Statistical Review of the Observational Studies of Aprotinin Safety Part II: The i3 Drug Safety Study

1. Statistical Review of the Observational Studies of Aprotinin Safety Part II:The i3 Drug Safety Study CRDAC and DSaRM Meeting September 12, 2007 P. Chris Holland, M.S. Quantitative Safety and Pharmacoepidemiology Group Office of Biostatistics

2. Outline • Objective • Background • Statistical Methods • Results • Aprotinin vs. Aminocaproic Acid and Tranexamic Acid • Comparisons between FDA and i3 results • Aprotinin vs. No Treatment • Summary

3. Objectives • To examine the statistical robustness of the conclusions from the i3 Drug Safety study by implementing an alternative methodology • To compare aprotinin patients to patients who receive no IV antifibrinolytic (AF) with respect to: • All-cause in-hospital mortality • Cardiovascular outcomes • Cerebrovascular outcomes • Renal outcomes

4. Background Preliminary Report Title: Mortality and Cardiovascular and Renal Outcomes in Recipients of Aprotinin, Aminocaproic Acid and Tranexamic Acid during CABG Surgery Report on Computerized Inpatient Data from the Premier Perspective Comparative Database

5. Sets of Analyses

6. Data Provided to and Analyzed by FDA Data Set 1: 66,435 patients in the preliminary study* Data Set 2: 69,176 patients received no IV AF during CABG 35,719 patients in aminocaproic acid group 1,358 patients in tranexamic acid group 17,588 patients received IV AFs after CABG 51,588 patients received no IV AF during or after CABG 29,358 patients in aprotinin group 37,077 patients in other IV AF group * Only includes patients who received a minimum sufficient dose of an IV AF agent

7. Study Limitations • Accuracy of derivations for outcomes and covariates has not been evaluated • For example, no medical chart review • Unavailable Covariates • Creatinine levels, hemoglobin levels, platelet counts, use of aspirin, use of heparin, use of antithrombotics, etc. • Outcome Definitions • Not all outcomes are explicitly collected • (e.g., dialysis is a surrogate for renal failure) • Heart failure and renal failure not evaluable on the day of CABG surgery • Unable to determine whether they occur before or after CABG • Only in-hospital deaths are available

8. Statistical Methods

9. Statistical Methods • Propensity score (PS) methods were used for all FDA-conducted statistical analyses • Subgroup and other sensitivity analyses were conducted in order to assess the statistical robustness of the overall results • Sensitivity: • Time adjusted rates (rates per patient weeks) • Analyses of patients with >1 and >3 hospital days prior to surgery • Analyses of patients in PS Deciles 1-9 • Subgroups: • Age (>65, <65) • Gender (Male, Female) • Race (White, Other)

10. Propensity Score Analysis • Stratified analysis: • Hospital characteristics that are predictive of aprotinin use were chosen to create strata • Resulted in: • 8 strata for aprotinin vs. other IV AFs • 4 strata for aprotinin vs. no treatment • PS modeling performed within each strata • allows for better PS estimates • PS deciles constructed within each stratum • Final estimates are weighted averages across strata (hospital characteristics and deciles)

11. Patients Excluded for FDA Analysis • Patients with 0 days of follow-up excluded from: • Heart failure outcome analysis • Renal failure outcome analysis • Patients who met the criteria for pre-existing renal failure excluded from: • Renal failure outcome analysis

12. Results

13. Demographics * p<0.001 compared to aprotinin

14. Selected Baseline Risk Factors * p<0.001 compared to aprotinin

15. Unadjusted Outcome Rates

16. Results FDA Analysis of Aprotinin vs. Aminocaproic Acid and Tranexamic Acid and Comparison with Preliminary Report Results

17. Risk Ratios – All Outcomes

18. Comparisons of Estimates forAprotinin vs. Other IV Antifibrinolytics • i3 analysis based on multivariate logistic regression results (OR used to estimate RR) • Crude and FDA analyses of renal failure excluded patients with pre-existing renal failure and with zero days of follow-up • Crude and FDA analyses of heart failure excluded patients with zero days of follow-up

19. Results FDA Analysis of Aprotinin vs. No Treatment

20. Strata Sample Sizes

21. Propensity Score Estimation and Assessment of Overlap and Balance • PS scores were estimated separately for each stratum • Patients were divided into PS deciles within each stratum • Assessment of Overlap and Balance: • PS overlap between treatment groups within each decile was good • Analyses showed good balance between treatment groups with respect to most risk factors

23. Good overlap between groups

24. Selected Baseline Risk Factors Pre- and Post-Adjustment

25. Risk Ratios – All Outcomes

26. Risk Differences – All Outcomes

27. Risk Ratios – Death

28. Risk Ratios – Renal Failure

29. Sensitivity Analyses – Risk Ratios

30. Summary

31. Summary • Compared to other IV antifibrinolytics and no treatment, aprotinin is associated with increased risks of: • Death • Renal failure • Heart failure • Stroke • Study limitations: • Unavailable confounders for which adjustments could not have been made • Accuracy of the derivations for covariates and outcomes has not been evaluated • No medical chart review

32. Acknowledgements Mark Levenson, Rita Ouellet-Hellstrom, Allen Brinker, George Shashaty, Tiffany Brown, Dwaine Rieves, Kathy Robie-Suh, Karen Weiss, Gerald Dal Pan