1. Results of the Analysis of Suicidality in Pediatric Trials of Newer Antidepressants Psychopharmacologic Drugs Advisory Committee and the Pediatric Advisory Committee September 13-14, 2004 Tarek A. Hammad, MD, PhD, MSc, MS Senior Medical Reviewer Division of Neuropharmacological Drug Products Center for Drug Evaluation and Research, FDA

2. Outline Objective Data Domain Findings Limitations Summary

3. Objective To investigate the relationship between antidepressants and pediatric suicidality based on: 1.Adverse events reported 1.Suicidality item(s) scores in pertinent depression questionnaires 3 3

4. Data Domain 25 trials conducted in pediatric patients in nine drug development programs + TADS trial Drugs - number of trials: –SSRIs Citalopram (Celexa) - 2 Fluoxetine (Prozac) – 4 + TADS Fluvoxamine (Luvox) - 1 Paroxetine (Paxil) - 6 Sertraline (Zoloft) - 3 –Atypical antidepressants Bupropion (Wellbutrin) - 2 Mirtazapine (Remeron) – 1 Nefazodone (Serzone) - 2 Venlafaxine (Effexor XR) - 4 4 4

5. Data Domain, continued... Indications - number of trials –Major Depressive Disorder (MDD) – 15 (+ TADS) –Anxiety Disorders Obsessive Compulsive Disorder - 5 Generalized Anxiety Disorder - 2 Social anxiety Disorder/Social Phobia - 1 –Attention Deficit Hyperactivity Disorder – 2 Two trials excluded (Paxil 453-relapse prevention, Wellbutrin 41-uncontrolled) Trial year: 1983 to 2004 Duration: 4 to 16 weeks 5 5

6. Outline Objective Data Domain Findings –Outcomes based on adverse events –Outcomes based on suicidality scores Limitations Summary

7. Outcomes Based on Adverse Events

8. Main Outcomes Used in the Analysis OutcomesDescriptionNumber Outcome 1 Suicidal behavior, codes 1 & 2 n=33 Outcome 2 Suicidal ideation, code 6 n=45 Outcome 3 (primary outcome) Suicidal behavior or ideation (codes 1, 2, & 6) n=78 Outcome 4 Possible suicidal behavior or ideation (codes 1, 2, 6 + 3, 10 ) n=109 8 8

9. 21 “not eligible” (outside d-b) Disposition of Events as Classified by Columbia 427 AEs narratives evaluated 260 others, not pertinent to the analysis (Codes 7, 8, 9, 12) 167 potential events (Codes 1, 2, 3, 4, 5, 6, 10, 11) 141 unique patients (47 events in 21 patients, most severe chosen as [1 or 2] > 6 > 3 > [4 or 5] > 10) 120 “eligible” --> 109 “pertinent” (double-blind) Codes 1, 2, 3, 6, 10) (double-blind)

10. AdjudicationAdjudication of Sponsors’ Original Events 113 possibly suicide-related events -11 “not pertinent” 29 new events from broader search & classification + 22 pertinent events (5 “not eligible”, 2 “not pertinent”) -15 “not eligible” 109 pertinent events (codes 1, 2, 3, 6, 10) 87 pertinent events 10

11. Caveats! Possibility for chance finding –Post- hoc analyses –Multiple outcomes –Many sub-analyses Difficult to compare across drugs –Low power for individual trials –Differences in size of drugs’ databases –Potential role of differences in level of ascertainment of events and completeness of narratives between trials/development programs 11

12. Examining Effect Modification and Confounding

13. Examining Effect Modification  Analysis was done by trialdone by trial  Examining effect modification (interaction) was difficult due to small number of events  Variables examined were: Age group (6-11, 12-18) Gender History of suicide attempt at baseline  None was found to meaningfully impact the risk estimates 13

14. Examining Confounding  Analysis was done by trialdone by trial  Examining confounding effect of many variablesconfoundingmany variables Demographics variables Trial-related variables Disease-related variables Drug-related variables Psychiatric history  None was found to meaningfully impact the risk estimates 14

