Issues in Projecting Increased Risk of Cardiovascular Events to the Exposed Population

1. Issues in Projecting Increased Risk of Cardiovascular Events to the Exposed Population Robert T. O’Neill, Ph.D. Director, Office of Biostatistics February 18, 2005

2. Issues for Discussion • The approach we used (one of several) and why? Rofecoxib as an example - data sets available • What do we know from the rofecoxib randomized controlled trials (RCTs) • Data sources for projections of events to the exposed population • Model assumptions • A range of estimates for different scenarios • Uncertainties in the projection strategies • Concluding remarks

3. Approach to estimation of excess eventsassociated with exposure • Estimate the absolute difference in cumulative incidence rates of events from the VIGOR and APPROVE studies (event definitions) at monthly intervals for 1 to 36 months of exposure - control and dose groups assumed • Estimate from IMS data the total prescriptions (usage) for 25mg and 50 mg between years 1999 thru 2004 • Define episodes of exposure from the Caremark database and estimate approximate patient numbers with different durations of successive episodes • Use a ratio multiplier to project up from Caremark database to IMS data base - provides U.S. estimate of number of patients with various successive episodes

4. Approach to estimation of excess events associated with exposure (cont.) • Multiply this estimate of numbers of patients with exposure episodes by the estimate of increased risk (incidence) to obtain number of excess events • Add estimates from all mutually exclusive episode groups to get total excess events over the control background risk (rate) • Discuss statistical variability of estimate (confidence interval) • Repeat for different sets of assumptions about control rates, relative risk by duration, and risk factors

5. For chronically used drugs, the duration of person exposure and the calculated risk at that time matterAs does the appropriate estimate of risk and increased risk at defined intervals of time or durations of exposure

6. What did we learn from two Rofecoxib clinical trials?(Studies were not prospectively designed for CV events)

7. Replication • Do we believe there is an effect to project? • Is there consistency of effects in different studies on same endpoints? • Alzheimer’s Disease • Mild Cognitive Impairment

8. Event Definitions Matter ! • Event rates are a function of event definitions and how many different subtypes are included in the definition • Blindly adjudicated vs. investigator reported events will (can) make a difference risk estimates and risk ratios

9. Event definitions in VIGOR & APPROVeWe will examine two of them: • Confirmed thrombotic events within 14 days of last dose of study drug: • Myocardial infarction (fatal and non-fatal), or sudden cardiac death APTC events within 14 days of last dose of study drug

10. VIGORPatients withdraw from exposure (trial) early !Impact on event rates and comparisons Completed: rofecoxib: 2862 / 4047 ( 70.7% ) naproxen: 2880 / 4029 ( 71.5% ) Discontinued (Reasons): Clinical AE’s: rofecoxib: 563 (13. 9%); naproxen: 492 (12.2%) Lack of efficacy: rofexocib: 256 (6.3%); naproxen: 263 (6.5%) Other: laboratory AE’s, moved, lost, withdrew consent

12. RR = 2.28, P =0.0035 (1.31,3.97 )

13. Myocardial infarction contributes most to the composite endpoint in the VIGOR Study

14. (Huque)

15. Data support risk and increased risk, increases with longer duration of exposure

16. RR = 1.92 , p = 0.008) (1.19 , 3.11)

18. Risk and increased risk, increases with exposure duration and time (Merck, 2005 )

19. (Merck, 2005 )

20. 25 mg Tablet strength - Increased Risk (Difference in rates)

21. 50 mg Tablet strength - Increased Risk (Difference in rates)

22. Assumptions for increased risk beyond 12 months

23. Assumptions for increased risk beyond 12 months: Using APPROVe estimates for 12 to 36 months

24. Projections to the Population Projections to the population involve: • Estimating how many persons are exposed • For how long • To what increased risk, and • Against what background rate We will rely on drug use prescription data – many assumptions are made

25. National Prescription Data Sources • IMS Health, National Prescription Audit Plus • Total number of prescriptions dispensed from U.S. retail pharmacies • Does not provide patient demographic information • IMS Health, National Disease and Therapeutic Index • Survey of 2000 - 3000 office-based physicians • Projections may be unstable given small sample size

26. Total number of prescriptions dispensed for Vioxx from 1999 - 2004 in outpatient settings IMS Health, National Prescription Audit Plus™, Years 1999 – 2004, Extracted January 2005, Original File: 0501viox.dvr

27. Indications for Use of Vioxx: 1999-2003

28. Estimating Patient Level Duration of Rofecoxib ExposureThe Caremark DataAn attempt to estimate how many persons are exposed for how long a duration to different tablet strengths

