Hospital Quality Indicators in Iowa Rural Hospitals
Pengxiang (Alex) Li, Marcia M. Ward, Paul James, John E. Schneider
2008 AHRQ Annual Meeting Bethesda, Maryland
Support grant: Agency for Healthcare Research and Quality Grant # HS015009
Background
Hospital quality indicators were used to provide a perspective on hospital quality of care AHRQ Inpatient Quality Indicators (IQIs) AHRQ Patient Safety Indicators (PSIs)
Our analyses focus on Acute Myocardial Infarction (AMI) in-hospital
mortality (IQI-15) Four PSIs (PSI-5, PSI-6, PSI-7, and PSI-15)
Outline
Comparison of Iowa urban and rural hospitals on AMI in-hospital mortality James PA, Li P, Ward MM. Myocardial infarction mortality in rural and urban hospitals:
Rethinking measures of quality of care. Annals of Family Medicine, 5:105-111, 2007
Association between Critical Access Hospital (CAH) conversion and patient safety indicator performance Li, P., Schneider, J. E. & Ward, M. M., (2007) Effect of Critical Access Hospital
Conversion on Patient Safety. Health Services Research, 42 (6): 2089-2108
Exploration of a potential reason of patient safety change associated with CAH conversion Li, P., Schneider, J. E. & Ward, M. M., Effects of Critical Access Hospital Conversion
on the Financial Performance of Rural Hospitals Inquiry (in press)
How do Iowa urban and rural hospitals compare on AMI in-
hospital mortality?
James PA, Li P, Ward MM. Myocardial infarction mortality in rural and urban hospitals: Rethinking
measures of quality of care. Annals of Family Medicine, 5:105-111, 2007
Introduction
Observational studies find that the quality of care for myocardial infarction (MI) patients admitted to rural hospitals is substandard (Sheikh 2001, Baldwin 2004) Lower volumes of MI patients in rural hospitals Lacking cardiologists Lacking support services
Introduction Validity of these observational studies has been questioned
Unbalanced comparison groups
Patients admitted to rural hospitals tend to be older, poorer,
in poorer health, and have greater number of comorbidities
(Baldwin 2004, Chen 2000, Frances 2000)
Referral patterns of rural provider
Empirical study showed that less severe patients were
referred to urban hospitals (Metha 1999)
Unmeasured confounding may account for differences in
patient outcomes
Objectives of the study
To compare characteristics of MI patients admitted to rural and urban hospitals
To examine in-hospital mortality between rural and urban hospitals among MI patients Using traditional risk adjustment techniques
(Logistic regression) Using instrumental variable methods (IV)
Methods: Data Discharge data from Iowa State Inpatient Dataset (2002 & 2003) Inclusion criteria
A principal diagnosis of MI (ICD-9-CM: 410.01-410.91) Eighteen years or older
Exclusion criteria The hospital identification number was missing (n=9) Patient’s whose home county was not in Iowa (n=1,248) Patients’ zip code was missing (n=14) Patients’ sex was missing (n=1) Our primary analyses also excluded patients discharged or
transferred to another short term general hospital for inpatient care (n=1,618)
Most of our analyses are based on 12,191 MI patients
Methods: Variables Dependent variable
In-hospital mortality Independent variables
Urban vs Rural hospitals that patients admitted to Urban: 27 hospitals Rural: 89 hospitals
Payer: e.g. Medicare, private insurance, self-pay Admission type: e.g. emergency Race Risk adjustment index
Charlson comorbidity index All Patient Refined DRGs (APR-DRGs) risk index
Methods: Traditional Analytic Approach (Logistic Regression)
Univariate analyses of group comparisons Chi-square tests for dichotomous data ANOVAs for continuous data
Logistic regressions for multiple regression analyses
Methods: Pitfalls with Logistic Regression
Using administrative inpatient data, one cannot control all patients’ risk
factors (e.g. severity of illness)
If unmeasured variables are related to selection of the hospital, the
estimates of the hospital-specific contribution to mortality will be
biased.
