Columbia–Suicide SeverityRatingScaleScoring and DataAnalysisGuide
************************************************************************ Contributors: MaryE. Nilsson, ShailajaSuryawanshi,Cristiana Gassmann-Mayer,SarahDubrava,Paul McSorley,and KaihongJiang.
Introduction
TheColumbia–Suicide SeverityRatingScale (C-SSRS) is anassessmenttoolthatevaluatessuicidal ideation andbehavior.Thisguide outlines the proposedsafetyoutcomes and statisticalanalysis strategyfor theC-SSRS foranindividualclinicaltrial.Theactual tables or combination of tablesmayvaryto beconsistent with asponsor’sstandards, orto beconsistent with the needs of aparticular regulatoryagency/division.As such this documentserves as a generalguideline and is not a proposal for mandatoryanalyses involvingthe C-SSRS.When analyzingtheC-SSRS acrossstudies, the samesafetyoutcomes can beused. However,differentstatisticalmethodologywould applytocombiningdataacross studies, and is not discussed in this document.Thebriefingdocumentfor the 13 December2006Foodand DrugAdministration(FDA)AdvisoryCommitteeMeetingcontains a discussion of meta-analyticalmethods to consider whenanalyzingsuicide-relatedoutcomesacrossstudies( be used to assessefficacy, but specificguidance related to efficacyassessment will beatopic for aseparate document.
As noted inthe August 2012draftguidance titled“SuicidalIdeationandBehavior:Prospective Assessmentof Occurrence in ClinicalTrials”, the FDA hasadopted the11categoriesdefined in theC-SSRS(fivesubtypesofsuicidalideation, fivesubtypes ofsuicidalbehavior,andself-injuriousbehavior without suicidal intent)as their standard.Some datacollected on theC-SSRS arenotincluded in these proposed displaysfor safetyanalysis(e.g.,suicidal behaviorlethalityandsuicidalideationintensity), but are usedforindividual clinicalmanagement,safetymonitoring, orotherresearch purposes.
Analysis Set:
Theanalysis set should be definedconsistentlywith the sponsor’sstandards forthe safetyanalysis set.Frequently,this meansincludingpatients who have been exposedto at leastonedoseof studydrug.When analyzingthe C-SSRS, itis recommended to includepatients having atleast 1post-baseline C-SSRSmeasurement,regardless ofwhethertheyhad a baseline C-SSRSmeasurement. For someanalyses of theC-SSRS (e.g., treatment-emergent assessments and shift tables), a pre-treatment C-SSRSmeasurement is alsorequired.
Outcomes:
Thereis current debateonwhethersuicidal ideation andsuicidal behaviorshould becombinedandanalyzedas a single outcome.Researchersagree on the needfor havinganalyses that keep suicidalideationandsuicidal behavior separate, but disagree on
whetherthereis valuetohavinganyoutcomethatcombines them.Thisdocumentcurrentlyincludes thecombined outcomesmeasure.
ThefollowingoutcomesareC-SSRS categories and have binaryresponses(yes/no). Thecategories havebeen re-orderedfrom theactual scale to facilitate thedefinitions of thecomposite andcomparative endpoints, and to enable clarityin the presentation of theresults.
Category1 – Wish to be Dead
Category2 – Non-specific Active Suicidal Thoughts
Category3 – Active SuicidalIdeation with AnyMethods (Not Plan)withoutIntent to ActCategory4 – Active SuicidalIdeation with SomeIntent to Act, without SpecificPlanCategory5 – Active SuicidalIdeation with SpecificPlan andIntent
Category6 – PreparatoryActs or BehaviorCategory7 – AbortedAttempt
Category8 –InterruptedAttemptCategory9 – Actual Attempt (non-fatal)Category10 – CompletedSuicide
Self-injuriousbehavior without suicidal intent is also a C-SSRS outcome (athough notsuicide-related) and has abinaryresponse(yes/no).
Thefollowingoutcomeis a numerical scorederived from theC-SSRS categories.Thescoreis createdateach assessment for each patientand is usedfordeterminingtreatmentemergence.
SuicidalIdeationScore:Themaximum suicidalideation category(1-5 onthe C-SSRS) present at the assessment. Assign ascore of0 if no ideation is present.
Endpoints:
Composite endpointsbased on the abovecategoriesare definedbelow.