15. Suicidal behavior or ideation (codes 1, 2, & 6) by Drug

16. Celexa

17. 17 Placebo is worse Drug is worse

18. Paxil

19. 19

20. Prozac

21. 21

22. TADS trial GroupEvent code *Number of events PROZ (109) 1 (suicidal behavior) 6 (suicidal ideation) 2727 PBO (112) 6 (suicidal ideation)2 * Events were classified by the Columbia expert group 22

23. 23

24. Zoloft

25. 25

26. Effexor XR

27. 27 EFFEX (GAD, 396)

28. Overall relative risks (RR) of suicidal behavior or ideation (codes 1, 2, & 6) by drug DrugRelative Risk (95% CI), MDD trials Relative Risk (95% CI), all trials, all indications Celexa1.37 (0.53, 3.50) LuvoxNo MDD trials5.52 (0.27, 112.55) Paxil2.15 (0.71, 6.52)2.65 (1.00, 7.02) Prozac *1.53 (0.74, 3.16)1.52 (0.75, 3.09) Zoloft2.16 (0.48, 9.62)1.48 (0.42, 5.24) Effexor XR8.84 (1.12, 69.51)4.97 (1.09, 22.72) Remeron1.58 (0.06, 38.37) SerzoneNo events WellbutrinNo MDD trialsNo events 28 * Note that TADS data are added to Prozac

29. Trial Design Attributes The following trial design attributes were examined:trial design attributes –Location (North Am/non-NA) –Setting (inpatient/outpatient) –Presence of an active control arm –Sample size –Rate of discontinuation –Number of centers –Extensive screening –Exclusion of placebo respondents –Exclusion of treatment-resistant patients –Exclusion of baseline suicide risk –Exclusion of history of suicide attempt –Exclusion of homicide risk None was found to consistently explain the observed differences in the risk estimates between trials within/between development programs 29

30. Components of suicidal behavior or ideation (codes 1, 2, & 6) 1. Suicidal Behavior (codes 1 & 2) 2. Suicidal Ideation (code 6)

31. Overall relative risks of suicidal behavior or ideation (codes 1, 2, & 6) by drug in MDD trials DrugRelative Risk (95% CI), suicidal behavior (codes 1 & 2)suicidal behavior (codes 1 & 2) Relative Risk (95% CI), suicidal ideation (code 6)suicidal ideation (code 6) Relative Risk (95% CI), suicidal behavior or ideation (codes 1, 2, & 6) Celexa 2.23 (0.59, 8.46)0.75 (0.19, 2.95)1.37 (0.53, 3.50) Paxil 2.30 (0.67, 7.93)1.09 (0.24, 5.01)2.15 (0.71, 6.52) Prozac * 2.15 (0.50, 9.26)1.30 (0.59, 2.87)1.53 (0.74, 3.16) Zoloft 0.98 (0.17, 5.68)3.88 (0.44, 34.54)2.16 (0.48, 9.62) Effexor XR 2.77 (0.11, 67.10)7.89 (0.99, 62.59)8.84 (1.12, 69.51) Remeron No events1.58 (0.07, 38.37)1.58 (0.06, 38.37) 31 * Note that TADS data are added to Prozac

32. Sensitivity Analysis

33. Robustness of the risk estimates of suicidal behavior or ideation (codes 1,2, & 6) to event ascertainment: results of outcome 4 “possible suicidal behavior or ideation” OutcomesOverall RR (95% CI), all trials, all indications Overall RR (95% CI), SSRI MDD trials Suicidal behavior or ideation (codes 1,2, & 6) 1.95 (1.28, 2.98)1.66 (1.02, 2.68) Possible suicidal behavior or ideation (codes 1, 2, 6 + 3, 10) 2.19 (1.50, 3.19)1.91 (1.27, 2.89) 33 The signal was slightly altered Note that TADS data are added to all analyses

34. Analysis of Risk Difference (RD)

35. This analysis estimates the absolute increase in the risk of the event of interest due to treatment Risk in the drug grp - risk in the placebo grp Overall RD for SSRIs in MDD trials = 2 to 3%Overall RD Out of 100 patients treated, we might expect 2 to 3 patients to have some increase in suicidality due to short-term treatment, i.e., beyond the risk that occurs with the disease being treated 35

36. Outcomes Based on Suicidality Scores

37. Worsening of Suicidality Score (Outcome 6, n=434) Increase in the suicidality item(s) score of pertinent depression questionnaires relative to baseline, regardless of subsequent change Questionnaires used: HAM-D, CDRS-R, MADRS, and K-SADS 37 Emergence of Suicidality (Outcome 7, n=349) Same concept as above, but with normal baseline score

38. Worsening of Suicidality (Outcome6) Effect modification and confounding were examined  none was found

39. 39 Results By discontin.

