development of an integrated prostate cancer research information system

Development of an Integrated Prostate Cancer Research Information System

Submitted: Mar 28, 2006; Revised: May 15, 2006; Accepted: May 22, 2006

Address for correspondence: William K. Oh, MDLank Center for Genitourinary Oncology, Dana-Farber Cancer Institute44 Binney StBoston, MA 02115 Fax: 617-632-2165; e-mail: [email protected]

William K. Oh,1 Julia Hayes,1 Carolyn Evan,1 Judith Manola,2 Daniel J. George,1 Helen Waldron,1 Meaghan Donovan,1 John Varner,3 John Orechia,3 Beth Katcher,3 Diana Lu,3 Arthur Nevins,3 Renée L. Wright,1 Lauren Tormey,1 James Talcott,4 Mark A. Rubin,5 Massimo Loda,5 William R. Sellers,1 Jerome P. Richie,6 Philip W. Kantoff,1* Jane Weeks7*

Original Contribution

Background: In this article, we describe the design and implementa-tion of a comprehensive prostate cancer database developed to collect, store, and access clinical, treatment, and outcomes data for research and clinical care. Patients and Methods: The Prostate Cancer Clinical Research Information System is a relational database. Data are entered from multiple sources, including medical records, institutional laboratory, patient registration, pharmacy systems, and clinician forms. The history, design, and operational characteristics of the database are described. Issues regarding necessary staffing and funding of databases are re-viewed. Results: Four thousand two hundred forty-six patients have in-formation in the Prostate Cancer Clinical Research Information System. Mean age of patients is 62 years, and 89% are white. Seventy-one percent of patients presented at diagnosis with T1 or T2 disease, and 78% had biopsy Gleason scores of ≤ 7, 8-10 in 18%. Median prostate-specific antigen level at diagnosis was 7 ng/mL, and 77% of patients presented with increased prostate-specific antigen as a trigger symptom. Sixty-four percent of patients presented to our clinic having had no previ-ous treatment for prostate cancer. The majority of approached patients provided consent for collection of clinical data, blood, and tissue. Quality control assessments demonstrate high levels of concordance among data entry personnel. Conclusion: Clinical databases are difficult to implement and maintain; however, they represent a valuable resource, particularly when linked to blood and tissue banks. Elements needed for a successful clinical database include engagement of clinicians, utility for research, and the ability to integrate with legacy systems. As cancer centers develop such databases, lessons learned from each experience should be shared in order to optimize the process.

Abstract

1Lank Center for Genitourinary Oncology2Department of Biostatistical Science3Department of Information ServicesDana-Farber Cancer Institute4Division of Medical Oncology, Massachusetts General Hospital5Department of Pathology 6Division of Urological SurgeryBrigham and Women’s Hospital7Center for Outcomes and Policy Research, Dana-Farber Cancer Institute, Boston, MA*Contributed equally to this work

Clinical Genitourinary Cancer, Vol. 5, No. 1, 61-66, 2006

Key words: Database, Gleason score, Prostate-specific antigen Introduction

Prostate cancer is the most common type of cancer in men in the United States, with an estimated 234,460 new cases in 2006.1 A reliable source of clinical information is essential to improve the efficiency and quality of treatment, to characterize the disease, and to investigate clini-cal and biologic markers of the disease.2-4 In recent years, several clinical research databases for prostate cancer have been developed. The Cancer of the Prostate Strategic Urologic Research Endeavor (CaPSURE) obser-vational database, managed by the University of California San Francisco Department of Urology and supported by TAP Pharmaceuticals, Inc, has enrolled > 11,000 patients and resulted in > 75 publications since 1995.2 This database incorporates clinical, quality of life, and economic outcomes based on data acquired primarily from urology office practices. The Department of Defense–funded Center for Prostate Disease Research is a relational multisite database consisting of data from > 15,000 patients with prostate cancer.3,4 Recognition of the value of integrating clinical information with research specimens has led to the creation of several

Clinical Genitourinary Cancer June 2006 • 61

Electronic forwarding or copying is a violation of US and International Copyright Laws.Authorization to photocopy items for internal or personal use, or the internal or personal use of specific clients, is granted by CIG Media Group, LP, ISSN #1558-7673, provided the appropriate fee is paid directly to Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923 USA 978-750-8400.

62 • Clinical Genitourinary Cancer June 2006

large databases incorporating research specimens, including the Specialized Programs of Research Excellence–supported Memo-rial Sloan-Kettering Cancer Center database and the multisite National Cancer Institute–supported Cooperative Prostate Can-cer Tissue Resource. These databases demonstrate the value of pooling data across many hospitals and offices. However, there are important weaknesses with these large databases, including a paucity of data on patients with advanced-stage prostate cancer.