29. Source of Data for Defining Episodes of Vioxx Use • Dataset obtained through Caremark™ • Large pharmacy benefits manager • Covers 70+ million patient lives and ~25% of total U.S. prescriptions • Dataset included all Vioxx claims (n=1,969,285)from 1 January 2002 through 31 December 2004 • Patients must have been in Caremark™ for entire study time period (~25% of all claims)

30. Analytical Caremark™ Dataset • We excluded prescriptions claims where: • Strength = 12.5 mg • Patients who were younger than 18 years • Suspension formulation • We also removed nonsensical data • Days supply dispensed equal to zero • Patients aging more than 2 years between subsequent prescriptions • Patients whose gender changed between prescriptions • Final dataset included 1,736,639 claims and 403,293 patients

31. Limitations of Caremark™ Data • Not nationally representative • Limited to insured persons with prescription drug coverage • Under-representation of elderly persons (due to Medicare coverage) • Poor, indigent persons underrepresented • Quality of prescription claims data • Data not collected for research purposes

32. Definition of exposure episode One exposure-episode Two exposure-episodes Caremark 36-month window) 1/1/2002 6/1/2002 12/31/2004 Exposure-episode included in analysis (≥ 3 months observed) Exposure-episode excluded (< 3 months observed)

33. Definition of Continuous Episode of Vioxx Use

36. Projection Ratio R IMS Rx relative increase over Caremark™ Rx: Assume constant multiplier R = (Total IMS Rx’s in 1999-2004) /(Total Rx’s in Caremark)

37. What is the increased risk in exposure sub-populations? • By dose • By duration of exposure • No other factors possible to examine: • Age • Gender • Indication, or • Other risk factors

38. CalculationAverage Increased Number (Events) Attributable to Vioxx • For each dose strength and length of exposure-episode L NL = (# of exposures) x (Increased Incidence) x R • To obtain the increased events for each dose strength Sum over all exposure lengths Sum over both dose strengths to obtain total

39. Estimates of Population Increased Events:MI & Sudden Cardiovascular Deaths (Events) 25 mg = 28,686 (95% CI upper limit = 100,159) 50 mg = 3,732 (9,256) – constant projection = 4,407 (w / APPROVe Projection ) Both 25mg and 50mg strengths combined 32,418 (109,415) events (w/constant projection) 33,093 (w/APPROVe Projection)

40. Estimates of Population Increased Events:Confirmed thrombotic events within 14 days of last study dose 25 mg = 41,406 (95% CI upper limit 145,046) 50 mg = 6304 (15,380) for constant projection = 8034 for APPROVe projection Both 25mg and 50mg strengths combined 47,710 (160,426) events for constant projection 49,440 events for APPROVe Projection

41. Sensitivity of the Estimate Estimates of the number of increased events are influenced by: • Validity of the assumption to use clinical trial incidence for population incidence and comparability of populations (cross trial data) • Relevance of observational data odds ratio to short term use, hazard rate • The Caremark inclusion criteria for left censored exposure-episode and the cutoff time of the first prescription • The variability of the ratio projected from Caremark to US population (nationwide) and its validity • Variability of the estimates of increased incidence • Assumption of projection from 13-month trial exposure data to 36-month of exposure

42. Where do some of the differences in estimates come from? • Assumption of constant hazard rate calculated as events per 100 person years - can overestimate early 1 - 2 month hazard, underestimates later ( > 8 months, 1.5 year ) hazard • Assumption of constant (proportional hazard ratio over exposure time) • Estimate of very large number of people exposed for short (1-2 months) episodes - RCT vs. Caremark vs. Actual US pop

43. Exposure Length Distribution Difference

46. Increased Incidence of MI & Sudden Cardiovascular Death by Kaplan-Meier Diff and Average Patient-Month Diff • 25 mg • By Kaplan-Meier diff 28,686 • By Patient-Month diff 39,187 • 50 mg • By Kaplan-Meier diff 7,542(w/APPROVe Projection) • By Patient-Month diff 5,843 • 25 mg & 50 mg Combined • By Kaplan-Meier diff 36,228 • By Patient-Month diff 45,030

47. Summary • Goal of the projection effort: • Provide a framework for considering estimates • Provide a range of estimates to show how assumption dependent each is - not to emphasize any one estimate • To describe the logic, process, assumptions and uncertainties in the estimates • To focus attention on duration of exposure and time related risk • Given the data limitations and uncertainties, argue that there is no best approach nor estimate and it is difficult to choose an estimate without qualifying its relevance

48. Acknowledgements • ODS Team • Laura Governale • Gerald Dal Pan • Aaron Mendelsohn • Judy Staffa • OB Team • Joanne Zhang • Yi Tsong • Mohammad Huque