Suicidalideation: A “yes” answeratanytime duringtreatment to anyoneofthe
fivesuicidalideation questions (Categories 1-5)on the C-SSRS.
Suicidalbehavior: A “yes” answeratanytime duringtreatment to anyoneof the
fivesuicidalbehavior questions (Categories 6-10)on the C-SSRS.
Suicidalideationor behavior: A “yes”answer at anytime duringtreatment toanyoneof theten suicidal ideationand behaviorquestions(Categories 1-10)onthe C-SSRS.
Comparative endpoints of interest are defined below. “Treatmentemergence” is used foroutcomes that includeeventsthat first emerge or worsen. “Emergence”isusedforoutcomes that includeeventsthat first emerge.
Treatment-emergentsuicidal ideationcompared to recenthistory:An increaseinthe maximum suicidalideationscore duringtreatment from the maximumsuicidalideationcategoryduringa specifiedpre-treatmentperiod(C-SSRSscalestakenduringthe specified pre-treatment period; excludes “lifetime”scoresfromtheBaseline C-SSRSscale or Baseline/ScreeningC-SSRS scale).
Treatment-emergentserious suicidalideationcompared to recenthistory: Anincrease in themaximumsuicidalideationscore to4 or5 on the C-SSRS duringtreatmentfrom not havingserious suicidal ideation (scoresof0-3) during aspecified pre-treatment period (C-SSRS scalestakenduring the specified pre-treatmentperiod; excludes “lifetime”scoresfromthe Baseline C-SSRS scale orBaseline/ScreeningC-SSRS scale).
Emergence ofserious suicidalideationcompared to recenthistory: An increasein themaximumsuicidalideationscore to 4 or 5 on the C-SSRS duringtreatmentfrom no suicidalideation(scoresof 0)duringaspecifiedpre-treatment period (C-SSRSscalestaken duringthe specified pre-treatment period; excludes “lifetime”scores from the BaselineC-SSRSscale orBaseline/ScreeningC-SSRS scale).
Improvementin suicidalideationat a timepoint ofinterestcompared to baseline:An improvement in this endpoint can beconsidered as a decrease in suicidalideationscoreat thetime point of interest(e.g., thelast measurement duringtreatment) from the baselinemeasurement(e.g., the measurementtaken justpriorto treatment.This analysis should onlybeperformed for studies in whichabaseline C-SSRScan be defined(i.e., havingimprovementfrom the worseeventover alifetime is not clinicallymeaningful).
Emergence ofsuicidal behaviorcompared toall prior history: Theoccurrenceofsuicidalbehavior(Categories 6-10)duringtreatment fromnot havingsuicidalbehavior(Categories 6-10) prior to treatment (includes “lifetime”and/or“screening” scoresfromthe Baseline C-SSRS scale,ScreeningC-SSRS scale, orBaseline/ScreeningC-SSRS scale,andany“SinceLast Visit”from the SinceLastVisit C-SSRSscalestaken priorto treatment).
Additionalcomparative endpoints forconsiderationaredefinedbelow.These arenotnecessarilyrecommendedforalltreatment programs, but if used, should follow thenomenclature.
Treatment-emergent suicidalideationcomparedtoall prior history: An increasein themaximumsuicidalideationscore duringtreatment fromthe maximumsuicidalideationscore prior to treatment(includes“lifetime”and/or“screening”scores from the BaselineC-SSRSscale,Screening C-SSRSscale, orBaseline/ScreeningC-SSRS scale,andany“SinceLast Visit”from the SinceLastVisit C-SSRSscalestaken priorto treatment).
Emergence ofserious suicidalideationcomparedto all prior history:An increasein themaximumsuicidalideationscore to 4 or 5 duringtreatmentfrom nosuicidalideation(scores of 0)priorto treatment (includes “lifetime” and/or“screening” scoresfromthe Baseline C-SSRS scale,ScreeningC-SSRS scale, orBaseline/ScreeningC-SSRS scale,andany“SinceLast Visit”from the SinceLastVisit C-SSRSscalestaken priorto treatment).
Treatment-emergentserious suicidalideationcompared to all prior history: Anincrease in themaximumsuicidalideationscore to4 or5 on the C-SSRS duringtreatmentfrom not havingserious suicidal ideation (scoresof0-3) prior totreatment(includes“lifetime” and/or“screening”scores from the Baseline C-SSRSscale,ScreeningC-SSRS scale, orBaseline/ScreeningC-SSRS scale,andany“SinceLast Visit”from theSince Last Visit C-SSRSscalestaken priortotreatment).