40. Emergence of Suicidality (Outcome 7) Effect modification and confounding were examined  none was found

41. 41

42. 42 Other Analyses Additional sensitivity analysis Fixed-effect vs. random-effects Completers analysisCompleters Time-to-event analysis Kaplan-Meier survival estimatesKaplan-Meier Hazard functions for pooled SSRIs in MDD trialsHazard functions Potential “activation” syndromesyndrome Post-hoc power analysisanalysis

43. Limitations Post-hoc analyses with multiple outcomes involved, in addition to many sub-analyses. Therefore, caution is warranted in the interpretation of the findings Observed differences between drugs –Chance finding –True differences between drugs (i.e. no class effect) –Differences in size of drugs’ databases –Differences in level of ascertainment of events and completeness of narratives –Differences in trial design attributes Short-term exposure (4-16 weeks) 43

44. Limitations, continued… Medication noncompliance may have influenced the occurrence of the events of interest. However, the determination of noncompliance was suboptimal Observed rates of suicidality associated with the use of antidepressants might not reflect actual rates among patients in the general population Most trials were conducted with a flexible dosing scheme eliminating our ability to examine the dose effect 44

45. Summary of Findings –The broader search for adverse events in various drug development programs and the blinded classification process identified many new events and also eliminated several events that were not appropriately classified –There were NO completed suicides –Many individual trials had a RR of 2 or more for suicidality and some CIs of overall estimates did not include 1 –The sensitivity analyses did not yield a meaningful difference in the evaluation of the estimated risks 45

46. Summary of Findings, continued… –None of the examined covariates was found to be an effect modifier or to meaningfully impact the risk estimates as a confounder –Among the examined trial design attributes, none was found to consistently explain the observed differences in the risk estimates between trials –No signal was observed in outcomes based on the suicidality scores 46

48. Backup Slides

49. 49 Meta-analysis Meta-analysis estimates of overall effects  Dealing with zero cellszero cells  Trial was used as the unit of analysisunit –“fixed” effects models using the Mantel-Haenszel (M-H) method“fixed” effects models (M-H) method  Sensitivity analysis –“Possible” events added –Fixed-effect vs. random-effectsFixed-effect vs. random-effects  Completers analysis Completers

50. Fixed Effect vs. Random Effects In the fixed effect approach, the premise is that the real effect that we are trying to estimate is fixed, and the observed variation between trials is by chance In the random effects approach, the premise is that the real effect vary around an average within a distribution reflected in the variation observed between trials The idea is that the results of various trials are given a different weight based on the underlying premise The random effects approach usually results in a wider CI Results are reported both ways for comparison 50

51. Power Curves 51 Assuming 100 patient per arm Incidence in placebo group 1% 5%

52. Sample Size Calculation 52 Incidence in placebo group Assuming 80% power 1% 5%

53. Phases of the 141 relevant events Trial end8 31 120 2 8 1 4 3 2 53

54. Definition of “phases” based on the timing of AEs PhasesDescription Phase 1Event occurred in double-blind acute treatment phase or within one day of the end of this phase[1]. The end of trials with a tapering period was set to be at the beginning of the tapering period.[1] Phase 2Event occurred during a taper phase following the end of the double-blind period Phase 3Event occurred during the discontinuation phase--this phase was defined as 2 to 8 days after the cessation of medication for all drugs except Prozac where it was 2 to 31 days after the cessation of medication because it has a long half life and active metabolites. For an event to be classified in this phase, the patient must not have been taking drug at the time of the event Phase 4Event occurred between 2 and 8 days (2 and 31 days for Prozac) after the cessation of double-blind acute phase study medication and the patient had continued in an extension phase or started on a prescription anti-depressant Phase 5Event occurred between 9 and 31 days after the cessation of double-blind acute phase study medication and the patient had continued in an extension phase or started on a prescription anti-depressant (this category would not apply to Prozac patients) Phase 6Event occurred more than 30 days after the cessation of double-blind medication in the acute phase 54

