`Food and Drug Administration
`Center for Drug Evaluation and Research
`Office of Translational Sciences
`Office of Biostatistics
`
`STATISTICAL REVIEW AND EVALUATION
`
` ANTIEPILEPTIC DRUGS AND SUICIDALITY
`
`Drug Class:
`Drug Names (NDA
`Numbers):
`
`Antiepileptic drugs
`Carbamazepine (21-710)
`Divalproex (18-723, 19-680, 21-168)
`Felbamate (20-189)
`Gabapentin (20-235, 20-882, 21-129, 21-216)
`Lamotrigine (20-241, 20-764)
`Levetiracetam (21-035, 21-505, 21-872)
`Oxcarbazepine (21-014, 21-285)
`Pregabalin (21-446)
`Tiagabine (20-646)
`Topiramate (20-505, 20-844)
`Zonisamide (20-789)
`Indication(s):
`Epilepsy, psychiatric disorders, other
`Date:
`23 May 2008
`Biometrics Division:
`Division of Biometrics 6
`Statistical Reviewer:
`Mark Levenson, Ph.D.
`Statistical Team Leader: C. George Rochester, Ph.D., RAC
`Medical Division:
`Division of Neurology Products
`Clinical Team:
`Evelyn Mentari, MD
`Alice Hughes, MD
`John Feeney III, MD
`Marc Stone, MD
`Jacqueline Ware, Pharm.D., RAC
`
`Project Manager:
`
`Keywords: Epilepsy, psychiatric, bipolar, suicide, suicidality, meta-analysis
`
`ARGENTUM Exhibit 1128
` Argentum Pharmaceuticals LLC v. Research Corporation Technologies, Inc.
`IPR2016-00204
`
`1
`
`Page 00001
`
`
`
`2
`
`TABLE OF CONTENTS
`List of Tables ...................................................................................................................... 3
`List of Figures..................................................................................................................... 4
`Executive Summary............................................................................................................ 5
`1.1
`Overview............................................................................................................. 5
`1.2
`Findings............................................................................................................... 5
`1.3
`Conclusions......................................................................................................... 6
`Introduction................................................................................................................. 7
`2.1
`Background......................................................................................................... 7
`2.2
`Review Objectives .............................................................................................. 7
`3 Data Sources ............................................................................................................... 7
`3.1
`Data Requests...................................................................................................... 7
`3.2
`Trial Summary .................................................................................................... 9
`4 Methods..................................................................................................................... 12
`4.1
`Endpoints .......................................................................................................... 12
`4.2
`Analysis Population .......................................................................................... 13
`4.3
`Subgroups and Special Populations.................................................................. 13
`4.4
`Statistical Methods............................................................................................ 14
`Patient Summary....................................................................................................... 16
`5.1
`Drugs and Demographics.................................................................................. 16
`5.2
`Discontinuation and Duration........................................................................... 20
`Findings..................................................................................................................... 22
`6.1
`Suicidal Behavior or Ideation ........................................................................... 22
`6.2
`Suicidal Behavior and Suicidal Ideation........................................................... 26
`6.3
`Sensitivity Analysis .......................................................................................... 27
`6.4
`Exploratory Analysis ........................................................................................ 32
`Findings in Special/Subgroup Populations ............................................................... 35
`7.1
`Drug Groups...................................................................................................... 35
`7.2
`Trial Indication.................................................................................................. 37
`7.3
`Demographics ................................................................................................... 39
`Post-Hoc Analyses.................................................................................................... 45
`8.1
`Lamotrigine Additional Data ............................................................................ 45
`8.2
`Alternative Age Subgroups............................................................................... 47
`Summary and Conclusions ....................................................................................... 47
`9.1
`Review Summary.............................................................................................. 47
`9.2
`Conclusions....................................................................................................... 49
`10
`References............................................................................................................. 49
`
`5
`
`6
`
`7
`
`8
`
`9
`
`2
`
`Page 00002
`
`
`
`LIST OF TABLES
`Table 1: Suicidality Events and Codes. .............................................................................. 8
