quality of care for
radical cystectomy in Ontario between 1992 and 2004. In the first paper, the associations
between provider volume and both operative and overall mortality were assessed. In the
second paper, potential factors that could explain the association between volume and
outcome were explored. In the final paper, the impact of waiting for cystectomy on
survival outcomes was evaluated.
Methods: A total of 3296 patients undergoing cystectomy for bladder cancer in Ontario
between 1992 and 2004 were identified using the Canadian Institute for Health
Information Discharge Abstract Database and the Ontario Cancer Registry. The effects of
hospital and surgeon volume on operative mortality and overall survival were assessed
using random effects logistic regression and marginal Cox Proportional Hazards
modeling, respectively. To elucidate the factors underlying the volume-outcome
association, the ability of a number of structure and process of care variables to attenuate
the impact of volume was assessed. The effect of waiting for care, from transurethral
resection to cystectomy, on overall survival was also assessed using marginal Cox
models.
Results: Neither hospital nor surgeon volume was significantly associated with operative
mortality; however, both were associated with overall mortality. Of the measured
iii
structure/process measures, hospital factors caused the greatest attenuation of the volume
hazard ratios, albeit to a limited degree. The wait time between the decision for surgery
and cystectomy was also significantly associated with overall survival. The impact of
delayed care was greatest for patients with lower stage disease. The data suggested a
maximum wait time of 40 days for cystectomy.
Conclusions: In this thesis, gaps in the quality of care for radical cystectomy in Ontario
were identified. Patients treated by low volume hospitals and surgeons or those with long
wait times all experienced worse outcomes. Since the underlying measures responsible
for provider volume remain elusive, additional work is required to understand what these
factors are. Initiatives to decrease wait times, however, are under way in Ontario.
Whether these interventions decrease wait times and benefit patients remains to be seen.
iv
ACKNOWLEDGEMENTS
There are a number of people I would like to thank whose contributions made this thesis
possible:
First and foremost, to Andreas Laupacis, my supervisor and mentor during the PhD
process, for keeping me on my toes with his pragmatic and insightful comments and
thoughts and for acting as a role model as a supervisor and researcher.
To Neil Fleshner, my co-supervisor and urology mentor, for facilitating my entry into the
world of urologic research and sharing with me the excitement that comes from clinical
research.
To my Thesis Committee (Peter Austin, Dave Urbach, Andreas Laupacis and Neil
Fleshner), for providing guidance and input at every stage of this thesis.
To the Institute for Clinical Evaluative Sciences (ICES), for supporting this research and
for providing an environment that fosters academic success.
To ICES personnel, who were warm and open and considerate of my inexperience as a
junior researcher: Lingsong Yun, Refik Saskin, Ruth Croxford, Gale Delaney, Pam
Slaughter, and Flora Lo.
To the Canadian Institutes of Health Research, the Division of Urology, University of
Toronto and the Surgeon Scientist Program, University of Toronto for supporting me
over the past 4 years.
To members of the Clinical Epidemiology program, Department of Health Policy,
Management and Evaluation, past and present (Gillian Hawker, Ahmed Bayoumi, Laurie
McQuarrie, Gladys Honein, Amber Gertzbein, Jennifer James), for their administrative
support and guidance for all things thesis-related.
To my fellow students and colleagues, Rob Quinn, Steven Lopushinsky, Nick Daneman,
Jensen Tan, whose humorous advice was always appreciated though not always solicited.
To my parents and parents-in-law, for their immeasurable support and help with
emergency child care that was required on a not-too-infrequent basis.
And, most important, to my loving wife, Jasmit Bhandal and our wonderful, beautiful
children, Nikhil and Naina who provided unequivocal and undying support at every stage
of this long and arduous journey.
v
TABLE OF CONTENTS
THESIS ABSTRACT ................................................................................................... II
ACKNOWLEDGEMENTS ....................................................................................... IV
LIST OF TABLES .................................................................................................... VII
LIST OF FIGURES................................................................................................. VIII
CHAPTER 1 : BACKGROUND ................................................................................... 1
BLADDER CANCER EPIDEMIOLOGY............................................................................................. 1 MANAGEMENT OF INVASIVE BLADDER CANCER...................................................................... 2 QUALITY OF CARE IN SURGERY ................................................................................................... 4 QUALITY OF CARE FOR RADICAL CYSTECTOMY ...................................................................... 6
Provider Volume and Structures/Processes of Care .......................................................................... 7 Wait Times .................................................................................................................................... 11
CURRENT KNOWLEDGE LIMITATIONS REGARDING THE QUALITY OF CARE FOR RADICAL
CYSTECTOMY ................................................................................................................................ 14 THESIS CONCEPTUAL FRAMEWORK .......................................................................................... 15 THESIS OVERVIEW ........................................................................................................................ 17 SPECIFIC OBJECTIVES ................................................................................................................... 17 DATA SOURCES AND VARIABLES ............................................................................................... 18 ETHICS STATEMENT ..................................................................................................................... 18 TABLES FOR CHAPTER 1 ............................................................................................................... 19
CHAPTER 2 : COHORT DEFINITIONS AND DESCRIPTIONS .......................... 26
PATIENT IDENTIFICATION ........................................................................................................... 26 VALIDATION .................................................................................................................................. 28 RELIABILITY (ABSTRACTOR AGREEMENT) .............................................................................. 30 THESIS COHORTS ........................................................................................................................... 31 FIGURES FOR CHAPTER 2 ............................................................................................................. 33 TABLES FOR CHAPTER 2 ............................................................................................................... 36
CHAPTER 3 : CYSTECTOMY VOLUME-OUTCOME ASSOCIATIONS IN
ONTARIO ................................................................................................................... 42
SUMMARY ...................................................................................................................................... 42 INTRODUCTION ............................................................................................................................. 44 METHODS ........................................................................................................................................ 46
Cohort Identification ..................................................................................................................... 46 Outcome Definitions ...................................................................................................................... 47 Exposure Definitions ..................................................................................................................... 48 Potential Confounding Variables ................................................................................................... 49 Statistical Analyses ........................................................................................................................ 50
RESULTS ......................................................................................................................................... 53 Patient and Provider Demographics .............................................................................................. 53 Operative Mortality ....................................................................................................................... 54 Overall Mortality........................................................................................................................... 54
DISCUSSION .................................................................................................................................... 57 CONCLUSIONS ............................................................................................................................... 62 FIGURES FOR CHAPTER 3 ............................................................................................................. 63 TABLES FOR CHAPTER 3 ............................................................................................................... 65
CHAPTER 4 : CYSTECTOMY VOLUME AND OVERALL MORTALITY –
UNDERLYING STRUCTURES AND PROCESSES OF CARE .............................. 78
vi
SUMMARY ...................................................................................................................................... 78 INTRODUCTION ............................................................................................................................. 80 METHODS ........................................................................................................................................ 82
Overview ....................................................................................................................................... 82 Cohort Identification ..................................................................................................................... 82 Volume-Overall Survival Analyses ................................................................................................. 84 Structures and Processes of Care ................................................................................................... 86 Statistical Analyses ........................................................................................................................ 86
RESULTS ......................................................................................................................................... 88 Univariate Analyses....................................................................................................................... 88 Multi-collinearity Assessment ........................................................................................................ 89 Hospital Volume and Structures/Processes of Care ........................................................................ 89 Surgeon Volume and Structures/Processes of Care ........................................................................ 90
DISCUSSION .................................................................................................................................... 91 CONCLUSIONS ............................................................................................................................... 96 FIGURES FOR CHAPTER 4 ............................................................................................................. 97 TABLES FOR CHAPTER 4 ............................................................................................................... 99
CHAPTER 5 : THE EFFECT OF WAIT TIMES FOR CYSTECTOMY ON
OVERALL MORTALITY IN ONTARIO: A POPULATION-BASED STUDY ... 105
SUMMARY .................................................................................................................................... 105 INTRODUCTION ........................................................................................................................... 107 METHODS ...................................................................................................................................... 109
Cohort Identification ................................................................................................................... 109 Wait Time Definition ................................................................................................................... 110 Confounding Variable Definitions ................................................................................................ 111 Statistical Analyses ...................................................................................................................... 112
RESULTS ....................................................................................................................................... 114 Baseline Demographic and Univariate Analyses .......................................................................... 114 Survival Analyses ........................................................................................................................ 115 Tumour Stage-Wait Time Interactions .......................................................................................... 116 Maximum Wait Time Recommendation ........................................................................................ 116
DISCUSSION .................................................................................................................................. 118 CONCLUSIONS ............................................................................................................................. 124 FIGURES FOR CHAPTER 5 ........................................................................................................... 125 TABLES FOR CHAPTER 5 ............................................................................................................. 130
CHAPTER 6 : DISCUSSION AND CONCLUSIONS ............................................. 137
THESIS SUMMARY ....................................................................................................................... 137 IMPLICATIONS AND RECOMMENDATIONS ............................................................................. 138
Clinical ....................................................................................................................................... 138 Methodological ........................................................................................................................... 140 Health Policy .............................................................................................................................. 142
Volume, Structure and Process of Care.................................................................................................... 142 Wait times .............................................................................................................................................. 144
THESIS LIMITATIONS .................................................................................................................. 146 FUTURE STUDIES ......................................................................................................................... 149
Volume, Structures and Processes of Care ................................................................................... 149 Wait times ................................................................................................................................... 150
CONCLUSIONS ............................................................................................................................. 151
APPENDIX A ............................................................................................................ 152
APPENDIX B ............................................................................................................ 156
REFERENCES .......................................................................................................... 161
vii
LIST OF TABLES
Table 1.1: Summary of hospital volume-outcome studies assessing postoperative mortality as
the outcome variable. ...................................................................................................... 19
Table 1.2: Summary of surgeon volume-outcome studies assessing postoperative mortality as
the outcome variable. ...................................................................................................... 21
Table 1.3: Summary of studies assessing the effect of cystectomy wait times on outcome. ... 22
Table 1.4: Data sources and their validation, where available. ............................................... 24
Table 2.1: Inter-rater reliability and agreement statistics for 2 raters extracting pathologic
variables from OCR radical cystectomy pathology reports. ........................................... 36
Table 2.2: Intra-rater reliability and agreement statistics for the primary abstractor. ............. 37
Table 2.3: Pathology variables for the Pathology cohort. ........................................................ 38
Table 2.4: Patient level variables by pathology report availability. ........................................ 39
Table 2.5: Physician and hospital level variables by pathology report availability................. 40
Table 3.1: General cohort characteristics based on average annual volume quartiles. ............ 65
Table 3.2: Patient level and pathologic variables by hospital volume quartile. ...................... 66
Table 3.3: Patient level and pathologic variables by surgeon volume quartile. ...................... 68
Table 3.4: Effect of Hospital Volume on Postoperative Mortality. ......................................... 70
Table 3.5: Effect of Surgeon Volume on Postoperative Mortality. ......................................... 71
Table 3.6: Effect of Hospital Volume on Overall Mortality. ................................................... 72
Table 3.7: Effect of Surgeon Volume on Overall Mortality. ................................................... 74
Table 3.8: Decrease in hazard of overall death by an incremental increase in the number of
cystectomy operations performed at the hospital or surgeon level. ................................ 76
Table 3.9: Simultaneous effect of Hospital and Surgeon Volume on Overall Mortality. ....... 77
Table 4.1: List of candidate structures and processes of care variables assessed for their
ability to define provider “volume.” ................................................................................ 99
Table 4.2: Preoperative, intraoperative and hospital structure and process of care variables by
hospital volume quartile from the full cohort. ............................................................... 100
Table 4.3: Preoperative, intraoperative and hospital structure and process of care variables by
surgeon volume quartile from the full cohort. ............................................................... 101
Table 4.4: Results of multi-collinearity assessment of all candidate structure/process of care
variables......................................................................................................................... 102
Table 4.5: Effect of structure and process of care variables on the hospital volume parameter
estimates for overall mortality. ...................................................................................... 103
Table 4.6: Effect of structure and process of care variables on the surgeon volume parameter
estimates for overall mortality. ...................................................................................... 104
Table 5.1: Patient characteristics by wait time. ..................................................................... 130
Table 5.2: Effect of Wait Time on Overall Mortality. ........................................................... 132
Table 5.3: Time-dependent effects of wait time on overall mortality. .................................. 134
Table 5.4: Hazard ratios for death and corresponding P values for a 30 day increase in
preoperative wait time for cystectomy. ......................................................................... 135
Table 5.5: Hazard ratios by Tumour stage and survival time. ............................................... 136
Table B.1: Patient level variable definitions. ......................................................................... 157
Table B.2: Pathology variable definitions. ............................................................................ 158
Table B.3: Physician level variable definitions. .................................................................... 159
Table B.4: Hospital level variable definitions. ...................................................................... 160
viii
LIST OF FIGURES
Figure 2.1: Cohort Identification flow diagram. ............................................................. 33
Figure 2.2: Kaplan-Meier survival curves stratified by local T stage. ............................. 34
Figure 2.3: Kaplan-Meier survival curves stratified by lymph node status. ..................... 35
Figure 3.1: Postoperative mortality by hospital volume quartile between 1992-2004. .... 63
Figure 3.2: Postoperative mortality by surgeon volume quartile between 1992-2004. .... 64
Figure 4.1: Effects of accounting for structure and process of care groups on the hazard
ratio of hospital volume. ........................................................................................ 97
Figure 4.2: Effects of accounting for structure and process of care groups on the hazard
ratio of surgeon volume. ........................................................................................ 98
Figure 5.1: Bladder cancer wait time intervals from symptom development to definitive
therapy................................................................................................................. 125
Figure 5.2: Histogram of wait times for radical cystectomy in Ontario, 1992-2004. ..... 126
Figure 5.3: Histogram of median wait times for radical cystectomy in Ontario by year,
1992-2004. .......................................................................................................... 127
Figure 5.4: Relative increase in the hazard of death for a 30 day preoperative wait by
tumour stage. ....................................................................................................... 128
Figure 5.5: Effect of waiting for radical cystectomy on the hazard ratio for death from
any cause. ............................................................................................................ 129
Figure A.1: Thesis conceptual framework.................................................................... 153
Figure A.2: Conceptual framework for objective 1. ..................................................... 153
Figure A.3: Conceptual framework for objective 2. ..................................................... 154
Figure A.4: Conceptual framework for objective 3. ..................................................... 155
1
CHAPTER 1 : BACKGROUND
BLADDER CANCER EPIDEMIOLOGY
Bladder cancer has the sixth highest incidence of all malignancies diagnosed in
Canada with 6600 new cases estimated in 2007.1 The urinary bladder ranks fourth
amongst males and twelfth amongst females with respect to incident cancer cases. Due
to its recurring nature, however, bladder carcinoma has the fourth highest prevalence of
all malignancies.2
Staging of bladder cancer follows the TNM classification as defined by the
American Joint Committee on Cancer (AJCC) Staging manual, 6th
edition (2002).
Briefly, superficial tumours are either on the surface of (Ta) or within (Tis – carcinoma-
in-situ) the bladder mucosa or invade the lamina propria (T1). Invasive bladder cancer
extends into the detrusor muscle (T2), the perivesical fat (T3) or adjacent organs (T4). At
the time of diagnosis, 30% of bladder cancers are classified as invasive. Approximately
20-30% of the remaining superficial lesions will progress to muscle invasion. Thus,
almost 50% of all diagnosed bladder cancer will ultimately be classified as muscle
invasive.
The vast majority (85%) of all bladder cancers are of the transitional cell
carcinoma (TCC) histologic variety.3 Squamous cell carcinoma and adenocarcinoma
comprise much of the remaining 15% of tumours. Stage for stage, these histologic
subtypes have similar mortality rates.4
2
MANAGEMENT OF INVASIVE BLADDER CANCER
Patients with bladder cancer usually present with gross or microscopic hematuria,
with or without irritative voiding symptoms.5 Cystoscopic evaluation of the bladder is
used to confirm the presence of a tumour. Final diagnosis, however, requires a
histological assessment of tissue retrieved either by biopsy or, more frequently, by
transurethral resection of the bladder tumour (TURBT).
Superficial bladder tumours are managed effectively via TURBT and/or
intravesical immuno- or chemotherapy.6 Cystectomy may be offered to patients with
superficial disease in cases where the tumour is refractory to intravesical therapy, multi-
focal or associated with poor prognostic features such as carcinoma-in-situ or in
situations where rapid tumour growth outpaces the ability to perform a complete TURBT.
For those individuals who progress to or who are diagnosed with muscle-invasive TCC,
however, the gold standard of treatment is radical cystectomy with creation of a urinary
diversion.7 The cystectomy procedure entails the en bloc removal of the anterior pelvic
organs, which includes the bladder, prostate and seminal vesicles in men and the bladder,
urethra, uterus, ovaries and vaginal cuff in women. Dissection and removal of regional
lymph nodes (pelvic lymphadenectomy) is routinely performed. Urinary diversion is
managed either via drainage into a non-continent ileal conduit (urostomy) or into a
continent reservoir.
In addition to surgical management, various regimens using chemotherapy and
radiation in primary, neoadjuvant or adjuvant settings have been utilized to treat patients
with invasive bladder cancer. The use of chemoradiation for the primary treatment of
invasive TCC is often limited to patients unfit for surgery, patients averse to surgical
3
therapy or rarely patients with initially unresectable disease. Should combined
chemoradiation fail to achieve a complete disease remission in these patients, however,
radical cystectomy may be pursued as „salvage‟ therapy.8 Adjuvant chemotherapy, after
completion of definitive local therapy (cystectomy), has been assessed in a number of
studies with mixed results.9-12
Specifically, consistent benefits with respect to disease-free
survival have been noted but these have not translated into significant changes in overall
survival. Furthermore, the design, interpretation and conduct of these trials has been
questioned.13
Neoadjuvant chemotherapy, before completion of definitive local therapy,
has recently been demonstrated to be an effective adjunct for patients with muscle
invasive TCC undergoing cystectomy.14,15
Despite apparent level-1 evidence for
neoadjuvant chemotherapy, however, many urologists and oncologists have been hesitant
to recommend its routine use, citing methodological flaws and design and interpretation
challenges inherent to the randomized controlled trials and meta-analyses studying the
issue.16,17
For example, in the most widely cited neoadjuvant chemotherapy randomized
controlled trial, surgical technique across treatment arms was not standardized, one-sided
p values were utilized and post-cystectomy therapies were not controlled for.14
Since
chemotherapy provision in the adjuvant setting has the advantage of having accurate
pathologic staging information available to inform chemotherapeutic decision-making
and of eliminating potential delays to definitive surgery secondary to chemotherapy
complications, many physicians favour it over neoadjuvant therapy. Based on these
arguments and the disease-free survival benefits associated with adjuvant chemotherapy,
many cancer centres have adopted a policy of offering patients adjuvant chemotherapy
for lymph node positive or locally advanced bladder cancer.
4
Patients with invasive bladder cancer who undergo cystectomy have significant
short term (postoperative) and long term mortality. With respect to the former, numerous
studies18,19
have demonstrated postoperative mortality rates between 2-6%. For the latter,
the overall 5-year survival for all patients undergoing radical cystectomy is 50%.14
Stratified by stage, this procedure yields a 60-75% and a 20-40% 5-year survival for
patients with T2 and T3/T4 disease, respectively.6 Given these results, the need for
improved care for these patients has been recognized.
QUALITY OF CARE IN SURGERY
Improving patient outcomes can broadly be achieved via two distinct routes. The
first involves the development of new treatment strategies proven to be more effective
than existing approaches. The second entails modifying the delivery of existing therapies
to maximize effectiveness and thus outcome. Implicit to this latter point is an existing
deficit in health care delivery. In fact, the extent to which high-quality health care is
delivered in North America has been questioned.20
Focusing on the quality of health care provides a means of optimizing current
medical care. Requisite to ameliorating deficiencies in care provision is the ability to
detect such short-comings. One method of investigating the quality of health care
delivery is based on Avedis Donabedian‟s classic paradigm of structure, process and
outcome domains.21
This model has been used for years and has served as one of the
bases for contemporary outcomes research. In addition to using outcomes, such as
mortality, length of hospital stay or quality of life, as indicators of the quality of care,
structure and process variables are gaining more recognition for their role in the
5
Donabedian model.22
Structures of care refer to a group of variables that reflect the
setting in which health care is delivered (e.g. hospital volume, type of surgeon, wait times
for care) whereas processes of care are measures of what health care provider‟s are doing
and what patients are receiving (e.g. antibiotic prophylaxis prior to surgery).
Structures and processes of care and their effects on outcome are now common
research themes in the surgical literature.23-25
For example, Birkmeyer et al. have
addressed the issue of provider volume at both the hospital and surgeon levels, for a
variety of surgical conditions, finding that high volume hospitals and surgeons are
associated with improved outcomes compared to their low volume counterparts.26,27
With these findings, this group of investigators has now shifted focus to the factors
underpinning the volume-outcome relationship as it is widely believed that “volume” is a
proxy measure for underlying structures and processes of care. Khuri and associates have
also adapted the structure and process model to improve care in the large Veterans‟
Administration medical system in the United States.28,29
Surgical quality of care improvement initiatives have also been initiated in
Canada. The Canadian Cardiovascular Outcomes Research Team has embarked on a
massive scientific endeavour focusing on the quality of care for patients from 5 Canadian
provinces who suffer from cardiovascular disease. This research group‟s surgical
objectives include the measurement and improvement of outcomes of invasive cardiac
procedures such as coronary artery bypass grafting (CABG).25
To this end, the group has
compiled a list of quality of care indicators for patients undergoing CABG using the
structure, process and outcome of care model. The derivation of these variables via
6
literature review, Delphi expert panel input and methodological critique of other
programs has provided a basis for future research activities.30-34
Despite the flurry of activity in North American surgical outcomes research, the
application to urology and specifically radical cystectomy is limited. The postoperative
mortality rates of radical cystectomy are similar to those of elective CABG and elective
abdominal aortic aneurysm repair.35
Since the latter two procedures have both been the
subject of outcomes research and quality of care initiatives, it is reasonable to suggest
that radical cystectomy might benefit from a similar approach. Also, the high long term
mortality rates for patients who have undergone cystectomy warrant research aimed at
improving these rates. Finally, given the resource-intensiveness36,37
, costs38,39
and the
burden of treatment for radical cystectomy40
, there is considerable impetus for evaluating
cystectomy quality of care.
QUALITY OF CARE FOR RADICAL CYSTECTOMY
The quality of care literature for radical cystectomy has often focused on
outcomes research. In particular, two major areas have been studied: 1) issues pertaining
to provider volume and their impact on cystectomy mortality and 2) the effects of delayed
therapy (wait times) and their impact on cystectomy mortality. Although other topics
related to the quality of care of cystectomy patients have been studied, these 2 themes
will underlie this thesis.
7
Provider Volume and Structures/Processes of Care
Volume-outcome associations assessing postoperative cystectomy mortality have
demonstrated improved mortality with higher volumes at both the hospital and surgeon
levels. A systematic MEDLINE search (1966 to present) of the health services literature
using the search terms “CYSTECTOMY” and “VOLUME” yielded 211 articles. Upon
review of the abstracts of these studies, 19 discussed volume-mortality outcome
associations for cystectomy. Of these 19, four were review articles41-44
, one addressed the
impact of both hospital and surgeon volume on outcome45
, two focused on the impact of
surgeon volume on outcome27,46
and the remaining 12 studied the effect of hospital
volume on cystectomy outcome. Of these 12, 1126,47-56
focused on short term
(postoperative) mortality and only one assessed the impact of hospital volume on long
term survival.57
Similar search strategies, using the terms “BLADDER NEOPLASM”,
“OUTCOME” and “VOLUME” did not reveal any missed articles nor did hand searches
of the reference sections of each publication. To reassure ourselves that these searches
captured all relevant cystectomy-related articles, the online search strategy published in a
report commissioned by the Canadian Institute for Health Information (CIHI), which
summarizes the surgical volume-outcome literature, was run.58
Both the MEDLINE and
EMBASE databases were searched from 1980 to November 2007 using the following
combinations of search terms: a) [volume.ti OR frequent.ti OR frequency.ti OR
statistics.ti] AND [outcome.ti OR outcomes.ti]; b) [volume.ti AND mortality.ti]; c)
[volume.ti AND survival.ti]. The same articles from the previous search strategies were
found and, more importantly, no relevant, unidentified articles pertaining to cystectomy
volume-outcome analyses were identified.
8
Table 1.1 summarizes the 12 studies (comprising 13 analyses) of hospital volume
and its relation to postoperative mortality. Eleven of the twelve studies (12/13 analyses)
were performed in the United States. Of the 13 analyses, 9 demonstrated significant
volume-outcome associations. The trends for the remaining 4 analyses, however, still
supported inverse volume-outcome relationships with higher mortality rates at lower
volume hospitals. Possible explanations for the lack of uniformly statistically significant
results included small sample sizes47,53
with insufficient power to detect a statistically
significant association, as suggested by some authors45,47
and/or misclassification of
hospital volume because the Surveillance, Epidemiology and End Results (SEER)
database only accounts for patients residing within SEER catchment areas and ignores
patient migration54
, or failure to appropriately account for hospital
restructuring.(Kulkarni et al., submitted) The lone non-U.S.-based study by McCabe et al.
originated from Britain.50
This group did not divide patients into equal sized a priori
defined categories based on caseload, but rather calculated the correlation coefficient for
hospital volume and inpatient mortality. Upon finding a significant correlation between
hospital cystectomy volume and inpatient mortality, they proceeded to define a caseload
cut-point at which a significant difference in mortality was present. Arbitrarily using
hospital volumes of 6, 8, 10, 11, 12 and 16 cases/year, these investigators only found a
significant mortality difference if their hospitals were split into less than 11 or 11 and
greater cases/year. They concluded that a minimum caseload of 11 cystectomy
procedures per year is required per institution to provide quality cystectomy care.
Surgeon volume and its influence on postoperative mortality has also been studied
(Table 1.2). Birkmeyer and colleagues have demonstrated a significant association
9
between a surgeon‟s cystectomy operative volume and short term mortality using the
Medicare database.27
High volume surgeons, defined by an annual volume of 3.5
cystectomy procedures or greater experienced a 30-day postoperative mortality of 3.1%
whereas low volume surgeons (< 2 procedures/year) had a postoperative death rate of
5.5%. Further analysis from this group demonstrated that surgeon volume possibly
accounts for 39% of the significant effect of hospital volume on postoperative mortality.
The same phenomenon was also observed for the effect of hospital volume on surgeon
volume. In other words, patient short term mortality was found to be related to both
surgeon level and hospital level variables. On the other hand, Konety et al., using the
Nationwide Inpatient Sample, observed a non-significant trend between surgeon volume
and in-hospital mortality.45
In this study, surgeon volume fully accounted for the
significant effects of hospital volume on mortality, suggesting that a major portion of the
hospital volume effect is secondary to surgeon volume. Finally, in the lone non-U.S.
based study, McCabe et al. applied the same methodology used previously for hospital
volume (see above), substituting surgeon for hospital volume.46
Using sequential surgeon
volume cutpoints between 6 and 15 cases per year, they determined that a statistically
significant surgeon volume effect on inpatient mortality existed after 8 annual cystectomy
procedures and therefore suggested 8 procedures as the minimum level to
maintain/achieve competence. Their results and recommendations, however, were based
on univariate analyses not corrected for patient level factors. Thus, given the paucity of
surgeon volume-outcome studies, their variable results and the methodological short-
comings of existing studies, further research is required to clarify the effect of surgeon
volume on postoperative mortality.
10
To date, only one group has investigated the impact of provider volume on long
term survival outcomes. Birkmeyer and colleagues assessed the late survival of patients
undergoing surgical resection for 6 separate malignant conditions, including cystectomy
for bladder cancer, by hospital volume tertiles.57
In this report, cystectomy hospital
volume was not significantly associated with long term survival in adjusted analyses (HR
0.90; 95% CI: 0.79-1.02). Incorporating process of care variables (provision of adjuvant
radiation and/or chemotherapy) did not alter the results.
Increasingly, investigators have begun to investigate the potential structure or
process of care variables that may underlie cystectomy volume-outcome relationship for
postoperative mortality. Elting and investigators assessed the effects of hospital bed
characteristics, nurse staffing, hospital critical care characteristics along with other
hospital and patient factors on postoperative patient mortality in Texas.49
In addition to
confirming previously reported hospital volume-outcome associations, a hospital‟s nurse-
to-patient ratio was significantly associated with postoperative death. Modeling hospital
volume with nurse-to-patient ratio eliminated the effect of hospital volume, suggesting
that nurse staffing may explain the hospital volume-outcome findings in this patient
population. Konety and colleagues assessed the impact of structural variables, postulating
that improved systems of care such as nursing support, anesthesia care, critical care unit
care, laboratory and radiological factors, etc. could mitigate the volume-outcome
relationship.51
Incorporating variables to identify hospitals that meet U.S. volume
thresholds, and thus likely have improved structural support systems, into multivariate
volume-outcome models did not attenuate the impact of hospital volume on inpatient
mortality. Konety et al. therefore speculated that perhaps processes of care are the
11
influential determinants underlying hospital volume. Expanding on this notion,
Hollenbeck et al. reported substantial variation in the processes of care, such as
preoperative cardiac testing, intraoperative arterial monitoring and use of continent
diversion, between high and low volume cystectomy institutions.55
Accounting for these
process measures reduced the OR for death from 1.48 (95% CI: 1.03-2.13) to 1.39 (95%
CI: 0.93-2.09), explaining 23% of the volume effect. Although these process measures
mitigated the significance of the volume-outcome association, the odds ratio of 1.39 led
the authors to conclude that a considerable component of the underlying
processes/structures of care remain undefined. In a separate analysis using a different
patient cohort, Hollenbeck and investigators assessed the impact of structural factors on
the cystectomy volume-outcome association and similarly reported a 59% reduction in
the odds ratio for death after accounting for structural factors such as hospital capacity
and staffing variables.53
To date, no cystectomy study has simultaneously assessed both
process and structural component variables in an attempt to explain the volume-outcome
phenomenon.
Wait Times
In addition to „volume‟ issues, waiting time for surgery has been identified as an
important quality indicator.59
Wait time can be considered either a process or a structure
variable25,60
although its classification likely depends upon whether it is a fixed function
of the health care system or a modifiable variable determined by clinical judgment.
Regardless of its classification, wait time for cystectomy can be regarded as an important
12
determinant of quality care since delay in therapy may have the dual effects of increasing
patient anxiety and increasing the propensity for tumour invasion and metastases.
Interest in the effect of waiting for care on outcomes has been increasing. A
recent systematic review on the impact of delayed surgical treatment of bladder cancer on
outcome revealed 13 papers published between 1950 and 2006. Of these 13, 7 evaluated
the wait time between TURBT and cystectomy, the former often cited as the time at
which a decision to pursue cystectomy is made.61
Updating this review to November
2007 using the authors‟ published OVID MEDLINE search criteria yielded 2 additional
articles.62,63
Manual search of the bibliographies of these 2 new publications failed to
reveal relevant overlooked articles.
Of the 9 total publications (Table 1.3), only two reported a statistically significant
inverse association between wait time and long-term survival.62,64
Specifically, Lee et al.
determined that a treatment delay greater than 93 days between TURBT and cystectomy,
in patients with T2 disease, negatively influenced overall survival.62
They also found a
moderate trend in a similar direction for disease-specific survival (p = 0.08). Hautmann
and colleagues, in a series of subgroup analyses, insinuated that treatment delay after the
diagnosis of muscle-invasive bladder cancer was important only for those individuals
with T3b (macroscopic fat invasion) or T4 disease.64
Unfortunately, this group failed to
define “treatment delay” in their manuscript, impeding interpretation of their results.
Four papers reported strong trends towards a statistically significant association
between wait times and survival.62,65-67
In addition to Lee‟s study quoted above, Mahmud
and colleagues, using population-based administrative data from the province of Quebec,
reported a p value of 0.051 for the association between wait time and overall survival,
13
suggesting a strong detrimental trend to prolonged waiting.65
Likewise, both Sanchez-
Ortiz et al.66
and May et al.67
found near significant associations between wait times and
overall (HR 1.93, 95% CI: 0.99-3.76) and disease-specific (HR 1.62, 95% CI: 0.99-2.66)
survival, respectively.
Of the 3 negative studies63,68,69
, the investigation by Liedberg‟s group yielded a
non-significant hazard ratio that implied patients with a wait time greater than 60 or 90
days had improved outcomes compared to those operated on expeditiously.69
Hazard
ratios for the other two studies were not reported making commentary on the
directionality of risk in these reports impossible.