55.  Pooling of data from different trials treating them as one large trial fails to preserve the randomization effect and might introduce bias and confounding  Maintaining the randomization guards against the foreseen (e.g. age and gender) and unforeseen (e.g. differences in medical practices or case ascertainment) imbalances between treatment groups  The issue of trial similarity is not only pertinent to having the same protocols, but is also pertinent to the implementation of those protocols (implementation of criteria of inclusion and exclusion, patients care in reality…etc) 55 Trial vs. Patient as the Unit of Analysis

56. DispositionDisposition of Sponsors’ Events 113 possibly suicide-related events (108+3+2) -11 taken out ( 2 events code 8 1 event code 4 5 events code 5 3 events code 11) 29 new events + 22 events added ( 1 event code 1 10 events code 6 4 events code 3 7 events code 10) + 5 outside d-b (2, 5, 3x 6) + 2 not pertinent (4, 11) -15 outside double-blind ( 6 events code 1 1 event code 3 1 event code 5 7 events code 6) 109 pertinent events 87 pertinent events (26 events code 1 6 events code 2 20 events code 3 35 events code 6)

57. Relationship Between Sponsors’ and Columbia’s Classifications ColumbiaSponsorTotal NoYes No event4418174435 Suicidal behavior (codes 1 & 2)13233 Suicidal ideation (code 6)103545 Suicidal behavior /ideation (codes 1, 2, & 6) 116778 Possible suicidal behavior/ideation (codes 1, 2, 3, 6, & 10) 2287109 Self-injurious behavior, non-suicidal (codes 4, 5, & 11) 2911 22 new events26 events out

58. Relationship Between Sponsors’ and Columbia’s Classification Sponsor Classification Change after Columbia Final count No event4420+264446 Suicidal behavior (codes 1 & 2) 32+133 Suicidal ideation (code 6) 35+1045 Possible events (code 3, 10) 20+1131 58 –Codes 1, 2, & 6 pooled as “suicidal behavior or ideation” and is the primary outcome

59. Dealing with Zero Cells in Meta-analysis –Trials with no outcomes in both arms were excluded For the primary outcome (outcome 3), four trials did not have any events (trials # 75 [Wellbutrin, ADHD], 141 & 187 [Serzone, MDD], and 396 [Effexor XR, GAD]) –Continuity correction: “0.5” was added to each of the 4 cells before proceeding with the analysis 59

60. Dealing with Zero Cells: Trial 676 60

61. Calculation of the weight used in the M-H method EventNo eventGroup size Drugabn1 Placebocdn2 The weight (W) assigned to the RR will be equal to: W= (c * n1)/N Where N = n1 + n2 61

62. What is “Confounding”? From the Latin confundere, to mix together “The distortion of the apparent effect of an exposure (e.g. an SSRI) on risk (e.g. suicidal behavior) brought about by the association with other factor[s] (e.g. history of suicide attempt) that can influence the outcome” A Dictionary of Epidemiology by John Last, 1995. History of Suicide Attempt Suicidal behaviorTreatment 62

63. What is “Confounding”?, continued... Confounding Principles: –Indirect (spurious or weaker than we think) association (A1) between exposure (E) and disease (D) that depends on A2 [and is weaker than direct association (A3) between confounder and disease] –Confounder (C) must associate with both E & D –More likely to explain small effects A1 A2A3 63 Treatment (E) Suicidal behavior (D) History of Suicide Attempt (C)