Outcomes that canbeused forclinicalmanagement and safetymonitoring or potentiallyforresearchpurposesaredescribed below.
Suicidalbehaviorlethalityratingtaken directlyfrom theC-SSRS Suicidalideationscore
Anyscoregreaterthan 0is important and mayindicate theneed formental healthintervention. Theprotocol proceduresrelated to clinicalcareofpatients withtreatmentemergent suicidal ideationand behaviorswill then beimplemented toensureproper management of the event and protection ofthe patient’ssafety. Forexample, a score of 4(activesuicidal ideation with some intent to act) or5(activesuicidalideation with specificplanandintent) on suicidal ideationcan beused toindicate serioussuicidal ideationand can beusedtotriggerfurtherevaluationandimmediatecontact with patient’s mental health practitioner (fornon-psychiatrytrials this maybeusedtotrigger aprompt referralto a mentalhealth professional)and/or possiblythe emergencyroom.
Suicidalideationintensity rating
Add the five intensityitem scores to createatotal score (range 0 to 25) torepresent theintensityrating. If the patient did not endorse anysuicidal ideationset theintensityratingto0.
Analytical andStatisticalMethods:
Thesponsor’s standard approaches to present safetydata should be used consistently.Suicidalideationandsuicidalbehaviorarerelativelyrare events in individual, double-blind, controlledclinicaltrials,except possiblyfor apopulation particularlysusceptible tothis safetyrisk.Ingeneral, no formalstatistical hypothesis testingis recommended forindividual studies as onlyfeweventsare typicallyobservedand descriptivesummarieswill then suffice [9].
See the mocktables, themock listing,and mock figureforsuggestedsummariesandpresentation of thedata. Notethat thesemock tables can beeasilyextendedto includemultiple treatmentgroups/comparisonsasneeded. Tablesshould be modified to fit theneeds of thespecificanalysesperformed.
For verysmall studies(e.g.,Phase 1 studies),orindicationswhere suicide-related eventsareexpected to be rare, itmaybe sufficient to planon providingalistingonly(seeListing1). Otherwise,Tables 1 and 2 as shown maybe provided.
Thespecificstatisticalmethod (or lackofaninferentialmethod) should beconsistentwith the sponsor’s standards. When inferentialanalysesarenot performedthe p-valuecolumn should be eliminated. Tables3-4(options forashift table) will likelybeusedwhenthereare findingstounderstand in more detail and/orwhenthe expectednumber ofsuicide-relatedevents is large either dueto havinga large studyor having apopulation athigh risk. Agraphicalpresentation, forexampleastacked bar chart,mayalso beusedwhenappropriate(seeFigure 1).
A summaryofpre-treatment datais generallydesirable and can beadded toexistingplannedtables forpatientcharacteristics(e.g., percent of subjects with lifetimesuicidalideation,baseline suicidal ideation,lifetimesuicidalbehavior,baseline suicidal behavior,lifetime self-injurious behavior withoutsuicidal intent, andbaseline self-injuriousbehaviorwithoutsuicidalintent). Alternatively, ifallcategories of suicidalideationandsuicidalbehavior areof interest for pre-treatmentsummaries,Table 1 canbe repeated forprovidingasummaryoflifetimeoutcomesand again for baseline outcoments.Thespecifictimeframeusedfordefiningbaseline willneed to be defined.
When inferentialstatisticsare used, methodsforconsiderationinclude:theMiettinenandNurminen method [1], anunconditional,asymptotic method;the Fisher'sexact test; 95%confidence intervals (CIs) based on Wilson's scoremethod; and forest plots to presenteffect across treatment arms or subgroups ofinterest.
A furtherconsideration forinferentialanalysis maybethe requirementthatatleast 4patients in anytreatmentgroupexhibit the event [2].Becausemany95%CIsmaybeprovidedwithout adjustment for multiplicity, theCIs should be regarded as a helpfuldescriptive measureto beusedin review, not aformal method for assessing the statisticalsignificance of the between-group differences in adverseexperiences. Appropriatecautionarystatements on the interpretation of inferentialstatistics forpurposes ofsummarizinginformation or signalingtrends should be made.