A number of reports have also note significant inverse associations between wait
time and pathologic outcome. For example, Chang and colleagues discovered that
prolonged waiting between TURBT and cystectomy resulted in poorer pathological stage
on final cystectomy examination.70
Confirming these results, May et al. analyzed the
results of 189 patients and determined that those with a wait time greater than 3 months
were more likely to have T4 disease compared to those operated on within 3 months.67
Sanchez-Ortiz et al. reported similar results with respect to pathological stage using a 12
week cutpoint in wait times.66
These studies provide evidence that waiting for care may
lead to worse pathology which in turn can affect long-term prognosis.
Clearly inconsistency exists in the literature over the true effect of wait time on
cystectomy outcome. Exacerbating the inconsistent results are the methodological
limitations inherent to the papers published to date.61
Small sample sizes, analyses
performed in subsets of selected patients, a lack of staging or comorbidity data for proper
risk adjustment or a failure to perform or report multivariate analyses all affect the
14
interpretability or generalizability of current studies. Additional research is required to
address some of these short-comings.
CURRENT KNOWLEDGE LIMITATIONS REGARDING THE QUALITY OF CARE
FOR RADICAL CYSTECTOMY
The health services literature about the quality of care for cystectomy is growing.
Nevertheless, a number of limitations to current studies exist. First, the majority of
studies performed to date have used databases that tend to represent restricted
populations. The Medicare database, for example, only contains data on patients aged 65
and over.71
The Nationwide Inpatient Sample only represents a selected sample of
hospital inpatient records from 8 to 37 different states, depending on the year(s) being
considered.72
The SEER database also only represents approximately 26% of the U.S.
population.73
Finally, the University HealthSystem Consortium Clinical Database (UHC)
is a database representing up to 90% of U.S. non-profit academic hospitals, depending on
the year(s) considered (59 academic medical centres in 1992, 97 centres in 2007).74
For
profit institutions are not included. The restrictive nature of these databases may
potentially limit the generalizability of results derived from them. Second, no cystectomy
volume-outcome study has been performed in a universal health care setting. All but one
set of studies have originated from the U.S. These restrictions may also affect the
generalizability of results to jurisdictions such as Canada which operate in a different
health care environment.75
Third, only one cystectomy study has assessed the impact of
hospital or surgeon volume on long term outcomes.57
Further work is required to clarify
the impact of provider volume on long term mortality beyond the single study published
15
to date. Fourth, research on the structures and processes of care potentially responsible
for volume-outcome associations is minimal and those papers published on the topic have
not delineated the important components behind provider volume. Fifth, controversy
exists over the true impact of wait times for cystectomy on mortality outcomes, as
described above. Methodological concerns with current studies hamper interpretation and
application of wait time-outcomes research results, particularly since few studies
recommend a maximum possible wait time.
THESIS CONCEPTUAL FRAMEWORK
Evaluating the quality of care for radical cystectomy is a complex process because
of the myriad inter-related variables that underlie quality. Figure A1 of Appendix A
depicts the conceptual framework that forms the basis of this study and illustrates the
factors involved in evaluating quality of care at a general level. According to the
Donabedian model, quality of care (QOC) for any medical condition is governed by both
structures and processes of care.21
The quality of delivered care is then manifest in terms
of measurable outcomes. In the case of radical cystectomy, these may include short and
long term mortality rates, postoperative complication rates, postoperative hospital length
of stay (LOS), patient quality of life (QOL) and patient satisfaction. The framework
presented forms the basis for two types of research questions. First, by following the flow
of the model, it is possible to identify structures and processes of care that affect quality
of care, as manifest by specific outcome measures in a chosen population. Second, once
significant structures/processes have been established for a particular outcome,
subsequent investigation can be pursued to determine whether these factors are present or
16
apply in other similar populations. In other words, the framework outlines a way of
identifying significant structures and processes of care and then determining whether
these structures or processes are being utilized in different cohorts.
The association between hospital or surgeon volume and short term mortality has
been consistently demonstrated for many different disease states.76-82
Thus, an important
structural variable which may affect quality of care for cystectomy patients is provider
volume. One theory explaining volume-outcome relationships is that volume is a
surrogate for surgical skill. Figure A2, Appendix A depicts this notion. In this
modification of the conceptual model, quality of care, as measured by short and long term
mortality outcomes, is affected by provider volume. A competing hypothesis states that
hospital or surgeon volume is a proxy measure for underlying, unidentified structures and
processes of care which are the important factors affecting quality of care. This concept is
illustrated in Figure A3, Appendix A. The impact of provider volume on cystectomy
quality of care may be mediated by factors that are associated with provider volume.
These potential factors, which can be broadly classified into preoperative, intraoperative
and postoperative structures and processes, may thus explain the volume-outcome
relationship.
Finally, Figure A4 of Appendix A adapts the conceptual framework to depict the
potential relationship of wait times on long-term mortality. Since many preoperative
factors influence wait time, such as delays in obtaining preoperative imaging or specialty
consultation, structures and processes of care are depicted as modifiers of the waiting
period for cystectomy. Furthermore, surgeon and hospital volume are additional key
determinants of a patient‟s wait to cystectomy as busier surgeons and busier hospitals
17
may have longer surgical lists. Ultimately, waiting for care may affect the quality of care
for radical cystectomy in terms of long-term mortality outcomes.
THESIS OVERVIEW
Using the conceptual framework outlined above, this body of work will attempt to
address many of the deficiencies and knowledge gaps present in the current cystectomy
quality of care literature. The over-arching aim of this research is to identify methods to
improve quality of care for radical cystectomy patients. Consequently, the following
three studies were performed:
1) A set of provider (hospital and surgeon) volume-outcome analyses for radical
cystectomy in Ontario examining both short and long term outcomes. In addition
to involving an entire population of patients, this paper is the first of its kind in a
not-for-profit, publicly-funded health care system.
2) A study aimed at determining the structure and process variables underpinning
volume-outcome associations in Ontario.
3) A well-designed, population-based study assessing the impact of cystectomy wait
times on overall mortality in Ontario.
SPECIFIC OBJECTIVES
a) To determine whether patients who undergo radical cystectomy for bladder cancer in
Ontario have lower long-term mortality and/or postoperative mortality rates if their
18
operation is performed at a high volume hospital or by a high volume surgeon
compared to those operated on at a low-volume hospital or by a low volume surgeon.
b) To identify those structure and process of care variables for patients undergoing
radical cystectomy which may potentially contribute to any observed volume-
outcome associations.
c) To determine whether a prolonged waiting time from transurethral resection of a
bladder tumour (TURBT) to radical cystectomy results in lower overall survival rates
for patients undergoing radical cystectomy for bladder cancer in Ontario.
DATA SOURCES AND VARIABLES
The data sources utilized in this thesis are listed in Table 1.4. Dates and relevant
results from studies supporting the validity of each database are also listed. Although the
general purpose of each database is provided in the table, details pertaining to each
dataset are discussed in subsequent chapters. A list of all variables used in the thesis is
provided in Appendix B (Tables B1-B4).
ETHICS STATEMENT
Data collection and analysis for this thesis only occurred after approval by the
Sunnybrook Health Sciences Centre and University of Toronto institutional review
boards. All data were uniquely labeled using encrypted health card numbers. No unique
identifiers such as patient name, OHIP number, postal code or address were recorded.
19
TABLES FOR CHAPTER 1
Table 1.1: Summary of hospital volume-outcome studies assessing postoperative
mortality as the outcome variable.
Mortality outcomes were defined as death within 30 days post-operatively (“30-day”),
death prior to discharge (“inpatient”) or death within 30 days post-operatively or prior to
discharge (“operative”). Adjusted odds ratios (OR) refer to extreme quantile
comparisons.
Study Database N Outcome Mortality (%) Adjusted
OR
(95% CI)
Highest
category‡
Lowest
category‡
Begg et al.
JAMA 199848
SEER-
Medicare
linkage
1984-1993
3,380 30-day 1.5% 3.7% Not reported
P = 0.05
Birkmeyer et al.
NEJM 200226
Medicare
1994-1999
22,349 Operative 2.9% 6.4% 0.46
(0.37-0.58)
Finlayson et al.
Arch Surg 200347
NIS 1995-
1997
4,937 Inpatient 2.5% 3.6% 0.7
(0.4-1.2)
Elting et al.
Cancer 200549
Multiple†
1999-2000
1,302 Inpatient 0.7% 3.1% 0.24
(0.07-0.80)
Konety et al.
J Urol 200545
NIS 1988-
1999
13,949 Inpatient 2.7% 4.7% 1.96*
(1.33-2.88)
McCabe et al. BJU Int 200550
HES 1998-2003
6,317 Inpatient 4.6% 8.1% Not reported P < 0.01
Konety et al.
JCO 200651
NIS 1998-
2002
6,577 Inpatient N/A N/A 0.53
(0.34-0.82)
Hollenbeck et al.
JCO 200752
NIS 1993-
2003
19,319 Inpatient 1.9% 3.7% 1.3*
(0.8-2.3)
Hollenbeck et al.
Urology 200755
SEER-
Medicare
linkage 1992-
1999
4,465# Operative 3.5% 4.9% 1.48*
(1.03-2.13)
Hollenbeck et al.
J Urol 200753
NIS 2003 1,847 Inpatient 1.1% 3.5% 3.2*
(0.8-13.4)
Barbieri et al.
J Urol 200756
UHC 2002-
2005
6,728 Inpatient 1.3% 2.4% Not reported
P = 0.03
Hollenbeck et al.
Surg Innov 200754
Medicare
1994-1999
SEER-
Medicare linkage 1994-
1999
2165
2165
Operative
Operative
3.4%
4.3%
6.2%
6.1%
1.82*
(1.17-2.84)
1.41*
(0.89-2.23)
‡Most studies categorized hospital cystectomy volume into tertiles. Exceptions:
Birkmeyer et al. (quintiles); McCabe et al. did not categorize a priori (see text for full
description); Hollenbeck et al. NIS 1993-2003 study (deciles); Barbieri et al. (tertiles
based on total hospital discharges).
†Texas Hospital Discharge Public Use Data File, Centre for Medicare and Medicaid
Services‟ Hospital Cost Report Information System, Provider of Services files and the
American Hospital Association Survey
*Reference group was high volume institutes.
20
#Included both partial (n=1375) and radical (n=3090) cystectomy patients.
Abbreviations: NIS = Nationwide Inpatient Sample; SEER = Surveillance,
Epidemiology and End Results database; HES = Hospital Episode Statistics; UHC =
University Health System Consortium Clinical Database.
21
Table 1.2: Summary of surgeon volume-outcome studies assessing postoperative
mortality as the outcome variable.
Mortality outcomes were defined as death prior to discharge (“inpatient”) or death within
30 days post-operatively or prior to discharge (“operative”). Adjusted odds ratios (OR)
refer to extreme quantile comparisons.
Study Database N Outcome Mortality (%) Adjusted OR
(95% CI) Highest
category‡
Lowest
category‡
Birkmeyer et al.
NEJM 200327
Medicare
1998-1999
6,340 Operative 3.1% 5.5% 1.83*
(1.37-2.45)
Konety et al. J Urol 200545
NIS 1988-1999
6,763 Inpatient 2.9% 3.9% Not reported P = NS
McCabe et al.
PMJ 200746
HES 1998-
2003
6,308 Inpatient 4.2% 6.7% Not reported
P < 0.01
‡Most studies categorized surgeon cystectomy volume into tertiles. Exception: McCabe
et al. did not categorize a priori (see text for full description).
*Reference group was high volume institutes.
Abbreviations: NIS = Nationwide Inpatient Sample; HES = Hospital Episode Statistics;
NS = Not significant (actual p value not reported).
22
Table 1.3: Summary of studies assessing the effect of cystectomy wait times on
outcome.
Survival outcomes, if available, are provided preferentially where multiple outcomes
were reported. Adjusted rate ratios refer to comparisons with the lower wait time period
referent (unless otherwise specified).
Study Country
N Median
(Mean)
WT (d)
WT
Cutpoint
Outcomes Adjusted Risk
Ratios (95% CI) /
% Change
Hautmann et al.
J Urol 199864
Germany
1986-1994
213#‡ N/A
(293)
Not defined DSS
(subgroup
analyses)
T2 + T4
T2 + T3a
T3b + T4
Not reported
P = 0.53
Not reported
P = 0.72 Not reported
P = 0.04
Hara et al.
Jpn JCO 200268
Japan
1985-2000
50# N/A
(83)
Binary
90 days
RFS
DSS
OS
Not reported
P = NS
Not reported
P = NS
Not reported
P = NS
Chang et al.
J Urol 200370
USA
1998-2001
153# 42
(63)
Binary
90 days
Pathology
≥T3
LN +
29% more pT3 or
higher
P = 0.01
17% more node positive disease
P value not reported
Sanchez-Ortiz et al.
J Urol 200366
USA
1987-2000
189# N/A
(55)
Binary
84 days
OS
1.93
(0.99-3.76)
P = 0.05
May et al.
Scand J Urol
Nephrol 200467
Germany
1992-2002
189# 54
(N/A)
Binary
90 days
DFS 1.62
(0.99-2.66)
P = 0.057
Liedberg et al.
J Urol 200569
Sweden
1990-1997
139 49
(N/A)
Binary
60 days
90 days
DSS
DSS
1.02*
(0.56-1.87)
0.72
(0.35-1.51)
Mahmud et al.
J Urol 200665
Canada
(Quebec) 1990-2002
1315 33
(N/A)
Continuous
Binary
84 days
OS
OS
1.0
(1.0-1.0) P = NS
1.2
(1.0-1.5)
P = 0.051
Lee et al.
J Urol 200662
USA
1990-2004
214# 61
(N/A)
Binary
93 days
DSS
OS
2.12
P = 0.08
1.96
P = 0.04
Nielsen et al.
BJU Int 200763
USA (3
institutions)
1984-2003
592 54
(N/A)
Continuous
RFS
DSS
Not reported
P = 0.213†
Not reported
P = 0.118†
23
Binary
90 days
OS
RFS
DSS
OS
Not reported
P = 0.105†
Not reported
P = 0.445†
Not reported
P = 0.323†
Not reported P = 0.833†
*Longer wait times referent.
#Restricted cohorts: Hautmann et al., Hara et al., Chang et al., Sanchez-Ortiz et al. and
May et al. restricted to patients with clinical stage ≥T2 disease; Lee et al. restricted to
patients with clinical stage T2 disease.
‡Only male patients evaluated.
†Based on univariate (unadjusted) analysis only. Direction of association not reported.
Abbreviations: WT = Wait time; LN = Lymph node; DSS = Disease-specific survival;
OS = Overall survival; RFS = Recurrence-free survival; NS = Not significant
24
Table 1.4: Data sources and their validation, where available.
Data source Use Dates and Validity
CIHI DAD Cohort
identification, Risk
adjustment, Outcome
measurement
1991-2004
Agreement between CIHI DAD and
hospital charts for procedures ranges
between 88-95%83,84
Agreement in administrative data better
for major versus minor procedures85
OHIP Cohort
identification,
Structures and
processes of care
1992-2006
Agreement between OHIP and CIHI for
hysterectomy or cholecystectomy
between 93-94%84
Billing claims typically provide
complete capture of procedure codes
Sensitivity and specificity of non-
procedure codes vary widely but can be
as high as 84% and 96%, respectively
(for ICU daily billing fee code)86
Clinical activity from physicians
remunerated via alternate funding plans
(AFP) (e.g. Kingston, Ontario) are not
captured
OCR Cohort
identification, Risk
adjustment
1992-2004
Estimated from both two and three
source capture-recapture methods. Data
completeness is estimated at 95.15% for
the three source method and 95.87% for
the two source method. The estimates
of completeness vary by organ site and
range from 91-100%87
Captures 97% of incident cases of
bladder cancer.88
RPDB Outcome
measurement, Risk
adjustment
1992-2006
Specific information regarding data
accuracy is not available.
Patient deaths are linked
probabilistically to the RPDB based on
the name and birth date listed on the
death certificate. Patient death
information from the CIHI DAD is used
to corroborate/supplement RPDB data.
If multiple individuals meet the linkage
criteria, a patient death is not recorded in
the RPDB. Thus, there are more people
in the RPDB than are alive in Ontario.
25
(ICES Intranet 2005)
Physician‟s
(PHYS) database
Risk adjustment,
Structures and
processes of care
1992-2004
Validated against the Ontario Physician
Human Resource Data Centre database,
which verifies this information through
periodic telephone interviews with
physicians. (ICES Intranet 2005)
Hospital
(INSTNUM)
database
Structures and
processes of care 1992-2004
Specific information regarding data
accuracy is not available.
1996 and 2001
Canadian Census
Risk adjustment 1996 and 2001
Socio-economic data are collected on
20% of the census sample and felt to be
representative (ICES Intranet 2005)
No verification of the data is performed
CCN Cardiac
catheterization
facility
identification,
Structures and
processes of care
1992-2004
Opening dates for cardiac care facilities
are available online.89
Diabetes Atlas Dialysis facility
identification,
Structures and
processes of care
1992-2002
The list of dialysis facilities found in the
ICES diabetes atlas is 100% accurate90
Data on dialysis facilities between 2002-
2004 were obtained from Ontario
Ministry of Health documents housed at
ICES
Information regarding data accuracy
post-2002 is not available.
Abbreviations: CIHI DAD – Canadian Institute for Health Information; OHIP – Ontario
Health Insurance Plan; OCR – Ontario Cancer Registry; RPBD – Registered Persons
Database; CCN – Cardiac Care Network of Ontario.
26
CHAPTER 2 : COHORT DEFINITIONS AND DESCRIPTIONS
PATIENT IDENTIFICATION
Patients were identified based on the outline depicted in Figure 2.1. Prior to 2002,
cystectomy patients were identified from the CIHI-DAD using the Canadian
Classification of Diagnostic, Therapeutic and Surgical Procedures (CCP) cystectomy
procedure code 69.51. From 2002-2004, with the conversion of the International
Classification of Diseases (ICD) coding system from version 9 (ICD-9) to version 10
(ICD-10), cystectomy procedures were identified from CIHI based on the Canadian
Classification of Health Interventions (CCI) codes 1.PM.91 and 1.PM.92. The CCP and
CCI complement the ICD-9 and ICD-10 coding systems for procedures, respectively. A
total of 3811 potential cases were identified between fiscal years 1992 and 2004 (April 1,
1992 to March 31, 2005). Since radical cystectomy can be performed for bladder cancer
or as part of large exenterative procedures for other pelvic malignancies (e.g. cervical,
vaginal, colon, etc.), linkage with the Ontario Cancer Registry (OCR) was performed to
identify only those cases in which the cystectomy was performed for bladder cancer. The
cohort for this thesis therefore consisted of only patients who underwent cystectomy for
bladder cancer.
Linkage with OCR yielded 3722 (97.7%) patients, of whom 2710 had pathology
reports housed at OCR. Each cystectomy pathology report was reviewed to determine
whether the cystectomy was performed for bladder cancer (2535 cases) or for non-
bladder cancer (175 cases). The reason only 71% of cystectomies recorded in CIHI had
pathology reports relates to the various methods by which cancers can be registered with
the OCR, which can occur in one of four ways: 1) the cancer is detected via a death
27
certificate; 2) the cancer diagnosis is submitted from a regional cancer centre, all of
which are required by law to submit cancer diagnoses to the OCR; 3) the cancer is
identified via pathology reports submitted from local hospitals and/or 4) the cancer
diagnosis is detected on a hospital discharge summary.91
Since the aim of the OCR is to
capture incident rather than prevalent cases of cancer, a number of cancers are registered
with the OCR without availability of pathology information. In this manner, OCR has
historically identified 97% of all bladder cancer diagnoses in the province of Ontario.88
For this study, pathology information (pathological stage, grade, margin status,
lymph node status, lymphovascular invasion status and perineural invasion status) was
collected on all 2535 bladder cancer cases via chart review.
Of the 1012 patients who underwent cystectomy but did not have pathology
reports available at OCR, we used the OCR diagnostic code for bladder cancer to retain
likely bladder cancer cases. To further increase the specificity for cystectomy procedures
in the remaining 831 patients, we limited our cohort to cases where an OHIP cystectomy
billing was filed (OHIP codes: S484, S485, S453, S440). One patient less than 19 years
of age was excluded based on predefined inclusion/exclusion criteria.
Using the OCR bladder cancer diagnosis and OHIP cystectomy billing codes as
additional “filters,” we attempted to achieve as specific a cystectomy cohort of patients
with bladder cancer as possible where confirmatory pathology reports were unavailable.
Our final cohort consisted of 3296 patients (circled values in Figure 2.1), for whom
pathology data was available on 2535 (77%). A total of 515 patients were excluded
(boxed values in Figure 2.1).
28
VALIDATION
To date, a formal validation study to assess the accuracy of cystectomy
ICD/CCP/CCI codes has not been performed. Nevertheless, prior abstraction studies
along with data from this thesis provide strong evidence for the validity of CIHI
cystectomy codes. For example, Quan et al. assessed the validity of procedure codes in
ICD-9 administrative data. In their reabstraction study of 600 randomly selected surgical
patients, the sensitivity of administrative data for detection of major surgical procedures
ranged between 41-94% with most above 80%. The specificity of surgical procedures,
however, was above 99% with positive and negative predictive values well above 90%.
Based on their data, they concluded that major procedures performed in the operating
theatre are relatively well-coded.85
Hawker and colleagues similarly assessed the
accuracy of CCP defined knee-replacement procedure codes, finding 99.4% (174/175)
agreement between CIHI files and the hospital record for the primary procedure.83
Finally, an internally-directed data quality review initiated by CIHI in 2000 found that
major procedures were inaccurately coded as other procedures 0.3% of the time, or were
missing 4.9% of the time, suggesting excellent specificity and sensitivity.92
While these
data do not directly validate cystectomy procedure codes, they provide indirect evidence
of the general validity of procedure codes in administrative data, particularly for primary,
operative procedures such as radical cystectomy.
A recent study commissioned by the Ontario Ministry of Health and Long Term
Care to assess the accuracy of coding at CIHI provides additional evidence that radical
cystectomy codes in CIHI are accurate (Juurlink and Croxford, 2005, Institutes for
Clinical Evaluative Sciences, unpublished). In this study, 1500 hospital charts were
29
reviewed by trained abstractors and the re-abstracted CIHI codes and diagnoses were
compared to the information previously recorded. A total of 25 radical cystectomy
procedures were identified by the re-abstractors. These same procedures were also
identified by the original abstractors resulting in an agreement of 100% (95% CI of 89-
100%). No additional radical cystectomy procedures were coded by the original
abstractors. All codes were based on the ICD-10 coding system. However, since the
codes for radical cystectomy in ICD-9 are similar to those in ICD-10 (one versus two
codes, respectively), it is fair to assume that the abstraction study results would be
applicable to the ICD-9 system for radical cystectomy.
Finally, based on our review of pathology reports at OCR, we can surmise a
specificity of CIHI cystectomy codes of at least 71% (2710/3811) since we were able to
confirm 2710 cystectomy cases by manual review of pathology reports. Unfortunately,
our data preclude comment on sensitivity; however, CIHI procedure code sensitivity has
historically been quite high as discussed above. Of the 2535 OCR-confirmed cystectomy
cases for bladder cancer, 2318 (91.4%) had an accompanying OCR bladder cancer
diagnostic code compared to 11 of the 175 (6.3%) cases performed for non-bladder
cancer. These data supported using the OCR bladder cancer diagnostic code for cohort
refinement in cases where pathology reports were unavailable. To further increase the
specificity of our non-pathology confirmed cohort, we limited cases to those where an
OHIP cystectomy billing code was available. The aim of this restriction was to use an
independent database (cross-validation) to remove CIHI-derived cases that may have
been potentially miscoded as cystectomy. Using OCR and OHIP “filters” reduced the
total number of potential cases (the denominator) to 3296. Since 98.7% (2503) of the
30
confirmed 2535 bladder cancer cystectomies and 98.6% (750/761) of the non-pathology
confirmed cases had a CIHI bladder cancer code, we felt confident that our denominator
was accurate. Thus, our final algorithm for defining cystectomy for bladder cancer using
CIHI, OCR and OHIP codes has a specificity of at least 77% (i.e. 2535/3296). This value
is the minimum specificity, with the true value almost certainly much higher. We are
confident that our administrative data algorithm has accurately identified bladder cancer
cystectomy patients.
RELIABILITY (ABSTRACTOR AGREEMENT)
Two abstractors were responsible for retrieving data from the 2710 pathology
reports (the primary author of this thesis and a trained nurse practitioner, the latter of
whom was the main reviewer responsible for abstraction of approximately 75% of the
reports). The first step in the process was to identify whether the cystectomy had been
performed for bladder cancer versus non-bladder cancer causes. In the second step,
bladder cancer cases were then abstracted in detail for staging, grading and other
pathologic variables. Based on published sample size recommendations aimed at
achieving kappa values with maximum confidence intervals of ± 0.193
, we randomly
selected 200 reports for each rater to abstract. Calculated agreement statistics included
linearly weighted kappas for ordinal categorical variables, unweighted kappas for
nominal categorical variables and intraclass correlation coefficients (ICC) for continuous
variables.
Inter-rater reliability results are presented in Table 2.1. “Substantial” (kappa 0.6-
0.8) to “almost perfect” (kappa 0.8-1.0) agreement statistics were obtained for almost all
31
variables.94
The actual agreement for all variables was greater than 88%. We also
assessed the intra-rater reliability of the main abstractor by reabstracting 104 consecutive
pathology records 6 weeks after the original abstraction had occurred. Intra-rater
reliability statistics (Table 2.2), with the exception of gradability (whether the abstractor
could assign a tumour grade or not), were all within the “substantial” to “almost perfect”
agreement range. Although gradability had a kappa value of 0.583, which can be
considered “moderate,” the actual agreement for this variable was very high (96%),
indicating that the collection of this variable was indeed precise.
Another evaluation of the accuracy of abstraction was obtained by assessing
whether the outcomes of selected pathology variables followed an expected pattern.
Figures 2.2 and 2.3 demonstrate worse overall survival with higher stage disease and for
patients with lymph node metastases, which is what would be expected based on the
prognosis of bladder cancer. These data therefore support our abstraction of pathology
data from radical cystectomy pathology reports. A summary of the pathologic variables
and their distributions among the 2535 patients is presented in Table 2.3.
THESIS COHORTS
Pathology data were available for approximately 77% of the radical cystectomy
procedures performed in Ontario between 1992 to 2004. Tables 2.4 and 2.5 compare
patients with and without pathology reports on all variables used in this thesis. With
respect to patient factors (Table 2.4) patients with pathology reports tended to be sicker
and were more often admitted urgently than those without pathology reports. As a result,
they also were significantly more likely to die than non-pathology report patients. There
32
were also geographic discrepancies in the place of residence as assessed by LHIN
between pathology report patients and non-report patients. Of note, a policy of submitting
pathology reports in cases of cancer deaths did not exist in the province of Ontario,
suggesting that reports were not missing because of outcome. Assessing surgeon and
hospital factors (Table 2.5) demonstrated that patients without pathology reports were
more likely to receive an anesthetic consult, be operated on by a less experienced
urologist and to have a urologist as the primary intraoperative assistant compared to those
with pathology reports. Patients who had pathology reports available at OCR were
statistically more likely to be operated on at high volume centers and centers with cardiac
catheterization capabilities.
Differences between patients with and without pathology reports are highlighted
because of the use of two cohorts in this thesis. For short term outcomes (post operative
mortality) assessed in Chapter 3, we used the “full” cohort of 3296 patients because we
hypothesized that pathology variables would have a limited impact on post-operative
deaths and thus would not have to be accounted for during risk adjustment. For long term
outcomes (overall survival), which were investigated in Chapters 3-5, we used the
“pathology” cohort because of the need to risk adjust overall mortality outcomes with
pathology variables, many of which are strong predictors of long-term mortality in
bladder cancer patients.95,96
33
FIGURES FOR CHAPTER 2
Figure 2.1: Cohort Identification flow diagram.
Abbreviations: CIHI – Canadian Institute for Health Information; OCR – Ontario
Cancer Registry; BCa – Bladder Cancer; OHIP – Ontario Health Insurance Plan.
34
Figure 2.2: Kaplan-Meier survival curves stratified by local T stage.
The outcome assessed was overall survival in the 2535 patients with pathology data
available.
Survival by local T stage
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 12 24 36 48 60 72 84 96 108 120 132 144 156 168 180
Time (months)
Pro
ba
bil
ity
of
su
rviv
al
T0, Ta, Tis
T1
T2
T3
T4
Log Rank: p<0.001
35
Figure 2.3: Kaplan-Meier survival curves stratified by lymph node status.
The outcome assessed was overall survival in the 2535 patients with pathology data
available.
Survival by nodal status
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 12 24 36 48 60 72 84 96 108 120 132 144 156 168 180
Time (months)
Pro
ba
bil
ity
of
su
rviv
al
LN -
LN +
Nx
Log Rank: p<0.001
36
TABLES FOR CHAPTER 2
Table 2.1: Inter-rater reliability and agreement statistics for 2 raters extracting
pathologic variables from OCR radical cystectomy pathology reports.
VARIABLE AGREEMENT KAPPA/ICC (95% CI)
Presence of bladder cancer 95.5% 0.744 (0.585, 0.904)
T Stage
Stagability*#
Stage†
98.3%
88.3%
0.719 (0.415, 1.000)
0.824 (0.734, 0.906)
Grade
Gradability*#
Grade†
98.2%
99.4%
0.719 (0.416, 1.000)
0.986 (0.959, 1.000)
Adverse Pathology
Perineural invasion#
Vascular invasion#
Lymphatic invasion#
97.0%
94.5%
98.2%
0.919 (0.849, 0.989)
0.916 (0.862, 0.969)
0.970 (0.937, 1.000)
Total Lymph Nodes
Separate LN packets#
Extent of LN dissection#
Preciseness of LN count#
Number of LN‡
97.0%
96.9%
97.8%
--
0.938 (0.884, 0.991)
0.652 (0.289, 1.000)
0.891 (0.831, 0.951)
0.971 (0.964, 0.978)
Positive LN
Preciseness of positive LN count#
Number of positive LN‡
95.0%
--
0.903 (0.843, 0.963)
0.937 (0.921, 0.953)
*Stagability and gradability – refer to the presence of enough information on the
pathology to assign a stage and grade, respectively.
#Unweighted kappa statistic
†Linearly weighted kappa statistic
‡Intra-class correlation coefficient – the ICC [2,1] was calculated where 2 refers to the
fact that the two raters assessed each case and were considered a random sample from a
potential population of raters and the 1 refers to the fact that the reliability of single rating
(e.g. number of LN for a given patient) as opposed to the mean of several ratings was
assessed.97
Abbreviations: LN = lymph node
37
Table 2.2: Intra-rater reliability and agreement statistics for the primary
abstractor.
VARIABLE AGREEMENT KAPPA/ICC (95% CI)
Presence of bladder cancer 100% 1.000 (1.000, 1.000)
T Stage
Stagability*#
Stage†
100%
96.0%
1.000 (1.000, 1.000)
0.972 (0.945, 1.000)
Grade
Gradability*#
Grade†
96.0%
98.9%
0.583 (0.218, 0.947)
0.953 (0.868, 1.000)
Adverse Pathology
Perineural invasion#
Vascular invasion#
Lymphatic invasion#
100%
99.0%
98.0%
1.000 (1.000, 1.000)
0.985 (0.955, 1.000)
0.970 (0.929, 1.000)
Total Lymph Nodes
Separate LN packets#
Extent of LN dissection#
Preciseness of LN count#
Number of LN‡
100%
98.5%
97.0%
---
1.000 (1.000, 1.000)
0.662 (0.043, 1.000)
0.947 (0.889, 1.000)
0.999 (0.999, 1.000)
Positive LN
Preciseness of positive LN count#
Number of positive LN‡
100%
---
1.000 (1.000, 1.000)
0.963 (0.950, 0.973)
*Stagability and gradability – refer to the presence of enough information on the
pathology to assign a stage and grade, respectively.
#Unweighted kappa statistic
†Linearly weighted kappa statistic
‡Intra-class correlation coefficient – the ICC [2,1] was calculated where 2 refers to the
fact that the two raters assessed each case and were considered a random sample from a
potential population of raters and the 1 refers to the fact that the reliability of single rating
(e.g. number of LN for a given patient) as opposed to the mean of several ratings was
assessed.97
Abbreviations: LN = lymph node
38
Table 2.3: Pathology variables for the Pathology cohort.
