Antidepressants and Suicidality in Adults: Statistical Evaluation

1. Antidepressants and Suicidality in Adults: Statistical Evaluation Mark Levenson, Ph.D.* and Chris Holland, M.S. Statistical Safety Reviewers Quantitative Safety and Pharmacoepidemiology Assessment Team Division of Biometrics 6/CDER/FDA Psychopharmacologic Drugs Advisory Committee December 13, 2006 * Presenter

2. Outline • Objectives • Analysis plan • Populations • Endpoints • Methods • Results • Primary and secondary • Sensitivity • Subgroup • Summary

3. Objectives Primary Objective To estimate the effect of antidepressant drugs versus placebo on suicidality in adults in double-blind, randomized, placebo-controlled clinical trials Secondary Objective To explore the effect for various subgroups defined by subject-level and trial-level characteristics

4. Analysis Plan

5. Study Indication Groups • Major depressive disorder (MDD) • Other depressive disorders • Other psychiatric disorders • Behavioral disorders • Other disorders Non-MDD Indications

6. Analysis Populations Primary: “Psychiatric Indications” Major depressive disorder Other depressive disorders Other psychiatric disorders Secondary The indication groups considered individually (major depressive disorder, other depressive disorders, other psychiatric disorders, behavioral disorders, other disorders)

7. Primary Endpoint Suicidal Behavior and Ideation • Completed suicide • Suicide attempt • Preparatory acts • Suicidal ideation

8. Secondary Endpoints Suicidal Behavior • Completed suicide • Suicide attempt • Preparatory acts Suicidal Ideation Only • Suicidal ideation

9. Primary Analysis Method “Exact method” for common odds ratio • Stratified method • Handles low event counts and small trial sizes • Assumes a common odds ratio across trials • Does not make use of trials with no events

10. Sensitivity Analysis Traditional and model-based methods • Mantel-Haenszel odds ratio • With and without continuity correction • Logistic regression • Unconditional and conditional estimates Methods that allow “trial-to-trial” treatment variation “Random effects methods” • Generalized linear mixed model (GLMM) • DerSimonian-Laird

11. Sensitivity Analysis(Continued) Method that makes use of trials with no events • Mantel-Haenszel risk difference Bayesian methods • Encompass fixed- and random-effect models and hierarchical models • Make use of trials with no events • Kaizar et al. (2006) models

12. Subgroup Analysis Performed on subject- and trial-level characteristics • Age group • Gender • Race • Drug type: SSRI vs. non-SSRI • Location: North America vs. other • Setting: in-patient/out-patient vs. out-patient only

13. Results: Trial and Subject Summaries

14. Trial Indication Groups Psychiatric Indications

15. Location of TrialsPsychiatric Indications

16. Trial Duration Psychiatric Indications

17. Subject Characteristics:Psychiatric Indications • No notable differences between test drug subjects and placebo subjects for: • Age • Gender • Race • Baseline history of suicide attempts • Baseline history of suicide ideation • Treatment exposure • Subject (not subject-years) is unit of analysis

18. Events Psychiatric Indications

19. Suicidal Behavior and Ideation Unadjusted Rates

20. Suicidal Behavior and Ideation Psychiatric Indications Placebo: 0.72% of subjects with event Test Drug: 0.62% of subjects with event 174/295 = 59% trials had reported events

21. Results: Primary and Secondary Analyses

25. Results: Sensitivity Analysis

26. * * Per 1000 subjects

28. Results: Subgroups

29. Suicidal Behavior and Ideation Unadjusted RatesPsychiatric Indications

30. * * Reanalysisof FDA/Hammad 2004 data

31. * † * Per 1000 Subjects. † Reanalysis of FDA/Hammad 2004 data.

32. Additional Subjects with Suicidal Behavior and Ideation (Per 1000 Subjects)

34. Other Subgroups No notable differences in other subgroups • Gender • Race • Location of trial • Setting of care (in-patient vs. out-patient) • SSRI vs. non-SSRI drug class

35. Summary Primary analysis population and endpoint OR = 0.84 (95% CI: 0.69, 1.02) Clear pattern in the estimates with increasing age Other subgroups (gender, race, location, setting, drug class) do not have notable effect Results are not sensitive to method