`Table 2: Antiepileptic Drugs under Review. ...................................................................... 9
`Table 3: Indication Categories.......................................................................................... 10
`Table 4: Trials by Comparator Type and Drug................................................................. 11
`Table 5: Trials by Indication Group and Therapy (Monotherapy, Adjunctive Therapy,
`Other). ............................................................................................................................... 12
`Table 6: Patients by Treatment Arm and Comparator Type............................................. 16
`Table 7: Patients by Treatment Arm and Drug, Placebo-Controlled Trials. .................... 17
`Table 8: Patients by Indication Group and Drug, Placebo-Controlled Trials................... 18
`Table 9: Demographics by Treatment Arm, Placebo-Controlled Trials........................... 19
`Table 10: Patients by Drug Class and Treatment Arm, Placebo-Controlled Trials.......... 20
`Table 11: Patient Treatment Discontinuation and Duration by Treatment Arm, Placebo-
`Controlled Trials. .............................................................................................................. 21
`Table 12: Events by Type and Treatment Arm, Placebo-Controlled Trials. .................... 22
`Table 13: Suicidal Behavior or Ideation Events and Patients by Drug, Placebo-Controlled
`Trials. ................................................................................................................................ 23
`Table 14: Suicidal Behavior or Ideation Hazard Estimates by Treatment Arm, Placebo-
`Controlled Trials. .............................................................................................................. 32
`Table 15: Events from Patients with Multiple Events, Placebo-Controlled Trials........... 34
`Table 16: Placebo and Drug Suicidal Behavior or Ideation Event Rates and Risk
`Difference by Indication, Placebo-Controlled Trials........................................................ 38
`
`3
`
`Page 00003
`
`
`
`LIST OF FIGURES
`Figure 1: Mean Trial Duration by Treatment Arm, Placebo-Controlled Trials................ 21
`Figure 2: Suicidal Behavior or Ideation Odds Ratio Estimates, Placebo-Controlled Trials.
`........................................................................................................................................... 25
`Figure 3: Suicidal Behavior versus Suicidal Ideation Odds Ratio Estimates, Placebo-
`Controlled Trials. .............................................................................................................. 26
`Figure 4: Suicidal Behavior or Ideation Risk Difference Estimates, Placebo-Controlled
`Trials. ................................................................................................................................ 28
`Figure 5: Suicidal Behavior or Ideation Rate Ratio Estimates, Placebo-Controlled Trials.
`........................................................................................................................................... 31
`Figure 6: Kaplan-Meier Suicidal Behavior or Ideation Incidence Curves by Treatment
`Arm, Placebo-Controlled Trials........................................................................................ 33
`Figure 7: Suicidal Behavior or Ideation Odds Ratio Estimates by Drug Group, Placebo-
`Controlled Trials. .............................................................................................................. 36
`Figure 8: Suicidal Behavior or Ideation Odds Ratio Estimates by Indication Group,
`Placebo-Controlled Trials. ................................................................................................ 37
`Figure 9: Suicidal Behavior or Ideation Odds Ratio Estimates by Age Group, Placebo-
`Controlled Trials. .............................................................................................................. 40
`Figure 10: Suicidal Behavior or Ideation Odds Ratio Estimates by Gender, Placebo-
`Controlled Trials. .............................................................................................................. 41
`Figure 11: Suicidal Behavior or Ideation Odds Ratio Estimates by Race Group, Placebo-
`Controlled Trials. .............................................................................................................. 42
`Figure 12: Suicidal Behavior or Ideation Odds Ratio Estimates by Setting, Placebo-
`Controlled Trials. .............................................................................................................. 43
`Figure 13: Suicidal Behavior or Ideation Odds Ratio Estimates by Location, Placebo-
`Controlled Trials. .............................................................................................................. 44
`Figure 14: Suicidal Behavior or Ideation Odds Ratio Estimates, Placebo-Controlled and
`Low-Dose-Controlled Trials............................................................................................. 45
`Figure 15: Suicidal Behavior or Ideation Odds Ratio Estimates with Additional
`Lamotrigine Data, Placebo-Controlled Trials................................................................... 46
`Figure 16: Suicidal Behavior or Ideation Odds Ratio Estimates by Post-Hoc Age Group,
`Placebo-Controlled Trials. ................................................................................................ 47
`
`4
`
`Page 00004
`
`
`
`EXECUTIVE SUMMARY1
`1.1 Overview
`The Food and Drug Administration (FDA) concerned about the potential for elevated risk
`of suicidality (suicidal behavior or ideation) from the use of antiepileptic drugs carried
`out a meta-analysis of 11 drugs. Antiepileptic drugs are also used for indications other
`than epilepsy including psychiatric disorders.