These values were computed via direct assessment of each patient‟s cystectomy
pathology report (n=2535). Percentages for “tumour stage” may not add to 100 due to
rounding.
Variable Number (%) / Mean (SD)
Tumour Stage
Tx
T0
Ta
Tis
T1
T2
T3
T4
8 (0.3%)
47 (1.9%)
51 (2.0%)
127 (5.0%)
238 (9.4%)
646 (25.5%)
896 (35.4%)
522 (20.6%)
Grade
Not specified
Grade 1
Grade 2
Grade 3
182 (7.2%)
43 (1.7%)
330 (13.0%)
1980 (78.1%)
Positive Margin Status 414 (16.3%)
Lymphovascular invasion (LVI) 1019 (40.2%)
Perineural invasion* 398 (15.7%)
Lymphadenectomy performed† 1580 (62.5%)
Positive lymph node status
Nx
N0
N+
778 (30.7%)
1193 (47.1)
564 (22.3%)
*2 patients missing
†5 patients missing
39
Table 2.4: Patient level variables by pathology report availability.
(“Full Cohort” = entire cohort, “Pathology Cohort” = patients where pathology reports
were available; “Missing Pathology” = patients with missing pathology reports). P values
reflect comparisons between the groups with and without pathology reports. Continuous
variables were assessed using the Wilcoxon Mann Whitney test and categorical variables
were assessed via a Chi square test. The total number of patients in each cohort is: full –
3296, pathology – 2535 and missing pathology – 761.
Variable Full Cohort
Mean (SD) /
N (%)
Pathology
Cohort
Mean (SD) /
N (%)
Missing
Pathology
Mean (SD) / N
(%)
P value
Age 67.61 (9.96) 67.76 (9.90) 67.09 (10.15) 0.055
Sex
Male
2656 (80.6%)
2058 (81.2%)
598 (78.6%)
0.111
Charlson Comorbidity
Index score
2.47 (2.50)
2.56 (2.52)
2.17 (2.40)
<0.001
Socioeconomic status*
Quintile 1
Quintile 2 Quintile 3
Quintile 4
Quintile 5
579 (17.6%)
707 (21.5%) 631 (19.1%)
609 (18.5%)
685 (20.8%)
445 (17.6%)
548 (21.6%) 490 (19.3%)
454 (17.9%)
531 (21.0%)
134 (17.6%)
159 (20.9%) 141 (18.5%)
155 (20.4%)
154 (20.2%)
0.765
Admission status#
Urgent/Emergent
459 (13.9%)
376 (14.8%)
83 (10.9%)
0.006
LHIN
1 (Erie St. Clair)
2 (South West)
3 (Waterloo
Wellington)
4 (Hamilton Niagara
Haldimand Brant) 5 (Central West)
6 (Mississauga
Halton)
7 (Toronto Central)
8 (Central)
9 (Central East)
10 (South East)
11 (Champlain)
12 (North Simcoe
Muskoka)
13 (North East) 14 (North West)
213 (6.5%)
293 (8.9%)
194 (5.9%)
490 (14.9%)
156 (4.7%)
171 (5.2%)
287 (8.7%)
334 (10.2%)
396 (12.0%)
117 (3.6%)
258 (7.8%)
146 (4.4%)
192 (5.8%) 44 (1.3%)
153 (6.0%)
226 (8.9%)
148 (5.9%)
368 (14.5%)
123 (4.9%)
139 (5.5%)
238 (9.4%)
254 (10.0%)
327 (12.9%)
102 (4.0%)
174 (6.9%)
104 (4.1%)
145 (5.7%) 31 (1.2%)
60 (7.9%)
67 (8.8%)
46 (6.1%)
122 (16.1%)
33 (4.4%)
32 (4.2%)
49 (6.5%)
80 (10.5%)
69 (9.1%)
15 (2.0%)
84 (11.1%)
42 (5.5%)
47 (6.2%) 13 (1.7%)
<0.001
Adjuvant Chemotherapy 433 (13.1%) 348 (13.7%) 85 (11.1%) 0.066
Mortality
Postoperative
Overall
126 (3.8%)
2230 (67.7%)
104 (4.1%)
1796 (70.9%)
22 (2.9%)
434 (57.0%)
0.126
<0.001
Percentages may not add to 100 due to rounding.
* Quintile 5 refers to the highest socioeconomic status whereas quintile 1 is the lowest. A
total of 85 patients were missing socioeconomic information in the full cohort
#Urgent or Emergent admission status versus Elective admission
40
Table 2.5: Physician and hospital level variables by pathology report availability.
(“Full Cohort” = entire cohort, “Pathology Cohort” = patients where pathology reports
were available; “Missing Pathology” = patients with missing pathology reports). P value
reflect comparisons between groups with and without. Continuous variables were
assessed using the Kruskal Wallis test and categorical variables were assessed via a Chi
square test. The total number of patients in each cohort is: full – 3296, pathology – 2535
and missing pathology – 761. Percentages may not add to 100 due to rounding.
Variable Full Cohort
Mean (SD) /
N (%)
Pathology
Cohort
Mean (SD) /
N (%)
Missing
Pathology
Mean (SD) /
N (%)
P value
PHYSICIAN LEVEL
General Surgeon Volume* Quartile 1
Quartile 2
Quartile 3
Quartile 4
811 (25.9%)
749 (23.9%)
793 (25.3%)
783 (25.0%)
640 (27.0%)
560 (23.6%)
594 (25.0%)
581 (24.5%)
171 (22.5%)
189 (24.8%)
199 (26.2%)
202 (26.5%)
0.105
Wait time (days)# 64.47 (52.59) 64.53 (53.20) 64.25 (50.59) 0.604
Preoperative Anesthesia Consult 1558 (47.3%) 1167 (46.0%) 391 (51.4%) 0.010 Medical Consult 1740 (52.8%) 1350 (53.3%) 390 (51.3%) 0.331
Preoperative Imaging 2661 (80.7%) 2048 (80.8%) 613 (80.6%) 0.884
Intraoperative Anesthetic
specializationA 2970 (96.1%) 2278 (96.3%) 692 (95.6%) 0.393
Urologist –experience
(years)B 20.8 (9.4) 21.0 (9.5) 20.0 (9.2) 0.022
Urologist – international
medical graduateC
348 (11.1%) 259 (10.9%) 89 (11.7%) 0.551
Urologist as surgical
assistantD 1229 (40.0%) 907 (38.5%) 322 (45.2%) 0.001
Continent diversion 407 (13.0%) 301 (12.7%) 106 (13.9%) 0.369
HOSPITAL LEVEL
Hospital volume*
Quartile 1
Quartile 2
Quartile 3
Quartile 4
830 (25.2%)
794 (24.1%)
823 (23.6%)
849 (25.8%)
639 (25.2%)
604 (23.8%)
598 (23.6%)
694 (27.4%)
191 (25.1%)
190 (25.0%)
225 (29.6%)
155 (20.4%)
<0.001
Cardiac Catheterization
availability
1416 (43.0%) 1117 (44.1%) 299 (39.3%) 0.019
Regional Dialysis Centre 2013 (61.1%) 1566 (61.8%) 447 (58.7%) 0.131
Teaching status 1370 (41.6%) 1061 (41.9%) 309 (40.6%) 0.539
*Quartile 1 refers to the lowest volume surgeons/hospitals whereas quartile 4 is
comprised of the highest volume surgeons/hospitals. Surgeon volume values are
missing160 patients in the full cohort
#162 missing in full cohort (138 in pathology cohort and 24 in missing pathology group).
Median values for wait time: Full cohort – 51.0 d; Pathology cohort – 50.0 d; Missing
pathology cohort – 52.0 d. A206 missing in full cohort
41
C160 patients missing in full cohort
D225 missing in full cohort
B177 missing in full cohort
42
CHAPTER 3 : CYSTECTOMY VOLUME-OUTCOME ASSOCIATIONS IN
ONTARIO
SUMMARY
INTRODUCTION: Hospital and surgeon volume are often used as proxy measures of
quality of care for radical cystectomy. Studies published to date have primarily originated
from privately funded health care systems and have focused on post-operative mortality
rates. We assessed the effect of provider cystectomy volume on both postoperative and
overall mortality in a publicly funded health care setting.
METHODS: Patients undergoing cystectomy in Ontario, Canada, between 1992-2004
were identified via the Canadian Institute for Health Information Discharge Abstract
Database, a population-based administrative database of all inpatient hospital admissions.
The effects of hospital volume and surgeon volume on postoperative mortality rates were
assessed with multilevel, random effects logistic regression models. Analyses were
adjusted for patient characteristics. The effects of hospital volume and surgeon volume
on overall survival were assessed using Cox proportional hazards models designed to
account for patient clustering within hospital or surgeons, respectively. In addition to
patient factors, overall survival analyses were adjusted for tumour characteristics
extracted from cystectomy pathology reports gathered via linkage to the Ontario Cancer
Registry.
RESULTS: Of 3296 cystectomy patients identified, 126 (3.8%) experienced a
postoperative death. In separate models, neither hospital volume (Odds Ratio 0.98, 95%
CI: 0.95-1.00; p=0.074) nor surgeon volume (Odds Ratio 0.96, 95% CI: 0.90-1.02;
p=0.143) were significantly associated with postoperative cystectomy mortality.
43
However, both hospital volume (Hazard Ratio 0.995, 95% CI: 0.990-1.000; p=0.044) and
surgeon volume (Hazard Ratio 0.984, 95% CI: 0.975-0.994; p=0.002) were significantly
associated with overall survival. With both hospital volume and surgeon volume in the
Cox model, neither was statistically significant, indicating that the high volume benefit
was attained by receiving care from either high volume hospitals or high volume
surgeons.
CONCLUSIONS: In a publicly funded health care system, provider volume was not
significantly associated with postoperative mortality. High volume providers, however,
experienced improved overall mortality rates compared to low volume providers. Future
research should focus on the underlying processes that contribute to the overall mortality
benefit of high volume providers.
44
INTRODUCTION
Patients who undergo radical cystectomy for bladder cancer carry a significant
risk for postoperative death and/or an attenuated life expectancy. As a result, the quality
of care for patients undergoing radical cystectomy has become an important focus of
research. With postoperative mortality rates for these patients ranging between 1-
4%26,49,52
and long term overall mortality rates between 40-50%65,95,96
, many investigators
have initiated efforts to identify gaps in quality of care delivery in an attempt to improve
patient outcomes. A popular model used to study quality of care is Donabedian‟s
Structure-Process-Outcome framework.21,22
Using this model, current outcomes research
has focused on volume of care as a starting point for quality of care assessment.
The association between provider volume and cystectomy outcome has been
described previously in the medical literature.26,27,45
These studies have demonstrated
that higher volume hospitals and surgeons tend to have improved outcomes compared to
their lower volume counterparts. While the mechanisms underlying this relationship
remain unclear, it is likely that differences in the structures and/or processes of care are
responsible for the effects of “volume.” Consequently, some researchers have attempted
to unearth these structures/processes to further understand “volume,”53,55
whereas others
have supported regionalization of care to higher volume providers.98
The impact of hospital and surgeon volume on cystectomy outcomes, however,
may not be completely generalizable. With few exceptions46,50
, almost all studies on the
topic have arisen from the United States.41,43
Furthermore, the databases used to study
provider volume represent samples of the U.S. population. For example, Medicare
datasets are restricted to patients aged 65 and older71
, the Surveillance, Epidemiology and
45
End Results (SEER) database is based on representative samples from SEER areas and
captures 26% of the American population73
and the NIS represents a 20% stratified
sample of U.S. community hospitals.72
The potential for selection bias, therefore, exists
and additional whole-population studies outside of the U.S. healthcare setting are
required to demonstrate the generalizability of the volume-outcome phenomenon at the
international level.
The effect of volume on long term cystectomy outcomes, on the other hand, has
only been studied to a limited extent.57
Ample evidence, however, supports differences
in long term survival based on surgical technique and the performance of an optimal
tumour-clearing operation.99-101
Higher case loads may facilitate refined intraoperative
techniques, supporting the possibility that volume may be related to long term survival.
Furthermore, since care of the cancer patient is not restricted to the perioperative period,
it is reasonable to hypothesize that volume may serve as a quality of care indicator for
long term outcomes at the hospital level as well. Although evidence is lacking,
differences in tumour surveillance, medical oncology involvement and/or screening for
complications of urinary diversion could all potentially contribute to differences in long
term survival post-cystectomy.
Based on these limitations, we investigated the impact of both hospital and
surgeon cystectomy volume on post-operative mortality rates in a publicly-funded
(Canadian) health care setting. We also studied the effect of cystectomy provider volume
on long-term outcomes to address the knowledge deficit in the literature on this topic.
46
METHODS
Cohort Identification
After ethics approval from the Sunnybrook Health Sciences Centre and University
of Toronto institutional review boards, we evaluated the effect of hospital and surgeon
volume for radical cystectomy on both operative mortality and overall survival in the
province of Ontario. Between 1992 and 2004, radical cystectomy patients were identified
from the Canadian Institute for Health Information Discharge Abstract Database (CIHI
DAD) using Canadian Classification of Diagnostic, Therapeutic and Surgical Procedures
(CCP) and Canadian Classification of Health Interventions (CCI) procedure codes (from
1992-2002 CCP: 69.51; from 2003-2004 CCI: 1.PM.91 and 1.PM.92;). The CIHI DAD is
a population-based database that contains information on all inpatient hospital admissions
in Ontario. In addition to identifying cystectomy patients, the CIHI DAD in conjunction
with the provincial Registered Person‟s Database, provides demographic details for each
cystectomy patient including age, sex, comorbidity, urgency of admission, region of
residence and vital status. Comorbidity in the form of the Charlson Comorbidity Index,
was derived based on CIHI DAD International Classification of Diseases (ICD)
diagnostic codes from each patient‟s index admission and from any hospital admissions
in the year prior to cystectomy.85,102,103
Comorbid status was divided into 4 categories
(Charlson 0, 1, 2 and > 2) and classified as none, mild, moderate and severe,
respectively.104
Because radical cystectomy can be performed for both bladder cancer and for
non-bladder malignancies, the latter as part of larger exenterative procedures for
colorectal, prostate or gynecological malignancies, we linked the CIHI data to the Ontario
47
Cancer Registry (OCR) to select only those cystectomy patients with a diagnosis of
bladder cancer. The OCR contains information on all incident cancers detected in the
province of Ontario with 97% capture of incident cases of bladder cancer.88
A total of
3296 patients undergoing cystectomy for bladder cancer were identified. These patients
served as the cohort for analyses in which short-term mortality served as the outcome
measure.
Because of the importance of pathological variables in assessing survival
outcomes, we limited analyses assessing long term outcomes to those individuals who
had pathology reports available for review at OCR. The cohort used to assess the impact
of volume on overall survival was thus composed of 2535 patients who represent 77% of
all patients that underwent cystectomy for bladder cancer in the province of Ontario
between 1992 and 2004. The pathology reports of all 2535 patients were reviewed for
important pathologic variables including pathologic stage, grade, margin and lymph node
status and the presence of lymphovascular invasion or perineural invasion. Pathologic
staging was based on the 2002 American Joint Committee on Cancer system.105
Outcome Definitions
Two outcomes were assessed in this study: 1) Short term (“operative”) mortality
was defined as death within 30 days of discharge or death prior to discharge (based on
prior convention in the volume-outcome literature55
); 2) Long term mortality was defined
by survival during the study period (and called “overall survival”).
48
Exposure Definitions
Hospital volume was defined as the average annual number of cystectomy cases
performed at an institution during the study time period. In situations where a hospital
closed or newly opened, only the years of the hospital‟s existence during the study span
were used for volume calculations. Hospitals were identified by using CIHI DAD
institution unique identifiers. Between 1996 and 2000, hospital mergers and
amalgamations occurred frequently in Ontario, resulting in changes to the hospital
identifying numbers in CIHI DAD. For the purposes of volume measurement, hospitals
that underwent a corporate amalgamation where medical services were not transferred
were treated the same way pre- and post-amalgamation with respect to identifying
institution numbers. Hospitals that underwent a merger or closure, however, where
medical services were transferred and cystectomy volumes changed, were treated as
separate institutions after the merger/closure to reflect changes in volume status. Details
surrounding hospital restructuring in Ontario were derived from a local Institution
database and from each hospital‟s website. The importance of properly accounting for
hospital restructuring in hospital volume-outcome analyses has been outlined in a
previous report from our group (Kulkarni et al, submitted).
Surgeon volume was defined as the average annual number of cystectomy cases
performed by a surgeon during his/her active years of clinical activity. This definition
enabled accurate calculation of volume in situations where a surgeon retired or started
practice in Ontario during the study time period. Surgeons were identified based on their
Ontario Health Insurance Plan (OHIP) unique identifiers. Because of the fee-for-service
nature of Canadian health care, each cystectomy identified in CIHI is linkable to an OHIP
49
billing fee-code (S484, S485, S453, S440) and thus a specific surgeon. Small pockets of
care in Ontario, however, are remunerated via salary and thus lack billing codes and
accompanying surgeon identifiers. Consequently, 160 (4.9%) cystectomy cases were
missing surgeon identifiers.
Potential Confounding Variables
Analyses in which provider volume was regressed against operative mortality
were risk adjusted for age, sex, admission status (urgent/emergent vs. elective), Charlson
comorbidity score and socioeconomic status (SES). Socioeconomic status was based on
neighbourhood-specific quintiles of income (higher quintiles corresponding with higher
income) as derived from the Canadian Census. For patients operated on between 1992
and 1998, the 1996 census was used for SES derivation whereas the 2001 census was
referenced for patients operated on between 1999 and 2004. The above factors represent
common risk adjustment variables used in volume-outcome studies.26,27
Analyses assessing overall survival outcomes were adjusted for the patient factors
listed above in addition to pathology variables, use of adjuvant chemotherapy, patient
location of residence (Local Health Integration Network – LHIN) at the time of operation
and year of operation. These additional variables were included for adjustment of overall
survival because of their potential impact on long term outcomes. Patient LHIN and year
of operation were obtained from the CIHI DAD. Adjuvant chemotherapy, determined
from OHIP billing codes for systemic chemotherapy (G381, G281, G339, G345, G382),
was defined by the initiation of chemotherapy in the first 6 months postoperatively. We
chose a 6 month time period because this allowed ample time for patient discharge,
50
postoperative followup, referral to medical oncology and initiation of chemotherapy. We
did not account for the use of neoadjuvant chemotherapy since this treatment was not
widely used during the study time period (<1% of patients).
Statistical Analyses
Due to the hierarchical nature of the data, with patients clustered within surgeons
and surgeons clustered within hospitals, we used statistical software with multi-level
modeling capabilities where applicable.106
The program MLwiN v2.02 (Centre for
Multilevel Modeling, Bristol, UK) was used to fit random effects logistic regression
models. All remaining statistical analyses were performed using SAS version 9.1.3 (SAS
Institute, Cary, North Carolina). A two-sided p value of 0.05 was defined as statistically
significant. For descriptive statistics, the data were divided into quartiles of hospital
volume and surgeon volume. Comparisons across quartiles were assessed using the
Kruskal Wallis test for continuous variables and the Chi square or Fisher‟s Exact test for
categorical variables. Multicollinearity, defined as a variance inflation factor (VIF) >
10107
, was determined for all variables to ensure collinear covariates were not added to
the subsequent regression models.
The effect of hospital and surgeon volume on operative mortality was determined
using fully adjusted, 3-level random intercept models. We fit 2 separate models: 1) A
hospital volume model without including surgeon volume; 2) A surgeon volume model
without including hospital volume. Goodness of fit of both logistic models was
determined via the Hosmer-Lemeshow test, with p values > 0.05 signifying adequate
model fitting.108
51
Multivariable Cox proportional hazards modeling was performed to assess the
effect of volume (surgeon and hospital) on overall mortality. A total of 3 separate models
was fit: 1) Hospital volume alone; 2) Surgeon volume alone; 3) Hospital and surgeon
volume together in the same model. We used marginal („variance-corrected‟) survival
models designed to account for non-independent observations at either the hospital or
surgeon levels for hospital volume-outcome and surgeon volume-outcome analyses,
respectively.109,110
Patients alive as of March 31, 2007, the last day of follow-up, were
censored. This ensured a minimal follow up of 2 years and a maximum potential follow
up of 15 years. Observations with identical follow up times (ties) were handled by the
method of Efron.111
In all analyses, volume was modeled as a continuous variable. To
avoid survivor treatment bias when adjusting for adjuvant chemotherapy, we modeled use
of adjuvant chemotherapy as a time-dependent covariate. Evaluation of the proportional
hazards assumption for all models was performed by incorporating volume into the
model as a time dependent covariate (volume*loge(survival_time)).112
Since risk adjustment using administrative datasets may not be fully accurate, we
performed sensitivity analyses, reproducing the multivariable Cox proportional hazards
models using only the healthiest patients (Charlson comorbidity index score of 0 or 1, ) to
“level the playing field” and potentially eliminate unmeasured confounding.113
We were
unable to perform this sensitivity analysis where operative mortality was the outcome
because of a paucity of short term events (33 deaths) in low Charlson score patients. A
second sensitivity analysis was performed by excluding patients with an operative
mortality outcome and subsequently assessing the effect of hospital/surgeon volume on
overall mortality. This type of analysis removed the impact of operative deaths and
52
allowed us to determine the effect of volume on long term outcomes after the
perioperative period. In a third and final sensitivity analysis, we assessed the impact of
provider volume on overall survival in the entire cohort of 3296 patients using only
covariates derived from administrative data. This analysis was aimed at assessing the
volume-overall mortality association in all cystectomy patients, albeit limited by a lack of
risk-adjustment for pathologic information. We hypothesized that consistent
demonstration of a volume-outcome relationship in the entire population would support
the results of the fully-adjusted model in which 761 patients with missing pathology
reports were omitted.
53
RESULTS
Patient and Provider Demographics
From 1992 to 2004, radical cystectomy was performed by 199 surgeons in 90
hospitals across the province of Ontario. A breakdown of the number of cases, hospitals,
physicians and volume cutpoints by quartile of hospital and surgeon volume is provided
in Table 3.1. Hospitals and surgeons in the highest volume quartiles performed equal to
or greater than 19.43 and 8.11 cases per year respectively. Baseline information for the
entire cohort, divided into quartiles of hospital volume and surgeon volume, are presented
in Tables 3.2 and 3.3, respectively. Higher volume hospitals tended to treat younger
patients with a higher SES. Significant differences in patient region of residence also
existed with marked differences across quartiles for all LHIN‟s. Patients at high volume
centres were also more likely to receive adjuvant chemotherapy. With respect to
pathologic variables, a trend (p=0.054) to higher stage disease was noted at high volume
institutes. Lower rates of perineural invasion along with higher rates of
lymphadenectomy and lymph node negative disease were noted at high volume hospitals.
Low volume hospitals had higher rates of lymph node positive disease. Assessment of
baseline patient variables across surgeon volume quartiles revealed similar findings as
those observed across hospital volume quartiles. One exception was a trend (p=0.056) to
more urgent/emergent admission by high versus low volume surgeons. For pathologic
variables, statistically significant differences for tumour stage or perineural invasion were
not seen. However, lymphadenectomy and lymph node status were significantly different
across surgeon volume quartiles, with the same pattern seen with hospital volume
quartiles.
54
Operative Mortality
A total of 126 patients experienced operative mortality. Figures 3.1 and 3.2 depict
the operative mortality rates for all 3296 patients across hospital and surgeon volume
quartiles. The highest volume hospitals had an operative mortality rate of 2.9% compared
to 4.3% for the lowest volume quartile. Similarly, the highest volume surgeons had an
operative mortality rate of 2.9% compared to 4.3% for the lowest surgeons. Tables 3.4
and 5 provide results from random effects logistic regression models regressing volume
against operative mortality. In both crude (unadjusted) and adjusted analyses, neither
hospital volume nor surgeon volume were statistically significantly associated with
operative mortality, although the odds ratio and p value for hospital volume (OR: 0.975,
p=0.074; Table 3.4) suggested a trend towards improved outcomes at higher volume
centres. For both multivariate models, the Hosmer-Lemeshow goodness of fit test was
non-significant, indicating adequate model fit (Hospital volume model: p=0.9057;
Surgeon volume model: p = 0.8538).
Overall Mortality
Of the 2535 patients with available pathology information, 1796 died during the
study time period. The mean (SD) and median (range) follow up for the cohort was 1260
days (SD: 1276) and 786 days (Range: 0-5441), respectively. The 5-year overall survival
rate was 35%. Both hospital and surgeon volume were statistically significant in both
unadjusted and adjusted Cox proportional hazards models (Tables 3.6 and 3.7,
respectively). The adjusted hazard ratio for hospital volume (HR=0.995) implied that for
every additional cystectomy performed at a hospital, the risk of overall mortality
55
decreased by 0.5%. Likewise, for every additional cystectomy performed by an
individual surgeon, the risk of overall death diminished by 1.6% (HR=0.984). The
decrease in the instantaneous hazard of death for selected cystectomy volume thresholds,
based on the Cox multivariate model results, is presented in Table 3.8. For each increase
in hospital caseload by 10 cystectomy operations per year, a decrease in the hazard of
death of 5.2% would be expected (i.e. e-0.0053*10
= 0.948). At the surgeon level, an
increase of 10 cases per year would result in a 14.7% (e-0.0159*10
= 0.853) decreased risk of
death, indicating that the benefits of volume on mortality risk reduction are greatest at the
surgeon level. Incorporating both hospital volume and surgeon volume in the same
regression model nullified the significance of both variables (Table 3.9), demonstrating
that adjustment for surgeon volume removed the effect of hospital volume and vice versa.
Since the VIF for both of these variables implied non-collinearity (VIF<5), the beneficial
effect of high volume was attainable at either the hospital or the surgeon level.
The Proportional Hazards assumption for hospital volume was tested using a
time-dependent covariate [loge(survival_time)*hospital_volume] in the final model. This
parameter was not statistically significant (p=0.2191) indicating that the proportional
hazards assumption was not violated. The same assumption was tested for surgeon
volume using the time-dependent covariate [loge(survival_time)*surgeon_volume] in the
final surgeon volume model. Since this parameter was not statistically significant
(p=0.0693), the proportional hazards assumption was met for the surgeon volume Cox
model as well.
A series of sensitivity analyses were then performed to assess the robustness of
the results. First, repeating our analyses in only low-risk patients (i.e. Charlson scores 0
56
and 1) did not alter the conclusions (HR (95% CI) for hospital volume: 0.985 (0.976,
0.994), p=0.002; HR (95% CI) for surgeon volume: 0.972 (0.953, 0.993), p=0.008),
suggesting that inaccuracies inherent to risk-adjustment with administrative data were not
responsible for the reported results. In a second sensitivity analysis, removal of patients
who suffered an operative death enabled confirmation of the long term impact of volume
on overall survival. Without these patients, both hospital volume and surgeon volume
remained significant (or near-significant) predictors of overall survival (HR (95% CI) for
hospital volume: 0.995 (0.991, 1.000), p=0.051; HR (95% CI) for surgeon volume: 0.986
(0.976, 0.996), p=0.007), implying that the impact of volume on long term survival
occurred after the perioperative period. Third, an analysis of all 3296 patients without
adjustment for pathology variables yielded similar results to fully-adjusted analyses,
indicating that omission of the 761 patients without pathology reports did not introduce
significant selection bias (HR (95% CI) for hospital volume: 0.993 (0.989, 0.998),
p=0.003; HR (95% CI) for surgeon volume: 0.979 (0.970, 0.989), p<0.001).
57
DISCUSSION
Outcomes researchers have used provider procedural volume as a means of
uncovering gaps in quality of care delivery. For radical cystectomy, the majority of such
studies have arisen from restricted datasets in a private health care system. Using a
dataset with full population coverage from a publicly funded health care system, we
demonstrated that neither hospital nor surgeon volume were significantly associated with
operative mortality in the province of Ontario. However, the point estimate for the effect
of hospital volume on operative mortality was in keeping with a modest effect, and the p
value of 0.074 was close to conventional statistical significance. This suggests that this
study may have been underpowered to detect a true association between hospital volume
and operative mortality. For long term outcomes, provider volume was significantly
associated with overall survival, with fewer deaths attributable to high volume providers.
The beneficial impact of surgeon volume was approximately 3 times greater than that of
hospital volume. Nevertheless, hospital and surgeon volume dually (within the same
model) were not significantly associated with overall mortality, suggesting that patients
treated at a high volume centre or by a high volume surgeon could expect improved
outcomes.
A significant association between provider volume and long term cystectomy
survival has not been reported by others. Birkmeyer and colleagues assessed this question
for hospital volume using the Surveillance, Epidemiology and End Results (SEER)-
Medicare linked database but failed to demonstrate a significant association in adjusted
analyses.57
Nevertheless, they were able to detect differences in late mortality by hospital
volume for other operatively-treated primary neoplasms (esophagus, lung, pancreas and
58
stomach), supporting the basis for the cystectomy volume-outcome association found
with our data. The reasons for the relationship between volume and late outcome are
currently unclear. Possibilities for high volume surgeons include superior operative
technique (i.e. optimal cancer resections or improved lymph node harvesting and
staging), vigilant follow-up and surveillance for recurrent disease and/or timely and
appropriate use of ancillary medical services such as medical oncology. Explanations for
the impact of high volume hospitals are less clear but could include improved pathways
to detect recurrent disease and thus the timely provision of treatment, superior access to
chemotherapy and/or better treatment of comorbid diseases. Clearly, future research is
required to elucidate the pathways and mechanisms responsible for the volume-overall
survival relationship.
Contrary to the work of others, our data did not support a volume perioperative
mortality association at either the hospital or surgeon level. Only a handful of other
studies have failed to detect significant volume outcome associations for perioperative
cystectomy mortality.47,52
In all of these studies, however, a trend to improved outcomes
with higher volumes was seen, with some authors suggesting a lack of statistical
significance secondary to small sample sizes and thus diminished power.47,114
This
explanation could also apply to our hospital volume-operative mortality results. Another
potential explanation for our results could lie in the differences inherent to the health care
systems of Canada and the United States. For example, Urbach et al. observed that
Canadian volume-outcome studies were significantly less likely to report statistically
significant volume-outcome associations compared to U.S.-based studies.75
They
hypothesized that less inter-hospital competition and the potential for coordinated health
59
services in a public, single-payer health care system may decrease variability in the
delivery of quality care compared to a market-based model where competition could
potentially exacerbate such differences. Indirect evidence supporting this claim comes
from the Veterans‟ Administration system in the United States, which operates as a
publicly funded system in a private health care environment.115
Few volume-outcome
studies from the VA system have been positive.116
A final possible reason why short
term mortality was not significantly associated with volume in Ontario is that the data in
this study (for short term outcomes) represent an entire population of cystectomy cases.
Since many positive volume-outcome studies use databases of representative population
samples, it is possible that selection bias in the representative samples may have
contributed to the significant results from these studies. Variations in the results of
volume outcome studies based on samples of populations compared to complete datasets
have been reported by others.71
Our study is not without limitations. First, as alluded to above, the lack of a
significant hospital volume-outcome association for operative mortality may have been
secondary to decreased statistical power to detect an actual mortality difference (a type II
error). A power calculation using commercially available statistical software (PASS –
Power and Sample Size software, NCSS, Kaysville, Utah) while accounting for the
multilevel nature of the data117
suggested that our dataset may not have had enough
statistical power to detect a significant association, making this limitation a possible
concern (i.e. for a single standard deviation increase in hospital volume, our data had
enough statistical power (assuming a power of 80%) to detect an odds ratio of 0.674. The
actual data yielded an odds ratio of 0.797 for a standard deviation increase).
60
Nevertheless, other investigators have been able to detect significant differences in
similar analyses with far fewer patients118
, implying larger quality gaps in their patient
populations. Second, our analysis of overall survival outcomes was limited to 2535
patients with pathology data. We restricted our patient population because of the
importance of adjusting long term cancer-survival outcomes for pathologic variables such
as stage and grade. The 761 omitted patients tended to be healthier with lower overall
mortality compared to the patients with reported pathology. Furthermore, there were also
statistically significant differences in hospital volume (but not surgeon volume) between
patients with and without pathology data, with the latter preferentially treated at lower
volume centres. Consequently, it remains possible that our results are confounded by
omission of these 761 patients. However, an adjusted analysis with all patients (pathology
data excluded), supported our fully-adjusted findings, suggesting that this limitation may
not have materially affected our overall conclusions. Third, while we were able to assess
the effect of volume on operative and overall mortality, lack of validity in administrative
data precluded assessment of other outcome measures such as disease-specific or
recurrence-free survival.91
Additional work is necessary to validate these measures in
administrative data prior to their widespread use.