Methods for assessingdose-response relationshipsshouldbe consistent with the sponsor'sstandards. Calculatingand presentingincidencerates (e.g., n/patient-years)should beconsidered,especiallyfor studies with longpatient exposures and expecteddifferentialdrop-outratesbetweentreatmentgroups.
Inlong-termtrials where the suicidalityscale is administeredfrequently,a time-to-eventanalysis wherethe eventis anysuicidalevent(eitherideationorbehavior whicheveroccurs first)canbe performed usingmethods suchasKaplan-Meiermethod[3],log-rank
test [4],generalized Wilcoxon test[5]andCoxproportionalhazardsmodel [6]adjustingforalimited number of covariates. It is noted thateventtype-specificanalysis (i.e., time-to-first-ideationandtime-to-first-behavior analyses) maynot be interpretable dueto thepresence of-competingrisks[7]andinformativecensoring[8]. In situations that warrantanassessmentbyage,genderand/orother risk factors summarystatisticsofscores onsuicidalityendpoints of interestcanbe presented byvisit. For studies withlongerduration,scorescan besummarized byvisit or time period(treatment phase, post-treatment phase etc.).
Based on theindividual programstrategy,active treatmentgroupsmaybepooled forcomparisons to placebo/active comparator, orforperformingall pair-wisedosecomparisons. Analysismethods forassessingeffects overseveral trials,subgroups orindicationsarepresentedin [9].
Notethat missingdata should not beimputed.
Table 1: Number of PatientswithSuicidalIdeation,SuicidalBehavior,andSelf-Injurious BehaviorwithoutSuicidalIntentBased onthe C-SSRSDuringTreatment;[DefineAnalysisSet]
Eventsduringtreatment / DrugNameN=xxn(%) / ComparatorNameN=xx
n(%) / p-valuesa(tocomparepercentages)
SuicidalIdeation(1-5) / x(%) / x(%) / 0.xxx
1) Wishtobedead / x(%) / x(%)
2) Non-specificactivesuicidalthoughts / x(%) / x(%)
3) Activesuicidalideationwithanymethods(notplan)withoutintenttoact / x(%) / x(%)
4) Activesuicidalideation withsomeintenttoact,withoutspecificplan / x(%) / x(%)
5) Activesuicidalideation withspecificplanandintent / x(%) / x(%)
SuicidalBehavior(6-10) / x(%) / x(%) / 0.xxx
6) Preparatoryactsorbehavior / x(%) / x(%)
7) Abortedattempt / x(%) / x(%)
8) Interruptedattempt / x(%) / x(%)
9)Non-fatalsuicideattempt / x(%) / x(%)
10)Completedsuicide / x(%) / x(%)
SuicidalIdeationorBehavior(1-10) / x(%) / x(%) / 0.xxx
Self-injuriousbehaviorwithoutsuicidalintent / x(%) / x(%)
NOTE: thep-value columnshouldbeeliminatedwheninferentialanalysesare notperformeddependingon theSponsor's standardapproachforsafety datapresentations.
ap-valuesarefrom[specifytest].
Notes: N=numberofenrolledpatientswith atleastonepost-baseline C-SSRS assessment. Inthistable,n and(%)referto thenumber andpercentofpatientswhoexperience theeventatleast onceduringtreatment.For the compositeendpointofsuicidalideation(1-5), n and (%)refertothe numberandpercentof patientswho experienceany oneof the five suicidal ideationeventsat leastonceduringtreatment. For thecompositeendpointof suicidalbehavior(6-10),n and(%) refertothenumberandpercentofpatients whoexperience anyone ofthe fivesuicidalbehavioreventsatleast onceduring treatment. Forthecompositeendpointofsuicidalideationorbehavior(1-10),n and (%)refertothe numberandpercentofpatients whoexperience any one ofthetensuicidalideationor behavioreventsatleast onceduring treatment.