`
`In March 2005, FDA sent letters to sponsors of antiepileptic drugs requesting that they
`submit data from placebo-controlled trials for the FDA to review the possible association
`of suicidality events and antiepileptic drugs. Letters in July 2005, May 2006, and January
`2007 requested additional information to obtain the data necessary for the review. The
`letters specified detailed instructions for the identification of suicidality events and the
`format of the data to be submitted.
`
`Prior to the analysis of the data, medical reviewers in the Division of Neurology and
`statistical reviewers in the Quantitative Safety and Pharmacoepidemiology Group agreed
`upon the definition of the research objectives, endpoints, study population, and subgroups
`and upon the specification of the statistical methods. These elements were incorporated
`into a statistical analysis plan prior to the review. The statistical methods maintained the
`integrity of placebo-controlled trials. This allowed for trials to have different background
`rates of events.
`
`1.2 Findings
`There were 199 placebo-controlled trials consisting of 27,863 patients in drug arms and
`16,029 patients in placebo arms from 11 drugs that formed the primary analysis
`population.
`
`The average age of patients was 42 years. The majority of patients were female (55%),
`white (79%), and from North American locations (61%). The placebo patients had
`statistically higher treatment duration (77 days for placebo versus 73 days for drug).
`There were no statistical differences among the baseline characteristics of the drug and
`placebo patients for age, gender, race, and location.
`
`There were 4 completed suicides among drug patients and none among placebo patients.
`The majority of suicidality events for both drug and placebo patients were Suicidal
`Ideation. The second most frequent type of event was Suicide Attempt. Without adjusting
`for differences among trials, 0.37% of the drug patients had a Suicidal Behavior or
`Ideation event versus 0.24% of the placebo patients.
`
`1 This review replaces the March 5, 2008 version. Two small discrepancies in the data have been corrected
`for this version.
`
`5
`
`Page 00005
`
`
`
`Overall, patients who received an antiepileptic drug had statistically significant increased
`risk of Suicidal Behavior or Ideation relative to placebo patients. The estimated overall
`odds ratio (OR) of a drug patient experiencing a Suicidal Behavior or Ideation event
`versus a placebo patient was 1.80 (95% CI: 1.24, 2.66). The results for individual drugs
`were generally consistent with the overall result. Suicidal Behavior had a larger estimated
`odds ratio [2.92 (95% CI: 1.44, 6.47)] than Suicidal Ideation [1.45 (95% CI: 0.93, 2.30)].
`Sensitivity analyses showed that the results were robust to statistical methods and
`differences in the treatment durations between the treatment groups.
`
`Indication and location appeared to have the largest effects on the odds ratio among the
`subgroups considered. The epilepsy indication subgroup had the largest estimated odds
`ratio [3.53 (95% CI: 1.28, 12.10)] compared to the psychiatric indication subgroup [1.51
`(95% CI: 0.95, 2.45)] and the other indication subgroup [1.87 (95% CI: 0.81, 4.76)].
`However, the psychiatric indication subgroup had the largest placebo risk and the risk
`difference for the psychiatric indications subgroup was the largest. The estimated odds
`ratio for the Non-North American subgroup [4.53 (95% CI: 1.86, 13.18)] was notably
`larger than that of the North American subgroup [1.38 (95% CI: 0.90, 2.13)].
`
`The higher risk of events for the drug-treated patients was observed as early as 1 week
`from initiating treatment until at least 24 weeks. After 24 weeks, it was not possible to
`draw conclusions due to the scarcity of data beyond 24 weeks.
`
`There was no obvious pattern in the drug effect with respect to age subgroups. Likewise,
`there were no patterns with respect to subgroups based on gender, race, setting, and
`prespecified drug groups (sodium channel blocking, GABAergic and GABAmimetric,
`and carbonic anhydrase inhibitors).