Our results raise an important health policy question. Now that a deficit in quality
of care for radical cystectomy has been established (manifest by variations in long term
survival outcomes), how do we proceed to narrow the gap? Regionalization of health
services has been proposed as one possible solution. Private insurers in the U.S. have
already promoted volume-based referral patterns, implementing minimum volume
thresholds for a number of complex operations based on published volume-outcome
61
studies.119
Many physicians, researchers and health policy makers, however, question the
utility of regionalization, given the additional burden of excess travel time for
patients120,121
, the potential marginalization of lower-volume physicians and the logistical
difficulties inherent to implementing system-wide change.122
Furthermore, the
directionality of the volume-outcome association has not been definitively proven.123
In
this study we have assumed, as have others, that provider procedure volume is a mediator
of quality of care. It remains possible that quality may actually be the driver of volume as
opposed to volume being the driver of quality. In other words, selective referral to high
quality hospitals and/or surgeons may cause higher volumes to be associated with
improved outcomes. Such reverse-causation would argue against regionalized care
because minimum volume thresholds in that setting would not necessarily result in higher
quality care.
Given the practical and theoretical concerns regarding regionalization of health
care, a growing movement is underway to truly understand what “volume” means.
Evidence is mounting that volume is actually a surrogate for underlying structures and/or
processes of care which, in turn, affect quality of care.124
Identifying relevant
structure/process measures and then implementing them in the form of best practice
guidelines could ultimately improve the outcomes associated with low volume service
providers. To date, little is known about the important structures and processes of care
underlying the volume-outcome association. Additional research is necessary to elucidate
these factors and eventually adopt them for radical cystectomy patients.
62
CONCLUSIONS
Provider volume is widely used as a surrogate for quality of care. During the
modern era in the province of Ontario, a statistically significant association between
radical cystectomy provider volume and operative mortality did not exist. However, both
hospital and surgeon volume were significantly associated with overall mortality in this
patient cohort, with the effect of surgeon volume being three times larger than that of
hospital volume. Unfortunately, the mechanisms behind the volume-overall survival
relationship remain unclear. Further research to identify the structures and processes of
care underlying the impact of provider volume is necessary to improve the quality of care
afforded to cystectomy patients.
63
FIGURES FOR CHAPTER 3
Figure 3.1: Postoperative mortality by hospital volume quartile between 1992-2004.
Increasing quartiles indicate increasing hospital volume. (n=3296)
Postoperative Mortality by Hospital Volume Quartile
between 1992-2004
4.33.7
4.4
2.9
-1
1
3
5
7
1 2 3 4
Hospital Volume Quartile
Pe
rce
nt
64
Figure 3.2: Postoperative mortality by surgeon volume quartile between 1992-2004.
Increasing quartiles indicate increasing surgeon volume. (n=3136)
Postoperative Mortality by Surgeon Volume Quartile
between 1992-2004
4.35.1
3.3 2.9
-1
1
3
5
7
1 2 3 4
Surgeon Volume Quartile
Perc
en
t
65
TABLES FOR CHAPTER 3
Table 3.1: General cohort characteristics based on average annual volume quartiles.
Increasing quartile indicates increasing cystectomy volume.
Volume Measure Quartile 1 Quartile 2 Quartile 3 Quartile 4
Hospital Volume
Number of patients 830 794 823 849
Number of hospitals 58 17 11 4
Volume cut-points 0.77 – 3.22 3.23 – 5.85 6.00 – 17.00 19.43 – 32.63
Surgeon Volume
Number of patients 811 749 793 783
Number of surgeons 128 42 21 8
Volume cut-points 0.77 – 1.54 1.67 – 2.54 2.63 – 8.08 8.11 – 16.71
66
Table 3.2: Patient level and pathologic variables by hospital volume quartile.
Hospital volume increases with quartiles. Values listed are counts (percentages) or means
(standard deviations). Patient level variables are determined based on the full cohort of
3296 patients while pathology variables are based on the 2535 patients with pathology
information available. P values reflect comparisons across quartiles.
Variable Hospital Volume P value
Quartile 1
(n=830)
Quartile 2
(n=794)
Quartile 3
(n=823)
Quartile 4
(n=849)
PATIENT LEVEL
Age 68.4 (9.39) 67.5 (10.00) 68.0 (9.59) 66.5 (10.71) 0.006
Sex
Males
677 (81.6%)
636 (80.1%)
665 (80.8%)
678 (79.9%)
0.816
Comorbidity†
None
Mild
Moderate
Severe
282 (34.0%)
79 (9.5%) 197 (23.7%)
272 (32.8%)
266 (33.5%)
77 (9.7%) 184 (23.2%)
267 (33.6%)
276 (33.5%)
73 (8.9%) 195 (23.7%)
279 (33.9%)
247 (29.1%)
69 (8.1%) 189 (22.3%)
344 (40.5%)
0.087
Socioeconomic status*
Quintile 1
Quintile 2
Quintile 3 Quintile 4
Quintile 5
156 (18.8%)
168 (20.2%)
169 (20.4%) 142 (17.1%)
171 (20.6%)
104 (13.1%)
185 (23.3%)
158 (19.9%) 156 (19.7%)
173 (21.8%)
172 (20.9%)
186 (22.6%)
167 (20.3%) 157 (19.1%)
122 (14.8%)
147 (17.3%)
168 (19.8%)
137 (16.1%) 154 (18.1%)
219 (25.8%)
<0.001
Admission status
Urgent/Emergent
102 (12.3%)
104 (13.1%)
127 (15.4%)
126 (14.8%)
0.216
Adjuvant
chemotherapy
97 (11.7%)
115 (14.5%)
67 (8.1%)
154 (18.1%)
<0.001
LHIN
1 (Erie St. Clair)
2 (South West)
3 (Waterloo
Wellington)
4 (Hamilton Niagara Haldimand Brant)
5 (Central West)
6 (Mississauga
Halton)
7 (Toronto Central)
8 (Central)
9 (Central East)
10 (South East)
11 (Champlain)
12 (North Simcoe
Muskoka) 13 (North East)
14 (North West)
50 (6.0%)
48 (5.8%)
37 (4.5%)
149 (18.0%)
16 (1.9%)
62 (7.5%)
37 (4.5%)
119 (14.4%)
55 (6.6%)
25 (3.0%)
79 (9.5%)
38 (4.6%)
82 (9.9%)
32 (3.9%)
1 (0.1%)
50 (6.3%)
112 (14.1%)
72 (9.1%)
102 (12.9%)
42 (5.3%)
95 (12.0%)
82 (10.4%)
90 (11.4%)
4 (0.5%)
83 (10.5%)
52 (6.6%)
6 (0.8%)
1 (0.1%)
83 (10.1%)
58 (7.1%)
26 (3.2%)
239 (29.1%)
3 (0.4%)
9 (1.1%)
20 (2.4%)
20 (2.4%)
162 (19.7%)
84 (10.2%)
96 (11.7%)
8 (1.0%)
9 (1.1%)
5 (0.6%)
79 (9.3%)
137 (16.2%)
19 (2.2%)
30 (3.5%)
35 (4.1%)
58 (6.8%)
135 (15.9%)
113 (13.3%)
89 (10.5%)
4 (0.5%)
0 (0%)
48 (5.7%)
95 (11.2%)
6 (0.7%)
<0.001
PATHOLOGY
Quartile 1
(n=639)
Quartile 2
(n=604)
Quartile 3
(n=598)
Quartile 4
(n=694)
Tumour Stage
Tx
T0
3 (0.5%)
13 (2.0%)
1 (0.2%)
7 (1.2%)
4 (0.7%)
13 (2.2%)
0 (0%)
14 (2.0%)
0.054
67
Ta
Tis
T1
T2
T3
T4
13 (2.0%)
28 (4.4%)
65 (10.2%)
163 (25.5%)
237 (37.1%)
117 (18.3%)
13 (2.2%)
38 (6.3%)
47 (7.8%)
165 (27.3%)
228 (37.8%)
105 (17.4%)
13 (2.2%)
37 (6.2%)
58 (9.7%)
147 (24.6%)
197 (32.9%)
129 (21.6%)
12 (1.7%)
24 (3.5%)
68 (9.8%)
171 (24.6%)
234 (33.7%)
171 (24.6%)
Grade
Not specified
Grade 1
Grade 2
Grade 3
42 (6.6%)
6 (0.9%)
74 (11.6%)
516 (80.9%)
42 (7.0%)
13 (2.2%)
79 (13.1%)
470 (77.8%)
46 (7.7%)
12 (2.0%)
87 (14.6%)
453 (75.8%)
51 (7.4%)
12 (1.7%)
90 (13.0%)
541 (78.0%)
0.641
Positive Margin Status 106 (16.6%) 92 (15.2%) 101 (16.9%) 115 (16.6%) 0.866
Lymphovascular
invasion (LVI)
258 (40.4%) 241 (39.9%) 221 (37.0%) 299 (43.1%) 0.168
Perineural invasion 122 (19.2%) 79 (13.1%) 83 (13.9%) 114 (16.4%) 0.014
Lymphadenectomy# 324 (50.8%) 363 (60.3%) 337 (56.4%) 556 (80.4%) <0.001
Positive Lymph node
status
Nx
N0
N+
254 (39.8%)
244 (38.2%)
141 (22.1%)
210 (34.8%)
263 (43.5%)
131 (21.7%)
209 (35.0%)
282 (47.2%)
107 (17.9%)
105 (15.1%)
404 (58.2%)
185 (26.7%)
<0.001
†Comorbidity scale based on Charlson scores: None = Charlson 0; Mild = Charlson 1;
Moderate = Charlson 2 and Severe = Charlson > 2.
*Quintile 5 refers to the highest socioeconomic (neighbourhood income) status whereas
quintile 1 is the lowest.
#Percentages refer to those who have undergone a lymphadenectomy.
Percentages may not add to 100 due to rounding.
68
Table 3.3: Patient level and pathologic variables by surgeon volume quartile.
Surgeon volume increases with quartiles. Values listed are counts (percentages) or means
(standard deviations). Patient level variables are determined based on the full cohort of
3136 patients while pathology variables are based on the 2375 patients with pathology
information available. P values reflect comparisons across quartiles.
Variable Surgeon Volume P value
Quartile 1
(n=811)
Quartile 2
(n=749)
Quartile 3
(n=793)
Quartile 4
(n=783)
PATIENT LEVEL
Age 68.22 (9.27) 67.83 (9.77) 68.08 (9.90) 66.61 (10.64) 0.021
Sex
Males
659 (81.3%)
586 (78.2%)
644 (81.2%)
637 (81.4%)
0.339
Comorbidity†
None
Mild
Moderate
Severe
281 (34.7%)
85 (10.5%) 187 (23.1%)
258 (31.8%)
246 (32.8%)
70 (9.4%) 166 (22.2%)
267 (35.7%)
251 (31.7%)
67 (8.5%) 181 (22.8%)
294 (37.1%)
237 (30.3%)
68 (8.7%) 186 (23.8%)
292 (37.3%)
0.391
Socioeconomic status*
Quintile 1
Quintile 2
Quintile 3 Quintile 4
Quintile 5
131 (16.2%)
192 (23.7%)
178 (22.0%) 127 (15.7%)
162 (20.0%)
138 (18.4%)
154 (20.6%)
135 (18.0%) 156 (20.8%)
145 (19.4%)
150 (18.9%)
172 (21.7%)
141 (17.8%) 139 (17.5%)
167 (21.1%)
129 (16.5%)
155 (19.8%)
146 (18.7%) 158 (20.2%)
182 (23.2%)
0.058
Admission status
Urgent/Emergent
105 (13.0%)
103 (13.8%)
96 (12.1%)
130 (16.6%)
0.056
Adjuvant
chemotherapy
100 (12.3%)
77 (10.3%)
105 (13.2%)
128 (16.4%)
0.004
LHIN
1 (Erie St. Clair)
2 (South West)
3 (Waterloo
Wellington)
4 (Hamilton Niagara Haldimand Brant)
5 (Central West)
6 (Mississauga
Halton)
7 (Toronto Central)
8 (Central)
9 (Central East)
10 (South East)
11 (Champlain)
12 (North Simcoe
Muskoka) 13 (North East)
14 (North West)
48 (5.9%)
59 (7.3%)
41 (5.1%)
103 (12.7%)
49 (6.1%)
71 (8.8%)
85 (10.5%)
105 (13.0%)
72 (8.9%)
19 (2.4%)
90 (11.1%)
39 (4.8%)
24 (3.0%)
5 (0.6%)
59 (7.9%)
12 (1.6%)
83 (11.1%)
97 (13.0%)
41 (5.5%)
39 (5.2%)
51 (6.8%)
103 (13.8%)
123 (16.5%)
7 (0.9%)
65 (8.7%)
32 (4.3%)
32 (4.3%)
2 (0.3%)
23 (2.9%)
99 (12.5%)
45 (5.7%)
125 (15.8%)
43 (5.4%)
24 (3.0%)
68 (8.6%)
48 (6.1%)
148 18.7(%)
20 (2.5%)
2 (0.3%)
45 (5.7%)
75 (9.5%)
47 (3.4%)
77 (9.8%)
119 (15.2%)
22 (2.8%)
152 (19.4%)
19 (2.4%)
33 (4.2%)
71 (9.1%)
63 (8.1%)
41 (5.2%)
6 (0.8%)
97 (12.4%)
25 (3.2%)
50 (6.4%)
8 (1.0%)
<0.001
PATHOLOGY
Quartile 1
(n=640)
Quartile 2
(n=560)
Quartile 3
(n=594)
Quartile 4
(n=581)
Tumour Stage
Tx
T0
0 (0%)
9 (1.4%)
3 (0.5%)
13 (2.3%)
3 (0.5%)
9 (1.5%)
0 (0%)
15 (2.6%)
0.197
69
Ta
Tis
T1
T2
T3
T4
9 (1.4%)
32 (5.0%)
58 (9.1%)
181 (28.3%)
234 (36.6%)
117 (18.3%)
12 (2.1%)
25 (4.5%)
54 (9.6%)
138 (24.6%)
214 (38.2%)
101 (18.0%)
15 (2.5%)
26 (4.4%)
62 (10.4%)
145 (24.4%)
205 (34.5%)
129 (21.7%)
15 (2.6%)
33 (5.7%)
55 (9.5%)
141 (24.3%)
186 (32.0%)
136 (23.4%)
Grade
Not specified
Grade 1
Grade 2
Grade 3
36 (5.6%)
10 (1.6%)
84 (13.2%)
509 (79.7%)
41 (7.3%)
9 (1.6%)
80 (14.3%)
430 (76.8%)
48 (8.1%)
8 (1.4%)
62 (10.4%)
476 (80.1%)
46 (7.9%)
14 (2.4%)
91 (15.7%)
430 (74.0%)
0.130
Positive Margin Status 106 (16.6%) 83 (14.8%) 101 (17.0%) 94 (16.2%) 0.770
Lymphovascular
invasion (LVI)
250 (39.1%) 225 (40.2%) 247 (41.6%) 230 (39.6%) 0.826
Perineural invasion 93 (14.6%) 105 (18.8%) 90 (15.2%) 90 (15.5%) 0.207
Lymphadenectomy# 341 (53.5%) 286 (51.3%) 394 (66.4%) 457 (78.7%) <0.001
Positive Lymph node
status
Nx
N0
N+
255 (39.8%)
243 (38.0%)
142 (22.2%)
224 (40.0%)
226 (40.4%)
110 (19.6%)
166 (28.0%)
284 (47.8%)
144 (24.2%)
86 (14.8%)
357 (61.5%)
138 (23.8%)
<0.001
†Comorbidity scale based on Charlson scores: None = Charlson 0; Mild = Charlson 1;
Moderate = Charlson 2 and Severe = Charlson > 2.
*Quintile 5 refers to the highest socioeconomic (neighbourhood income) status whereas
quintile 1 is the lowest.
#Percentages refer to those who have undergone a lymphadenectomy.
Percentages may not add to 100 due to rounding.
70
Table 3.4: Effect of Hospital Volume on Postoperative Mortality.
P values derived from a 3-level random effects logistic regression model. Hospital
volume was modeled as a continuous variable. Random effects for both hospitals and
surgeons were included in the models.
Variable Beta
coefficient
Standard
Error
Odds
Ratio
95% C.I. P value
Crude (Unadjusted) Model
Hospital Volume -0.0267 0.0158 0.974 (0.943, 1.005) 0.091
Adjusted Model
Hospital Volume -0.0258 0.0144 0.975 (0.947, 1.003) 0.074
Age (per yr) 0.1076 0.0138 1.114 (1.083, 1.145) <0.001
Gender -0.1934 0.2347 0.824 (0.515, 1.318) 0.410
Charlson
Comorbidity Score
0.1206 0.0371 1.128 (1.047, 1.215) 0.001
Admission Status 0.7285 0.2302 2.072 (1.307, 3.283) 0.002
Socioeconomic
Status Quintile*
1 (reference)
2
3
4
5
---
-0.2122
-0.3674
-0.4018
-0.1999
---
0.2868
0.3118
0.3290
0.2944
---
0.809
0.693
0.669
0.819
---
(0.456, 1.435)
(0.371, 1.292)
(0.347, 1.292)
(0.454, 1.475)
---
0.459
0.239
0.222
0.497
*Complete data on all 3296 patients was available for the Crude Model. The Adjusted
Model is based on 3211 patients (85 missing).
71
Table 3.5: Effect of Surgeon Volume on Postoperative Mortality.
P values derived from a3-level random effects logistic regression model. Surgeon volume
was modeled as a continuous variable. Random effects for both hospitals and surgeons
were included in the models.
Variable Beta
coefficient
Standard
Error
Odds
Ratio
95% C.I. P value
Crude (Unadjusted) Model
Surgeon Volume -0.0432 0.0315 0.958 (0.899, 1.020) 0.170
Adjusted Model
Surgeon Volume -0.0452 0.0309 0.956 (0.899, 1.017) 0.143
Age (per yr) 0.1077 0.0138 1.114 (1.083, 1.145) <0.001
Gender -0.1873 0.2330 0.829 (0.520, 1.321) 0.421
Charlson
Comorbidity Score
0.1199 0.0368 1.127 (1.047, 1.213) 0.001
Admission Status 0.7362 0.2276 2.088 (1.324, 3.292) 0.001
Socioeconomic
Status Quintile*
1 (reference)
2
3
4
5
---
-0.2059
-0.3654
-0.4015
-0.1973
---
0.2841
0.3093
0.3264
0.2912
---
0.814
0.694
0.669
0.821
---
(0.461, 1.437)
(0.374, 1.288)
(0.348, 1.286)
(0.459, 1.470)
---
0.469
0.238
0.219
0.498
*Complete data on 3136 patients was available for the Crude Model. The Adjusted
Model is based on 3057 patients (3296-3057 = 239 total missing).
72
Table 3.6: Effect of Hospital Volume on Overall Mortality.
P values derived from Cox Proportional Hazards model after accounting for clustered
data at the hospital level. Hospital volume was modeled as a continuous variable.
(n=2454 for the adjusted model)
Variable Beta
coefficient
Standard
Error
Hazard
Ratio
95% C.I. P value
Crude (Unadjusted) Model
Hospital Volume -0.0062 0.0025 0.994 (0.989, 0.999) 0.015
Adjusted Model
Hospital Volume -0.0053 0.0026 0.995 (0.990, 1.000) 0.044
Age (per yr) 0.0233 0.0034 1.024 (1.017, 1.030) <0.001
Gender -0.0840 0.0782 0.919 (0.789, 1.072) 0.282
Comorbidity†
None (ref)
Mild
Moderate
Severe
---
0.1239
0.1203
0.3584
---
0.0809
0.0754
0.0813
---
1.132
1.128
1.431
---
(0.966, 1.326)
(0.973, 1.307)
(1.220, 1.678)
---
0.126
0.111
<0.001
Admission Status 0.1558 0.0767 1.169 (1.006, 1.358) 0.042
Socioeconomic
Status Quintile
1
2
3
4
5 (ref)
0.1825
0.1096
-0.0436
0.0112
---
0.0748
0.0588
0.0617
0.0762
---
1.200
1.116
0.957
1.011
---
(1.037, 1.390)
(0.994, 1.252)
(0.848, 1.080)
(0.871, 1.174)
---
0.015
0.063
0.480
0.883
---
Tumour Stage
T0, Ta, Tis (ref)
T1
T2
T3
T4
---
0.3000
0.4017
0.8582
1.0183
---
0.0988
0.1086
0.1109
0.1075
---
1.350
1.494
2.359
2.768
---
(1.112, 1.638)
(1.208, 1.849)
(1.898, 2.932)
(2.243, 3.418)
---
0.002
<0.001
<0.001
<0.001
Margin 0.4023 0.0590 1.495 (1.332, 1.679) <0.001
Nodal Status
N0 (ref)
N+
Nx
---
0.2748
0.2123
---
0.0815
0.0960
---
1.316
1.236
---
(1.122, 1.544)
(1.024, 1.493)
---
<0.001
0.027
Lymphadenectomy -0.1105 0.0888 0.895 (0.752, 1.066) 0.214
Adjuvant Chemo -0.1404 0.0753 0.869 (0.750, 1.008) 0.063
LVI 0.4785 0.0426 1.614 (1.484, 1.754) <0.001
PNI 0.0031 0.0662 1.003 (0.881, 1.142) 0.963
Tumour Grade
1 (ref)
2
3
X (missing/T0)
---
0.1101
0.1573
0.2260
---
0.2283
0.2086
0.2356
---
1.116
1.170
1.254
---
(0.714, 1.746)
(0.778, 1.761)
(0.790, 1.989)
---
0.630
0.451
0.337
73
Year
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004 (ref)
-0.0196
-0.0026
-0.0198
-0.1223
0.1183
-0.0908
-0.0082
0.0697
0.1138
0.0979
-0.1728
-0.1662
---
0.1720
0.1831
0.1844
0.1662
0.1733
0.1519
0.1513
0.1806
0.1512
0.1592
0.1801
0.1680
---
0.981
0.997
0.980
0.885
1.126
0.913
0.992
1.072
1.120
1.103
0.841
0.847
---
(0.700, 1.374)
(0.697, 1.428)
(0.683, 1.407)
(0.639, 1.226)
(0.801, 1.581)
(0.678, 1.230)
(0.737, 1.334)
(0.753, 1.528)
(0.833, 1.507)
(0.807, 1.507)
(0.591, 1.198)
(0.609, 1.177)
---
0.909
0.989
0.914
0.462
0.495
0.550
0.957
0.699
0.452
0.538
0.338
0.323
---
Local Health
Integration
Network (LHIN)
1
2
3
4
5
6
7
8
9
10
11
12
13
14 (ref)
-0.3680
-0.4713
-0.1691
-0.3080
-0.2433
-0.2844
-0.3200
-0.4759
-0.3599
-0.4737
-0.4133
-0.3208
-0.1902
---
0.1663
0.1125
0.1209
0.0962
0.1443
0.1393
0.1184
0.1531
0.1229
0.1239
0.1512
0.1576
0.1090
---
0.692
0.624
0.844
0.735
0.784
0.752
0.726
0.621
0.698
0.623
0.661
0.726
0.827
---
(0.500, 0.959)
(0.501, 0.778)
(0.666, 1.070)
(0.609, 0.887)
(0.591, 1.040)
(0.573, 0.989)
(0.576, 0.916)
(0.460, 0.839)
(0.548, 0.888)
(0.488, 0.794)
(0.492, 0.890)
(0.533, 0.988)
(0.668, 1.024)
---
0.027
<0.001
0.162
0.001
0.092
0.041
0.007
0.002
0.003
<0.001
0.006
0.042
0.081
---
74
Table 3.7: Effect of Surgeon Volume on Overall Mortality.
P values derived from Cox Proportional Hazards model after accounting for clustered
data at the surgeon level. Surgeon volume was modeled as a continuous variable.
(n=2302 for the adjusted model)
Variable Beta
coefficient
Standard
Error
Hazard
Ratio
95% C.I. P value
Crude (Unadjusted) Model
Surgeon Volume -0.0188 0.0044 0.981 (0.973, 0.990) <0.001
Adjusted Model
Surgeon Volume -0.0159 0.0050 0.984 (0.975, 0.994) 0.002
Age (per yr) 0.0245 0.0033 1.025 (1.018, 1.032) <0.001
Gender -0.0876 0.0684 0.916 (0.801, 1.048) 0.200
Comorbidity†
None (ref)
Mild
Moderate
Severe
---
0.1191
0.1552
0.3834
---
0.0897
0.0763
0.0752
---
1.126
1.168
1.467
---
(0.945, 1.343)
(1.006, 1.356)
(1.266, 1.700)
---
0.185
0.042
<0.001
Admission Status 0.1540 0.0716 1.166 (1.014, 1.342) 0.032
Socioeconomic
Status Quintile
1
2
3
4
5 (ref)
0.1923
0.1048
-0.0330
0.0263
---
0.0809
0.0673
0.0707
0.0799
---
1.212
1.110
0.968
1.027
---
(1.034, 1.420)
(0.973, 1.267)
(0.842, 1.111)
(0.878, 1.201)
---
0.018
0.119
0.641
0.742
---
Tumour Stage
T0, Ta, Tis (ref)
T1
T2
T3
T4
---
0.2837
0.3721
0.8340
1.0127
---
0.1201
0.1201
0.1150
0.1315
---
1.328
1.451
2.303
2.753
---
(1.049, 1.681)
(1.146, 1.836)
(1.838, 2.885)
(2.128, 3.562)
---
0.018
0.002
<0.001
<0.001
Margin 0.4362 0.0708 1.547 (1.347, 1.777) <0.001
Nodal Status
N0 (ref)
N+
Nx
---
0.2680
0.2011
---
0.0822
0.0987
---
1.307
1.223
---
(1.113, 1.536)
(1.008, 1.484)
---
0.001
0.042
Lymphadenectomy -0.0784 0.0921 0.925 (0.772, 1.108) 0.395
Adjuvant Chemo -0.1567 0.0717 0.855 (0.743, 0.984) 0.029
LVI 0.4947 0.0524 1.640 (1.480, 1.817) <0.001
PNI 0.0004 0.0731 1.000 (0.867, 1.155) 0.995
Tumour Grade
1 (ref)
2
3
X (missing/T0)
---
0.1378
0.1854
0.2666
---
0.2278
0.2248
0.2488
---
1.148
1.204
1.306
---
(0.734, 1.794)
(0.775, 1.870)
(0.802, 2.126)
---
0.545
0.410
0.284
75
Year
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004 (ref)
0.0665
0.0363
-0.0240
-0.0976
0.1147
-0.0997
0.0293
0.0785
0.0759
0.0860
-0.1811
-0.1674
---
0.1581
0.1498
0.1626
0.1531
0.1473
0.1458
0.1536
0.1526
0.1500
0.1459
0.1510
0.1587
---
1.069
1.037
0.976
0.907
1.122
0.905
1.030
1.082
1.079
1.090
0.834
0.846
---
(0.784, 1.457)
(0.773, 1.391)
(0.710, 1.343)
(0.672, 1.224)
(0.840, 1.497)
(0.680, 1.204)
(0.762, 1.391)
(0.802, 1.459)
(0.804, 1.448)
(0.819, 1.450)
(0.621, 1.122)
(0.620, 1.155)
---
0.674
0.809
0.883
0.524
0.436
0.494
0.849
0.607
0.613
0.555
0.231
0.292
---
Local Health
Integration
Network (LHIN)
1
2
3
4
5
6
7
8
9
10
11
12
13
14 (ref)
-0.3144
-0.4551
-0.1461
-0.2531
-0.2852
-0.3009
-0.3255
-0.4374
-0.3509
-0.4168
-0.4108
-0.3004
-0.1733
---
0.1805
0.1561
0.1344
0.1290
0.1591
0.1572
0.1542
0.1349
0.1299
0.2262
0.1653
0.1562
0.1473
---
0.730
0.634
0.864
0.776
0.752
0.740
0.722
0.646
0.704
0.659
0.663
0.740
0.841
---
(0.513, 1.040)
(0.467, 0.862)
(0.664, 1.124)
(0.603, 1.000)
(0.550, 1.027)
(0.544, 1.007)
(0.534, 0.977)
(0.496, 0.841)
(0.546, 0.908)
(0.423, 1.027)
(0.480, 0.917)
(0.545, 1.006)
(0.630, 1.112)
---
0.082
0.004
0.277
0.050
0.073
0.056
0.035
0.001
0.007
0.065
0.013
0.054
0.239
---
76
Table 3.8: Decrease in hazard of overall death by an incremental increase in the
number of cystectomy operations performed at the hospital or surgeon level.
Hazard ratios based on fully adjusted analyses (Tables 6 and 7 above).
Incremental increase in the
annual number of
cystectomy procedures
Hospital Volume
Hazard Ratio
Surgeon Volume
Hazard Ratio
1 0.995 0.984
5 0.974 0.924
10 0.948 0.853
20 0.899 0.728
77
Table 3.9: Simultaneous effect of Hospital and Surgeon Volume on Overall
Mortality.
P values derived from a Cox Proportional Hazards model after accounting for clustered
data at the surgeon level. The model was fully adjusted using the same covariates listed in
Tables 3 and 4 (covariate parameters not shown). Hospital and surgeon volume were
modeled as continuous variables. Similar results were obtained when accounting for
clustering at the hospital level. (n=2302)
Variable Beta
coefficient
Standard
Error
Hazard
Ratio
95% C.I. P value
Adjusted Model*
Hospital Volume -0.0022 0.0042 0.998 (0.990, 1.006) 0.606
Surgeon Volume -0.0124 0.0089 0.988 (0.971, 1.005) 0.166
78
CHAPTER 4 : CYSTECTOMY VOLUME AND OVERALL MORTALITY –
UNDERLYING STRUCTURES AND PROCESSES OF CARE
SUMMARY
INTRODUCTION: High hospital and surgeon volumes are associated with improved
long term mortality outcomes following radical cystectomy. The mechanisms behind this
phenomenon are unclear. We assessed the preoperative processes, physician
characteristics and hospital-level factors that may underlie the cystectomy volume-
outcome relationship.
METHODS: All patients undergoing cystectomy in Ontario, Canada, between 1992-
2004 were identified via the Canadian Institute for Health Information Discharge
Abstract Database. Linkage to the Ontario Cancer Registry enabled review of each
patient‟s cystectomy pathology. Baseline Cox proportional hazards models, designed to
account for patient clustering within hospital or surgeons and adjusted for patient factors
and pathologic factors, were created. Sequential addition of preoperative physician
processes (consultations, imaging studies) intraoperative physician variables (experience,
specialization and choice of diversion) and hospital-level factors (teaching status, cardiac
catheterization and dialysis capabilities) as variable blocks to the model was performed to
assess which set of variables, if any, attenuated the effect of volume on overall mortality.
Assessment of each individual variable on the volume hazard ratio (HR) was also
performed to elucidate the single most influential factor underlying the volume-outcome
relationship.
RESULTS: A total of 2535 patients were included in the analysis. Both baseline hospital
and surgeon volume models, adjusted for patient characteristics, were statistically
79
significant (Hospital volume HR: 0.995, p=0.044; Surgeon volume HR: 0.984, p=0.002).
Addition of preoperative process or physician characteristic variables attenuated the
significance of hospital volume but not surgeon volume in the models. However, the
point estimate (HR) for hospital volume did not change after accounting for these
variables. Introduction of hospital factors attenuated the significance of volume in both
models with the HR for hospital volume (0.997) moving closer to 1.0 (null effect) than
the HR for surgeon volume (0.988). The most influential hospital characteristic was the
presence of on-site cardiac catheterization facilities.
CONCLUSIONS: Hospital factors, specifically the presence of on-site cardiac
catheterization facilities, were the most influential determinants underlying the hospital
volume and surgeon volume association with overall survival. The effect of hospital
factors was greatest on the hospital volume-outcome association. Structures and process
of care measures underlying the surgeon volume-outcome association require further
elucidation.