64. Demographics variables Age, gender, race, & BMI Trial-related variables Trial location (North America vs. not) Trial setting (inpatient vs. outpatient vs. both) Disease-related variables Baseline severity score Suicidality score at baseline Duration of illness prior to treatment Drug-related variables Discontinuation Erratic compliance (defined as: not taking drug as prescribed during RCT) Prior history of: Suicide attempt or ideation Psychiatric hospitalization Substance abuse Hostility or aggressive behavior Irritability or agitation Insomnia 64 List of Covariates

65. Completeness of Explanatory Variables Explanatory variables that were completely reported in all trials are age, gender, race, setting of trial, location of trial, baseline severity score, and all outcomes Variables that were notably missing in many trials were duration of illness prior to randomization (10), and history of psychiatric hospitalization (21), substance abuse (9), and hostility or aggressive behavior (8) 65

66. Suicidal Behavior or Ideation (Codes 1,2, & 6), Overall

67. Random-effects 1.75 (1.11, 2.76) 67

68. Suicidal Behavior or Ideation (Codes 1,2, & 6) by Indication and Drug Class

69. SSRIs as a Class in MDD trials: Celexa, Prozac*, Paxil, & Zoloft 68 events, 2033 patients, 11 MDD trials * Note that TADS data are added

70. 70 Random-effects 1.56 (0.94, 2.59)

71. 71 Random-effects 1.99 (0.58, 6.85)

72. Outcome 4 and Sponsor’s Events, Overall

73. 73

74. 74

75. Components of suicidal behavior or ideation (codes 1, 2, & 6) Suicidal Behavior (codes 1 & 2) Suicidal Ideation (code 6)

76. 76

77. 77

78. Trials Design Attributes

79. Trials Design Attributes: Celexa Trial design attributesTrial CIT_18Trial 94404 LocationNorth AmericaNon-North America SettingOutpatientInpatient/outpatient Excluded placebo responders YesNo Extensive screeningNo Excluded Tx. resistantYesNo Excluded baseline suicide risk YesNo Excluded history of suicide attempt YesNo 79

80. Trials Design Attributes: Paxil Trial design attributesTrial 329Trial 377Trial 701 Excluded placebo responders NoYesNo Extensive screeningYesNo Excluded tx-resistantNo Yes Excluded baseline suicide risk Yes Excluded history of suicide attempt No Other attributesActive controlLargest 80

81. Trials Design Attributes: Prozac Trial design attributesTrial HCCJTrial X065Trial HCJE TADS Excluded placebo responders Yes No Extensive screeningNoYes Excluded tx-resistantNo Yes Excluded baseline suicide risk YesNoYes Excluded history of suicide attempt No Other attributesEarly termination 81

82. Trials Design Attributes: Zoloft Trial design attributesTrial 1001Trial 1017 Excluded placebo responders (placebo lead-in) No Extensive screeningNo Excluded tx-resistantNoYes Excluded baseline suicide risk Yes Excluded history of suicide attempt Yes 82

83. Trials Design Attributes: Effexor XR Trial design attributesTrial 382Trial 394 Excluded placebo responders Yes Extensive screeningNo Excluded tx-resistantNo Excluded baseline suicide risk Yes Excluded history of suicide attempt No 83

84. Worsening of Suicidality (Outcome 6) Stratified by Premature Discontinuation: Completers Analysis

85. 85

86. 86

87. Suicidal behavior or ideation (codes 1,2, & 6) Stratified by Premature Discontinuation: Completers Analysis

88. 88

89. Overall RRs of suicidal behavior or ideation (codes 1,2, & 6) in completers and non-completers DrugRR and 95% CI in completers RR and 95% CI in non-completers Prozac *1.17 (0.30, 4.61)0.84 (0.29, 2.44) Paxil2.79 (0.47, 16.53)1.86 (0.70, 4.95) Zoloft0.34 (0.01, 8.16)1.35 (0.34, 5.40) Celexa0.94 (0.20, 4.50)1.67 (0.52, 5.33) Luvox2.85 (0.12, 67.68)4.20 (0.18, 97.89) Effexor XR3.12 (0.13, 75.39)6.22 (0.81, 47.94) RemeronNo events1.73 (0.07, 40.32) All SSRIs in MDD trials 1.08 (0.45, 2.60)1.40 (0.76, 2.56) 89 * Note that TADS data are NOT added to Prozac

