Analysis of Suicidality Data from Adult Antidepressant Trials Litigated by Law Firm Baum Hedlund December 13, 2006

1991 PDAC Conclusion: “More research and data is needed” Although the committee members ultimately voted that there was insufficient data to conclude that Prozac caused suicide (the question posed was: “Is there credible evidence to support a conclusion that antidepressant drugs cause the emergence and/or intensification of suicidality and/or other violent behaviors?”), the FDA stated that it did “not dismiss the possibility that antidepressants in general or fluoxetine in particular may have the capacity to cause untoward injurious behaviors and acts, and/or to intensify them.” The FDA concluded that “more research is needed” and “asked [Lilly] to develop plans to conduct new studies, including clinical trials and epidemiological studies, studies that could provide more direct answers to the questions that have been raised” in the advisory committee meeting. (September 1991 PDAC Transcript at 128:18-24: ​VIEW FULL DOCUMENT​) The FDA advisory committee only saw data from the Prozac clinical trials. Had the committee seen the data from the Paxil clinical trials, things might have ended quite differently. After the FDA asked GSK to submit an analysis of its clinical trial data in order to respond to the public’s concerns about the risk of suicidality, GSK responded by inappropriately adding suicide events that occurred in the placebo run-in/wash-out period, thus masking Paxil’s suicidality risk. When the placebo run-in/wash-outs are removed, Paxil users were over 8 times more likely to have engaged in suicidal behavior than those on placebo. At the time, had GSK properly reported this to FDA and had FDA actually noticed the actual Paxil suicidality risk, this would have had a devastating impact on Paxil’s capacity to compete with Prozac, an SSRI with an already established market, and, under proper FDA supervision, may have even impacted Paxil’s ability to have obtained FDA approval. Moreover, it would likely have had a substantial impact upon the PDAC’s conclusions and the whole history of the hidden risk of SSRI-induced suicidality.

rsm ,I DEPARTMENT OF HEALTH AND m S'ERVICES PUBLIC HEALTH SERVICE .'. .' FOOD AND DRUG AJ3MI@TtiTIQN' : ADVISORY ,Cq,MMITTEE \" Conference Rooms D/E Pazklawn +ilding Rockville;, Maryland MUER RRPGRllNG CCL 1%. 507 C Succr. N.E. Wuhiigton. D-C. 20002 (202) 546-6666

rsm ,, -i 1 taken - _.,.- 2 Prozac meets the standards 3 required 4 and Cosmetic 5 6 7 conclusion E collectively, s risk-free. 1c 13 1; 1: 14 marketed 1: 1t 1; 1Z example, 15 mention, 2c 2. 2: 2. - 21 MILLER REPORTING CO.. IN6! ! >07 CSuccr. N.E. Wuhingcon. D.C. 20002 (202) 146.6666 128 : as a whole, continues to support the conclusion that I of drug product safety and efficacy for marketing approval under the federal Food, Drug, Act that is our national drug domestic regulator; law. I want to emphasize again, to be fair, that this does not mean that we believe, individually or that antidepressants, or Prozac, are absolutely Neither does this conclusion mean that the agency is going to lessen its vigilance or will cease to review and assess the significance of adverse reports it receives on Prozac now or in the future. Reports of Prozac, like those received on all drugs, are regularly monitored and evaluated. When a signal of potential concern is identified, as it has been i the case of Prozac, we take additional actions, and urge manufacturers to do so as well. In the present case, for the sponsor, Eli Lilly, was asked -- and, I want to expeditiously complied with the request -- to examin data from previously conducted controlled investigations and was also asked to develop plans to conduct new studies, including clinical trials and epidemiological studies, studie that could provide more direct answers to the questions that have been raised in the open session earlier. Unfortunately, it is very difficult to tell, from n I