80
INTRODUCTION
Tremendous interest has emerged in assessing volume of care as a marker of
quality of care. Support for volume as a quality surrogate comes from many studies
investigating the impact of provider volume on outcome.58,125
Consistent demonstration
of inverse volume-outcome relationships, with improved outcomes linked to high volume
providers, has led to efforts aimed at understanding why high volume providers
experience better outcomes. One dominant theory states that high volume providers have
superior underlying structures and processes of care in place which, in turn, translate into
better outcomes.23
A number of studies have demonstrated improved postoperative mortality rates
for radical cystectomy procedures performed by high volume providers.26,27,114
Investigation of the processes and structures of care responsible for these relationships is
under way.51,53,55
However, research has only just begun to explore the relationship
between volume and long term outcome57
, making the underlying structures/process
measures for this phenomenon even less defined. At present, additional work is required
to pinpoint the important factors that contribute to volume-outcome associations.
In Chapter 3, we found that both hospital and surgeon cystectomy volume were
significantly associated with overall survival in the province of Ontario. Since the
structures/processes of care underlying cystectomy “volume,” particularly for long term
outcomes, remain unclear and research on the topic is only in its infancy, we set out to
identify which variables could potentially explain the association between volume and
overall survival. We hypothesized that key explanatory variables, when added to the
existing statistically significant volume-outcome regression models, would attenuate the
81
volume hazard ratio thereby indicating that these factors act as important mediators of
volume.
82
METHODS
Overview
After ethics approval from the Sunnybrook Health Sciences Centre and University
of Toronto institutional review boards, we retrospectively explored the impact of various
pre-defined structures and processes of care on the hospital and surgeon volume-survival
association for radical cystectomy in the province of Ontario. We restricted our analyses
to long term survival outcomes because, based on work from Chapter 3 of this thesis, we
only detected a significant association between provider volume and outcome for overall
survival.
Cohort Identification
Between 1992 and 2004, radical cystectomy patients were identified from the
Canadian Institute for Health Information Discharge Abstract Database (CIHI DAD)
using Canadian Classification of Diagnostic, Therapeutic and Surgical Procedures (CCP)
and Canadian Classification of Health Interventions (CCI) procedure codes (from 1992-
2002 CCP: 69.51; from 2003-2004 CCI: 1.PM.91 and 1.PM.92;). The CIHI DAD is a
population-based database that contains information on all inpatient hospital admissions
in Ontario. In addition to identifying cystectomy patients, the CIHI DAD in conjunction
with the provincial Registered Person‟s Database, provided demographic details for each
cystectomy patient including age, sex, comorbidity, urgency of admission, region of
residence and vital status. Comorbidity in the form of the Charlson Comorbidity Index,
was derived based on CIHI DAD International Classification of Diseases (ICD)
diagnostic codes from each patient‟s index admission and from any hospital admissions
83
in the year prior to cystectomy.85,102,103
Comorbid status was divided into 4 categories
(Charlson 0, 1, 2 and > 2) and classified as none, mild, moderate and severe,
respectively.104
Because radical cystectomy can be performed for both bladder cancer and for
non-bladder malignancies, the latter as part of larger exenterative procedures for
colorectal, prostate or gynecological malignancies, we linked the CIHI data to the Ontario
Cancer Registry (OCR) to select only those cystectomy patients with a diagnosis of
bladder cancer. The OCR contains information on all incident cancers detected in the
province of Ontario with 97% capture of incident cases of bladder cancer.88
A total of
3296 patients undergoing cystectomy for bladder cancer were identified.
Because of the importance of pathological variables in assessing survival
outcomes, we limited analyses to those individuals who had pathology reports available
for review at OCR. The cohort used to define the relevant structure and process measures
for volume and overall survival was thus composed of 2535 patients who represented
77% of all patients that underwent cystectomy for bladder cancer in the province of
Ontario between 1992 and 2004. The pathology reports of all 2535 patients were
reviewed for important pathologic variables including pathologic stage, grade, margin
and lymph node status and the presence of lymphovascular invasion or perineural
invasion. Pathologic staging was based on the 2002 American Joint Committee on
Cancer system.105
84
Volume-Overall Survival Analyses
Hospital volume was defined as the average annual number of cystectomy cases
performed at an institution during the study time period. In situations where a hospital
closed or newly opened, only the years of the hospital‟s existence during the study period
were used for volume calculations. Hospitals were identified via CIHI DAD institution
unique identifiers. Between 1996 and 2000, hospital mergers and amalgamations
occurred with regularity in Ontario resulting in changes to the hospital identifying
numbers in CIHI DAD. For the purposes of volume measurement, hospitals that
underwent a corporate amalgamation where medical services were not transferred were
treated the same way pre- and post-amalgamation with respect to identifying institution
numbers. However, hospitals that underwent a merger or closure, where medical services
were transferred and cystectomy volumes changed, were treated as separate institutions
after the merger/closure to reflect changes in volume status. Details surrounding hospital
restructuring in Ontario were derived from a local Institution database and from each
hospital‟s website. The importance of properly accounting for hospital restructuring in
hospital volume-outcome analyses has been outlined in a previous report from our group
(Kulkarni et al, submitted).
Surgeon volume was defined as the average annual number of cystectomy cases
performed by a surgeon during his/her active years of clinical activity. This definition
enabled accurate calculation of volume in situations where a surgeon retired or started
practice in Ontario during the study time period. Surgeons were identified based on their
Ontario Health Insurance Plan (OHIP) unique identifiers. Because of the fee-for-service
nature of Canadian health care, each cystectomy identified in CIHI is linkable to an OHIP
85
billing fee-code (S484, S485, S453, S440) and thus a specific surgeon. Small pockets of
care in Ontario, however, are remunerated via salary and thus lack billing codes and
accompanying surgeon identifiers. Consequently, 160 (4.9%) cystectomy cases were
missing surgeon identifiers.
All baseline analyses in which provider volume was regressed against overall
mortality were risk adjusted for age, sex, admission status (urgent/emergent vs. elective),
Charlson comorbidity score, socioeconomic status (SES), pathology variables, use of
adjuvant chemotherapy, patient location of residence (Local Health Integration Network
– LHIN) at the time of operation and year of operation. Socioeconomic status was based
on neighbourhood-specific quintiles of income (higher quintiles corresponding with
higher income) as derived from the Canadian Census. For patients operated on between
1992 and 1998, the 1996 census was used for SES derivation whereas the 2001 census
was referenced for patients operated on between 1999 and 2004. Adjuvant chemotherapy,
determined from OHIP billing codes for systemic chemotherapy (G381, G281, G339,
G345, G382), was defined by the initiation of chemotherapy in the first 6 months
postoperatively. We chose a 6 month time period because this allowed ample time for
patient discharge, postoperative followup, referral to medical oncology and initiation of
chemotherapy. Although adjuvant chemotherapy can be considered a process of care
measure, its effect in that capacity has been evaluated elsewhere57
without significant
influence on the volume-outcome relationship. Consequently, we incorporated it as a
risk-adjustment variable because of its impact on survival after radical cystectomy. We
did not account for the use of neoadjuvant chemotherapy since this treatment was not
widely used during the study time period (<1% of patients). Patient location of residence
86
(Local Health Integration Network – LHIN) and year of operation were obtained from the
CIHI DAD.
Structures and Processes of Care
Structures/processes that could be defined using administrative data alone and
which could potentially be associated with long term outcomes were defined a priori.
These variables were derived by an expert panel consisting of 2 urologic oncologists, one
internist, one general surgeon and one urology resident (the latter the primary author of
this thesis). Deriving measures directly from patient records was beyond the scope of this
thesis. A total of 11 distinct candidate processes and structures of care were identified
and divided into the following categories: physician preoperative processes, physician
intraoperative variables and hospital-level factors. A list of these variables, their
definitions and the sources from which they were derived is provided in Table 4.1.
Statistical Analyses
All statistical analyses were performed using SAS version 9.1.3 (SAS Institute,
Cary, North Carolina). A two-sided p value of 0.05 was defined as statistically
significant. For descriptive statistics, the data were divided into quartiles of hospital
volume and surgeon volume. Comparisons across quartiles were assessed using the
Kruskal Wallis test for continuous variables and the Chi square or Fisher‟s Exact test for
categorical variables. Multicollinearity, defined as a variance inflation factor (VIF) >
10107
, was determined for the potential structure/process variables to ensure collinear
covariates were not added to the subsequent regression models.
87
Multivariable Cox proportional hazards modeling was performed to assess the
baseline effect of volume on overall mortality. We fit 2 separate baseline models: 1) A
hospital volume model without including surgeon volume; 2) A surgeon volume model
without including hospital volume. We used marginal („variance-corrected‟) survival
models designed to account for non-independent observations at either the hospital or
surgeon levels for hospital volume-outcome and surgeon volume-outcome analyses,
respectively.109,110
To avoid survivor treatment bias when adjusting for adjuvant
chemotherapy, we modeled use of adjuvant chemotherapy as a time-dependent covariate.
Patients alive as of March 31, 2007, the last day of follow-up, were censored. This
ensured a minimal follow up of 2 years and a maximum potential follow up of 15 years.
Observations with identical follow up times (ties) were handled by the method of
Efron.111
In all analyses, volume was modeled as a continuous variable.
To determine which structure or process of care variables, if any, were
responsible for the effect of volume on long term outcome, we first inserted the potential
variables as category blocks (A: physician – preoperative, B: physician – intraoperative
and C: hospital) and assessed the impact of each block of variables on the hazard ratio
(HR) and p value of volume. Next, we used all possible combinations of these variable
blocks (i.e. (i) A and B; (ii) A and C; (iii) B and C; (iv) A, B and C) to determine which
combination of blocks resulted in the greatest attenuation of the hospital and surgeon HR.
Finally, to determine the individual effect of each variable on hospital volume, we
inserted each variable individually into the fully adjusted baseline model and noted the
effect of the variable on the volume beta coefficient (hazard ratio).
88
RESULTS
Univariate Analyses
Of the 2535 patients with available pathology information, 1796 died during the
study time period. The mean (SD) and median (range) follow up for the cohort was 1260
days (SD: 1275) and 786 days (Range: 0-5441), respectively. The 5-year overall survival
rate was 35%. Tables 4.2 and 4.3 illustrate differences in the structure and process
variables across hospital and surgeon volume quartiles, respectively. With the exception
of preoperative anesthetic consults and, across hospital volume quartiles, use of
preoperative imaging, there were statistically significant differences across provider
volume quartiles for all of the putative structures and processes of care. Higher volume
hospitals were associated with higher rates of preoperative medical consult usage
(compared to the lowest volume quartile) and were more likely to have onsite cardiac
catheterization facilities, serve as regional dialysis centres and function as teaching
institutes. With respect to intraoperative physician characteristics, high volume centres
tended to employ specialty-trained anesthesiologists, slightly older surgeons and surgeons
trained in North America or countries with similar training systems. Use of continent
diversions was more common in high volume hospitals as was use of non-urologists as
surgical assistants. Similar patterns were seen across surgeon volume quartiles.
Exceptions included surgeon age, as high volume surgeons tended to be younger, and use
of preoperative imaging which was more commonly associated with higher volume
surgeons.
89
Multi-collinearity Assessment
To ensure the candidate structure and process variables were not measuring the
same construct, we calculated the VIF for each variable including hospital and surgeon
volume (Table 4.4). The highest VIF was associated with hospital volume (4.01),
indicating that none of the variables were collinear.
Hospital Volume and Structures/Processes of Care
The hazard ratio (95% CI) for hospital volume, adjusted for patient and
pathologic factors, was 0.995 (0.990, 1.000). This model served as the baseline hospital
volume-outcome model (Figure 4.1). Addition of preoperative or intraoperative physician
level variables led to no attenuation of the hospital volume hazard ratio but did nullify the
statistical significance of hospital volume. Addition of hospital level variables increased
the HR for volume modestly to 0.997 and resulted in the loss of statistical significance,
indicating that, of the factors we measured, those at the hospital level had the greatest
effects on the hospital volume HR. The combination of variable categories that attenuated
the HR most (i.e. moved the HR closest to 1.0) were intraoperative and hospital factors
together (HR .999). Examination of each structure and process variable individually
revealed that the presence of onsite cardiac catheterization facilities (a structural
variable), affected the volume regression coefficient more than any other variable (Table
4.5).
90
Surgeon Volume and Structures/Processes of Care
For surgeon volume, the HR (95% CI) for the adjusted, baseline model was 0.984
(0.975, 0.994). Figure 4.2 depicts the effect of adding preoperative, intraoperative or
hospital variable blocks on the surgeon volume hazard ratio. Neither preoperative nor
intraoperative processes/structures attenuated the statistical significance or the HR for
surgeon volume. However, the addition of hospital-level structural variables did result in
a loss of the statistical significance of surgeon volume and an increase in the volume HR
to 0.988, again demonstrating that, of the variables we measured, hospital level factors
were the most important factors influencing the surgeon volume HR. The combination of
categories that caused the most attenuation of the HR was the dual inclusion of both
intraoperative and hospital level variables, although the rise in the HR with the additional
intraoperative factors was negligible (HR increase of 0.0001) compared to the model with
hospital factors alone. This combination of variables had less of an impact nullifying the
HR of surgeon volume compared to the effect they exerted on the HR for hospital
volume. Finally, assessing the impact of each structure/process variable individually on
the surgeon volume beta (Table 4.6) demonstrated that the availability of cardiac
catheterization facilities was the most influential variable affecting surgeon volume.
91
DISCUSSION
With mounting evidence associating provider volume and outcome, health
outcomes researchers have begun to investigate factors that may explain why high
volume hospitals and surgeons experience improved outcomes. We attempted to uncover
the structures and processes of care potentially associated with radical cystectomy
provider volume in Ontario, which in Chapter 3 was found to be associated with overall
survival. Assessing both hospital and surgeon structure and process measures, we
demonstrated marked differences in candidate variables across provider volume quartiles,
suggesting that differences in quality of care could potentially be attributable to these
variables. However, the only group of factors that independently attenuated the point
estimates of both hospital and surgeon volume, albeit to a modest degree, was hospital
structural variables. Assessing both hospital and intraoperative variables together
appeared to explain all of the effect of hospital volume but only a small proportion of the
surgeon volume effect. The most influential hospital structural variable was the presence
of onsite cardiac catheterization facilities.
The relationship between cystectomy volume and overall survival has not been
well described in the medical literature. To date, only Birkmeyer and colleagues have
studied cystectomy hospital volume and its association with long term survival.57
In their
study, cystectomy volume was not significantly associated with survival (High volume
tertile versus low volume tertile: HR 0.90, 95%CI: 0.79-1.02). Nevertheless, due to the
trend suggesting benefit with high volume hospitals, they assessed the impact of one
process of care measure, provision of adjuvant chemotherapy, as a potential explanatory
variable. Rather than attenuating the effect of hospital volume on late survival, inclusion
92
of this variable had virtually no effect on the HR of hospital volume (HR 0.89, 95% CI:
0.79-1.01). Based on this observation and its potential impact on overall survival6,13,126
,
we included adjuvant chemotherapy as a risk-adjusting variable rather than as a process
of care measure in our analyses.
For both hospital volume and surgeon volume, we found that hospital-level
structural variables were the most influential group of variables affecting the volume-
outcome relationship and, in particular, the presence of onsite cardiac catheterization was
the most influential hospital factor. A number of arguments, however, suggest that
cardiac catheterization capacity, like provider volume, acted as a surrogate for underlying
structures/processes important for the care of bladder cancer patients. First, since the
majority (81.5%) of our patients had T2 disease or worse, most of these patients likely
died from bladder cancer causes96
rather than cardiac causes, making it unlikely that any
improved cardiac care associated with cardiac catheterization capabilities accounted for
the long term beneficial effect of hospital volume. Second, prior work has demonstrated
no difference in survival for cardiac patients in hospitals with versus those without on-
site revascularization facilities.127
Finally, cystectomy performance at a hospital with on-
site catheterization facilities doesn‟t necessarily mean subsequent cardiac care delivery at
the same location.
After accounting for hospital factors and intraoperative factors, the effect of
hospital volume on overall survival was completely attenuated (HR 0.999),
demonstrating that we were able to completely account for the effect of hospital volume
with structure and process of care variables. Unfortunately, as mentioned, we still do not
fully understand which hospital factors are truly at play since the most influential variable
93
(cardiac catheterization) is also a surrogate measure. Our data also preclude determining
which intraoperative factors were most important because none of the component
variables minimized the hospital volume HR in a meaningful way. Hospital and
intraoperative variables technically represented the strongest combination of
structure/process categories for surgeon volume, although the magnitude of change in the
surgeon volume HR was minimal. These data suggest that additional, unmeasured
variables are important determinants of both the hospital and surgeon volume effect on
overall survival.
Our study does not allow us to make broad health policy recommendations. As of
yet, the actual process and structural characteristics underlying provider volume, in the
context of long term outcomes, remain elusive. However, this work does provide
reassurance that cystectomy hospital volume is a surrogate for underlying processes and
structures of care, since the latter variables were able to attenuate the hospital volume-
outcome association. Ameliorating gaps in quality of care between high and low volume
centres could therefore theoretically be achieved by identifying these important
structure/process variables. Future research is required to uncover the variables actually
underlying volume. Our study also provides direction for future studies aimed at
uncovering volume-defining measures. For example, focusing on cardiac catheterization
centres that perform radical cystectomy could reveal important variations in structural
and process variables that could then be widely adopted by all centres performing
cystectomy. With respect to surgeon volume, our data suggest that some of the effect of
volume is mediated at the hospital level. Much of what surgeon volume encompasses,
however, is still unclear.
94
At present, we can only speculate on the structures and processes potentially
underlying provider volume. High volume surgeons, for example, may perform more
oncologically sound resections, harvest more lymph nodes or make more adept
intraoperative decisions that affect long term cancer control. They may also provide more
vigilant follow-up and surveillance for recurrent disease and/or timely and appropriate
use of chemotherapy. Explanations for the impact of high volume hospitals are less clear
but could include improved pathways to detect recurrent disease and thus the timely
provision of treatment, superior access to chemotherapy and/or better treatment of
comorbid diseases. Ultimately, additional research is required to gain a stronger
understanding of the provider volume-overall survival relationship.
Our study is not without limitations. First, as mentioned above, our data preclude
specific recommendations regarding provider volume. We still do not know the important
variables responsible for volume. Second, our analysis was limited to variables
measurable using administrative data only. It will likely be necessary in future studies to
obtain detailed, clinical information to ultimately disentangle the reasons for the effect of
provider volume. This may be particularly true for surgeon volume since our
administratively derived variables were unable to negate the volume HR. Third, we did
not examine postoperative physician processes such as frequency of follow up or
appropriate use of diagnostic imaging to detect recurrence. Differences in these variables
could potentially mediate the effect of provider volume. This aspect of post-cystectomy
care, which was beyond the scope of this thesis, serves as an avenue for future clinical
research. Fourth, we could not control for the number of lymph nodes harvested during
radical cystectomy due to the large number of missing values (approximately 40% of
95
patients). Since the extent and number of lymph nodes harvested is a known predictor of
long term survival128,129
, differences in these factors could also explain quality of care
differences between provider volume quartiles. An ongoing knowledge translation
strategy at our institution aimed at increasing rates of lymphadenectomy during radical
cystectomy may allow us to revisit this process measure in the context of a volume-
outcome study in the near future.
96
CONCLUSIONS
An inverse association between cystectomy provider volume and overall survival in the
province of Ontario points to a gap in the quality of care of bladder cancer patients. Using
preoperative, intraoperative and hospital structures and processes of care to further define
this quality gap, we found that only hospital factors, and specifically the presence of on-
site cardiac catheterization capabilities, were able to attenuate the effects of both hospital
and surgeon volume. A combination of both intraoperative and hospital structure and
process measures led to the greatest hazard ratio attenuation for both surgeon and hospital
volume and actually accounted for the entire effect of hospital volume. Nevertheless,
many of the underlying variables responsible for provider volume remain unaccounted
for and their elucidation should serve as a focus for additional research.
97
FIGURES FOR CHAPTER 4
Figure 4.1: Effects of accounting for structure and process of care groups on the
hazard ratio of hospital volume.
Each group was initially assessed individually to determine the most influential set of
process/structure variables. Below the dashed line is the combination of factors that
resulted in the most substantial attenuation of the hospital volume hazard ratio.
0.985 0.99 0.995 1 1.005 1.01 1.015
Adjusted Hazard Ratio (95% CI)
Volume alone
Volume + Preop
Volume + Intraop
Volume + Hospital
Favours High
Volume
Hospitals
Favours Low
Volume
Hospitals
Volume + Intraop
+ Hospital
98
Figure 4.2: Effects of accounting for structure and process of care groups on the
hazard ratio of surgeon volume.
Each group was initially assessed individually to determine the most influential set of
process/structure variables. Below the dashed line is the combination of factors that
resulted in the most substantial attenuation of the surgeon volume hazard ratio.
0.96 0.97 0.98 0.99 1 1.01 1.02 1.03 1.04
Adjusted Hazard Ratio (95% CI)
Volume alone
Volume + Preop
Volume + Intraop
Volume + Hospital
Favours High
Volume
Surgeons
Favours Low
Volume
Surgeons
Volume + Intraop
+ Hospital
99
TABLES FOR CHAPTER 4
Table 4.1: List of candidate structures and processes of care variables assessed for
their ability to define provider “volume.”
Variable Definition Source*
Physician – level
Preoperative
Anesthesia Consult Presence of an out-patient anesthetic billing
code in the 6 months prior to cystectomy
OHIP (A015)
Medical Consult Presence of an out-patient medical (internal
medicine, respirology or cardiology) billing
code in the 6 months prior to cystectomy
OHIP (A605, A675,
A606, A601, A603,
A604, A135, A145,
A435, A136, A133,
A134, A138, A475,
A575, A476, A473,
A474, A471, 478,
Preoperative Imaging Presence of an (abdo and/or pelvic) MRI or CT
billing code in the 3 months prior to cystectomy
OHIP (X409, X410,
X126, X231, X232,
X233, X451, X455, X461, X465)
Physician – level
Intraoperative
Anesthetic specialization Provision of cystectomy anesthesia by a board-
certified anesthetist.
IPDB (Cystectomy
OHIP code with fee
suffix “C”: S484, S485,
S453, S440)
Urologist –experience
(years) Time, in years, between year of graduation and
year of cystectomy.
IPDB
Urologist – international
medical graduate
Medical graduate of a country outside of North
America excluding the United Kingdom,
Ireland, Australia or New Zealand.
IPDB
Urologist as surgical
assistant Presence of an assistant fee code billed by a
board-certified urologist. Surgical assist
assumed to be a resident at teaching institutions
unless billed by a urologist.
OHIP (Cystectomy
OHIP code with fee
suffix “B”: S484, S485,
S453, S440)
Continent diversion Presence of a billing code for a continent urinary
diversion.
OHIP (S440)
Hospital - level Cardiac Catheterization
availability
Presence of cardiac catheterization facilities at
the cystectomy institution during the year of
operation.
CCN
Regional Dialysis Centre Presence of a regional dialysis facility at the
cystectomy institution during the year of
operation.
Diabetes Atlas
Teaching status Teaching hospital classification of the institution
at which the cystectomy occurred.
ICES Source file
*OHIP – Ontario Health Insurance Plan
ICES – Institute for Clinical Evaluative Sciences
IPDB – ICES (internal) Physician Database
CCN – Cardiac Care Network of Ontario
100
Table 4.2: Preoperative, intraoperative and hospital structure and process of care
variables by hospital volume quartile from the full cohort.
Hospital volume increases with quartiles. Values listed are counts (percentages) or means
(standard deviations). P values reflect comparisons across quartiles. (n=2535)
Variable Hospital Volume* P value
Quartile 1
(n=639)
Quartile 2
(n=604)
Quartile 3
(n=598)
Quartile 4
(n=694)
Preoperative
Anesthesia Consult 296 (46.3%) 301 (49.8%) 258 (43.1%) 312 (45.0%) 0.118
Medical Consult 302 (47.3%) 353 (58.4%) 321 (53.7%) 374 (53.9%) 0.001
Preoperative Imaging 504 (78.9%) 504 (83.4%) 492 (82.3%) 548 (79.0%) 0.086
Intraoperative
Anesthetic
specialization‡
556 (90.9%) 564 (97.1%) 506 (98.6%) 652 (98.8%) <0.001
Urologist experience± 21.23 (10.94) 19.65 (9.07) 20.77 (8.00) 22.09 (9.33) <0.001
Urologist –
international medical
graduate#
107 (17.4%) 118 (20.2%) 32 (6.2%) 2 (0.3%) <0.001
Urologist as surgical
assistant† 274 (45.1%) 306 (54.2%) 163 (33.1%) 164 (23.6%) <0.001
Continent diversion 18 (2.9%) 40 (6.9%) 40 (7.7%) 203 (31.0%) <0.001
Hospital
Cardiac
Catheterization
availability
95 (14.9%) 110 (18.2%) 218 (36.5%) 694 (100.0%) <0.001
Regional Dialysis
Centre
208 (32.6%) 252 (41.7%) 412 (68.9%) 694 (100.0%) <0.001
Teaching status 46 (7.2%) 104 (17.2%) 217 (36.3%) 694 (100.0%) <0.001
Percentages may not add to 100 due to rounding.
*Quartile 1 refers to the lowest volume hospitals whereas quartile 4 is comprised of the
highest volume hospitals.
‡Evaluable in 2366 patients
± Evaluable in 2368 patients
#Evaluable in 2375 patients
†Evaluable in 2359 patients
101
Table 4.3: Preoperative, intraoperative and hospital structure and process of care
variables by surgeon volume quartile from the full cohort.
Surgeon volume increases with quartiles. Values listed are counts (percentages) or means
(standard deviations). P values reflect comparisons across quartiles. (n=2375)
Variable Surgeon Volume* P value
Quartile 1
(n=640)
Quartile 2
(n=560)
Quartile 3
(n=594)
Quartile 4
(n=581)
Preoperative Anesthesia Consult 288 (45.0%) 286 (51.1%) 290 (48.8%) 268 (46.1%) 0.151
Medical Consult 348 (54.4%) 316 (56.4%) 271 (45.6%) 369 (63.5%) <0.001
Preoperative Imaging 501 (78.3%) 452 (80.7%) 506 (85.2%) 469 (80.7%) 0.019
Intraoperative Anesthetic
specialization‡
582 (93.9%) 509 (95.9%) 556 (97.0%) 545 (98.9%) <0.001
Urologist experience± 24.52 (10.91) 18.72 (7.54) 20.13 (10.36) 20.12 (7.22) <0.001
Urologist –
international medical
graduate
161 (25.2%) 51 (9.1%) 47 (7.9%) 0 (0%) <0.001
Urologist as surgical
assistant† 312 (50.2%) 245 (48.3%) 203 (36.1%) 120 (23.2%) <0.001
Continent diversion 20 (3.1%) 34 (6.1%) 62 (10.4%) 185 (31.8%) <0.001
Hospital Cardiac
Catheterization
availability
103 (16.1%) 117 (20.9%) 334 (56.2%) 443 (76.3%) <0.001
Regional Dialysis
Centre
268 (41.9%) 272 (48.6%) 382 (64.3%) 581 (100.0%) <0.001
Teaching status 120 (18.8%) 66 (11.8%) 271 (45.6%) 490 (84.3%) <0.001
Percentages may not add to 100 due to rounding.
*Quartile 1 refers to the lowest volume surgeons whereas quartile 4 is comprised of the
highest volume surgeons.
‡Evaluable in 2275 patients
±Evaluable in 2368 patients
†Evaluable in 2208 patients
102
Table 4.4: Results of multi-collinearity assessment of all candidate structure/process
of care variables.
Variable Variance Inflation Factor (VIF)
General
Hospital volume 4.01
Surgeon Volume 2.87
Physician - preoperative
Anesthesia Consult 1.10
Medical Consult 1.10
Preoperative Imaging 1.04
Physician - intraoperative
Anesthetic specialization
1.04
Urologist –experience (years)
1.33
Urologist – international medical graduate 1.21
Urologist as surgical assistant
1.15
Continent diversion 1.23
Hospital
Cardiac Catheterization availability 2.13
Regional Dialysis Centre 1.74
Teaching status 3.10
103
Table 4.5: Effect of structure and process of care variables on the hospital volume
parameter estimates for overall mortality.
The effect of adding each process/structure variable individually on the beta coefficient
(and standard error) of hospital volume is presented. For example, in a model with
hospital volume adjusted for patient factors, the addition of preoperative anesthesia
consultation increased the beta coefficient by 4.35% (i.e. from -0.00529 to -0.00506). The
standard error for the hospital volume coefficient decreased by 3.42% and the p value for
hospital volume increased to 0.065. Anesthesia consult was then removed from the model
and the process was repeated for medical consultation, preoperative imaging, etc. P
values derived from Cox Proportional Hazards model after accounting for clustered data
at the hospital level. Hospital volume was modeled as a continuous variable.
Explanatory
Variable
Beta for
Hospital
Volume
Percentage
change from
baseline
SE for
Hospital
Volume
Percentage
change from
baseline
Hospital
Volume
P Value
Reference
(adjusted hospital
volume)
-0.00529 --- 0.00263 --- 0.044
Anesthesia Consult -0.00506 +4.35% 0.00254 -3.42% 0.065
Medical Consult -0.00548 -3.59% 0.00294 +11.79% 0.062
Preoperative
Imaging
-0.00535 -1.13% 0.00261 -0.76% 0.040
Anesthetic
specialization
-0.00517 +2.27% 0.00249 -5.32% 0.039
Urologist –
experience (years)
-0.00586 -10.78% 0.00246 -6.46% 0.017
Urologist –
international
medical graduate
-0.00585 -10.59% 0.00253 -3.80% 0.021
Urologist as
surgical assistant
-0.00584 -10.40% 0.00267 +1.52% 0.029
Continent diversion -0.00587 -10.96% 0.00274 +4.18% 0.032
Cardiac
Catheterization
availability
-0.00153 +71.088% 0.00272 +3.42% 0.575
Regional Dialysis
Centre
-0.00426 +19.47% 0.00302 +14.83% 0.159
Teaching status -0.00670 -26.65% 0.00319 +21.29% 0.036
104
Table 4.6: Effect of structure and process of care variables on the surgeon volume
parameter estimates for overall mortality.
The effect of adding each process/structure variable individually on the beta coefficient
(and standard error) of surgeon volume is presented. For example, in a model with
surgeon volume adjusted for patient factors, the addition of preoperative anesthesia
consultation increased the beta coefficient by 1.89% (i.e. from -0.01586 to -0.01556). The
standard error for the surgeon volume coefficient increased by 1.59% and the p value for
surgeon volume remained unchanged at 0.002. Anesthesia consult was then removed
from the model and the process was then repeated for medical consultation, preoperative
imaging, etc. P values derived from Cox Proportional Hazards model after accounting for
clustered data at the surgeon level. Surgeon volume was modeled as a continuous
variable.
Explanatory
Variable
Beta for
Surgeon
Volume
Percentage
change from
baseline
SE for
Surgeon
Volume
Percentage
change from
baseline
Surgeon
Volume
P Value
Reference
(adjusted surgeon
volume)
-0.01586 --- 0.00503 --- 0.002
Anesthesia Consult -0.01556 +1.89% 0.00511 +1.59% 0.002
Medical Consult -0.01662 -4.79% 0.00535 +6.36% 0.002
Preoperative
Imaging
-0.01580 +0.39% 0.00501 -0.40% 0.002
Anesthetic
specialization
-0.01733 -9.27% 0.00487 -3.18% <0.001
Urologist –
experience (years)
-0.01668 -5.17% 0.00506 +0.60% 0.001
Urologist –
international
medical graduate
-0.01546 +2.52% 0.00534 +6.16% 0.004
Urologist as
surgical assistant
-0.01436 +9.46% 0.00543 +7.95% 0.008
Continent diversion -0.01567 +1.20% 0.00553 +9.94% 0.005
Cardiac
Catheterization
availability
-0.00983 +38.02% 0.00627 +24.65% 0.117
Regional Dialysis
Centre
-0.01372 +13.49% 0.00636 +26.44% 0.031
Teaching status -0.01542 +2.77% 0.00666 +32.41% 0.021
105
CHAPTER 5 : THE EFFECT OF WAIT TIMES FOR CYSTECTOMY ON
OVERALL MORTALITY IN ONTARIO: A POPULATION-BASED STUDY
SUMMARY
INTRODUCTION: The impact of waiting for radical cystectomy is controversial. While
some studies have determined that extended wait times lead to tumour progression and
decreased survival, others have failed to corroborate these results. We used population-
level data incorporating tumour pathology variables and factors that influence
preoperative waiting to inform the debate.