Table 2: Number of Patientswith Suicide-RelatedTreatment-EmergentEventsBased on theC-SSRSDuringTreatment;[DefineAnalysisSet]
DrugName / ComparatorName / p-valuesaTreatment-emergent(TE)Events / N / n(%) / N / n(%)
TEsuicidalideation(1-5)comparedtorecenthistoryb / xx / x(%) / xx / x(%) / 0.xxx
TEserioussuicidalideation(0-3to4-5)comparedtorecenthistoryc / xx / x(%) / xx / x(%) / 0.xxx
Emergenceofserioussuicidalideation(0to4-5)comparedtorecenthistoryd / xx / x(%) / xx / x(%) / 0.xxx
Improvementinsuicidalideationatendpointcomparedwithbaselinee / xx / x(%) / xx / x(%) / 0.xxx
Emergenceofsuicidalbehavior(6-10)comparedtoallpriorhistoryf / xx / x(%) / xx / x(%) / 0.xxx
Notes:For thecompositeendpointofsuicidalideation(1-5),n and(%)refertothe numberandpercentof patientswhoexperience treatment-emergentsuicidalideationduringtreatment.For thecompositeendpointofsuicidalbehavior (6-10),n and(%) refer tothenumber andpercentofpatients whoexperience treatment-emergentsuicidalbehaviorduringtreatment.
ap-valuesarefrom[specifytest].
bN=Numberofenrolledpatientswith atleast onepost-baselinesuicidalideationscore andwhose maximumC-SSRSsuicidalideation scoreduringthecomparisonperiodisnon-missingand<5.
cN=Number ofenrolled patientswith atleast onepost-baselinesuicidalideationscoreand whose maximumC-SSRS
suicidalideation scoreduringthecomparisonperiodis0-3.
dN=Numberofenrolledpatientswith atleast onepost-baselinesuicidalideationscoreand whose maximumC-SSRSsuicidalideation scoreduringthecomparisonperiodis0.
eN=Numberofenrolledpatientswhosesuicidal ideationscore isnon-missingand>0 just prior totreatment.
fN=number ofenrolled patientswith atleastonepost-baselineC-SSRSassessment andwho didnothave suicidalbehavior (6-10)prior to treatment.
NOTE: thep-value columnshouldbeeliminatedwheninferentialanalysesare notperformeddependingon theSponsor's standardapproachforsafety datapresentations.
Table 3. Shift-tableto DemonstrateChangesin C-SSRSCategoriesfromBaselineDuring Treatment;[DefineAnalysisSet]
Treatment / BaselineCategory / TreatmentCategoryNosuicidalideationorbehavior
n(%) / Suicidalideationn(%) / Suicidalbehaviorn(%)
DrugName(N=xxx) / Nosuicidalideationorbehavior / x(%) / x(%) / x(%)
SuicidalIdeation / x(%) / x(%) / x(%)
SuicidalBehavior / x(%) / x(%) / x(%)
ComparatorName(N=xxx) / Nosuicidalideationorbehavior / x(%) / x(%) / x(%)
SuicidalIdeation / x(%) / x(%) / x(%)
SuicidalBehavior / x(%) / x(%) / x(%)
Notes: N=numberofpatientswith a baselineand post-baselineC-SSRSassessment,n=numberof patients incategory. %=100*n/N.
Baselinerefers to[specifydefinition]
SuicidalIdeationincludesanyoneofthefivesuicidalideationevents (Categories1-5).Suicidalbehaviorincludesanyone ofthe fivesuicidalbehaviorevents(Categories6-10).
Eachpatientis countedinonecellonly. Patientswith bothSuicidalIdeationandSuicidalBehavior areincluded intheSuicidalBehaviorcategory.
Table4. Shift-tableto DemonstrateChangesin C-SSRSSuicidalIdeationScoresfromBaselineDuringTreatment; [DefineAnalysisSet]
Treatment / MaximumBaselineScore / MaximumSuicidalIdeationScoreDuringTreatment0
n(%) / 1
n(%) / 2
n(%) / 3
n(%) / 4
n(%) / 5
n(%)
DrugName(N=xxx) / 0 / x(%) / x(%) / x(%) / x(%) / x(%) / x(%)
1 / x(%) / x(%) / x(%) / x(%) / x(%) / x(%)
2 / x(%) / x(%) / x(%) / x(%) / x(%) / x(%)
3 / x(%) / x(%) / x(%) / x(%) / x(%) / x(%)
4 / x(%) / x(%) / x(%) / x(%) / x(%) / x(%)
5 / x(%) / x(%) / x(%) / x(%) / x(%) / x(%)
ComparatorName(N=xxx) / 0 / x(%) / x(%) / x(%) / x(%) / x(%) / x(%)
1 / x(%) / x(%) / x(%) / x(%) / x(%) / x(%)
2 / x(%) / x(%) / x(%) / x(%) / x(%) / x(%)
3 / x(%) / x(%) / x(%) / x(%) / x(%) / x(%)
4 / x(%) / x(%) / x(%) / x(%) / x(%) / x(%)
5 / x(%) / x(%) / x(%) / x(%) / x(%) / x(%)
Notes: N=numberofpatientswith a baselineand post-baselineC-SSRSsuicidalideationscore, n=number ofpatientsincategory,
%=100*n/N.