For example, elderly MI patients with severe comorbid conditions,
which are unmeasured in administrative data, might prefer to remain in
the rural hospitals.
As a result, a higher risk-adjusted mortality rate in rural hospitals
might simply be due to more severe patients in rural hospitals.
Approaches to Minimize Bias Collect all the relevant patient-level variables: very costly Randomized controlled trial
Not feasible to this study Instrumental variable (IV) estimation
An econometric technique which enables us to obtain unbiased estimates of treatment effects in observational studies
An example: Wehby (2006) found that using the logistic regression model, early initiation of prenatal care is associated with a higher probability of low birth weight (LBW) Unmeasured confounders: women at a higher risk
demand more (or early) prenatal care compared to those at lower risk.
IV estimations showed that early time to prenatal care initiation is associated with a lower probability of LBW.
The Instrumental Variable (IV) estimation
IVs are used to achieve a “pseudo-randomization” The instrumental variable technique can extract variation
in the focal variable (rural hospital selection) that is unrelated to unmeasured confounders, and employ this variation to estimate the causal effect on an outcome
Assumptions for IV(s)1. IV(s) should correlate with treatment variable (choice of
rural hospital)2. IV(s) should not be correlated with the unmeasured
confounders
1** 210 uPmaIVsaaurban
2*)(* 210 uPmburbanpredictedbbdead
Methods: Instrumental Variable Technique
Instrumental Variable = Patients’ distance to the nearest urban hospital The distances between each patient’s home and all
urban hospitals in Iowa were obtained by calculating the distances between the centroids of each patient’s resident zip code and all urban hospitals’ zip codes.
Similar to Brooks (2003) approach, instrumental variables in the study are dummy variables that group patients based on the their distance to the nearest urban hospital.
Methods: IV Technique: First assumption
Patients who live closer to an urban hospital are more likely to choose an urban hospital than those who live farther away. Partial F-statistics for the IVs in the first stage
regression Small values of first-stage F-statistics imply failure
of assumption 1 Rule of thumb: F>10 indicates good association
(Staiger 1997)
Methods: IV Technique Second Assumption:
Distance to the nearest urban hospital is not associated with the severity or pre-morbid risks of patients with MI Descriptive comparison between two groups of
patients classified by IV If the instrument is independent of the
unmeasured confounders, it should also be independent of observed risk factors (e.g. age, and comorbidity index).
Over-identifying restrictions tests The null hypothesis is that the IV is not correlated
with unmeasured confounders
Methods: IV Technique
To examine the robustness of our findings: We used a range of patients’ groups for the
instrumental variable (2, 4, 8, and 12 groups). We varied the independent variables.
The syslin two-stage least squares (2SLS) procedure in SAS 9.1 was used to do IV estimation.
Results: Table 1: Baseline characteristics of MI patients* admitted
to rural and urban hospitals
Variables Rural (N= 1,426)
Urban(N= 10,765)
p-value
Age 82.35 68.89 <.0001
Male (%) 45.02 59.76 <.0001
Black (%) 0.14 1.13 0.0004
Number of secondary diagnoses 5.66 5.61 0.43
Charlson comorbidity index 0.96 0.69 <.0001
APR-DRG risk index 0.09 0.06 <.0001
In-hospital Mortality 0.14 0.06 <.0001
* Excluding patients discharged or transferred to another short term general hospital for inpatient care.