METHODS: Patients undergoing cystectomy in Ontario, Canada, between 1992-2004
were identified via the Canadian Institute for Health Information Discharge Abstract
Database, a population-based administrative database of all inpatient hospital admissions.
Linkage with the Ontario Cancer Registry yielded cystectomy pathology reports for 2535
patients, which were then reviewed for tumour characteristics such as stage, grade, lymph
node and margin status, amongst others. Wait time was defined as the period between
cystectomy and antecedent transurethral bladder tumour resection (TURBT). Cox
proportional hazards modeling was performed to assess the impact of wait time on
overall survival. The model was adjusted for patient factors, tumour variables and for
factors that could influence preoperative waiting (consultation, staging investigations,
surgeon and hospital volume). The tumour stage-specific impact of waiting for
cystectomy was also assessed. Cubic splines Cox regression analysis was used to
determine a maximum wait time within which optimal care can be provided.
RESULTS: The median wait time for cystectomy was 50 days. On univariate analysis,
wait time was significantly associated with overall mortality (p=0.015). The significant
106
effect of wait time on mortality remained after adjusting for patient, tumour and wait time
factors (p=0.042). For each incremental increase in wait time by 50 days, the risk of long
term death increased by 5.1%. Assessing the impact of wait time by tumour stage
revealed that wait times increased the relative hazard of death more for low stage lesions
(an 11-25% increase for stages T1 and lower) compared to high stage tumours (a 3-4%
increase for stage T3 or higher). Plotting the hazard ratio for death by increasing wait
time using cubic splines regression revealed that the risk of death begins to increase after
40 days.
CONCLUSIONS: Treatment delay between TURBT and radical cystectomy results in
worse overall survival. The wait time effect was most influential on lower stage lesions,
suggesting that delays facilitate further tumour invasion and micrometastases. The ideal
maximum time from TURBT to cystectomy was found to be 40 days.
107
INTRODUCTION
The path to a bladder cancer diagnosis is composed of multiple steps, each with a
potential wait time (Figure 5.1). Patients undergoing radical cystectomy for muscle-
invasive or high risk superficial bladder cancer are subject to an additional wait time
between the transurethral resection (TUR) and the cystectomy. The length of this time
period is influenced by a number of factors, including the surgeon‟s wait list, the
availability of hospital resources and the need for preoperative interventions such as
medical and/or anesthetic consultation and staging investigations. Delays related to any
of these factors increase the wait time to definitive cystectomy.
Prolonged wait times between TUR and cystectomy are important from a quality
of care perspective. Waiting for cancer treatment is associated with psychological stress
and anxiety and can therefore impact upon a patient‟s perioperative mental health.130-132
Furthermore, given the aggressive biology of invasive bladder cancer, even short delays
could theoretically diminish a patient‟s chances of survival by enabling the tumour to
invade further or spread systemically. Minimization of the waiting times for cystectomy
could therefore improve quality of care delivery for cystectomy patients by improving
short and long term outcomes.
At present, controversy exists about the effect of preoperative waiting on long
term bladder cancer outcomes. A number of studies have demonstrated that delayed
therapy results in poorer cancer-specific66
and overall survival.65
Other investigators,
however, have been unable to demonstrate any statistically significant association
between wait times and outcomes.63,68,133
Inconsistencies in the literature may be due to
variations in patient cohort definitions, small sample sizes with insufficient power to
108
detect significant associations or varied inclusion of appropriate confounding variables
such as pathological, hospital or surgeon specific factors that may influence both wait
times and outcomes.61
Based on these short-comings and the lack of consensus on the
effect of waiting from TUR to cystectomy on survival in bladder cancer patients, we
sought to determine the effect of wait time on overall survival in the province of Ontario
using population-level data while adjusting for important pathological, patient and
provider level variables.
109
METHODS
Cohort Identification
After ethics approval from the Sunnybrook Health Sciences Centre and University
of Toronto institutional review boards, we retrospectively evaluated the effect of waiting
for radical cystectomy on overall survival in the province of Ontario between 1992 and
2004. Radical cystectomy patients were identified from the Canadian Institute for Health
Information Discharge Abstract Database (CIHI DAD) using Canadian Classification of
Diagnostic, Therapeutic and Surgical Procedures (CCP) and Canadian Classification of
Health Interventions (CCI) procedure codes (from 1992-2002 CCP: 69.51; from 2003-
2004 CCI: 1.PM.91 and 1.PM.92;). The CIHI DAD is a population-based database that
contains information on all inpatient hospital admissions in Ontario. In addition to
identifying cystectomy patients, the CIHI DAD in conjunction with the provincial
Registered Person‟s Database, provided demographic details for each cystectomy patient
including age, sex, comorbidity, urgency of admission, region of residence and vital
status. Comorbidity in the form of the Charlson Comorbidity Index, was derived based on
CIHI DAD International Classification of Diseases (ICD) diagnostic codes from each
patient‟s index admission and from any hospital admissions in the year prior to
cystectomy.85,102,103
Comorbid status was divided into 4 categories (Charlson 0, 1, 2 and
> 2) and classified as none, mild, moderate and severe, respectively.104
Because radical cystectomy can be performed for both bladder cancer and for
non-bladder malignancies, the latter as part of larger exenterative procedures for
colorectal, prostate or gynecological malignancies, we linked the CIHI data to the Ontario
Cancer Registry (OCR) to select only those cystectomy patients with a diagnosis of
110
bladder cancer. The OCR contains information on all incident cancers detected in the
province of Ontario with 97% capture of incident cases of bladder cancer.88
A total of
3296 patients undergoing cystectomy for bladder cancer were identified. Because of the
importance of pathological variables in assessing survival outcomes, we limited our
analysis to those individuals who had pathology reports available for review at OCR. Our
final cohort was thus composed of 2535 patients who represent 77% of all patients that
underwent cystectomy for bladder cancer in the province of Ontario between 1992 and
2004. The pathology reports of all 2535 patients were reviewed for important
pathological variables including pathologic stage, grade, margin and lymph node status
and the presence of lymphovascular invasion or perineural invasion. Pathologic staging
was based on the 2002 American Joint Committee on Cancer system.105
Wait Time Definition
The use of administrative data to define wait times for surgery has been
previously validated by others.134
We defined wait time as the time between antecedent
TUR/biopsy and cystectomy. We chose this definition because the decision to pursue
radical cystectomy usually occurs based on the TUR procedure and/or once pathology
information is available on the TUR specimen. Furthermore, the interval between TUR
and cystectomy has been used in prior reports investigating the effect of treatment delays
and cystectomy outcomes, thereby allowing us to compare our results with those in the
published literature.62,66,70
The antecedent TUR or biopsy for cystectomy patients was
identified with CCP/CCI procedure codes (TUR: CCP 69.0, 69.2, 69.29, 69.3; CCI
1.PM.87, 1.PM.59,; Biopsy: CCP 69.81, 69.82; CCI 1.PM.58) via the CIHI DAD or from
111
the Ontario Health Insurance Plan (OHIP) physician‟s billing database (OHIP TUR
codes: Z632, Z633, Z634; OHIP Biopsy codes: E776, E784).
Confounding Variable Definitions
In addition to patient and pathological factors, potential hospital- and surgeon-
related confounding variables included in the analyses were hospital and surgeon volume,
surgeon experience, use of adjuvant chemotherapy, preoperative medical and anesthetic
consultation and the use of abdominal/pelvic imaging (CT and/or MRI) within 3 months
preoperatively. Average annual hospital volume was calculated using CIHI DAD hospital
unique identifiers. We accounted for hospital mergers and amalgamations, which
occurred frequently during the study time period, using methodology based on the details
of each institutional merger/amalgamation as described in a prior report (Kulkarni et al,
submitted). Surgeon experience, derived from a local Physician‟s database, was defined
as the number of years the operating surgeon was in practice. The remaining confounders
were extracted from OHIP billing data. Average annual surgeon volume was calculated
based on surgeon-specific unique identifiers. Adjuvant chemotherapy was defined by the
initiation of chemotherapy in the first 6 months postoperatively. We chose a 6 month
time period because this allowed ample time for patient discharge, postoperative
followup, referral to medical oncology and initiation of chemotherapy. We did not
account for the use of neoadjuvant chemotherapy since this treatment was not widely
used during the study time period (<1% of patients). Anesthetic and medical consults
were defined by OHIP billing codes for either consultation in the 6 months prior to
cystectomy. A 6 month time period for preoperative consultation was used to capture
112
consults related to bladder cancer surgery including those that may have occurred prior to
diagnostic TUR. We hypothesized that “timely” preoperative abdominal/pelvic imaging
(within 3 months of surgery) would facilitate intraoperative planning and rule out
metastatic disease and thus lead to improved patient care compared to imaging that
occurred much earlier than the cystectomy date. Since waiting for imaging could also
influence the wait time to cystectomy, we incorporated imaging as a potential
confounder.
Statistical Analyses
All statistical analyses were performed using SAS version 9.1.3 (SAS Institute,
Cary, North Carolina). A two-sided p value of 0.05 was defined as statistically significant
for all analyses. For descriptive statistics, the data was dichotomized using a 90 day wait
time period. The choice of a 90 day (3 month) cut point was based on convention in
previously published reports.70
Continuous variables were assessed using the Kruskal
Wallis test and categorical variables were assessed via a Chi square test. Multivariable
Cox proportional hazards modeling was then performed to assess the effect of wait time,
modeled as a continuous variable, on overall mortality. To avoid survivor treatment bias
when adjusting for adjuvant chemotherapy, we modeled use of adjuvant chemotherapy as
a time-dependent covariate. Patients alive as of March 31, 2007, the last day of follow-
up, were censored. This ensured a minimal follow up of 2 years and a maximum potential
follow up of 15 years. Observations with identical follow up times (ties) were handled by
the method of Efron.111
113
Since risk adjustment using administrative datasets may not be fully accurate, we
performed a sensitivity analysis, reproducing the multivariable Cox proportional hazards
model using only the healthiest patients (Charlson comorbidity index score of 0 or 1, ) to
“level the playing field” and potentially eliminate unmeasured confounding.113
Due to the
multilevel nature of the data, with patients clustered within surgeons and hospitals, we
assessed the robustness of our results using marginal („variance-corrected‟) survival
models designed to account for non-independent observations.109,110
Evaluation of the
proportional hazards assumption was performed by incorporating wait time into the
model as a time dependent covariate (wait_time*survival_time). The effect of wait time
according to TNM tumour stage was assessed via a series of interaction terms (i.e.
wait_time*stage, wait_time*survival_time, wait_time*survival_time*stage in the model
together). Finally, to recommend a maximum wait time within which a patient should
undergo cystectomy, we created a Cox model with wait time input as a cubic spline
function using 5 knots as per Harrell.135
With typical regression models including Cox
models, an important assumption is that the independent variables are linearly related to
the outcome (i.e. Y = B0 + B1X1 + B2X2 + ….). Cubic spline functions incorporate
multiple cubic polynomial terms and thus relax the assumption that the predictor
variables are linear. By allowing wait time to assume non-linear forms, we were able to
assess a time point after which the hazard of death due to waiting began to increase.
114
RESULTS
Baseline Demographic and Univariate Analyses
Of 2535 patients, the wait time between TUR and cystectomy could be evaluated
in 2397 (94.6%). Patients were excluded (138 total) if they did not have a pre-cystectomy
TUR or biopsy in CIHI or a similar billing code in OHIP. The distribution of wait times
is depicted in Figure 5.2. The median (range) and mean (standard deviation (SD)) wait
times for the cohort were 50 days (range: 0-363) and 64.5 days (SD: 53.2), respectively.
Figure 5.3 illustrates the trend for wait times by year in Ontario. The wait for patients
increased from a median of 42 days in 1992 to 65.5 days in 2004, a median increase of
23.5 days. Although gradual, this trend of increased wait times primarily occurred from
1997 onwards.
Patient, pathologic and hospital/surgeon factors for the entire cohort are listed in
Table 5.1. Univariate statistical analyses demonstrated a number of statistically
significant differences between patients who waited < 90 days for cystectomy compared
to those waiting > 90 days. Specifically, those with the longest wait times were older with
more comorbid disease, less likely to receive adjuvant chemotherapy and had a higher
proportion of stage T1 disease with lower proportions of T2 and T3 cancer. There were
also geographic differences in patients‟ places of residence based on wait time. For
example, patients who waited longer tended to reside in the Toronto Central or Central
LHIN‟s whereas patients with shorter waits tended to reside in the Central East LHIN.
With respect to hospital and physician factors, patients with > 90 day waits were more
likely to be seen by high volume providers, receive preoperative consultations and were
115
less likely to receive timely staging imaging studies. A non-significant trend for males
was also observed in the longer wait time group.
Survival Analyses
A total of 1796 patients died during the study time period. The mean (SD) and
median (range) follow up for the cohort was 1260 days (SD: 1275) and 786 days (Range:
0-5441), respectively. The 5-year overall survival rate was 35%. Results of the Cox
proportional hazards models are provided in Table 5.2. Wait time was a significant
predictor of overall survival in both crude (unadjusted) and adjusted models (Crude
analysis: HR (95% CI) = 1.001 (1.000-1.002), p = 0.015; Adjusted analysis: HR (95%
CI) = 1.001 (1.000-1.002), p = 0.042). Accounting for clustering of patients within
surgeons or hospitals did not alter the results (HR (95% CI) = 1.001 (1.000-1.002), p =
0.045). The hazard ratio of 1.001 represents the increased hazard of death for each day a
patient waits for cystectomy. Thus, for an increase in wait time of 50 days, the increased
hazard of long term death would be 5.1% (e50*0.0010
– 1.000). Reproducing the above
analysis in patients with a Charlson score of 0 and 1 (n = 955) revealed an even stronger
association between wait time and overall mortality (Adjusted analysis: HR (95% CI) =
1.003 (1.001-1.005), p = <0.001), suggesting that inaccurate risk adjustment was not
responsible for the noted effect.
Testing for proportional hazards revealed that the final multivariate model in
Table 5.2 violated the assumptions of proportional hazards since the time dependent wait
time covariate was statistically significant (Table 5.3). Violation of this assumption,
however, does not invalidate the model in Table 5.2. Rather, the effect of wait time on
116
survival, including the calculations presented above, in the setting of non-proportional
hazards can be interpreted as a mean effect (mean hazard) as opposed to an instantaneous
hazard at any followup time point.112
However, the instantaneous hazard of death for a
given wait time can be calculated as presented in Table 5.4 for a 30 day preoperative wait
time. The hazard ratio for a 30 day increment in wait time increases with survival time
(i.e. it is time-dependent) and becomes significant at 2 years with a 4.1% increase in the
hazard of death. The hazard ratio continues to increase with survival time indicating that
the impact of waiting for cystectomy is manifest at later survival time points.
Tumour Stage-Wait Time Interactions
Modeling the impact of wait time by tumour stage on overall survival using a
stage by wait time interactions in the time-dependent model revealed a stage-specific
effect (Table 5.5; p values for all stage interactions <0.05). The time-dependent hazard
ratios increased for all stages of disease as survival time increased. The biggest impact of
wait time, measured by the relative increase in the hazard ratio with time, was on patients
with lower stage disease (stages T2 and lower; Figure 5.4).
Maximum Wait Time Recommendation
To suggest an optimal wait time within which cystectomy should be offered,
cubic splines Cox regression analysis was used to generate a plot of the hazard ratio for a
given wait time by wait time (Figure 5.5). Wait times between 0 and 40 days were
associated with an elevated hazard of death which may represent a triage effect.
Substantiating this claim, patients operated on within 40 days of TUR had an
117
urgent/emergent admission rate of 17.3% compared to 12.8% of those admitted after a 40
day wait time (p value 0.003 for comparison). At 40 days (red line) the hazard of death
gradually began to increase again. At 150 days (blue line), the hazard of death increased
at an exponential rate. These data suggest that the wait time maximum should ideally be
set at 40 days but should definitely not exceed 150 days.
118
DISCUSSION
Radical cystectomy is the current gold standard treatment for invasive bladder
cancer and is a reasonable option for high risk superficial disease.8,136
While the decision
to pursue radical cystectomy is unique to each patient, in the majority of cases the
pathology results of the antecedent transurethral resection (TUR) play a large role in
determining treatment. Delay between TUR and cystectomy can be influenced by many
factors, including patient and tumour factors, as well as hospital and physician factors.
Waiting for cancer care is anxiety-provoking for patients because of the fear of further
tumour invasion and spread. For bladder cancer, these fears may not be groundless. In
this study, we demonstrated that treatment delays due to waiting for cystectomy are
associated with adverse long-term outcomes. Furthermore, the effect of waiting for care
was most pronounced for patients diagnosed with lower stage lesions.
Our results substantiate those previously published. In a subset of 214 patients
with clinical T2 disease, Lee and colleagues demonstrated a statistically significant
improvement in overall survival and a strong trend towards improved disease-specific
survival (p = 0.08) for patients operated on within a 93 day TUR to cystectomy time
period.62
Sanchez-Ortiz et al. evaluated 189 patients with clinical T2-T4 disease using
multivariate Cox regression analysis correcting for stage and nodal status and reported
that time lags greater than 12 weeks were associated with a 93% increased risk of 3-year
overall mortality.66
They also found that 84% of patients waiting 12 weeks or more,
compared to 48% who waited less than 12 weeks, had advanced stage (T3 or greater)
disease, suggesting that the prolonged wait may have resulted in tumour progression.
Chang and investigators corroborated Sanchez-Ortiz‟s findings in 153 patients with
119
clinical T2 or greater disease, demonstrating a higher rate of T3 or worse disease in
patients who waited more than 90 days between TUR and cystectomy.70
While others
have also detected a deleterious effect associated with waiting for cystectomy65,67
, not all
studies have done so. Liedberg et al., for example, failed to demonstrate any impact of
waiting for care on disease-specific survival after cystectomy, acknowledging that a
sample size of 141 patients may have precluded detection of a significant association.133
Nielsen and colleagues, investigating 592 patients from 3 large U.S. medical centers, did
not find a significant association between waiting and outcome but it is difficult to
comment on their findings because they did not present detailed multivariate analysis
results.63
We sought to address some of the controversy surrounding delayed treatment and
outcome for bladder cancer by performing a population-based study evaluating the
impact of wait time on overall survival after radical cystectomy. To date, only one other
population-based study addressing this question has been published. Performed using the
province of Quebec‟s physician‟s billing database, Mahmud et al. reported on 1592
cystectomy patients and found a near-significant association (p=0.051) between a
surgical delay of 12 weeks or more and worse overall survival.65
Major limitations to this
study, however, were a lack of potential confounding variables such as comorbidity and
pathology. In contrast, our study has many advantages. First, it represents the largest
contemporary cystectomy series to study the effect of wait time on outcome. Since the
data are population-based, with excellent population coverage, the results are likely
generalizable to most urologic practices. Second, we reviewed 2535 pathology reports
and thus could adjust for pathology factors in addition to patient, hospital and physician
120
factors. Third, we were able to extend our analyses to determine the impact of wait time
in a stage-specific manner and to suggest a pragmatic maximum wait time based on the
hazard of death.
A number of findings in this study bear comment. The effect of wait time on
survival was time-dependent (non-proportional hazards) and not unexpected. As
presented in Table 5.4, the true effect of waiting on overall survival manifests between 18
and 24 months after cystectomy. From an oncologic perspective, this finding supports the
purported mechanism of an increased risk of micrometastatic disease with prolonged
waits for definitive therapy. Since micrometastatic cancer would likely take time to
present because of the time required for tumour cell growth and further systemic spread
leading to death, it is not surprising that the “disease-specific” effects of delayed therapy
appeared after 18 months. The greater impact of wait time on lower stage lesions also
makes biological sense. Patients with less invasive disease, many of whom do not have
microscopic tumour spread and are thus potentially curable with cystectomy, may be put
at risk for developing micrometastases with longer waiting periods. On the other hand,
since many patients with T3 or T4 disease likely already have some form of
micrometastases due to the aggressive nature of their disease, prolonged waiting may not
be as detrimental to their outcomes as it would be for patients with potentially curable
disease. The concept of preferentially triaging lower stage bladder cancer patients rather
than those with higher stage lesions is new in urologic oncology. Current
recommendations by experts in urology137
suggest that patients with higher stage disease
may benefit more from expeditious treatment compared to patients with low stage
disease. Our data contradict this advice and thus have important policy implications since
121
current practices may negatively impact quality of care. Finally, the sensitivity analysis
depicting a stronger association between wait time and overall survival in healthy patients
suggests that either inaccuracies in comorbidity measurement masked the effect of
waiting on outcome for the entire cohort and that the true effect was in fact stronger, or
that patients with the lowest risk for long-term mortality were most susceptible to the
detriments of delayed therapy (or both). Unfortunately, our data do not enable us to
differentiate between these two competing theories.
The time to cystectomy is an important quality of care issue for patients with
bladder cancer. Mounting evidence, including this study, supports the concept of
improved outcomes with shorter wait times. Our data suggest that a 40 day window
between TUR and cystectomy is an ideal maximum wait time. Beyond that period of time
survival, and thus delivery of quality care, suffers. Since two-thirds of delays between
TUR and cystectomy are generally attributable to physician scheduling and patient
decision-making62,66
, an opportunity exists to improve patient care via patient and
physician education of the implications of extended wait times. Policy interventions
aimed at expediting preoperative staging, consultation and scheduling could potentially
improve patient outcomes. Further research exploring these hypotheses is warranted.
Our study has limitations. First, it is retrospective in nature and prone to selection
biases. For example, of 3296 cystectomy patients, we could only assess 2535 because
pathology information was not available for 761 (23%) patients. The excluded patients
were younger and healthier with subsequent lower overall mortality rates compared to the
patients with pathology data (57.0% vs. 70.9% mortality, p<0.001). Nevertheless,
significant differences in wait time were not present between these two groups (mean
122
64.5 days for the pathology group, mean 64.2 days for the non-pathology group,
p=0.604), suggesting that exclusion of these patients may not have introduced significant
bias. Second, the 5-year overall survival rate of 35% is much lower than that reported in
the U.S. literature.96
The reason for this discrepancy may be because of later patient
presentation and thus worse disease at the time of cystectomy. A 21% rate of T4 disease
compared to published rates of 11-14% supports this claim.62,63,66,70,96
Third, due to
limitations in the data, we did not distinguish between patients undergoing cystectomy as
primary therapy versus those undergoing cystectomy as salvage therapy after primary
chemoradiation. Since patients undergoing salvage cystectomy were ultimately deemed
to be surgical candidates, however, we felt their inclusion was warranted because their
receipt of ineffective primary therapy may have extended their wait time and thus
diminished their chances of cure. Fourth, we did not have information on disease
recurrence or cause of death and thus could not comment on recurrence-free survival or
cause-specific survival. Although overall survival as an outcome measure may be
susceptible to unmeasured confounders, differences in which could explain our results
(i.e. surgeon hesitation to operate on sicker patients), a sensitivity analysis in healthy
patients, where comorbidity measurement error and selection bias were less likely,
supported our conclusions. Unmeasured confounders also do not explain the time-
dependent effect of wait time. Finally, our study may be subject to lead time bias.
Patients operated on earlier (shorter wait times) may have seemingly improved results not
because of the effect of cystectomy on survival but rather because of the longer survival
time afforded by an earlier operation. Evidence against this possibility is the described
time-dependent association of wait time with survival and the stage-specific effect of wait
123
time on survival. In the setting of lead time bias, the systematic error would be uniform
regardless of the survival time or the tumour stage.
124
CONCLUSIONS
Expeditious timing of tumour resection is a key tenet in surgical oncology. We
demonstrated that shorter wait times between TUR and cystectomy are significantly
associated with improved overall survival in patients undergoing radical cystectomy for
bladder cancer. The effect of waiting was most pronounced for patients with lower stage
disease. Our data suggest a wait time of 40 days would yield maximum benefit to
patients, and that the effect of waiting markedly increases after a wait time of 150 days.
Surgeon and patient awareness, in addition to health policy interventions, could
potentially facilitate reductions in wait time below this target level.
125
FIGURES FOR CHAPTER 5
Figure 5.1: Bladder cancer wait time intervals from symptom development to
definitive therapy.
The decision to undergo cystectomy usually occurs at the time of TURBT or shortly
thereafter, once pathology results become available. Patients not requiring or not offered
cystectomy after TURBT remain at risk for developing recurrent disease that may
eventually warrant radical cystectomy.
Abbreviations: GP = General Practitioner; TURBT: Transurethral Resection of Bladder
Tumour.
126
Figure 5.2: Histogram of wait times for radical cystectomy in Ontario, 1992-2004.
Each interval represents a 20 day time period.
Distribution of wait times for radical cystectomy in
Ontario, 1992-2004
0
5
10
15
20
25
30
0 40 80 120 160 200 240 280 320 360
Wait Time (days)
Pe
rce
nt
127
Figure 5.3: Histogram of median wait times for radical cystectomy in Ontario by
year, 1992-2004.
Histogram of median wait times for radical
cystectomy in Ontario by year
25
35
45
55
65
75
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
Year
Wa
it T
ime
(d
ay
s)
128
Figure 5.4: Relative increase in the hazard of death for a 30 day preoperative wait
by tumour stage.
Due to the time-dependent nature of the wait time and mortality association, relative
hazards depicted are for a 4 year survival period.
Relative increase in Hazard of Death by Stage
0
5
10
15
20
25
30
T0/Ta/Tis T1 T2 T3 T4
Tumor stage
Perc
en
tag
e
129
Figure 5.5: Effect of waiting for radical cystectomy on the hazard ratio for death
from any cause.
Hazard ratios derived from a fully adjusted, Cox Proportional Hazards model with cubic
splines. At 40 days (left line) the hazard of death begins to increase again. At 150 days
(right line), the hazard of death increases at an exponential rate. Thus the wait time
maximum could ideally be set at 40 days but should definitely not exceed 150 days.
Effect of wait time on the hazard ratio for death
0
2
4
6
8
10
0 20 40 60 80 100 120 140 160 180 200 220 240 260 280
Wait time (days)
Ha
za
rd R
ati
o
Triage effect
130
TABLES FOR CHAPTER 5
Table 5.1: Patient characteristics by wait time.
Patient characteristics divided into patient factors, pathologic factors and variables
potentially influencing wait times for surgery. Wait time has been dichotomized into ≤ 90
days and > 90 days.
Variable Wait time ≤ 90 days
Mean (SD) / N (%)
(n=1916)
Wait time > 90 days
Mean (SD) / N (%)
(n=481)
P value
PATIENT FACTORS
Age 67.44 (10.00) 69.21 (9.26) <0.001
Sex
Males
1540 (80.4%)
403 (83.8%)
0.088
Comorbidity*
None
Mild
Moderate Severe
619 (32.3%)
169 (8.8%)
429 (22.4%) 699 (36.5%)
124 (25.8%)
43 (8.9%)
131 (27.2%) 183 (38.1%)
0.024
Socioeconomic status**
Quintile 1
Quintile 2
Quintile 3
Quintile 4
Quintile 5
319 (16.7%)
412 (21.5%)
376 (19.6%)
355 (18.5%)
404 (21.1%)
118 (18.7%)
134 (21.6%)
119 (18.5%)
106 (16.6%)
133 (21.6%)
0.828
Admission status
Urgent/Emergent
274 (14.3%)
72 (15.0%)
0.709
Adjuvant chemotherapy 281 (14.7%) 44 (9.2%) 0.002
LHIN
1 (Erie St. Clair) 2 (South West)
3 (Waterloo Wellington)
4 (Hamilton Niagara
Haldimand Brant)
5 (Central West)
6 (Mississauga Halton)
7 (Toronto Central)
8 (Central)
9 (Central East)
10 (South East)
11 (Champlain) 12 (North Simcoe
Muskoka)
13 (North East)
14 (North West)
118 (6.2%) 176 (9.2%)
120 (6.3%)
280 (14.6%)
95 (5.0%)
99 (5.2%)
175 (9.1%)
186 (9.7%)
268 (14.0%)
55 (2.9%)
122 (6.4%) 87 (4.5%)
110 (5.7%)
24 (1.3%)
27 (5.6%) 41 (8.6%)
25 (5.2%)
63 (13.2%)
22 (4.6%)
28 (5.9%)
49 (10.2%)
61 (12.7%)
44 (9.2%)
26 (5.4%)
42 (8.8%) 13 (2.7%)
33 (6.9%)
5 (1.0%)
0.011
TUMOUR FACTORS
Tumour Stage
Tx
T0
Ta
Tis
T1
T2
T3
7 (0.4%)
35 (1.8%)
35 (1.8%)
96 (5.0%)
163 (8.5%)
506 (26.4%)
704 (36.7%)
1 (0.2%)
10 (2.1%)
15 (3.1%)
21 (4.4%)
65 (13.5%)
112 (23.3%)
147 (30.6%)
0.003
131
T4 370 (19.3%) 110 (22.9%)
Grade
Not specified
Grade 1
Grade 2
Grade 3
134 (7.0%)
32 (1.7%)
246 (12.9%)
1503 (78.5%)
34 (7.1%)
8 (1.7%)
66 (13.7%)
373 (77.6%)
0.965
Positive Margin Status 303 (15.8%) 80 (16.6%) 0.662
Lymphovascular invasion
(LVI)
762 (39.8%) 201 (41.8%) 0.420
Perineural invasion 314 (16.4%) 70 (14.6%) 0.322
Lymphadenectomy 1209 (63.2%) 298 (62.0%) 0.604 Positive Lymph node status
Nx
N0
N+
565 (29.5%)
916 (47.8%)
435 (22.7%)
155 (32.2%)
232 (48.2%)
94 (19.5%)
0.258
HOSPITAL AND PHYSICIAN FACTORS
Hospital Volume†
Quartile 1
Quartile 2
Quartile 3
Quartile 4
489 (25.5%)
490 (25.6%)
442 (23.1%)
495 (25.8%)
111 (23.1%)
94 (19.5%)
107 (22.3%)
169 (35.1%)
<0.001
Surgeon Volume#†
Quartile 1
Quartile 2
Quartile 3
Quartile 4
486 (26.7%)
448 (24.6%)
469 (25.8%)
418 (23.0%)
111 (25.1%)
90 (20.3%)
103 (23.3%)
139 (31.4%)
0.003
Surgeon experience (yrs)‡ 20.66 (9.45) 21.70 (9.50) 0.040
Anesthesia Consult 862 (45.0%) 254 (52.8%) 0.002
Medical Consult 980 (51.2%) 302 (62.8%) <0.001
Preoperative Imaging 1629 (85.0%) 314 (65.3%) <0.001
*Comorbidity scale based on Charlson scores: None = Charlson 0; Mild = Charlson 1;
Moderate = Charlson 2 and Severe = Charlson > 2.
**The highest quintile represents the highest socioeconomic status.
# Number of patients for whom surgeon volume quartiles were compared were: Wait
time < 90 days = 1821; Wait time > 90 days = 443.
†Increasing quartile represents increasing procedure volume.
‡Number of patients for whom surgeon experience was compared were: Wait time < 90
days = 1816; Wait time > 90 days = 441.
132
Table 5.2: Effect of Wait Time on Overall Mortality.
P values derived from Cox Proportional Hazards models for both crude (unadjusted) and
adjusted analyses. Wait time was modeled as a continuous variable. (n=2397 for the
crude analysis and 2187 for the adjusted analysis).