`
`1.3 Conclusions
`In conclusion, antiepileptic drugs are associated with increased risk of suicidality relative
`to placebo in randomized placebo-controlled trials. The effect appears consistent among
`the group of 11 drugs. There are 1.9 per 1000 (95% CI: 0.6, 3.9) more antiepileptic drug
`patients than placebo patients who experience Suicidal Behavior or Ideation. In terms of
`adjusted risk estimates for the treatment groups, 0.43% of the drug patients experience
`Suicidal Behavior or Ideation compared to 0.24% of the placebo patients.
`
`There is no obvious subgroup of patients to which the increased risk is specifically
`attributed. The increased risk was seen in almost all subgroups, although epileptic and
`Non-North American patients may have higher relative risks.
`
`6
`
`Page 00006
`
`
`
`INTRODUCTION
`2
`2.1 Background
`The Food and Drug Administration (FDA) concerned about the potential for elevated risk
`of suicidality (suicidal behavior or ideation) from the use of antiepileptic drugs carried
`out a meta-analysis of 11 drugs. Antiepileptic drugs are also used for indications other
`than epilepsy including psychiatric disorders. In March 2005 FDA initiated requests to
`the sponsors of antiepileptic drugs for data to address the suicidality concern.
`
`Prior to the analysis of the antiepileptic data, medical reviewers in the Division of
`Neurology and statistical reviewers in the Quantitative Safety and
`Pharmacoepidemiology Group agreed upon the definition of the research objectives,
`endpoints, study population, and subgroups and upon the specification of the statistical
`methods. These elements were incorporated into a statistical analysis plan prior to the
`review.
`
`This review replaces the March 5, 2008 version. Two small discrepancies in the data have
`been corrected for this version.
`
`2.2 Review Objectives
`1. Examine whether 11 antiepileptic drugs as a group are associated with increased
`risk of suicidality relative to placebo in randomized placebo-controlled trials.
`2. Examine whether the risk of suicidality varies by (a) individual drug, (b) drug
`subgroups, (c) indication subgroups, and (d) demographic subgroups.
`3 DATA SOURCES
`3.1 Data Requests
`In March 2005, FDA sent letters to sponsors of antiepileptic drugs requesting that they
`submit data from placebo-controlled trials for the FDA to review the possible association
`of suicidality events and antiepileptic drugs. Sponsors of all drugs with available
`registration trials were contacted. Letters in July 2005, May 2006, and January 2007
`requested additional information to obtain the data necessary for the review. The letters
`specified detailed instructions for the identification of suicidality events and the format of
`the data to be submitted.
`
`3.1.1 Trial Inclusion Criteria
`The final directions to the sponsors called for the submission of data for all randomized
`parallel-arm, placebo-controlled trials, regardless of indication and duration, with at least
`30 patients total. Trials may have had active-control arms as well. In additional to
`parallel-arm trials, data from the first period of cross-over trials that otherwise met the
`trial inclusion criteria were also included. In addition to placebo-controlled trials, trials
`with subtherapeutic comparator arms, known as “low-dose placebo” were to be included.
`The low-dose controlled studies were not included in the primary analysis. The July 2005
`
`
`
`7
`
`Page 00007
`
`
`
`FDA letter specified that studies with ongoing blinded treatment phases should not be
`included.
`
`3.1.2
`Identification of Suicidality Events
`FDA specified the procedure for the identification of suicidality events. The procedure
`called for a search of “possibly suicide-related” adverse events (PSRAEs). The search
`was to be strictly limited to events that occurred during the double-blind phase of
`treatment, or within 1 day of stopping randomized treatment. All deaths and serious
`adverse events (SAEs) were to be included as PSRAEs. In addition, events were
`identified through a search of specified text-strings in the adverse event data. For each
`PSRAE, a narrative was to be prepared.
`
`Based on blinded versions of the narratives, the PSRAEs were to be classified into
`mutually exclusive suicidality events using the approach employed in classification of
`outcomes as implemented in the pediatric antidepressant analysis (Posner et al. 2007).
`Table 1 gives the suicidality events. Sponsors were responsible for the classification of
`events.
`
`Table 1: Suicidality Events and Codes.