Originally implemented in 1995, the Clinical Research Information System (CRIS) was developed at Dana-Farber Cancer Institute (DFCI) to provide a research database integrated with legacy information systems. In the absence of well-developed research tools, clinical research was being conducted using individual desktop databases. Such sys-tems have many flaws, including lack of technical support and security, minimal data model design, lack of integration with standard hospital-based clinical information systems, and a lack of systematic quality control. The CRIS wasThe CRIS was was established with support from Information Services to ad-dress these issues on an institutional level.

The prostate CRIS was designed to collect, store, and ac-cess clinical, treatment, and outcomes data on patients withpatients with

prostate cancer treated at DFCI and Dana-Farber/Partners Cancer Care hospitals. Although there are currently 5 CRIS applications in use at DFCI: breast, lung, prostate, renal, and gastrointestinal (multidisease application), we will be focusing on Prostate CRIS. This paper describes the implementation and management issues revolving around the creation and maintenance of a large clinical database at a single tertiary care cancer center. Because of the large un-dertaking represented in this effort, our hope is to provide some background and lessons for others who are interested in creating such databases and to provide baseline data on patients represented in our database.

Patients and MethodsCRIS/STIP

The CRIS system consists of the following components: data-entry software, a central data repository, collection of patient data including comprehensive follow-up of all patients, and tight-ly integrated security measures. Data are stored in an Oracle re-lational database. PowerBuilder was used to develop the database environment. The Specimen Tracking Inventory Program (STIP) is a separate but integrated product designed to record collection,

Patient Summary Report Figure 1

Integrated Prostate Cancer Research Information System

William K. Oh et al

storage status, and limited clinical information on research speci-mens. The STIP is directly linked to CRIS and other institutional systems through the reporting tool Business Objects™. Although the CRIS system is not currently available on the web, plans are under way to make certain features internet-accessible.

Various features were implemented to assist with data collec-tion, assure quality control, facilitate clinical care, and enhance reporting capabilities. First, data in CRIS are validated at the time of entry; free text fields have been avoided, and business rules in the application prevent data entry errors. A dictionary created in Microsoft® Access documents data entry rules and definitions. Lastly, real-time interfaces with preexisting legacy systems allow information regarding patient registration and clinical trials management and from laboratories and pharma-cies to be fed directly into CRIS.

An additional CRIS requirement was that it efficiently support multiple cancer diagnoses. In order to facilitate sharing of the data model, information systems created a core set of tables that could be used across cancer types, thus reducing programming time when a new disease group is added. In addition, all CRIS and STIP data rows are marked with a research group. This allows users to view only data from their own disease group. However, with the agree-ment of group leaders, data can also be shared across disease types (eg, in evaluating family history of cancer across groups).

Prostate Cancer Clinical Research Information System Development and Operations

The first Prostate CRIS application went “live” in February 1997. The general structure of the Prostate CRIS database was based on the previously developed Breast CRIS at DFCI. The current version of Prostate CRIS became available for prospective data entry in No-vember 2001. The following data elements are currently captured: demographics, initial diagnosis date, symptoms, progressions, pros-tate-specific antigen (PSA) values, radiologic studies, surgical proce-dures, pathology (> 80% of which are secondarily reviewed), TNM staging, physical examination, disease and treatment status, family history, and research protocol enrollment. The categories reflect the unique characteristics of prostate cancer progression. At this time, quality of life information is not prospectively collected.

Patients included in Prostate CRIS are recruited at the Lank Center for Genitourinary Oncology at DFCI or in the Urology Clinic at Brigham and Women’s Hospital (BWH). Current staff-

ing for the Prostate CRIS team includes a physician leader, a full-time project manager, 3 full-time study coordinators, a half-time biostatistician, and a half-time programmer. Medical oncologists, urologists, radiation oncologists, and nurse practitioners participate in data collection.

The CRIS coordinator first enters all outside medical record information into CRIS. The study coordinator then prints out a Patient Summary Report (PSR; Figure 1) and addends a blank CRIS form to the patient’s chart to be completed by the clinician on the initial visit day. When completed, the study coordinator edits the information in CRIS. Information regarding patient registration, clinical trials management, clinical laboratory results, and pharmacy information is transferred directly from established legacy systems into CRIS as previously described. At each subse-quent patient visit, a revised PSR is printed and provided to the clinician for updating (Table 1).