Variable Beta
coefficient
Standard
Error
Hazard
Ratio
95% C.I. P value
Crude (unadjusted)
Analyses
Wait Time 0.0011 0.0005 1.001 (1.000,1.002) 0.015
Adjusted Analyses
Wait Time 0.0010 0.0005 1.001 (1.000,1.002) 0.042
Hospital Volume -0.0010 0.0042 1.000 (0.992, 1.008) 0.983
Surgeon Volume -0.0186 0.0097 0.982 (0.963, 1.001) 0.057
Surgeon Experience -0.0035 0.0030 0997 (0.991, 1.002) 0.251
Anesthesia Consult 0.0899 0.0582 1.094 (0.976, 1.226) 0.122
Medical Consult 0.1325 0.0539 1.142 (1.027, 1.269) 0.014
Preoperative
Imaging
0.0425 0.0682 1.043 (0.913, 1.193) 0.533
Age (per yr) 0.0233 0.0030 1.024 (1.018, 1.030) <0.001
Gender -0.1203 0.0671 0.887 (0.777, 1.011) 0.073
Comorbidity†
None (ref)
Mild
Moderate
Severe
---
0.0849
0.1803
0.3871
---
0.1007
0.0751
0.0725
---
1.089
1.198
1.473
---
(0.894, 1.326)
(1.034, 1.388)
(1.278, 1.697)
---
0.399
0.016
<0.001
Admission Status 0.1411 0.0685 1.152 (1.007, 1.317) 0.039
Socioeconomic
Status Quintile
1
2
3
4
5 (ref)
0.1794
0.1021
-0.0204
0.0146
---
0.0815
0.0766
0.0808
0.0821
---
1.197
1.107
0.980
1.015
---
(1.020, 1.404)
(0.953, 1.287)
(0.836, 1.148)
(0.864, 1.192)
---
0.028
0.182
0.801
0.859
---
Tumour Stage
T0, Ta, Tis (ref)
T1
T2
T3
T4
---
0.3070
0.4259
0.8807
1.0293
---
0.1432
0.1258
0.1248
0.1305
---
1.359
1.531
2.413
2.799
---
(1.027, 1.800)
(1.197, 1.959)
(1.889, 3.081)
(2.167, 3.615)
---
0.032
<0.001
<0.001
<0.001
Margin 0.4736 0.0697 1.606 (1.401, 1.841) <0.001
Nodal Status
N0 (ref)
N+
Nx
---
0.2820
0.1961
---
0.0785
0.0958
---
1.326
1.217
---
(1.137, 1.546)
(1.008, 1.468)
---
<0.001
0.041
Lymphadenectomy -0.0939 0.0885 0.910 (0.765, 1.083) 0.288
Adjuvant Chemo -0.1733 0.0845 0.841 (0.713, 0.992) 0.040
133
LVI 0.5058 0.0588 1.658 (1.478, 1.861) <0.001
PNI -0.0040 0.0701 0.996 (0.868, 1.143) 0.954
Tumour Grade
1 (ref)
2
3
X (missing/T0)
---
0.2452
0.3163
0.3735
---
0.2505
0.2425
0.2653
---
1.278
1.372
1.453
---
(0.782, 2.088)
(0.853, 2.207)
(0.864, 2.444)
---
0.328
0.192
0.159
Year
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004 (ref)
0.2649
0.2116
0.1215
0.0366
0.2654
0.0543
0.1902
0.2124
0.1944
0.2018
-0.0859
-0.0523
---
0.1795
0.1707
0.1824
0.1716
0.1716
0.1696
0.1657
0.1620
0.1621
0.1641
0.1582
0.1589
---
1.303
1.236
1.129
1.037
1.304
1.056
1.210
1.237
1.215
1.224
0.918
0.949
---
(0.917, 1.853)
(0.884, 1.727)
(0.790, 1.614)
(0.741, 1.452)
(0.931, 1.825)
(0.757, 1.472)
(0.874, 1.674)
(0.900, 1.699)
(0.884, 1.669)
(0.887, 1.688)
(0.673, 1.251)
(0.695, 1.296)
---
0.140
0.215
0.505
0.831
0.122
0.749
0.251
0.190
0.230
0.219
0.587
0.742
---
Local Health
Integration Network
(LHIN)
1
2
3
4
5
6
7
8
9
10
11
12
13
14 (ref)
-0.3146
-0.4462
-0.1596
-0.2552
-0.3242
-0.2514
-0.3580
-0.4190
-0.3540
-0.4176
-0.3955
-0.3166
-0.1495
---
0.2298
0.2251
0.2252
0.2151
0.2334
0.2312
0.2198
0.2182
0.2147
0.2915
0.2271
0.2382
0.2287
---
0.730
0.640
0.852
0.775
0.723
0.778
0.699
0.658
0.702
0.659
0.673
0.729
0.861
---
(0.465, 1.145)
(0.412, 0.995)
(0.548, 1.325)
(0.508, 1.181)
(0.458, 1.143)
(0.494, 1.224)
(0.454, 1.076)
(0.429, 1.009)
(0.461, 1.069)
(0.372, 1.166)
(0.431, 1.051)
(0.457, 1.162)
(0.550, 1.348)
---
0.171
0.047
0.478
0.235
0.165
0.277
0.103
0.055
0.099
0.152
0.082
0.184
0.513
---
134
Table 5.3: Time-dependent effects of wait time on overall mortality.
Results presented were derived from a fully adjusted Cox Proportional Hazards model
using the variables listed in Table 2.
Variable Beta
coefficient
Standard
Error
Hazard
Ratio
95% C.I. P value
Wait time*Time 2.12 x 10-6
6.50 x 10-7
1.000 (1.000, 1.000) 0.001
Wait time -0.00022 0.00065 1.000 (0.999, 1.001) 0.734
135
Table 5.4: Hazard ratios for death and corresponding P values for a 30 day increase
in preoperative wait time for cystectomy.
Illustrated is the time-dependent effect of waiting for cystectomy. For every 30 days a
patient has to wait for cystectomy, their hazard for death from all-causes increases during
the post-operative period. At 2 years (24 months) the hazard of death for a 30 day wait
increases by a statistically significant 4.1%. A 30 day (1 month) increment in wait time
was selected as a clinically relevant and pragmatic time frame.
Survival Time Hazard Ratio P value
3 months 0.999 0.968
6 months 1.005 0.772
12 months 1.017 0.297
18 months 1.029 0.062
24 months 1.041 0.009
36 months 1.065 <0.001
48 months 1.090 <0.001
136
Table 5.5: Hazard ratios by Tumour stage and survival time.
A wait time increment of 30 days was used for the analysis. Hazard ratios were derived
from a fully adjusted Cox Proportional Hazards model with a stage by wait time
interaction. Absolute and Relative increases in hazard ratio depict the Tumour stages
where wait time is most influential. The lowest stage category (T0/Ta/Tis) was referent in
the Cox model.
Survival Time Stage T0,
Ta, Tis
Stage T1 Stage T2 Stage T3 Stage T4
3 months 0.797 0.937 1.221 1.945 2.401
6 months 0.809 0.943 1.229 1.948 2.407
12 months 0.833 0.956 1.244 1.954 2.420
18 months 0.859 0.970 1.260 1.959 2.432
24 months 0.885 0.983 1.276 1.965 2.444
36 months 0.939 1.011 1.309 1.976 2.469
48 months 0.997 1.040 1.343 1.988 2.494
Absolute
increase in
hazard ratio
20.1%
10.3%
12.2%
4.3%
9.3%
Relative increase
in hazard ratio
25.2%
11.0%
10.0%
2.2%
3.9%
137
CHAPTER 6 : DISCUSSION AND CONCLUSIONS
THESIS SUMMARY
Chapters 3, 4 and 5 of this thesis describe three studies pertaining to the quality of
care for cystectomy patients in the province of Ontario. All three studies were based on a
cohort of patients accrued between 1992 and 2004 using various administrative
databases.
The first study assessed the impact of hospital and surgeon volume on operative
mortality and overall survival. Although neither hospital nor surgeon volume were
significantly associated with operative mortality, both had a statistically significant
association with overall survival. Patients treated by high volume hospitals and high
volume surgeons tended to have better long-term survival rates. The impact of high
volume surgeons on overall survival was three times higher than that for high volume
hospitals. Thus, potentially modifiable gaps in the quality of care provided to cystectomy
patients were identified.
In the second study, we tried to further understand why patients treated by
high volume providers experienced improved long term outcomes compared to low
volume providers. By incorporating a number of structures and processes of care
measured with administrative databases into our analyses, we were able to attenuate the
hazard ratio associated with hospital volume completely, implying that hospital volume is
a surrogate for underlying structure and process measures. The set of variables that
attenuated the hazard ratio of hospital volume most were, not surprisingly, hospital
structural factors, with the presence of on-site cardiac catheterization being the most
influential variable. A combination of intraoperative and hospital level structural and
138
process variables, however, were responsible for the greatest hospital volume HR
attenuation. These same factors also caused attenuation and loss of significance of the
surgeon volume hazard ratio albeit to a much lesser extent than for hospital volume.
Although we were able to attenuate the impact of surgeon volume and completely
account for the impact of hospital volume on overall survival, this study was unable to
specifically identify structures and processes that could be widely adopted to address the
provider volume quality gap.
In the third study, we shifted our focus away from volume-outcome associations
and identified another potential quality of care concern for cystectomy patients in
Ontario. The impact of waiting for cystectomy, from the time of transurethral resection or
biopsy, on overall survival was evaluated. Longer wait times were found to be negatively
associated with overall survival outcomes. While this finding was true across all stages of
disease, there existed a stage-specific interaction whereby waiting for care was most
detrimental for patients with low stage disease. Upon calculating the instantaneous hazard
of death in the postoperative period based on the delay to surgery, an ideal wait time
maximum of 40 days was recommended.
IMPLICATIONS AND RECOMMENDATIONS
Clinical
Direct application of the research in Chapters 3 and 4 may be difficult. Although
regionalization of care is one potential approach to dealing with an established volume-
outcome relationship, it has only gained traction in a few settings138
(see Health Policy
section below for discussion regarding regionalization of cystectomy care). A focus on
139
the underlying structures and processes of care mediating provider volume is an
appealing second option. Unfortunately, these factors remain elusive, particularly in the
context of long term mortality outcomes.
Given the existence of a quality of care differential across providers, what is a
urologist to do? High cystectomy volume urologists or urologists working in high
cystectomy volume centres need not modify their clinical practice since they generally
experience acceptable outcomes. Urologists working in high cystectomy volume centres
generally experience good outcomes as well, but uniform outcomes across surgeons are
unlikely. A process whereby urologists in high volume centres audit their outcomes to
identify under-performing surgeons would facilitate improved quality of care. These
poorer-performing surgeons, along with low volume surgeons in low volume centres
have impetus to act. To better serve their patients until the factors underlying volume
become clearer, low volume surgeons at low volume centres or under-performing
surgeons at high volume hospitals should adhere to published best practices for radical
cystectomy. Standardization of cystectomy performance99
and actionable intraoperative
techniques known to improve survival, such as extended lymph node dissection139
,
should be adhered to. Adopting evidenced-based perioperative protocols140
also has the
potential to benefit cystectomy patients. Following these recommendations, low volume
providers would be, at the very least, using the best available evidenced-based medicine
for their patients. Furthermore, by implementing these guidelines poor performing
urologists may inadvertently apply the as of yet unidentified structure/process variable(s)
responsible for volume and improve the quality of care of their patients.
140
Application of the thesis results pertaining to wait times is more practical because,
in theory, it is easily actionable. Waiting for care is not a proxy or surrogate measure as is
provider volume. Consequently, physicians can directly modify this quality indicator for
each patient. In light of our findings, urologists should strive to operate earlier on
cystectomy patients. Once the decision to proceed with cystectomy is made, urologists
should triage their bladder cancer patients, giving higher priority to those with low stage
disease than is the current practice. These patients, in particular, should not automatically
be placed at end of the surgical cue. A scarcity of operating room (OR) time may make
this recommendation difficult to implement since unfilled OR slots are uncommon. A
potential solution, made at the hospital level with administrative buy-in, could involve
open urological OR days to help facilitate patient triage. Maneuvers such as these could
potentially improve bladder cancer long term outcomes.
Methodological
The methodological challenges of this thesis relate primarily to the structure of
the data. With patients clustered within surgeons and surgeons clustered within hospitals,
our data did not conform to the assumption of independent observations implicit in
traditional regression analyses. We accounted for the hierarchical nature of the data (3
levels: hospital, surgeon, patient) using the statistical program MLwiN v2.02 to perform
random effects logistic regression analysis for operative outcomes in Chapter 3.141
However, creating and running multilevel Cox proportional hazards survival models
(frailty models) to investigate overall survival outcomes in Chapters 3, 4 and 5 was
problematic in MLwiN because of frequent program termination errors and generation of
141
implausible results. Creating accelerated failure time models using MLwiN, a parametric
means of analyzing time-to-event data, was equally fruitless since these models failed to
converge. A search of available software revealed that no programs, other than MLwiN,
claimed to be able to run 3-level survival models. Clearly, multilevel survival analysis is
in its infancy.
To take into account clustering in our Cox survival models we used SAS v9.1.3
statistical software using the COVS(AGGREGATE) option in the PROC PHREG
command to generate variance-corrected estimates.109
This approach enabled us to
account for clustering at 2 levels (e.g. hospital and patient or surgeon and patient).
Although this method did not allow us to account for all 3 levels, it did enable us to
account for some degree of clustering in the data which is more methodologically sound
than running traditional survival models. Failure to account for clustering at all levels of
analysis, however, may have decreased the standard errors around our model parameter
estimates and thus decreased the corresponding p values (Type I error). Until reliable,
commercial software capable of modeling multilevel survival analysis is developed, we
cannot overcome this limitation.
In addition to being hierarchical, our data are also cross-classified, with lower
level units associated with multiple higher level units. In other words, surgeons
sometimes operated in more than one hospital. Specifically, of 199 urologists who
performed cystectomy during the study time period, 40 (20.1%) operated in more than
one hospital. Attempting to run cross-classified logistic regression models in MLwiN
resulted in repeated non-convergence. We could not run cross-classified survival models
due to the limitations in currently available software, as discussed above. Consequently,
142
we were not able to take into account the cross-classified nature of the data. As a result,
the p values listed in Chapters 3, 4 and 5 may in fact be too low. Unfortunately, this
methodological concern can only be addressed with refinement and development of
statistical software capable of running multi-level, cross-classified data.
Health Policy
Volume, Structure and Process of Care
Broad policy recommendations based on the data presented in Chapters 3 and 4 of
this thesis may be premature. At present, there is ample evidence supporting gaps in
quality of care across cystectomy provider volume thresholds but data are limited
regarding the cause of these inequities. Nevertheless, regionalization of health care
services has emerged as a potential health policy application of volume-outcomes
research. Proponents of regionalized care cite studies suggesting the life-saving benefits
of specialized care centres.142,143
Not all studies, however, support an over-arching health
policy recommendation of regionalized care. Hollenbeck and colleagues, for example,
presented evidence of regionalization of radical cystectomy care in the United States
between 1988 and 2000 in the absence of legislation directing cystectomy
regionalization.144
Despite regionalization of cystectomy care, volume-outcome
associations persisted during the same time period51
implying that a policy measure
requiring selective referral to high volume centres may not ameliorate differences in
quality of care.
Concentrating resources in “centres of excellence” may not be practical,
particularly for large sparsely populated areas such as some regions in Ontario. At the
143
system level, regionalization is a complex undertaking that would likely be resource-
intensive in the short term, place logistical burdens inherent to restructuring care on the
health care system and place additional strain on resources at high volume centres.122
At
the physician level, regionalization may increase the wait lists of high volume surgeons,
particularly if resources are not re-allocated to account for the expected influx of patients.
Urologists may also hesitate to refer to high volume colleagues because of the risk of
losing technical proficiency for cystectomy and/or the risk of loss of patient referrals
from general practitioners. Finally at the patient level, regionalization may potentially
impact on patient quality of life since it would force many individuals to travel longer
distances to receive care.120,145
The burden of travel to specialized centres may also
introduce the risk of less vigilant follow up for cystectomy patients, a potential process of
care important for long term survival.
Based on these arguments, a shift in policy to regionalized care may not be
appealing. Understanding what “volume” actually means and applying this knowledge
across the province would avoid the trials and tribulations of regionalized bladder cancer
care. Assuming low volume providers were willing to improve their processes of care, it
would not matter if patients received their care at high or low volume centres, an
appealing concept at the system, physician and patient level. Unfortunately, our research
has not revealed enough regarding the underlying factors that explain “volume,” which is
acting as a proxy for quality care. From a health policy point of view, funding initiatives
to further research aimed at understanding processes and structures of care important to
bladder cancer patients may help address this knowledge deficit (see Future Studies
section below for details regarding potential future research). Determining the structure
144
and process measures important for cystectomy patients may also prove beneficial to
other surgical procedures where volume-outcome associations have been noted146,147
since factors governing quality of care may overlap and be common to different surgical
procedures.
Wait times
Increasingly, wait times are seen as an indicator of the quality of the health care
patients are receiving. Unlike “volume of care” and the debate regarding volume and
quality of care, waiting for care is not an abstract concept to patients. Each individual
patient has experienced some form of waiting for care. Delays to visit general
practitioners, obtain specialist referral, complete specialist work-ups and undergo
definitive therapy (e.g. cystectomy) are just a few examples. In light of patient
experiences with many parts of the Canadian health care system, waiting for medical care
has become an important health issue in Canada over the past few years.148
Waiting for care is an important issue for cystectomy patients. Most invasive
bladder cancers are aggressive, fast-growing lesions and consequently, as demonstrated
in Chapter 5, each day of waiting beyond 40 days entails an increased risk of long term
death. Waiting for cancer therapy also heightens patient anxiety. Fear of metastatic
spread and concern about harbouring a malignant tumour during the wait time period can
negatively impact patient mental well-being.130-132
Nevertheless, wait times for radical
cystectomy increased between 1992 and 2004, rising from a median wait of 42 days to 66
days, respectively. A number of reasons for this rise can be postulated: 1) population
growth and demographic shifts towards an older population led to an increasing number
145
of patients developing and thus requiring treatment for bladder cancer.1; 2) broader
criteria for performing cystectomy149,150
and the advent of more cosmetically appealing
urinary diversions151
may have led to more liberal use of radical cystectomy; 3) health
care reform and limitations on the growth of health care spending implemented during
the 1990‟s152
may have decreased resources such as OR and surgical bed availability and;
4) a shortage of urologists in Ontario, particularly those performing cystectomy.153
A number of policy interventions have recently been introduced to decrease
waiting times for cystectomy. Addressing the potential future shortage of urologists, the
number of urology residency training positions in Canada has doubled from 15 in 2002 to
30 in 2007.154
From the healthcare expenditure point of view, beginning in 2004/2005 a
substantial infusion of funding at both the federal155
and provincial156
levels has occurred
in an attempt to decrease wait times for many procedures and tests. In Ontario
specifically, these funds have been directed to priority health care services with
unacceptably long wait times including cancer surgery (and thus radical cystectomy).157
Recently implemented policy and funding initiatives have had some success.
Since September 2005, Ontario cancer patients have experienced a 22% decrease in wait
time.157
For genitourinary cancers (excluding prostate cancer), median (mean) wait times
in the 3rd
quarter of 2007 were 28 days (39 days). The wait time within which 90% of
these patients received treatment was 73 days, which fell below the provincial target of
84 days. While an 84 day target may be reasonable for certain urological malignancies137
,
our data suggest that a 12 week wait for cystectomy for bladder cancer is too long.
Unfortunately, data broken down by tumour type are unavailable at present and it is
therefore difficult to determine recent mean and median wait times for bladder cancer
146
patients. Nevertheless, because of the overall drop in wait times, it is likely that bladder
cancer patients received improved quality of care as a result of policy interventions aimed
at improving wait time.
In Chapter 5, we identified a potential maximum cutpoint of 40 days within which
cystectomy should occur. Our recommendation is similar to that published by the surgical
wait times initiative in urologic oncology which advocated performance of cystectomy
within 28 days of the decision to proceed with surgery.137
With these recommendations in
place, additional information in the form of contemporary bladder cancer-specific wait
times data is needed to confirm we are meeting or at least moving towards these targets
for patients with bladder cancer. Given the beneficial impact of governmental funding
and strategies directed at reducing waits, an ultimate goal of performing cystectomy
within 40 days for 90% of patients is probably realistic. Meeting this goal would improve
upon the quality of care provided to patients undergoing radical cystectomy in Ontario.
THESIS LIMITATIONS
A number of general limitations must be acknowledged regarding this thesis:
Risk adjustment – many of the variables used for risk adjustment were derived from
administrative databases held at ICES. These data are known to have limitations in
the context of risk adjustment.158
Specifically, the Charlson Comorbidity Index, as
derived from administrative data, is known to underestimate patient comorbidity
compared to chart-abstracted clinical data and thus may have provided less robust risk
adjustment. We attempted to address this concern by repeating our analyses, where
applicable, in patients with very low Charlson scores. This methodology, advocated
147
by experts in observational research113
, theoretically attenuates the measurement error
associated with comorbid status assignment by restricting analyses to patients with
minimal comorbid disease. Reassuringly, these sensitivity analyses did not alter our
results.
Cohort identification – the codes used to identify the cohort assembled in this study
have never been formally validated. However, as pointed out in Chapter 2, plenty of
direct (pathology report abstraction) and indirect (prior validation studies of
procedure codes in CIHI) evidence supports our use of CIHI and OCR codes to
identify bladder cancer cystectomy patients.
Outcomes – only mortality was assessed in this thesis because of its relative ease of
measurement with administrative data and because of its relevance to patients and
physicians. Although the conceptual framework for this study (Figure A1, Appendix
A) lists 4 additional outcomes that can be used to measure quality of care, these
measures were not assessed because they are not easily measured using
administrative data (e.g. quality of life) or are not as relevant an outcome as death
(e.g. hospital length of stay).
Power – since only 3296 patients underwent radical cystectomy during the study time
period it is possible that analyses involving short term (operative) mortality outcomes,
where an event rate of 126/3296 (3.8%) was observed, may have been subject to a
type II statistical error. In Chapter 3, the p value for the hospital volume-operative
mortality outcome analysis was 0.074 (OR 0.98, 95% CI: 0.95-1.00) which supports
the premise of a lack of statistical power to detect a significant association.
Subsequent analyses with additional patients may help clarify this issue.
148
Structure and process of care variables – the structure and process variables assessed
in Chapter 4 were limited to those entities measurable using ICES administrative
data. Specific clinical data were not available for this study. As a result, many
variables such as the use of perioperative antibiotics, thrombosis prophylaxis and
perioperative beta-blocker use in high-risk patients, which affect postoperative
outcomes in surgical patients, were not assessed.159-164
Future studies using chart-
abstracted clinical data are required to clarify whether these variables explain
provider volume.
Missing pathology data – 761 of the 3296 cystectomy cases did not have available
pathology. Although reasons for these unavailable reports are unclear, selective
omission of reports may have introduced bias into our analyses. For example, patients
with missing reports tended to have lower mortality rates (both operative and overall
with the latter statistically significant) than those who had their reports sent to OCR
(Chapter 2, Table 3). Despite the association between missing reports and outcome, it
is unlikely that the cause of the missing pathology reports was directly related to
outcome for two major reasons. First, a policy to collect pathology reports
preferentially from patients who died was not in place at OCR. Second, the mortality
difference between patients with and without available data can be explained by the
fact that missing reports tended to come from lower volume hospitals which tended to
treat patients with less comorbidity.
149
FUTURE STUDIES
Volume, Structures and Processes of Care
A key study that should follow this thesis is a focused attempt to understand the
factors that explain the provider-volume relationship. In Ontario, the relationship between
hospital and surgeon cystectomy volume and long term associations requires
clarification. Possible variables to consider include postoperative structures and processes
of care such as appropriateness of follow up, tumour recurrence testing and
chemotherapy administration. Of course, definitions of appropriateness for each of these
measures will have to be created prior to embarking on such a study. To maximize the
probability of identifying key structural and process variables, this type of study will
likely require primary chart abstracted data in conjunction with administrative data.
Primary data collection will enable more accurate risk-adjustment, allow for extensive
recording of perioperative events and pathology details and facilitate collection of
information regarding follow up and imaging. Patients who move or switch hospitals
could be tracked using administrative data along with chart review data. At the
population level, this type of study will be large, costly and time-consuming. Its
feasibility, however, should not be in question since other groups have successfully
embarked upon quality of care initiatives on grander scales in both Canada25
and the
United States.165,166
Another potential study includes a follow up investigation of the study proposed
above where, after identification of important volume-related processes/structures, a
knowledge translation endeavour aimed at implementing these measures at low volume
centres would be pursued. A reassessment of the volume-overall survival outcome
150
association could then be performed to determine whether the knowledge translation
strategies closed the provider volume quality gap. In the event that important structure or
process of care variables are not identified and regionalization of cystectomy care is
implemented, reassessment of the volume-overall survival outcome relationship would be
useful to determine the effectiveness of regionalization.
Wait times
A number of future studies extending the work in Chapter 5 are possible.
Although we only evaluated the time between TUR/biopsy and cystectomy as our
definition of wait time, a number of other time intervals could be evaluated. For example,
the time from onset of patient symptoms to GP referral, the time from GP referral to
surgical referral, the time from surgical referral to the TUR/biopsy triggering a decision
for cystectomy and the total time from onset of symptoms to cystectomy are all important
wait times. All of these scenarios could be assessed using the same methodology as in
Chapter 5. An ancillary study to Chapter 5 could also describe significant patient and
provider factors associated with long wait times. Information on the predictors of waiting
could then be used to identify at-risk populations and direct resources at those with long
wait times.
Since our cohort was accrued just prior to federal and provincial funding
initiatives aimed at reducing wait times, a logical research priority would be to assess
whether the funding injections succeeded in reducing cystectomy wait times and, as a
corollary, overall mortality. Reproducing our wait time study for cystectomy patients
from 2005 onwards would help address the effectiveness of these policy interventions.
151
CONCLUSIONS
The focus of this thesis was the quality of care provided to radical cystectomy
patients in the province of Ontario. Two major quality themes were addressed: A) volume
of care and its impact on outcome and B) waiting for care and its impact on outcome. We
determined that cystectomy patients treated by high volume hospitals and surgeons had
improved long term, but not short term, outcomes compared to low volume providers.
We also discovered that cystectomy patients who had a longer preoperative wait had
worse outcomes compared to those operated on expeditiously. Although both of these
indicators suggest potential substandard care for many cystectomy patients in Ontario,
research and funding aimed at ameliorating these deficiencies have the potential to
ultimately improve care for these patients.
152
APPENDIX A
153
Figure A.1: Thesis conceptual framework.
Figure A.2: Conceptual framework for objective 1.
154
Figure A.3: Conceptual framework for objective 2.
155
Figure A.4: Conceptual framework for objective 3.
156
APPENDIX B
157
Table B.1: Patient level variable definitions.
Variable Definition Source*
Age Patient age at time of cystectomy in years. RPDB
Sex Patient gender. RPDB
Charlson
Comorbidity Index
score
Patient comorbidity at time of cystectomy
calculated based on a 1-year look-back
period from the date of cystectomy.
CIHI
Socioeconomic
status
Patient socioeconomic status, divided into
quintiles, based on neighbourhood income
from the 1996 (1992-1998 patients) and
2001 (1999-2004 patients) census.
1996 and 2001
Census
Admission status
(Urgent/Emergent)
Proportion of patients admitted with an
urgent or emergent status code in the
CIHI-DAD. Remainder are elective
admissions.
CIHI
LHIN Patient‟s Local Health Integration
Network (LHIN) of residence.
CIHI
Chemotherapy –
Adjuvant
Proportion of patients who received
adjuvant chemotherapy, defined as the
initiation of 3 or more chemotherapy
billing codes (minimum 3 cycles of
chemotherapy) in the 6 months following
cystectomy.
OHIP (G381, G281,
G339, G345, G382)
Post-operative
mortality
Post-cystectomy death prior to discharge
or within 30 days of operation.
RPDB
Overall mortality Death after cystectomy regardless of
cause.
RPDB
*RPDB – Registered Person‟s DataBase
CIHI – Canadian Institute of Health Information discharge abstract database
OHIP – Ontario Health Insurance Plan
158
Table B.2: Pathology variable definitions.
Pathology information abstracted directly from Ontario Cancer Registry pathology
reports at Cancer Care Ontario
Variable Definition
Tumour Stage
Tx
T0
Ta
Tis
T1
T2
T3
T4
Local tumour pathologic stage as per the 2002
TNM bladder cancer staging system:
Tx – Unable to stage
T0 – No tumour in specimen
Ta – Tumour confined to mucosa
Tis – Carcinoma-in-situ only
T1 – Invasion into the lamina propria
T2 – Invasion into the muscularis propria
T3 – Invasion into the perivesical fat
T4 – Extravesical invasion
Grade
Not specified
Grade 1
Grade 2
Grade 3
Tumour grade as per the WHO 1973 histologic
grading system. Increasing grade signifies worse
differentiation.
Positive Margin Status Percentage of cases with local margins involved
with tumour
Lymphovascular invasion (LVI) Percentage of cases in which tumour
cells/emboli are found within lymphatics and/or
capillaries
Perineural invasion (PNI) Percentage of cases in which tumour is invading
or surrounding vesicle nerve tissue
Lymphadenectomy performed Percentage of cases in which separate lymph
node packages were submitted for pathological
examination
Extent of nodal dissection
Conventional
Above bifurcation of iliac vessels
(Extended)
In cases where a lymphadenectomy was
performed, proportion where the upper limit of
dissection was the bifurcation of the common
iliac vessels (conventional) versus dissections
caudal to the bifurcation (extended)
Lymph node count
Mean number of lymph nodes in cases where an
exact lymph node count was provided
Positive lymph node status
Nx
N0
N+
Percentage of cases where lymph node
metastases were noted.
Nx – Not provided/evaluable
N0 – Negative for lymph node metastases
N+ - Positive for lymph node metastases
159
Table B.3: Physician level variable definitions.
Variable Definition Source*
General
Surgeon Volume Average annual cystectomy caseload for
each year a surgeon is in practice.
OHIP (S484, S485,
S453, S440)
Wait time Time, in days, between cystectomy and
antecedent biopsy/TUR.
CIHI (ICD-9: 69.0,
69.2, 69.29, 69.3,
69.81, 69.82; ICD-10:
1.PM.87, 1.PM.59,
1.PM.58) OHIP (Z632, Z633, Z634,
E776, E784)
Preoperative
Anesthesia Consult Presence of an out-patient anesthetic
billing code in the 6 months prior to
cystectomy
OHIP (A015)
Medical Consult Presence of an out-patient medical
(internal medicine, respirology or
cardiology) billing code in the 6 months
prior to cystectomy
OHIP (A605, A675,
A606, A601, A603,
A604, A135, A145, A435, A136, A133,
A134, A138, A475,
A575, A476, A473,
A474, A471, 478,
Preoperative Imaging Presence of an (abdo and/or pelvic) MRI
or CT billing code in the 3 months prior
to cystectomy
OHIP (X409, X410,
X126, X231, X232,
X233, X451, X455,
X461, X465)
Intraoperative
Anesthetic
specialization
Provision of cystectomy anesthesia by a
board-certified anesthetist.
IPDB (Cystectomy
OHIP code with fee
suffix “C”: S484, S485, S453, S440)
Urologist –experience
(years)
Time, in years, between year of
graduation and year of cystectomy.
IPDB
Urologist –
international medical
graduate
Medical graduate from a country outside
of Canada, the United State of America,
the United Kingdom, Ireland, Australia
or New Zealand.
IPDB
Urologist as surgical
assistant
Presence of an assistant fee code billed
by a board-certified urologist. Surgical
assist assumed to be a resident at
teaching institutions unless billed by a
urologist.
OHIP (Cystectomy
OHIP code with fee
suffix “B”: S484, S485,
S453, S440)
Continent diversion Presence of a billing code for a
continent urinary diversion.
OHIP (S440)
*IPDB – ICES Physician‟s DataBase
CIHI – Canadian Institute of Health Information discharge abstract database
OHIP – Ontario Health Insurance Plan
160
Table B.4: Hospital level variable definitions.
Variable Definition Source*
Hospital volume Average annual cystectomy caseload
for each year a hospital is providing
acute care services.
CIHI
Cardiac
Catheterization
availability
Presence of cardiac catheterization
facilities at the cystectomy institution
during the year of operation.
CCN
Regional Dialysis
Centre
Presence of a regional dialysis facility
at the cystectomy institution during
the year of operation.
Diabetes Atlas
Teaching status Teaching hospital classification of the
institution at which the cystectomy
occurred.
ICES Source file
*MIS – Management Information System
CIHI – Canadian Institute of Health Information discharge abstract database
CCN – Cardiac Care Network of Ontario
161
REFERENCES
1. Canadian Cancer Society/National Cancer Institute of Canada: Canadian Cancer
Statistics 2007, Accessed December 17, 2007 (URL: http://www.ncic.cancer.ca).
2. Sheridan W: Bladder cancer and Lack of awareness, Accessed September 1, 2005
(URL: http://blcwebcafe.org.myaccess.library.utoronto.ca/awarefacts.asp).