Results:
Table 2: Baseline characteristics of MI patients transferred
out of rural hospitals or staying in rural hospitals
Variables Stay in rural hospitals (N=1,426)
Transfer out* of rural hospitals
(N=730)
p-value
Age 82.35 71.46 <.0001
Male (%) 45.02 56.71 <.0001
Black (%) 0.14 0.14 0.99
Number of secondary diagnoses 5.66 4.24 <.0001
Charlson comorbidity index 0.96 0.67 <.0001
APR-DRG risk index 0.09 0.04 <.0001
* Patients discharged or transferred to another acute care hospital for inpatient care
Results: Table 3: Odds ratios of in-hospital mortality* among MI patients admitted to urban hospitals or to rural hospitals, using logistic
regression models (n=12,191)
Model components Odds ratio (Urban vs
Rural)
95% CI
p-value c-statistic
Unadjusted 0.42 0.36-0.50 <.0001 0.56
Adjusted for demographic variables (age, sex, race, admission type and source of payment)
0.70 0.59-0.84 <.0001 0.71
Adjusted for demographic variables and Charlson comorbidity index
0.70 0.59-0.84 0.0001 0.71
Adjusted for demographic variables and APR-DRG risk index
0.68 0.56-0.82 <.0001 0.86
* Excluding patients discharged or transferred to another short term general hospital for inpatient care
Results: Table 4: Characteristics among MI patients grouped by
distance to the nearest urban hospital Variables Distance to nearest
urban hospital <=14.08 miles*(N= 6,097)
Distance to nearest urban hospital >14.08 miles
(N= 6,104)
p-value
Mean Distance to the nearest urban hospital (miles)
4.94 34.20 <0.0001
Percent of patients admitted to urban hospitals (%)
99.54 77.07 <0.0001
Age 68.89 72.02 <0.0001
Male (%) 58.65 57.45 0.18
Black (%) 1.95 0.08 <0.0001
Number of secondary diagnoses 5.72 5.53 <0.0001
Charlson comorbidity index 0.72 0.72 0.67
APR-DRG risk index 0.07 0.07 0.48
In-hospital mortality rate (%) 7.07 7.52 0.34
*14.08 miles is the median distance from patient’s home to the nearest urban hospital
Results: Table 5: Instrumental variable estimates of the difference of in-patient mortality between urban and rural hospitals
* If a F-statistic is less than 10, the instrumental variables are weak.** If p-value is less than 0.05, one of the instrumental variables correlated with unmeasured confounders
IV models (n=12,191)
Number of groups for
instrumental variable
Tests for instrumental variablesIV estimates
of mortality difference
Instrument
P-value for overidentifying
restrictions tests** Coefficients P-valueF-statistic*
Unadjusted
2 1540.16 - -0.0199 0.34
4 642.65 0.65 -0.0269 0.16
12 184.31 0.13 -0.0288 0.13
Adjusted for demographic variables
2 1568.24 - 0.0127 0.58
4 652.86 0.80 0.0081 0.69
12 187.14 0.10 0.0065 0.75
Adjusted for demographic variables and Charlson comorbidity index
2 1539.9 - 0.0090 0.69
4 642.51 0.92 0.0053 0.80
12 184.29 0.12 0.0040 0.84
Adjusted for demographic variables and APR-DRG risk index
2 1694.27 - -0.0034 0.87
4 640.61 0.92 -0.0069 0.72
12 202.50 0.01 -0.0063 0.74
Results: Sensitivity analyses
Repeat analyses in different samples Excluding transferred in MI patients Three-year state inpatient datasets (2001 to 2003)
Different IV estimation method Two-stage residual inclusion method to account
for the endogeneity in nonlinear (logistic) model Bivariate Probit model (using Stata 9.0)
The results are consistent with IV estimation in Table 5
Discussion This study confirms earlier studies
MI patients admitted to rural hospitals were older and sicker than their urban counterparts
Traditional models all indicate significantly higher in-hospital mortality for those admitted to rural hospitals
Discussion Our findings suggest that the traditional logistic
regression models are biased Admissions to rural or urban hospitals are
likely to be confounded by unmeasured patient variables
Referral patterns in rural hospitals Younger and less sick patients are
transferred to urban hospitals The clinical judgment about transfer of
rural senior patients with MI may rely on different criteria
Discussion Patient preferences are likely to play a significant role
in transfer decisions for older MI patients May reflect personal choice or existing serious
comorbidities Serious cases may choose to remain close to
home The transfer patterns may reflect rural doctors
respecting their patients’ wishes Using in-hospital MI mortality to measure quality of
care in rural hospitals is problematic.