This project is financially supported by a combination of institu-tional and philanthropic funds. Initial capital expenses included the cost of building and customizing the database as well as purchasing software and licenses for reporting. In 1995, these costs were in excess of $1 million and were provided by the institution. Since then, institutional Information Services support has continued for database maintenance, backup, and programming enhancements. Philanthropic support and grant support account for the salaries of current staff, as well as collection and storage of tissue and blood.

Quality Control Three measures of quality control are employed to ensure

Prostate CRIS accuracy and completeness. The first is a monthly 5-point analysis, consisting of 2 sets of 5 questions used by the study coordinator when reviewing CRIS data. A second quality control is performed every other month by study coordinators during review of CRIS forms collected from clinic. All forms are reviewed, and the number of required corrections calculated. Errors are corrected upon data entry. Finally, chart review is used routinely to capture patient data not included in the 5-point question review.

Specimen Collection, Tracking, and ReportingMore than 5000 blood samples have been collected under

various specimen and data collection protocols since 1992. After obtaining informed consent, approximately 20 mL of blood is col-lected and stored at –80°C. In April 2001, a prostate tissue banking

Data Collection in Prostate Clinical Research Information System

Prostate CRIS FormProstate CRIS Form Source of DataSource of Data

Initial Encounter PSR

Follow-up Visit PSR

Patient medical record

Patient

Laboratory data, pharmacy information,clinical trials management

Outside medical record

Patient encounter

Laboratory data, pharmacy information,clinical trials management

Personnel Entering DataPersonnel Entering DataStudy coordinator

Clinician

Legacy systems

Study coordinator

Clinician

Legacy systems

Time Required to Enter DataTime Required to Enter DataUp to 2 hours

< 5 Minutes

0

< 5 Minutes

1-3 Minutes

0

Table 1




system was developed in conjunction with the BWH pathology department. We have acquired frozen prostate tissue and paraffin blocks from 1019 patients, both from consenting patients under-going radical prostatectomy at BWH and from outside hospitals.

All blood and tissue samples are logged into and tracked through STIP using a unique identifier code. Specimen data include specimen type and contents, collection method, and date collected and received. Specimen component data include component type, amount, storage date and status, times thawed, and pathology detail.

Protection of Human Subjects for ResearchThe Dana-Farber/Partners Cancer Center Institutional Re-

view Board (IRB) approved this research protocol. Patients can choose to consent to give (1) blood samples, (2) permission for researchers to have access to discarded tissue related to any surgery, (3) access to clinical information for research purposes, and/or (4) permission to recontact them or their families regard-ing future research. Before their visit, patients receive an IRB-approved letter by mail summarizing the study and stating that they will be approached in the clinic by a study coordinator.

Patient confidentiality is protected throughout the data col-lection and reporting process. The CRIS was implemented after the institution of Health Insurance Portability and Accountabil-ity Act (HIPAA) regulations and, thus, has been in compliance since its inception. The CRIS, STIP, and Business Objects™ are password-protected applications, with access limited to specified programmers, database staff, and physicians.

Data requests for research projects are reviewed at several levels. First, the investigator for the governing protocol must approve the work. An oversight committee then reviews requests for clinical, blood, or tissue samples from the database. If a request is approved, the project is reviewed by the IRB. If ap-proved by the IRB, data, blood, and/or tissue is provided to the investigator for the project. Per HIPAA requirements, investiga-tors can request the minimum amount of patient information needed to accomplish their goals. Finally, if patient identifiers are shared with researchers outside of Dana-Farber/Partners Cancer Care, a HIPAA accounting of disclosure form is com-pleted, and a data-use agreement is developed.

Statistical MethodsDescriptive statistics were used to summarize patient char-

acteristics and quality control scores, and t tests were used to test for differences between first-time and follow-up patients and between charts receiving a single review and multiple reviews. Analysis of variance was used to test for differences among reviewers. Exact 2-sided binomial confidence intervals were calculated for concordance rates among reviewers and for correction rates. All tests were conducted at a 2-sided 0.05 significance level.

ResultsClinical Data

As of January 2006, 4246 patients with clinical information were enrolled in Prostate CRIS. Table 2 reflects the demographic and clinical features of patients in this cohort. An increased PSA level was the most common trigger symptom for diagnosis, followed by positive digital rectal examination. The majority of new patients seen at DFCI had untreated localized prostate cancer (Table 3); the majority of follow-up patients are receiv-ing treatment for recurrent or advanced-stage disease. Of 741 patients who developed metastases, bone was the most common site of disease progression (n = 558; 52%). Lymph node involve-ment was noted in 335 patients (31%), liver in 63 (6%), and lung in 51 (5%).