90. Overall RRs of suicidal behavior or ideation (codes 1,2, & 6) by drug, stratified by history of suicide attempt at baseline

91. 91 12 events 14 events 5 events 8 events 39/66=59% Exclusion

92. Overall RRs of suicidal behavior or ideation (codes 1,2, & 6) stratified by history of suicide attempt at baseline by drug DrugRR and 95% CI in patients with no history RR and 95% CI in patients with history Prozac *0.91 (0.30, 2.72)0.92 (0.21, 4.14) Paxil1.36 (0.18, 10.35)2.13 (0.66, 6.88) Zoloft2.42 (0.36, 16.06)1.37 (0.18, 10.40) Celexa1.39 (0.30, 6.49)1.16 (0.39, 3.44) Effexor XR5.67 (0.69, 46.68)4.56 (0.52, 39.72) Remeron1.63 (0.07, 39.57)No events All SSRIs1.26 (0.60, 2.64)1.40 (0.73, 2.72) All drugs1.61 (0.83, 3.13)1.60 (0.86, 2.98) 92 * Note that TADS data are NOT added to Prozac

93. 93

94. 94

95. Potential “Activation” Syndrome Defined as Treatment-Emergent Hostility or Agitation

96. Overall relative risks of treatment-emergent agitation or hostility by drug in MDD trials DrugRelative Risk (95% CI), MDD trialsMDD trials Celexa1.87 (0.34, 10.13) Paxil7.69 (1.80, 32.99) Prozac *1.01 (0.40, 2.55) Zoloft2.92 (0.31, 27.83) Effexor XR2.86 (0.78, 10.44) Remeron0.52 (0.03, 8.27) Serzone1.09 (0.53, 2.25) All drugs1.79 (1.16, 2.76) 96 * Note that TADS data are NOT added to Prozac

97. 97

98. 98

99. Analysis of Risk Difference (RD): Suicidal behavior or ideation (codes 1, 2, & 6) and possible suicidal behavior or ideation (codes 1, 2, 3, 6, & 10)

100. 100

101. 101

102. 102

103. 103

104. Sensitivity Analysis

105. Robustness of the risk estimates of suicidal behavior or ideation (codes 1,2, & 6) to meta-analysis method (results of random effects models)random effects models Suicidal behavior or ideation (codes 1,2, & 6), Overall RR (95% CI), fixed effect Overall RR (95% CI), random effects Overall 1.95 (1.28, 2.98)1.75 (1.11, 2.76) SSRI MDD 1.66 (1.02, 2.68)1.56 (0.94, 2.59) Other indications 2.17 (0.72, 648)1.99 (0.58, 6.85) 105 The signal was slightly altered Note that TADS data are added to all analyses

106. Time to Event Analysis: Hazard Functions for All SSRIs in MDD Trials 57 events, 1812 patients, ten trials Note that TADS data are NOT added

107. 107 10 trials, 57 events, 1812 patients 4 events (1 drug and 3 placebo)

108. HR=1.45 (0.85, 2.48) vs. RR=1.41 (0.84, 2.37) PlaceboDrug 108

109. 109 HR 2.20 (1.04, 4.61) HR 0.20 (0.02, 1.80) HR 1.33 (0.50, 3.50) 10 trials, 57 events, 1812 patients

110. Time-to-Event Analysis by Trial

111. Time-to-Event Analysis: Trial 94404 111

112. Time-to-Event Analysis: Trial HCJE 112

113. Time-to-Event Analysis: Trial 377 113

114. Time-to-Event Analysis: Trial 329 114

115. Codes for Columbia University Classification 1: suicide attempt (n=27) 2: preparatory actions (n=6) 3: self-injurious behavior, intent unknown (n=24) 4: self-injurious behavior, no intent, primarily to affect circumstance (n=2) 5: self-injurious behavior, no intent, primarily to affect internal state (n=5) 6: suicidal ideation (n=45) 7: other: accident 8: other: psychiatric 9: other: medical 10: not enough information (n=7) 11: self-injurious behavior, no suicidal intent (n=4) 12: “other” 115