Limitations of the study
The results of the IV estimation can only be generalized to patients for whom distance affects their choice The conclusion cannot be applied to MI patients
bypassing rural hospitals and seeking care in urban hospitals
The findings for hospitals in one state may not generalize to other states .
Analyses of in-hospital mortality rates may not generalize to mortality rates after hospitalization.
Conclusions
Mortality from MI in rural Iowa hospitals is not higher when controlled for unmeasured confounders.
Current risk-adjustment models may not be sufficient when assessing hospitals that perform different functions within the healthcare system.
Unmeasured confounding is a significant concern when comparing heterogeneous and undifferentiated populations.
Did conversion to Critical Access Hospital (CAH) status affect patient
safety indicator performance?
Li, P., Schneider, J. E. & Ward, M. M., (2007) Effect of Critical Access Hospital Conversion on
Patient Safety. Health Services Research, 42 (6): 2089-2108
Background In order to protect small, financially vulnerable rural
hospitals, the Medicare Rural Hospital Flexibility Program of the 1997 Balanced Budget Act allowed hospitals meeting certain criteria to convert to critical access hospitals (CAH)
This changed their Medicare reimbursement mechanism from prospective (PPS) to cost-based
One objective of the policy was to increase the quality of care in these hospitals
Timeframe for Conversion to CAH
0%
20%
40%
60%
80%
100%
1997 1998 1999 2000 2001 2002 2003 2004 2005
Rural PPS hospitals CAHs
Patient Safety
4 PSIs and Composite AHRQ recommends suppressing the estimates if fewer than 30
cases are in the denominator
Only five patient safety indicators are able to provide PSI measures for all rural Iowa hospitals PSI-5: foreign body left during procedure PSI-6: iatrogenic pneumothorax PSI-7: selected infections due to medical care PSI-15: accidental puncture or laceration PSI-16: transfusion reaction
Too rare to provide variability to differentiate hospitals in Iowa
A composite patient safety variable was created by summing the four PSIs (PSI-5, PSI-6, PSI-7, and PSI-15).
Number of Hospitals Having Better or Worse Performance after CAH Conversion
0
5
1015
20
25
3035
40
45
PSI-5 PSI-6 PSI-7 PSI-15 Compositescore of4PSIs
Better performance worse performance
Cross-sectional Analyses Cross-sectional comparisons showed that CAHs had
better performance than rural PPS hospitals on 4 of the 5 PSI measures.
However, the difference in patient safety indicators might be due to differences in patient mix, hospital characteristics besides CAH conversion, and differences in markets and environment.
Multivariable Analyses
We used multivariable Generalized Estimating Equations (GEE) models and sensitivity analyses to control for the impact of patient case mix, market variables, and time trend.
GEE models showed that CAH conversion was associated with significant better performance in PSI-6, PSI-7, PSI-15 and composite PSI.
Findings were robust among sensitivity analyses using different samples and different methods
Conclusions CAH conversion in rural hospitals resulted in enhanced
performance in PSIs
We speculate that the likely mechanism involved an increase in financial resources following CAH conversion to cost-based reimbursement for Medicare patients
How did Critical Access Hospital conversion affect rural hospital
financial condition?