Clinical Features of Patients with CompleteNew Patient Information (N = 4246)

Mean Age at Diagnosis, Years (Range)

Ethnicity

White

Other/unknown

Black

Asian

American Indian

Mean PSA Level at Diagnosis,ng/mL (Range)

Median PSA Level at Diagnosis,ng/mL (Range)

Clinical Stage at Diagnosis

T1

T2

T3

T4

T(any) N1 M0

T(any) N(any) M1

Other

Gleason Score at Diagnosis

2-6

7

8-10

Missing

Trigger for Diagnosis

Increased PSA level

Abnormal DRE

Benign prostatic hyperplasia

Other

Pain

Hematuria

Missing

62 (39-89)

3790 (89)

275 (6)

149 (4)

30 (< 1)

2 (< 1)

41 (0-10,366)

7 (0-10,366)

2002 (47)

1009 (24)

64 (2)

3 (< 1)

69 (2)

199 (5)

900 (21)

1960 (46)

1381 (32)

755 (18)

150 (4)

3262 (77)

539 (13)

121 (3)

45 (1)

40 (< 1)

28 (< 1)

211 (5)

Table 2

Values in parentheses are percentages unless otherwise indicated.Abbreviation: DRE = digital rectal examination

FeatureFeature Value

Specimens and ConsentOf the 4246 patients with information in CRIS, 2690 (63%)

have consented to allow clinical information to be used for research purposes, 2196 (52%) have additionally given a blood sample, and 2293 (54%) have a blood or tissue sample available in storage. More recent data demonstrate that approximately 90% of new patients allow consent for acquisition of clinical data and use of blood and tissue (data not shown).

Quality Control Clinical Research Information System data for 60 patients

were reviewed using a 5-point analysis. Thirty new patients and 30 follow-up patients with ≥ 5 visits to our institution were reviewed. The mean score for charts reviewed by as many as 3 reviewers was 4.46/5 (95% confidence interval [CI], 4.26-4.65). There were no statistically significant differences among reviewers (P = 1) or between charts reviewed by 1 versus 3 reviewers (P = 0.36). Of the 30 records reviewed by 3 reviewers, complete concordance on all questions among all reviewers was achieved in 26 cases (13.3% discordance rate; 95% CI, 3.8%-30.7%). Finally, correction rates were calculated for CRIS forms collected from the clinic on 8 separate clinic dates. Fifty-one cli-nician and 27 data manager corrections were necessary, resulting in an overall correction rate of 15.4% (95% CI, 12.4-18.9%; data not shown). All corrections detected by this method were made before entry into the database.

In a separate quality control analysis, 42 consecutive Pros-tate CRIS reports were compared with data acquired routinely through our tumor registry. In 28 evaluable cases, overall con-cordance was high, with the primary source of discrepancies caused by TNM staging classification (ie, the use of “N0 M0” by the tumor registry vs. “Nx Mx” by CRIS coordinators if a bone and computed tomography scan were not performed) or to the version of American Joint Committee on Cancer Staging Guidelines version used. At least 9 dates noted on Prostate CRIS forms were not known by tumor registry personnel.

Clinician SatisfactionTwelve of the most active clinicians completed a survey

measuring the extent to which they found Prostate CRIS useful and of interest. The majority spent 1-3 minutes com-pleting a typical follow-up form and 1-5 minutes per New Patient form. All respondents noted that they “almost always” find the PSR helpful for understanding patient histories, and 92% use the PSR for discussion with patients at least “sometimes.” Most clinicians considered Prostate CRIS to be an important research tool. These results were encouraging, because we considered clinician participation and interest es-sential to the success of Prostate CRIS.

DiscussionClinical databases are expensive, labor-intensive, and fraught

with potential for error. In addition, prostate cancer is characterized by complexities of treatment and outcome patterns that make da-tabase development uniquely challenging. However, it is critical to understand prostate cancer at a clinical and molecular level. Prostate

CRIS was therefore developed following several principles that we believe are critical to the success of any clinical database: supporting clinical care, supporting research, and fostering exportability.