`Event Code
`Event
`0
`No Event
`1
`Completed suicide
`2
`Suicide attempt
`3
`Preparatory acts toward imminent suicidal behavior
`4
`Suicidal ideation
`5
`Self-injurious behavior, intent unknown
`6
`Not enough information, fatal
`7
`Not enough information, non-fatal
`
`3.1.3 Dataset Definition
`FDA specified the format of patient-level and trial-level datasets to be submitted. The
`patient-level dataset was to have one record per event. Patients with multiple events were
`to have multiple records in the dataset corresponding to each event. Patients without
`events were to be assigned an event code of 0. The patient-level dataset included
`variables for trial identification, patient identification, age, gender , race, setting of trial
`(inpatient, outpatient, both), location of trial (North America, Non-North America),
`treatment drug, event code, day of event, and discontinuation status. For location, FDA
`did not specify the meaning of North America.
`
`The trial-level dataset summarized and characterized the trial. This information included
`indication, nominal duration, treatment arm sizes, inclusion and exclusion criteria,
`dosage, and design features.
`
`8
`
`Page 00008
`
`
`
`3.2 Trial Summary
`Sponsors submitted datasets from 12 drug programs. One of these drugs, vigabatrin, is
`not currently approved in the United States and was not part of the review. Table 2 gives
`the names of the 11 drugs that were included in the review.
`
`Table 2: Antiepileptic Drugs under Review.
`Drug
`NDA Number
`Carbamazepine
`21-710
`Divalproex
`18-723, 19-680, 21-168
`Felbamate
`20-189
`Gabapentin
`20-235, 20-882, 21-129, 21-216
`Lamotrigine
`20-241, 20-764
`Levetiracetam
`21-035, 21-505, 21-872
`Oxcarbazepine
`21-014, 21-285
`Pregabalin
`21-446
`Tiagabine
`20-646
`Topiramate
`20-505, 20-844
`Zonisamide
`20-789
`
`The medical officer, Dr. Evelyn Mentari, Division of Neurology Products, performed
`some initial data processing of the submitted patient-level datasets including:
`1. Checking the correctness of the data submission to the FDA instructions
`2. Concatenating the data from the 11 drug programs into a single dataset
`3. Removing trials based on exclusion criteria (described below)
`4. Reducing multiple events in a single day to a single event (described below)
`5. Removing patients under the age of 5
`6. Creating indication categories (described below)
`
`Trials were excluded if the duration was less than 7 days, there were fewer than 20
`patients in any arm, all patients were less than 5-years old, or the trial had a randomized
`withdrawal design (including withdrawal to placebo).
`
`For patients with multiple events on a single day, only the most critical event (based on
`the event codes shown in Table 1) on the day was retained.
`
`9
`
`Page 00009
`
`
`
`The medical officer categorized the numerous indications into 21 indication categories.
`These 21 indication categories were further categorized into three categories: (1)
`epilepsy, (2) psychiatric, and (3) other. Table 3 gives the 21 indication categories and
`their further classification into three categories.
`
`Table 3: Indication Categories.
`Epilepsy
`Epilepsy
`
`Other
`Psychiatric
`Agitation
`Anxiety
`Chronic pain
`Binge eating disorder
`Fibromyalgia
`Bipolar disorder
`Impaired cognition
`Depression
`Insomnia
`Panic disorder
`Post-Traumatic Stress Disorder Migraine
`Schizophrenia
`Neuropathy
`Social phobia
`Obesity
`Radiculopathy
`Spasticity
`Tremor
`
`Note: The other indication category included volunteer studies.
`
`The review was based on the datasets prepared by Dr. Mentari and provided on 7
`November 2007.
`
`10
`
`Page 00010
`
`
`
`Table 4 gives the number of trials by comparator type and drug. There were 210 trials. Of
`these, 199 were placebo controlled and 11 were low-dose controlled. No study had both a
`placebo arm and a low-dose arm. There were 23 trials that also had an active-control arm.
`
`Table 4: Trials by Comparator Type and Drug.