Baselinerefers to[specifydefinition];Maximumrefers tothemaximumC-SSRSsuicidal ideationscoreduring treatment(0 =least severe,5=most severe)where 0=NoSuicidalIdeation,1=Wishto beDead,2=Non-specificActiveSuicidalThoughts, 3=ActiveSuicidal Ideationwith AnyMethods (NotPlan) withoutIntent toAct,4=Active SuicidalIdeationwithSomeIntent toAct,without SpecificPlan,and5=Active Suicidal IdeationwithSpecific PlanandIntent.
Listing 1: Listingof SubjectswithSuicidalIdeation,SuicidalBehavior,orSelf-Injurious BehaviorwithoutSuicidalIntentBased onthe C-SSRSDuringTreatmenta;[DefineAnalysisSet]
SuicidalIdeation / SuicidalBehaviorPatient / Trt / Visit / 1 / 2 / 3 / 4 / 5 / 6 / 7 / 8 / 9 / 10 / Self-InjBehwoSI
xxxx / Y / Y / Y / N / Y / N / N / N / N / N / N
Note:Onlypatientswithsuicidalideation, suicidalbehavior, orself-injuriousbehaviorwithout suicidalintentaredisplayed.For patientswithsuicidal ideation,suicidalbehavior,orself-injuriousbehaviorwithoutsuicidalintentatany time,datafromallvisitsare displayed. Self-Inj Behwo SI=Self-injurious BehaviorwithoutSuicidal Intent.
aKey: 1=Wishto be Dead,2=Non-specific ActiveSuicidalThoughts,3=ActiveSuicidal IdeationwithAnyMethods(NotPlan) withoutIntent toAct,4=ActiveSuicidalIdeationwith Some IntenttoAct,withoutSpecific Plan,5=Active SuicidalIdeation withSpecific Planand Intent,6=Preparatory ActsorBehavior,7=AbortedAttempt,8=InterruptedAttempt,9=ActualAttempt(non-fatal),10=CompletedSuicide.
Figure1: DistributionofSubjectsWith Suicide-RelatedEventsBased ontheC-SSRS ScoresOverTIme[DefineAnalysisSet]
50
40
30
20
10
0
DrugADrugBDrugADrugB
Visit1Visit2
References:
1.MiettinenO,NurminenM. Comparative analysisof tworates. Statistics inMedicine 1985; 4:213-226.
2.Crowe B,XiaA,BerlinJ, Watson D, Shi H, LinS,KueblerJ,SchriverR,Santanello N, Rochester G, Porter J, Oster M, MehrotraD,LiZ, KingE, HarpurE, Hall D. Recommendations forsafetyplanning,datacollection, evaluation andreportingduringdrug, biologic and vaccine development: a report ofthe safetyplanning, evaluation, andreportingteam.ClinicalTrials 2009; 6:430-440.
3.KaplanEL, Meier P. Nonparametric estimation forincomplete observations.AmStat Assoc1958; 53: 457-481.
4.Mantel N. Evaluation ofsurvival data and two newrank order statistics arisinginits consideration.CancerChemotherRep1966; 50(3): 163-170.
5.GehanEA.AgeneralizedWilcoxon test forcomparing arbitrarilysingly-censoredsamples.Biometrika1965; 52: 203-223.
6.CoxDR.Regression models and lifetables.JRoyal Stat Soc 1972; 34: 87-220.
7.Pintilie M. Competingrisks: a practical perspective. John Wileysons: WestSussex, England, 2006.
8.Satagopan JM, Ben-PoratL,Berwick M, et al.Anote on competing risks insurvival data analysis.Brit J Cancer2004; 91: 1229-1235.
9.Gassmann-Mayer C, JiangK, McSorleyP,AraniR, DuBrava S, Suryawanshi S,Webb DM, Nilsson M. Clinical and Statistical Assessment of SuicidalIdeationandBehavior in PharmaceuticalTrials. Clin Pharmacol Ther2011; 90(4):554-560.