Supporting Clinical Care Initially, CRIS failed to achieve high rates of form comple-

tion among practicing clinicians. However, by presenting clinicians with essential patient information in a concise and useful manner, we were able to improve form completion rates. For example, the patient summary report provides the clini-cian with an easily referenced summary of critical diagnostic information, dates of previous therapies, and sites of metastases. The individualized PSA graphs allow clinicians a simple way to

Disease and Treatment Status at the Time of First Visit

Disease StatusDisease StatusLocalized/Low Risk

Localized/Intermediate Risk

Localized/High Risk

Increasing PSA AfterLocal Therapy

Metastases, on PrimaryHormones: Refractory

Metastases, on PrimaryHormones: Remission

No Metastases, on PrimaryHormones: Refractory

No Metastases, on PrimaryHormones: Remission

No Therapy for Metastases

No Diagnosis of Prostate Cancer

Missing

No Previous ProstateCancer Treatment

Completed Localized Treatment

Receiving Primary HormoneTreatment for Advanced Disease

Receiving Neoadjuvant Treatment

Receiving Secondary HormoneTreatment for Advanced Disease

Completed Salvage Local Treatment

Receiving Chemotherapyfor Advanced Disease

Completed Adjuvant Treatment

Other Cancer Treatment

Receiving Adjuvant Treatment

Receiving Localized Treatment

Completed Neoadjuvant Treatment

Receiving Salvage Local Treatment

Palliative Care Only

Missing

1335 (31)

1004 (24)

594 (14)

378 (9)

298 (7)

163 (4)

94 (2)

84 (2)

49 (1)

59 (1)

188 (4)

2706 (64)

421 (10)

375 (9)

141 (3)

137 (3)

69 (2)

57 (1)

38 (< 1)

21 (< 1)

12 (< 1)

5 (< 1)

9 (< 1)

6 (< 1)

5 (< 1)

244 (6)

Table 3

N (%)N (%)

Treatment StatusTreatment Status N (%)N (%)

William K. Oh et al


communicate trends in disease states to their patients. Prostate CRIS thereby engages clinicians in this research endeavor by providing them with a useful tool, one for which they are will-ing to invest a few minutes of their time to ensure it is complete and accurate. In addition, this database encourages exploration of clinical research questions by physicians. Consistently, forms collected from physicians are completed at a rate > 90%.

Supporting ResearchThe ultimate value of this database lies in the connections it

permits between molecular mechanisms and clinical outcomes. By integrating the clinical database with our blood and tis-sue banks, we can elucidate mechanisms of disease. In addi-tion, prostate CRIS is designed to incorporate patient-related data from many sources, including preexisting hospital legacy systems. This design maximizes recruitment of patients with prostate cancer and minimizes the need for duplicate data entry, thereby broadening opportunities for research.

In an era of much concern regarding privacy of health-related information, the consenting process is critical to any research data-base endeavor. We have improved our consent rates by emphasizing patient education and hiring dedicated research coordinators.

Fostering Exportability

The CRIS and STIP data model designs have effectively sup-ported application and reporting requirements. The CRIS and STIP applications are user friendly and are optimized to support the prostate data collection process. As a result, the data collection process is efficient and accurate. The Business Objects™ report-ing system has proven to be a useful and maneuverable tool for

creating data reports. We have found that a dedicated statistician is critical for successful data acquisition and analysis.

Future DirectionsProstate CRIS is designed to capture prospective clinical and

specimen information on patients with prostate cancer seen at our hospitals. Clinical data are linked to blood and tissue specimens from the same patients under a unified consenting process that al-lows us to comply with patient privacy and safety. A project linking Prostate CRIS and STIP to a Web-based tissue microarray database established at the University of Michigan and BWH is under way. More than 4000 patients have been entered into our database over the past 5 years, and we anticipate continued real-time data entry for every patient with prostate cancer seen at DFCI and BWH. Mostpatient with prostate cancer seen at DFCI and BWH. Mostprostate cancer seen at DFCI and BWH. Most importantly, we are beginning to use our database to ask clinical and scientific questions that address the needs of our patients.

AcknowledgementsThis work was supported by the Arthur and Linda Gelb Cen-

ter for Translational Research, the Richard Almy Fund, and the C. Brendan Noonan, Jr, Charitable Foundation.

References 1. Jemal A, Siegel R, Ward E, et al. Cancer statistics, 2006. CA Cancer J Clin 2006;

56:106-130. 2. Lubeck DP, Litwin MS, Henning JM, et al. The CaPSURE database: a method-

ology for clinical practice and research in prostate cancer. CaPSURE Research Panel. Cancer of the Prostate Strategic Urologic Research Endeavor. Urology 1996; 48:773-777.

3. Smith CV, Bauer JJ, Connelly RR, et al. Prostate cancer in men age 50 years or younger: a review of the Department of Defense Center for Prostate Disease Re-search multicenter prostate cancer database. J Urol 2000; 164:1964-1967.

4. Moul JW. Outcome research: prostate cancer databases. Urol Oncol 2002; 7:39-42.



development of an integrated prostate cancer research information system

Documents