`Number of Trials
`Low-Dose
`Controlled
`0
`1
`3
`0
`2
`0
`1
`1
`0
`3
`0
`11
`
`Drug
`
`Carbamazepine
`Divalproex
`Felbamate
`Gabapentin
`Lamotrigine
`Levetiracetam
`Oxcarbazepine
`Pregabalin
`Tiagabine
`Topiramate
`Zonisamide
`Total
`
`Placebo-
`Controlled
`3
`13
`6
`28
`27
`21
`10
`38
`6
`42
`5
`199
`
`Total
`
`3
`14
`9
`28
`29
`21
`11
`39
`6
`45
`5
`210
`
`11
`
`Page 00011
`
`
`
`Table 5 gives the number of trials by indication group and therapy (monotherapy,
`adjunctive therapy, and other). In the majority of epilepsy trials (81%), the drug was used
`in combination with other therapies as adjunctive therapy. In contrast, in the majority of
`psychiatric trials (86%), the drug was used as monotherapy.
`
`Table 5: Trials by Indication Group and Therapy (Monotherapy, Adjunctive Therapy,
`Other).
`
`Indication Group
`Total
`Other
`Psychiatric
`Epilepsy
`N=210
`N=81
`N=56
`N=73
`n (%)
`n (%)
`n (%)
`n (%)
`Therapy
`123 (59)
`61 (75)
`48 (86)
`14 (19)
`Monotherapy
`79 (38)
`12 (15)
`8 (14)
`59 (81)
`Adjunctive Therapy
`8 (4)
`8 (10)
`0 (0)
`0 (0)
`Other
`Note: Other therapy includes trials with optional adjunctive therapy and a trial in which
`one patient cohort received adjunctive therapy and one patient cohort did not receive
`adjunctive therapy.
`
`4 METHODS
`The statistical analysis plan (SAP) including the definitions of the endpoints, study
`population, subgroups, and statistical methods were prespecified prior to conducting the
`review. As stated above, these definitions and specifications were chosen by medical
`reviewers in the Division of Neurology and statistical reviewers in the Quantitative
`Safety and Pharmacoepidemiology Group. Deviations from and additions to the SAP are
`noted.
`
`4.1 Endpoints
`
`4.1.1 Primary Endpoint
`The primary endpoint was Suicidal Behavior or Ideation. A patient had this endpoint if
`the patient had any of the following suicidality events:
`• Completed suicide
`• Suicide attempt
`• Preparatory acts toward imminent suicidal behavior
`• Suicidal ideation
`
`4.1.2 Secondary Endpoint
`There were two secondary endpoints. A patient had the endpoint Suicidal Behavior if the
`patient had any of the following suicidality events:
`• Completed suicide
`• Suicide attempt
`
`12
`
`Page 00012
`
`
`
`• Preparatory acts toward imminent suicidal behavior
`A patient had the endpoint Suicidal Ideation if the patient had only a Suicidal Ideation
`event. Note that the endpoint Suicidal Ideation was not part of the SAP.
`
`4.2 Analysis Population
`The primary analysis population was all patients in test drug and placebo arms from
`placebo-controlled trials that met the trial and patient inclusion criteria described in
`Section 3.2.
`
`4.3 Subgroups and Special Populations
`
`4.3.1 Drugs
`Each drug was considered separately
`
`4.3.2 Drug Groups
`Three groups of drugs were considered. These groupings were chosen by the medical
`officers from the Division of Neurology. Each group of drugs was compared to the
`complementary group of drugs. Note that the drug groups are not mutually exclusive or
`exhaustive.
`1. Sodium Channel Blocking Drugs
`• Carbamazepine
`• Lamotrigine
`• Oxcarbazepine
`• Topiramate
`• Zonisamide
`2. GABAergic Drugs and GABAmimetic Drugs
`• Divalproex
`• Gabapentin
`• Pregabalin
`• Tiagabine
`• Topiramate
`3. Carbonic Anhydrase Inhibitors
`• Topiramate
`• Zonisamide
`
`4.3.3 Trial Indication
`Three indication groups were considered as defined in Table 3:
`1. Epilepsy
`2. Psychiatric Indications
`3. Other Indications
`
`4.3.4 Demographics
`The following subgroup classes were considered:
`1. Age
`5-17
`•
`
`13
`
`Page 00013
`
`
`
`18-24
`•
`25-30
`•
`31-64
`•
`• ≥ 65
`2. Gender
`• Male
`• Female
`3. Race
`• White Caucasian
`• Other
`4. Setting
`Inpatient or Inpatient/Outpatient Combined
`•
`• Outpatient
`5. Location
`• North America
`• Non-North America
`
`The age subgroups were chosen to be the same as used in FDA analysis of the
`antidepressant suicidality. Only two race subgroups were used because the overwhelming
`majority of patients were white. “Other” for race included African American, Hispanic,
`Asian, and other. For location, FDA did not specify the meaning of North America.
