1. Databases or Registries? Points to Consider
Mary Lou Skovron, DrPH
Group Director, Global Epidemiology
Bristol-Myers Squibb
FDA/Industry Statistics Workshop
September 29, 2006

2. Overview Claims databases Registries Examples Conclusions Types
Advantages
Limitations
Registries
Examples
Conclusions

3. Claims Databases

4. Claims Databases Types
Open-Plan eg UHC, PharMetrics
Pharmacy-based
Practice-based
Hospital-based
HMO, eg Kaiser-Permanente
Government eg Medicare, Medicaid

5. Claims Databases: Advantages
Already exist, accruing patient information…useful when a relatively quick answer is needed
Potentially large populations from which to draw treated patient and comparison samples…statistical power not usually an issue
May include subgroups not included in clinical trials…expand knowledge

6. Claims Databases: Limitations
ICD-9 coding… potential for misclassification
Limited sensitivity/specificity
Coding affected by reimbursement
Short ‘residence’ in the database
Clinical, lab, imaging data not usually present
Rare events in rare diseases require huge “electronic” populations to identify adequate numbers treated
May not represent important sub-populations
Surveillance bias and channeling bias
Inpatient drug exposures not usually recorded

7. Claims Databases: Application
Acute events (short-term events)
Events that come to medical attention, eg CVA
Validated algorithm to identify indication and event of interest OR medical record review to verify events
Statistical approaches for confounders (eg propensity scores, instrumental variables, risk factor scores, multivariate analyses)

8. Registries

9. Registries
“A systematic collection of defined events or product exposures in a defined patient population for a defined period of time”1
“A registry per se is not a study. It is an organized collection of data in humans within a particular disease group or other special group…”2
1Arlett P, Moseley J, Seligman PJ p 119 in Pharmacoepidemiology Fourth Edition, Strom BL, ed 2005
2 CIOMS V Section II.h.

10. Registries: Types
Drug/Device Registry: Includes subjects receiving the drug or device regardless of indication
Disease Registry: Includes patients with the disease regardless of drug or device exposure

11. Registries: Strengths
Richer data than in electronic databases: Patient SES, history, treatment, clinical data can be collected
Define encounter frequency and follow-up duration
Event ascertainment does not depend on ICD-9 coding
Can address additional questions in the data

12. Registries: Limitations
Accrual can be slow if insufficient sites engaged
Generalizability must be established
Practical limits on number of patients followed

13. Registries: Application
Long-term outcomes
Rare diseases
Potential confounders important
Multiple objectives

14. Usefulness of Registries
Characteristics of patients in the target population for the new drug
Clinical course of the disease
Treatment patterns, health care utilization
Frequency of adverse events

15. Registry Lifecycle
Early: Describe patient demographics, clinical characteristics, practice patterns (usually cross-sectional analyses); incidence of AEs with short lag times
Intermediate: Analyze relationships pf patient characteristics, treatment with outcomes; incidence of less common AEs, AEs with longer lag times; assess risk factors for AE incidence
Advanced: Evaluate changes in practice patterns; impact on outcomes and AE incidence; assess rare AE incidence

16. Registry Quality
DeCIDE Network currently developing a reference document on developing, conducting and evaluating registries
Sponsored by AHRQ
Document in draft, Outline available on the web
Report currently in draft

17. Examples

18. Example: Cox-2 Inhibitors and Cardiac Events
Cox-2 use frequent in population
Primary care drug
Cardiac events not rare in target population
Short-term (< 2 years) events
Clinical history important

19. One Solution Cox-2 Inhibitors and Cardiac Events
Claims database analysis1
Advantages:
Data already accrued when study need identified
Large population
Limitation offset
Multivariate regression to control confounding
Verified cardiac events by chart review
Remaining limitations
Under-represented > 65 yo
1Velentgas P et al 2006

20. Key Feature of the Solution
Numbers readily available:
Exposure: ~ 425,000 eligible subjects available for study
at least one dispensing of the 5 study medications during preceding 18-month period
first dispensing after minimum of six months without any of the medications
Endpoint: ~ 725 confirmed MI/ACS
Verification of endpoints
Medial record review applying accepted criteria

21. Example: Thrombolytics And Bleeding
Important focus: subpopulations eg ethnicity, gender, age
Short-term event
Benefit and risk
Hospital-based treatment

22. Solution Thrombolytics and Bleeding
Registry approach: National Registry of Myocardial Infarctions
Salient strengths:
Clinical and lab data ascertained
Data from medical records
Limitation Offset
Validated completeness against another data source
Large number of hospitals to assure adequate subpopulations
Remaining limitations
Short-term data cannot answer long-term questions

23. Key Feature of the Solution
Large number of hospitals participate order to gather data on~200,000 MIs per year
Rich patient, clinical, treatment and outcome information
Answered the question about safety in subgroups under-represented in the clinical trials
In its > 10 year lifecycle has contributed to broad understanding and improvement of medical practice
External investigators can propose research; external advisory group reviews and approves all research

24. Example: Safety of Orencia®, a New Biological for RA
Short-term (infections) and longer-term (NHL and other malignancies) events
Low general population prevalence of RA candidates for biologics treatment
RA a specialty-treated disorder
Proactive approach

25. Solution: Complementary Approaches
Claims database study in UHC data
Event validation
Large pool of potential comparators
Population treatment patterns
Infections, other possible short-term events
Registry building on the independently conducted National databank for Rheumatic Diseases
Large pool of participating rheumatologists
5,000 Orencia® initiators comparison to >=15,000 initiators/switchers of other treatments
Patient self-report and event verification
Enrollment and retention enhancements
Short and long-term events

26. Key Features of the Solution
External scientific advisory group
Individual protocols
Statistical analysis plans
Interpretation of results including approaches to integration

27. Conclusions

28. Conclusions One size does not fit all: evaluate options
Registries: a useful alternative to electronic databases
Consider complementary approaches

29. Useful References
Strom BL, ed; Pharmacoepidemiology 4th Edition; UK; John Wiley and Sons Ltd;2005
AHRQ DeCIDE Network upcoming publication

30. Thank You