3. Humphrey PA: Urinary bladder pathology 2004: an update. Ann Diagn Pathol 2004; 8:
380.
4. Walsh PC, Retik AB and Darracott Vaughan Jr, E. et al: Campbell's Urology 8th
Ed.
Philadelphia: WB Saunders 2002; 3954.
5. Amling CL: Diagnosis and management of superficial bladder cancer. Curr Probl
Cancer 2001; 25: 219.
6. Raghavan D, Quinn D, Skinner DG and Stein JP: Surgery and adjunctive
chemotherapy for invasive bladder cancer. Surg Oncol 2002; 11: 55.
7. Bhojwani AG and Mellon JK: Contemporary cystectomy combined with ileal conduit
or bladder substitution. Surg Oncol 2002; 11: 65.
8. Cookson MS: The surgical management of muscle invasive bladder cancer: a
contemporary review. Semin Radiat Oncol 2005; 15: 10.
9. Skinner DG, Daniels JR, Russell CA, Lieskovsky G, Boyd SD, Nichols P et al: The
role of adjuvant chemotherapy following cystectomy for invasive bladder cancer: a
prospective comparative trial. J Urol 1991; 145: 459.
10. Stockle M, Meyenburg W, Wellek S, Voges GE, Rossmann M, Gertenbach U et al:
Adjuvant polychemotherapy of nonorgan-confined bladder cancer after radical
cystectomy revisited: long-term results of a controlled prospective study and further
clinical experience. J Urol 1995; 153: 47.
11. Studer UE, Bacchi M, Biedermann C, Jaeger P, Kraft R, Mazzucchelli L et al:
Adjuvant cisplatin chemotherapy following cystectomy for bladder cancer: results of
a prospective randomized trial. J Urol 1994; 152: 81.
12. Freiha F, Reese J and Torti FM: A randomized trial of radical cystectomy versus
radical cystectomy plus cisplatin, vinblastine and methotrexate chemotherapy for
muscle invasive bladder cancer. J Urol 1996; 155: 495.
13. Chester JD, Hall GD, Forster M and Protheroe AS: Systemic chemotherapy for
patients with bladder cancer--current controversies and future directions. Cancer
Treat Rev 2004; 30: 343.
162
14. Grossman HB, Natale RB, Tangen CM, Speights VO, Vogelzang NJ, Trump DL et
al: Neoadjuvant chemotherapy plus cystectomy compared with cystectomy alone for
locally advanced bladder cancer. N Engl J Med 2003; 349: 859.
15. Advanced Bladder Cancer Meta-analysis Collaboration.: Neoadjuvant chemotherapy
in invasive bladder cancer: a systematic review and meta-analysis. Lancet 2003; 361:
1927.
16. David KA, Milowsky MI, Ritchey J, Carroll PR and Nanus DM: Low incidence of
perioperative chemotherapy for stage III bladder cancer 1998 to 2003: a report from
the National Cancer Data Base. J Urol 2007; 178: 451.
17. Sternberg CN and Collette L: What has been learned from meta-analyses of
neoadjuvant and adjuvant chemotherapy in bladder cancer? BJU Int 2006; 98: 487.
18. Knap MM, Lundbeck F and Overgaard J: Early and late treatment-related morbidity
following radical cystectomy. Scand J Urol Nephrol 2004; 38: 153.
19. Rosario DJ, Becker M and Anderson JB: The changing pattern of mortality and
morbidity from radical cystectomy. BJU Int 2000; 85: 427.
20. Chassin MR and Galvin RW: The urgent need to improve health care quality.
Institute of Medicine National Roundtable on Health Care Quality. JAMA 1998;
280: 1000.
21. Donabedian A: Evaluating the quality of medical care. Milbank Mem Fund Q 1966;
44: Suppl:166.
22. Perrin EB: Some thoughts on outcomes research, quality improvement, and
performance measurement. Med Care 2002; 40: III89.
23. Birkmeyer JD, Dimick JB and Birkmeyer NJ: Measuring the quality of surgical care:
structure, process, or outcomes? J Am Coll Surg 2004; 198: 626.
24. Hannan EL, Popp AJ, Feustel P, Halm E, Bernardini G, Waldman J et al: Association
of surgical specialty and processes of care with patient outcomes for carotid
endarterectomy. Stroke 2001; 32: 2890.
25. Institute for Clinical Evaluative Sciences: Canadian Cardiovascular Outcomes
Research Team, Accessed October 12, 2005 (URL: http://www.ccort.ca/).
26. Birkmeyer JD, Siewers AE, Finlayson EV, Stukel TA, Lucas FL, Batista I et al:
Hospital volume and surgical mortality in the United States. N Engl J Med 2002;
346: 1128.
27. Birkmeyer JD, Stukel TA, Siewers AE, Goodney PP, Wennberg DE and Lucas FL:
Surgeon volume and operative mortality in the United States. N Engl J Med 2003;
349: 2117.
163
28. Khuri SF and Henderson WG: The case against volume as a measure of quality of
surgical care. World J Surg 2005; 29: 1222.
29. Khuri SF: Quality, advocacy, healthcare policy, and the surgeon. Ann Thorac Surg
2002; 74: 641.
30. Hassan A, Chiasson M, Buth K and Hirsch G: Women have worse long-term
outcomes after coronary artery bypass grafting than men. Can J Cardiol 2005; 21:
757.
31. Hassan A, Quan H, Newman A, Ghali WA, Hirsch GM and Canadian Cardiovascular
Outcomes Research Team.: Outcomes after aortic and mitral valve replacement
surgery in Canada: 1994/95 to 1999/2000. Can J Cardiol 2004; 20: 155.
32. Guru V, Fremes SE and Tu JV: Time-related mortality for women after coronary
artery bypass graft surgery: a population-based study. J Thorac Cardiovasc Surg
2004; 127: 1158.
33. Ghali WA, Quan H, Shrive FM, Hirsch GM and Canadian Cardiovascular Outcomes
Research Team.: Outcomes after coronary artery bypass graft surgery in Canada:
1992/93 to 2000/01. Can J Cardiol 2003; 19: 774.
34. Eisenberg MJ, Filion KB, Azoulay A, Brox AC, Haider S and Pilote L: Outcomes and
cost of coronary artery bypass graft surgery in the United States and Canada. Arch
Intern Med 2005; 165: 1506.
35. Goodney PP, Siewers AE, Stukel TA, Lucas FL, Wennberg DE and Birkmeyer JD: Is
surgery getting safer? National trends in operative mortality. J Am Coll Surg 2002;
195: 219.
36. Goodney PP, Stukel TA, Lucas FL, Finlayson EV and Birkmeyer JD: Hospital
volume, length of stay, and readmission rates in high-risk surgery. Ann Surg 2003;
238: 161.
37. Lee KL, Freiha F, Presti JC,Jr and Gill HS: Gender differences in radical cystectomy:
complications and blood loss. Urology 2004; 63: 1095.
38. Riley GF, Potosky AL, Lubitz JD and Kessler LG: Medicare payments from
diagnosis to death for elderly cancer patients by stage at diagnosis. Med Care 1995;
33: 828.
39. Ministry of Health and Long Term Care, Ontario Hospital Insurance Plan (OHIP):
Schedule of Benefits – 2005, Accessed October 15, 2005 (URL:
http://www.health.gov.on.ca/english/providers/program/ohip/sob/physserv/physserv_
mn.html).
40. Porter MP, Wei JT and Penson DF: Quality of life issues in bladder cancer patients
following cystectomy and urinary diversion. Urol Clin North Am 2005; 32: 207.
164
41. Joudi FN and Konety BR: The impact of provider volume on outcomes from
urological cancer therapy. J Urol 2005; 174: 432.
42. Nuttall M, van der Meulen J, Phillips N, Sharpin C, Gillatt D, McIntosh G et al: A
systematic review and critique of the literature relating hospital or surgeon volume to
health outcomes for 3 urological cancer procedures. J Urol 2004; 172: 2145.
43. Black PC, Brown GA and Dinney CP: Should cystectomy only be performed at high-
volume hospitals by high-volume surgeons? Curr Opin Urol 2006; 16: 344.
44. Singh R and O'Brien TS: The relationship between volume and outcome in urological
surgery. BJU Int 2004; 93: 931.
45. Konety BR, Dhawan V, Allareddy V and Joslyn SA: Impact of hospital and surgeon
volume on in-hospital mortality from radical cystectomy: data from the health care
utilization project. J Urol 2005; 173: 1695.
46. McCabe JE, Jibawi A and Javle PM: Radical cystectomy: defining the threshold for a
surgeon to achieve optimum outcomes. Postgrad Med J 2007; 83: 556.
47. Finlayson EV, Goodney PP and Birkmeyer JD: Hospital volume and operative
mortality in cancer surgery: a national study. Arch Surg 2003; 138: 721.
48. Begg CB, Cramer LD, Hoskins WJ and Brennan MF: Impact of hospital volume on
operative mortality for major cancer surgery. JAMA 1998; 280: 1747.
49. Elting LS, Pettaway C, Bekele BN, Grossman HB, Cooksley C, Avritscher EB et al:
Correlation between annual volume of cystectomy, professional staffing, and
outcomes: a statewide, population-based study. Cancer 2005; 104: 975.
50. McCabe JE, Jibawi A and Javle P: Defining the minimum hospital case-load to
achieve optimum outcomes in radical cystectomy. BJU Int 2005; 96: 806.
51. Konety BR, Allareddy V, Modak S and Smith B: Mortality after major surgery for
urologic cancers in specialized urology hospitals: are they any better? J Clin Oncol
2006; 24: 2006.
52. Hollenbeck BK, Dunn RL, Miller DC, Daignault S, Taub DA and Wei JT: Volume-
based referral for cancer surgery: informing the debate. J Clin Oncol 2007; 25: 91.
53. Hollenbeck BK, Daignault S, Dunn RL, Gilbert S, Weizer AZ and Miller DC: Getting
under the hood of the volume-outcome relationship for radical cystectomy. J Urol
2007; 177: 2095.
54. Hollenbeck BK, Hong J, Zaojun Y and Birkmeyer JD: Misclassification of hospital
volume with surveillance, epidemiology, and end results Medicare data. Surg Innov
2007; 14: 192.
165
55. Hollenbeck BK, Wei Y and Birkmeyer JD: Volume, process of care, and operative
mortality for cystectomy for bladder cancer. Urology 2007; 69: 871.
56. Barbieri CE, Lee B, Cookson MS, Bingham J, Clark PE, Smith JA,Jr et al:
Association of procedure volume with radical cystectomy outcomes in a nationwide
database. J Urol 2007; 178: 1418.
57. Birkmeyer JD, Sun Y, Wong SL and Stukel TA: Hospital volume and late survival
after cancer surgery. Ann Surg 2007; 245: 777.
58. Urbach DR, Stukel TA, Croxford R and MacCallum NL: Analysis of Current
Research Related to the Impact of Low-Volume Procedures/Surgery and Care on
Outcomes of Care. Canadian Institute for Health Information, 2005.
59. Eggertson L: Wait Time Alliance first to set benchmarks. CMAJ 2005; 172: 1277.
60. Shroyer AL, London MJ, Sethi GK, Marshall G, Grover FL and Hammermeister KE:
Relationships between patient-related risk factors, processes, structures, and
outcomes of cardiac surgical care. Conceptual models. Med Care 1995; 33: OS26.
61. Fahmy NM, Mahmud S and Aprikian AG: Delay in the surgical treatment of bladder
cancer and survival: systematic review of the literature. Eur Urol 2006; 50: 1176.
62. Lee CT, Madii R, Daignault S, Dunn RL, Zhang Y, Montie JE et al: Cystectomy
delay more than 3 months from initial bladder cancer diagnosis results in decreased
disease specific and overall survival. J Urol 2006; 175: 1262.
63. Nielsen ME, Palapattu GS, Karakiewicz PI, Lotan Y, Bastian PJ, Lerner SP et al: A
delay in radical cystectomy of >3 months is not associated with a worse clinical
outcome. BJU Int 2007; 100: 1015.
64. Hautmann RE and Paiss T: Does the option of the ileal neobladder stimulate patient
and physician decision toward earlier cystectomy? J Urol 1998; 159: 1845.
65. Mahmud SM, Fong B, Fahmy N, Tanguay S and Aprikian AG: Effect of preoperative
delay on survival in patients with bladder cancer undergoing cystectomy in Quebec:
a population based study. J Urol 2006; 175: 78.
66. Sanchez-Ortiz RF, Huang WC, Mick R, Van Arsdalen KN, Wein AJ and Malkowicz
SB: An interval longer than 12 weeks between the diagnosis of muscle invasion and
cystectomy is associated with worse outcome in bladder carcinoma. J Urol 2003;
169: 110.
67. May M, Nitzke T, Helke C, Vogler H and Hoschke B: Significance of the time period
between diagnosis of muscle invasion and radical cystectomy with regard to the
prognosis of transitional cell carcinoma of the urothelium in the bladder. Scand J
Urol Nephrol 2004; 38: 231.
166
68. Hara I, Miyake H, Hara S, Gotoh A, Okada H, Arakawa S et al: Optimal timing of
radical cystectomy for patients with invasive transitional cell carcinoma of the
bladder. Jpn J Clin Oncol 2002; 32: 14.
69. Liedberg F, Anderson H and Mansson W: Treatment delay and prognosis in invasive
bladder cancer. J Urol 2005; 174: 1777.
70. Chang SS, Hassan JM, Cookson MS, Wells N and Smith JA,Jr: Delaying radical
cystectomy for muscle invasive bladder cancer results in worse pathological stage. J
Urol 2003; 170: 1085.
71. Baxter NN and Urbach DR: Misclassification of hospital volume in surgical volume-
outcome analyses of persons aged 65 years or older: potential limitations of
Medicare data. Evidence-based Surgery 2003; 1: 1.
72. Overview of the Nationwide Inpatient Sample (NIS). Healthcare Cost and Utilization
Project (HCUP) June 2005, Accessed September 15, 2005 (URL: www.hcup-
us.ahrq.gov/nisoverview.jsp).
73. Overview of the SEER program. National Cancer Institute. Accessed September 15,
2005 (URL: http://seer.cancer.gov/about/).
74. University HealthSystem Consortium (UHC) Overview: The power of collaboration,
November 2007. Accessed December 20, 2007 (URL:
http://public.uhc.edu/UHCMail/About/UHCOverviewPublicVersion.pdf).
75. Urbach DR, Croxford R, MacCallum NL and Stukel TA: How are volume-outcome
associations related to models of health care funding and delivery? A comparison of
the United States and Canada. World J Surg 2005; 29: 1230.
76. Tu JV, Austin PC and Chan BT: Relationship between annual volume of patients
treated by admitting physician and mortality after acute myocardial infarction.
JAMA 2001; 285: 3116.
77. Welke KF, Barnett MJ, Sarrazin MS and Rosenthal GE: Limitations of hospital
volume as a measure of quality of care for coronary artery bypass graft surgery. Ann
Thorac Surg 2005; 80: 2114.
78. Fong Y, Gonen M, Rubin D, Radzyner M and Brennan MF: Long-term survival is
superior after resection for cancer in high-volume centers. Ann Surg 2005; 242: 540.
79. Durairaj L, Torner JC, Chrischilles EA, Vaughan Sarrazin MS, Yankey J and
Rosenthal GE: Hospital volume-outcome relationships among medical admissions to
ICUs. Chest 2005; 128: 1682.
80. Hannan EL, Wu C, Walford G, King SB,3rd, Holmes DR,Jr, Ambrose JA et al:
Volume-outcome relationships for percutaneous coronary interventions in the stent
era. Circulation 2005; 112: 1171.
167
81. Glance LG, Dick AW, Osler TM and Mukamel DB: The relation between surgeon
volume and outcome following off-pump vs on-pump coronary artery bypass graft
surgery. Chest 2005; 128: 829.
82. Curry WT, McDermott MW, Carter BS and Barker FG,2nd: Craniotomy for
meningioma in the United States between 1988 and 2000: decreasing rate of
mortality and the effect of provider caseload. J Neurosurg 2005; 102: 977.
83. Hawker GA, Coyte PC, Wright JG, Paul JE and Bombardier C: Accuracy of
administrative data for assessing outcomes after knee replacement surgery. J Clin
Epidemiol 1997; 50: 265.
84. Institute for Clinical Evaluative Sciences. Appendix - A summary of studies on the
quality of Health Care Administrative Databases in Canada. ICES Practice Atlas
1996; 339.
85. Quan H, Parsons GA and Ghali WA: Validity of procedure codes in International
Classification of Diseases, 9th revision, clinical modification administrative data.
Med Care 2004; 42: 801.
86. Scales DC, Guan J, Martin CM and Redelmeier DA: Administrative data accurately
identified intensive care unit admissions in Ontario. J Clin Epidemiol 2006; 59: 802.
87. Robles SC, Marrett LD, Clarke EA and Risch HA: An application of capture-
recapture methods to the estimation of completeness of cancer registration. J Clin
Epidemiol 1988; 41: 495.
88. Cancer Care Ontario: Bladder ICD9 (188), 2003. Accessed September 15, 2005
(URL: http://www.cancercare.on.ca/index_statisticsBladder.htm).
89. Cardiac Care Network of Ontario: Cardiac Centres, 2005. Accessed September 15,
2005 (URL: http://www.ccn.on.ca/).
90. Oliver MJ, Lok CE, Shi J and Rothwell DM: Chapter 8 - Dialysis therapy for persons
with diabetes. Diabetes in Ontario – An ICES Practice Atlas, 2003; 8.165.
91. Hall S, Schulze K, Groome P, Mackillop W and Holowaty E: Using cancer registry
data for survival studies: the example of the Ontario Cancer Registry. J Clin
Epidemiol 2006; 59: 67.
92. Richards J, Brown A and Homan C: The data quality study of the Canadian discharge
abstract database. Proceedings of Statistics Canada Symposium 2001 (Achieving
data quality in a statistical agency: a methodological perspective).
93. Flack VF, Afifi AA and Lachenbruch PA: Sample Size Determinations for the Two
Rater Kappa Statistic. Psychometrika 1988; 53: 321.
168
94. Viera AJ and Garrett JM: Understanding interobserver agreement: the kappa statistic.
Fam Med 2005; 37: 360.
95. Ghoneim MA, el-Mekresh MM, el-Baz MA, el-Attar IA and Ashamallah A: Radical
cystectomy for carcinoma of the bladder: critical evaluation of the results in 1,026
cases. J Urol 1997; 158: 393.
96. Stein JP, Lieskovsky G, Cote R, Groshen S, Feng AC, Boyd S et al: Radical
cystectomy in the treatment of invasive bladder cancer: long-term results in 1,054
patients. J Clin Oncol 2001; 19: 666.
97. Uebersax J: Intraclass correlation and related methods, April 2007. Accessed
December 15, 2007 (URL:
http://ourworld.compuserve.com/homepages/jsuebersax/icc.htm).
98. Birkmeyer JD, Finlayson EV and Birkmeyer CM: Volume standards for high-risk
surgical procedures: potential benefits of the Leapfrog initiative. Surgery 2001; 130:
415.
99. Herr H, Lee C, Chang S, Lerner S and Bladder Cancer Collaborative Group.:
Standardization of radical cystectomy and pelvic lymph node dissection for bladder
cancer: a collaborative group report. J Urol 2004; 171: 1823.
100. Herr HW, Faulkner JR, Grossman HB, Natale RB, deVere White R, Sarosdy MF et
al: Surgical factors influence bladder cancer outcomes: a cooperative group report. J
Clin Oncol 2004; 22: 2781.
101. Leissner J, Ghoneim MA, Abol-Enein H, Thuroff JW, Franzaring L, Fisch M et al:
Extended radical lymphadenectomy in patients with urothelial bladder cancer: results
of a prospective multicenter study. J Urol 2004; 171: 139.
102. Preen DB, Holman CD, Spilsbury K, Semmens JB and Brameld KJ: Length of
comorbidity lookback period affected regression model performance of
administrative health data. J Clin Epidemiol 2006; 59: 940.
103. Nuttall M, van der Meulen J and Emberton M: Charlson scores based on ICD-10
administrative data were valid in assessing comorbidity in patients undergoing
urological cancer surgery. J Clin Epidemiol 2006; 59: 265.
104. Charlson ME, Pompei P, Ales KL and MacKenzie CR: A new method of classifying
prognostic comorbidity in longitudinal studies: development and validation. J
Chronic Dis 1987; 40: 373.
105. American Joint Committee on Cancer: AJCC Cancer Staging Manual, 6th edition,
2002. Accessed November 21, 2007 (URL:
http://www.cancerstaging.org/education/tnmschema/tnmschema.html#8).
169
106. Austin PC, Goel V and van Walraven C: An introduction to multilevel regression
models. Can J Public Health 2001; 92: 150.
107. Christensen LA: Introduction to building a linear regression model. Proceedings of
the Twenty-Second Annual SAS Users Group International Conference - SUGI 22
1997.
108. Hosmer DW and Lemeshow S: Applied Logistic Regression. New York: J Wiley
2000.
109. Ying G and Liu C: Statistical analysis of clustered data using SAS system, 2006.
Accessed October 20, 2007 (URL:
http://www.nesug.org/proceedings/nesug06/an/da01.pdf).
110. Chuang SK, Tian L, Wei LJ and Dodson TB: Kaplan-Meier analysis of dental
implant survival: a strategy for estimating survival with clustered observations. J
Dent Res 2001; 80: 2016.
111. Smith T and Smith B: Survival analysis and the application of cox's proportional
hazards modeling using SAS. SAS Users Group International – SUGI 26. Accessed
October 20, 2007 (URL: http://www2.sas.com/proceedings/sugi26/p244-26.pdf).
112. Allison P: Survival analysis using SAS: A practical guide. Cary: SAS Press 1995;
304.
113. Naylor CD and Guyatt GH: Users' guides to the medical literature. X. How to use an
article reporting variations in the outcomes of health services. The Evidence-Based
Medicine Working Group. JAMA 1996; 275: 554.
114. Konety BR, Dhawan V, Allareddy V and Joslyn SA: Impact of hospital and surgeon
volume on in-hospital mortality from radical cystectomy: data from the health care
utilization project. J Urol 2005; 173: 1695.
115. Fooks C and Decter M: The transformation experience of the Veterans Health
Administration and its relevance to Canada. Healthc Pap 2005; 5: 60.
116. Khuri SF and Henderson WG: The case against volume as a measure of quality of
surgical care. World J Surg 2005; 29: 1222.
117. Snijders TAB: Power and sample size in multilevel modeling. Encycolpedia of
Statistics in Behavioral Sciences, Volume 3. Chicester: Wiley 2005; 1570.
118. Elting LS, Pettaway C, Bekele BN, Grossman HB, Cooksley C, Avritscher EB et al:
Correlation between annual volume of cystectomy, professional staffing, and
outcomes: a statewide, population-based study. Cancer 2005; 104: 975.
119. Shahian DM and Normand SL: The volume-outcome relationship: from Luft to
Leapfrog. Ann Thorac Surg 2003; 75: 1048.
170
120. Finlayson SR, Birkmeyer JD, Tosteson AN and Nease RF,Jr: Patient preferences for
location of care: implications for regionalization. Med Care 1999; 37: 204.
121. Birkmeyer JD, Siewers AE, Marth NJ and Goodman DC: Regionalization of high-
risk surgery and implications for patient travel times. JAMA 2003; 290: 2703.
122. Casebeer A: Regionalizing Canadian healthcare: the good--the bad--the ugly.
Healthc Pap 2004; 5: 88.
123. Murray GD and Teasdale GM: The relationship between volume and health
outcomes--a review. Scott Med J 2006; 51: 17.
124. Urbach DR and Baxter NN: Does it matter what a hospital is "high volume" for?
Specificity of hospital volume-outcome associations for surgical procedures: analysis
of administrative data. Qual Saf Health Care 2004; 13: 379.
125. Halm EA, Lee C and Chassin MR: Is volume related to outcome in health care? A
systematic review and methodologic critique of the literature. Ann Intern Med 2002;
137: 511.
126. Ruggeri EM, Giannarelli D, Bria E, Carlini P, Felici A, Nelli F et al: Adjuvant
chemotherapy in muscle-invasive bladder carcinoma: a pooled analysis from phase
III studies. Cancer 2006; 106: 783.
127. Alter DA, Naylor CD, Austin PC and Tu JV: Long-term MI outcomes at hospitals
with or without on-site revascularization. JAMA 2001; 285: 2101.
128. Konety BR, Joslyn SA and O'Donnell MA: Extent of pelvic lymphadenectomy and
its impact on outcome in patients diagnosed with bladder cancer: analysis of data
from the Surveillance, Epidemiology and End Results Program data base. J Urol
2003; 169: 946.
129. Herr HW, Bochner BH, Dalbagni G, Donat SM, Reuter VE and Bajorin DF: Impact
of the number of lymph nodes retrieved on outcome in patients with muscle invasive
bladder cancer. J Urol 2002; 167: 1295.
130. Risberg T, Sorbye SW, Norum J and Wist EA: Diagnostic delay causes more
psychological distress in female than in male cancer patients. Anticancer Res 1996;
16: 995.
131. Gray RE, Fitch MI, Phillips C, Labrecque M and Klotz L: Presurgery experiences of
prostate cancer patients and their spouses. Cancer Pract 1999; 7: 130.
132. Rapoport Y, Kreitler S, Chaitchik S, Algor R and Weissler K: Psychosocial
problems in head-and-neck cancer patients and their change with time since
diagnosis. Ann Oncol 1993; 4: 69.
171
133. Liedberg F, Anderson H, Mansson A and Mansson W: Diagnostic delay and
prognosis in invasive bladder cancer. Scand J Urol Nephrol 2003; 37: 396.
134. Simunovic M, Theriault ME, Paszat L, Coates A, Whelan T, Holowaty E et al:
Using administrative databases to measure waiting times for patients undergoing
major cancer surgery in Ontario, 1993-2000. Can J Surg 2005; 48: 137.
135. Harrell FE: Regression Modeling Strategies: With Applications to Linear Models,
Logisitic Regression and Survival Analysis. Springer Series in Statistics. New York:
Springer-Verlag 2001.
136. Manoharan M and Soloway MS: Optimal management of the T1G3 bladder cancer.
Urol Clin North Am 2005; 32: 133.
137. Canadian Surgical Wait Times (SWAT) Initiative: Consensus document:
recommendations for optimal surgical wait times for patients with urological
malignancies. Can J Urol 2006; 13 Suppl 3: 62.
138. Birkmeyer JD and Dimick JB: Potential benefits of the new Leapfrog standards:
effect of process and outcomes measures. Surgery 2004; 135: 569.
139. Stein JP: Lymphadenectomy in bladder cancer: how high is "high enough"? Urol
Oncol 2006; 24: 349.
140. Polk HC, Lewis JN, Garrison RN, Van Vlack JP, Houck PM and Hunt DR: Process
and outcomes measures in specialty surgery: Early steps in defining quality. Bulletin
of the American College of Surgeons 2005; 90: 8.
141. Centre for Multilevel Modeling: MLwiN version 2.02. University of Bristol, 2005.
(URL: http://www.cmm.bristol.ac.uk/index.shtml).
142. Urbach DR, Bell CM and Austin PC: Differences in operative mortality between
high- and low-volume hospitals in Ontario for 5 major surgical procedures:
estimating the number of lives potentially saved through regionalization. CMAJ
2003; 168: 1409.
143. Dudley RA, Johansen KL, Brand R, Rennie DJ and Milstein A: Selective referral to
high-volume hospitals: estimating potentially avoidable deaths. JAMA 2000; 283:
1159.
144. Hollenbeck BK, Taub DA, Miller DC, Dunn RL, Montie JE and Wei JT: The
regionalization of radical cystectomy to specific medical centers. J Urol 2005; 174:
1385.
145. Birkmeyer JD, Siewers AE, Marth NJ and Goodman DC: Regionalization of high-
risk surgery and implications for patient travel times. JAMA 2003; 290: 2703.
172
146. Bach PB, Cramer LD, Schrag D, Downey RJ, Gelfand SE and Begg CB: The
influence of hospital volume on survival after resection for lung cancer. N Engl J
Med 2001; 345: 181.
147. Simunovic M, Rempel E, Theriault ME, Coates A, Whelan T, Holowaty E et al:
Influence of hospital characteristics on operative death and survival of patients after
major cancer surgery in Ontario. Can J Surg 2006; 49: 251.
148. Geddes J: Bypassing your doctor: Can the web fix the wait-time crisis, and
ultimately even save medicare? Rogers Publishing Ltd. Macleans, October 12 2005;
149. Malavaud B: T1G3 bladder tumours: the case for radical cystectomy. Eur Urol
2004; 45: 406.
150. Kulkarni GS, Finelli A, Fleshner NE, Jewett MA, Lopushinsky SR and Alibhai SM:
Optimal management of high-risk T1G3 bladder cancer: a decision analysis. PLoS
Med 2007; 4: e284.
151. Freeman JA, Esrig D, Stein JP, Simoneau AR, Skinner EC, Chen SC et al: Radical
cystectomy for high risk patients with superficial bladder cancer in the era of
orthotopic urinary reconstruction. Cancer 1995; 76: 833.
152. Government of Ontario: Health Services Restructuring Commission web page,
2000. Accessed November 1, 2005 (URL:
http://www.health.gov.on.ca/hsrc/e_menu.htm).
153. Pace KT, Provan JL and Jewett MA: The urology work force in Ontario for the 21st
century: feast or famine? Can J Surg 1999; 42: 181.
154. Canadian Residency Matching Service: Canadian Residency Matching Service web
page, 2008. Accessed January 10, 2008 (URL: http://www.carms.ca/).
155. Health Canada: Wait times in Canada, 2007. Accessed January 10, 2008 (URL:
http://www.hc-sc.gc.ca/hcs-sss/qual/acces/wait-attente/index_e.html).
156. Government of Ontario: Ontario's Wait Time Strategy, 2005. Accessed January 10,
2008 (URL: http://www.premier.gov.on.ca/news/Product.asp?ProductID=106).
157. Ministry of Health and Long Term Care: Ontario Wait Times, 2007. Accessed
January 10, 2008 (URL:
http://www.health.gov.on.ca/transformation/wait_times/wait_mn.html).
158. Needham DM, Scales DC, Laupacis A and Pronovost PJ: A systematic review of the
Charlson comorbidity index using Canadian administrative databases: a perspective
on risk adjustment in critical care research. J Crit Care 2005; 20: 12.
173
159. Bratzler DW, Houck PM and Surgical Infection Prevention Guideline Writers
Workgroup.: Antimicrobial prophylaxis for surgery: an advisory statement from the
National Surgical Infection Prevention Project. Am J Surg 2005; 189: 395.
160. Bratzler DW, Houck PM, Surgical Infection Prevention Guidelines Writers
Workgroup., American Academy of Orthopaedic Surgeons., American Association
of Critical Care Nurses., American Association of Nurse Anesthetists. et al:
Antimicrobial prophylaxis for surgery: an advisory statement from the National
Surgical Infection Prevention Project. Clin Infect Dis 2004; 38: 1706.
161. Gutt CN, Oniu T, Wolkener F, Mehrabi A, Mistry S and Buchler MW: Prophylaxis
and treatment of deep vein thrombosis in general surgery. Am J Surg 2005; 189: 14.
162. Tooher R, Middleton P, Pham C, Fitridge R, Rowe S, Babidge W et al: A systematic
review of strategies to improve prophylaxis for venous thromboembolism in
hospitals. Ann Surg 2005; 241: 397.
163. McGory ML, Maggard MA and Ko CY: A meta-analysis of perioperative beta
blockade: what is the actual risk reduction? Surgery 2005; 138: 171.
164. Pearse RM, Rhodes A and Grounds RM: Clinical review: how to optimize
management of high-risk surgical patients. Crit Care 2004; 8: 503.
165. Khuri SF, Daley J, Henderson W, Hur K, Demakis J, Aust JB et al: The Department
of Veterans Affairs' NSQIP: the first national, validated, outcome-based, risk-
adjusted, and peer-controlled program for the measurement and enhancement of the
quality of surgical care. National VA Surgical Quality Improvement Program. Ann
Surg 1998; 228: 491.
166. Fink AS, Campbell DA,Jr, Mentzer RM,Jr, Henderson WG, Daley J, Bannister J et
al: The National Surgical Quality Improvement Program in non-veterans
administration hospitals: initial demonstration of feasibility. Ann Surg 2002; 236:
344.