`
`4.3.5 Comparator Type
`The group of patients from low-dose-controlled trials was considered. This group was
`compared to the primary analysis group of placebo-control trial patients and the group of
`patients from both placebo-controlled and low-dose-controlled trials. For the analysis of
`patients from both types of trials, the test drug patients were compared to the patients in
`the corresponding trial control arm patients (placebo or low-dose).
`
`4.4 Statistical Methods
`
`4.4.1 Primary Method
`The primary analysis method was the exact method for a stratified odds ratio and
`associated 95% confidence interval (Cytel 2005, Ch. 19). The odds ratio was in terms of
`patient units. The stratification factor was the trial.
`
`4.4.2 Sensitivity Methods
`Three sensitivity analyses were employed to examine the robustness of the primary
`method.
`
`4.4.2.1 Zero-Event Trials
`The first sensitivity analysis examined the consequences of the fact that a large number of
`the trials were expected to have no events. The exact method for a stratified odds ratio
`does not make use of these trials. The Mantel-Haenszel risk difference and associated
`confidence interval (Greenland and Robins 1985), which makes use of these trials, was
`
`14
`
`Page 00014
`
`
`
`used for this sensitivity analysis. However, if there are no events for any trials, for
`example in a subgroup, then the estimated variance will be zero. In this case, it is not
`appropriate to use the variance estimate, and no estimate and confidence intervals were
`presented.
`
`4.4.2.2 Trial Heterogeneity
`The second sensitivity analysis examined between-trial heterogeneity of the effect
`measure. Zelen’s test (Cytel 2005, Ch. 19), an exact test, was used to test the hypothesis
`of a common odds ratio. However, because of the small number of events, it was
`expected that there would be little power to detect heterogeneity of the odds ratio across
`trials. The result of the test was intended for qualitative purposes.
`
`The trial weight of the Mantel-Haenszel odds ratio estimator was used to quantitatively
`identify trials with large influence. The weight was equal to (control patients with
`events)*(test patients without events)/(total patients). Trials with no events had a weight
`of zero. For trials with events in one arm only, the weight was equal to (control patients
`with events +0.5)*(test patients without events +0.5)/(total patients +2). Note that the
`SAP incorrectly specified “+1” rather than “+2” in the denominator.
`
` A
`
` generalized linear mixed model (GLMM) (McCulloch and Searle 2001) was used to
`estimate the overall odds ratio in the presence of trial heterogeneity of the odds ratio. The
`model used the binomial error distribution and logit link function. The model included
`fixed effects for the trial and treatment effects and a random effect on the trial-level for
`the treatment-trial interaction. The estimate and the 95% confidence interval of the
`treatment effect were qualitatively compared to those from the primary method to
`examine the effect of trial heterogeneity. The confidence interval of the variance
`component of the random effect was also examined to evaluate trial heterogeneity.
`
`4.4.2.3 Duration Differences
`The third sensitivity method, which was not part of the SAP, examined the consequences
`of the observed difference in treatment duration between the treatment arms. The method
`was similar to the primary method, but used person-time rather than patients as the unit of
`analysis (Cytel 2005, Ch. 15). Because the duration difference was small, an assumption
`of constant hazards was not key.
`
`4.4.3 Exploratory Methods
`
`4.4.3.1 Time Pattern
`Kaplan-Meier incidence curves were used to examine the time-pattern (hazard function)
`of the Suicidal Behavior or Ideation events. For patients with multiple events, only the
`most critical event was used. No stratification was employed in the analysis.
`
`
`
`15
`
`Page 00015
`
`
`
`4.4.3.2 Demographics, Duration and Discontinuation
`Differences in treatment arms within trials of demographics, treatment duration, and
`premature discontinuation of patients were examined. For categorical variables, p-values
`for differences between treatment groups were based on the Cochran-Mantel-