Li, P., Schneider, J. E. & Ward, M. M., Effects of Critical Access Hospital Conversion on the
Financial Performance of Rural Hospitals Inquiry (in press)
Objectives
To study the effects of CAH conversion on Iowa rural hospitals’ operating revenue, cost, and profit margin
Study Sample and Study design
Sample Eight year (1997-2004) panel data for 89 Iowa
rural hospitals (rural PPS hospitals and CAHs) Unit of analysis is hospital-year
Study design Quasi-experimental designs that use both control
groups and pretests Panel data regression with fixed hospital effects
Models
Ad hoc models: Revenueit=f(CAHit,Pjt,Yit,Xit) Costit=f(CAHit,Wjt,Yit,Xit) Marginit=f(CAHit,Wjt, Pjt,Yit, Xit)
Variables: CAHit: hospital status (CAH or rural PPS) for ith hospital in
year t Pit: output prices for ith hospital in year t Wit: input prices for ith hospital in year t Yit: output volume for ith hospital in year t Xit: other variables for ith hospital in year t that empirically
affect dependent variables
CAH variables
One dummy variable CAH=1, if the hospital is in CAH status
Three dummy variables CAH1it=1, if the hospital is in the first year of CAH
status, otherwise CAH1it=0 CAH2it=1, if the hospital is in the second year of
CAH status, otherwise CAH2it=0 CAH3it=1, if the hospital is in CAH status for more
than 2 years, otherwise CAH3it=0 Comparison group: Rural PPS
Other covariates Pit: output prices for ith hospital in year t
Medicare Part A (hospital) adjusted average per capita cost (AAPCC) as proxy of hospital output price (county level)
Wit: input prices for ith hospital in year t Hourly wages for registered nurses (county level)
Yit: output volume for ith hospital in year t Total number of acute discharges, total number of outpatient
visits, and average length of stay of acute discharges The squared and cubed output measures and interaction terms
will be included
Others Xit: other variables for ith hospital in year t that empirically affect dependent
variables Hospital size (number of beds) Hospital case-mix
Hospital mean DRG weight, percent of emergency visits, and percent of Medicare and Medicaid days among acute inpatient days
Variables reflecting the hospital market (we assumed the county to be the relevant geographic market of hospital services.) Herfindahl-Hirschman Index (HHI), per capita income, and
population density in the county in which the hospital is located Year dummy variables which will adjust the effects of unmeasured, time-
specific factors Revenue and expense functions were log transformed
Data Sources
Iowa Hospital Association Profiles Iowa State Inpatient datasets Area Resource File Centers for Medicare and Medicaid Services American Hospital Association Annual Survey
Database Bureau of Labor Statistics
Result:Table 1: Changes in rural hospital patient care revenue, expense, and operating margin associated with CAH conversion, 1998-2004
Log(operating revenue) Log(operating expense) Operating margin Covariate
Coefficient Standard
error Coefficient
Standard error
Coefficient Standard
error
Hospital status
Rural PPS: CAH =0 Reference Reference Reference
CAH 0.0288** 0.0110 0.0199** 0.0089 0.0020 0.0078
…
Observations 623 623 623
Groups 89 89 89
R-Squared (within) 0.8833 0.9081 0.2177
* P-value< 0.1 ** P-value< 0.05
Table 2: Changes in rural hospital patient care revenue, expense, and operating margin during the first, second and third plus years of CAH conversion,
1998-2004 Log(operating revenue) Log(operating expense) Operating margin
Covariate Coefficient
Standard error
Coefficient Standard
error Coefficient
Standard error
Hospital status
Rural PPS: CAH =0 Reference Reference Reference
CAH1 0.0034 0.0114 0.0175* 0.0096 -0.0206** 0.0079
CAH2 0.0712** 0.0137 0.0206* 0.0114 0.0386** 0.0097
CAH3 0.0934** 0.0159 0.0483** 0.0133 0.0543** 0.0110
Observations 623 623 623
Groups 89 89 89
R-Squared (within) 0.8929 0.9101 0.3101
* P-value< 0.1 ** P-value< 0.05
Results Operating revenue
No change in the first year of conversion (paid an interim rate) Significant increases since the second year of CAH conversion
Operating expenses CAH conversion is associated with significant increase in hospital
operating expenses Hospitals increase expenses in the first year of conversion
Operating Margin Significant drop in the first year of conversion Significant increase since the second year of conversion
Sensitivity analyses showed similar results
Conclusions CAH conversion in rural hospitals resulted in better
patient safety.
Rural hospital CAH conversion was associated with significant increases in hospital operating revenues, expenses and margins
Summary: Limitations of measures
In-hospital mortality Substantial unmeasured confounders
Patient Safety Indicators Only small number of indicators can be applied to
rural hospitals Changes of indicators might reflect changes in
coding or reporting in administrative data We need hospital quality indicators specifically for
rural hospitals
Thank you Questions?
Contact information Pengxiang (Alex) Li University of Pennsylvania [email protected]