Post 1991 – None of the Companies Conducted Follow-up Research, Nor Did FDA Push Them to Do So Despite the PDAC’s mandate that further research be conducted, no such studies were ever conducted by Lilly or any other SSRI producing company. In fact, Lilly proposed a protocol for a “rechallenge” study of patients who developed suicidal ideation while under Prozac treatment, but never performed the study. In addition, a more sensitive scale for detecting treatment emergent suicidality was developed that could have been utilized in ongoing and future clinical trials, however, the scale was never implemented. According to Lilly’s answers to Interrogatories submitted under oath in the lawsuit ​Biffle v. Eli Lilly: ​“Discussions were had between Lilly and the FDA regarding possible data analyses or clinical trial designs which would test whether the Teicher assertions are in fact real. The FDA did not request or require any action from Lilly nor suggest a particular analytical of study approach. ​The discussions and question as to whether additional studies be done were mooted by the findings of the FDA Psychopharmacological Drug Advisory Committee on September 20, 1991. No additional studies were conducted.” (​VIEW FULL DOCUMENT​)









This statement is demonstrably false. According to a letter the FDAsent to Public Citizenin June 1992: “There was a consensus [amongstthe PDAC above] thatmore research is neededto further explore therelationship between suicidality and the use of not only Prozac, butother antidepressants as well.” The FDA further stated that it would“continue our careful evaluation of data in our spontaneous reportingsystem andencourage additional researchon this matter.”(VIEW FULL DOCUMENT)None of the antidepressant manufacturers at the time nor since then,has conducted a single safety oriented study to examine the risk norhave they utilized more sensitive measures to detecttreatment-emergent suicidality.







Lilly’s Analysis of Prozac Clinical Trial Data Criticized Lilly published its meta-analysis of Prozac clinical trial data in 1991, which meta-analysis was hotly criticized. As one scientist explained it, “the term meta-analysis sounds rather grand, but it is worth no more than the quality of the original data collection… What was needed was a critical assessment, independent of the manufacturers, that included assessment of the quality of data collection–and not Eli Lilly’s employees deciding which clinical comments should be ‘eliminated.’” Oswald, “Fluoxetine and suicide” ​BMJ ​1991 Oct 26;303(6809):1058-9. He concluded: “A negative result of research, a failure to find something, can arise from lack of sensitive research techniques.” ​Id. According to an internal Pfizer memorandum obtained in litigation and written by Pfizer’s top scientist at the time: “I do not think fluoxetine are ‘out of the woods’ as regards their association with violence/ suicidality. The recent meta-analysis of controlled clinical trials (Beasley et al, BMJ 303: 685-692, 1991) was initially followed by favorable comment but skeptical voices remain. A recent re-analysis of the data from this study using Monte-Carlo simulations demonstrates the conclusions of the Beasley paper to be invalid as this original meta-analysis had low power (LiWan PoA. Pharmacoepidemiology and Drug Safety 2: 78-84, 1994).” January 20, 1994 memo from Roger Lane to Giller regarding Use of Zoloft in Impulsive/ Aggressive Behaviour. (​VIEW FULL DOCUMENT​)

RL.eiaCC:W. FieesselsI. ffenryC. TruchanA .Wisa^*~~-~-`P-31CONNFIDEN T IALInternatioi3al'PharmaceuiieaLCAMJanuar,; 20, 1994T11Dr. E. Gi11er -GrotonFRGM.Roger Lane, M.D.SU8.1E?:II83 07 3CLCFT IiiI_'is~ULS.T.Y~/AGGr'2f38I~E 3AYICURThank youfar yourmemorandum to Mr.W. ffeesselswhich he has gassed to me farcosmoat_'ragree this is anintaratinqarea witha great deal ofevidencetosuggest utility far the SSZIs . Z have received numerous requests for.studiesirxmany of the areas you mention frcmaggressivebehaviour in mentaZlg ha=d=cappedchildraa toaggressive behavioural disturbancis in demented elderlygitientz. .However,research in these areas, if it ic=- .zrct pracluced by methc~vlogic:l .differences, is a very difficult \"sell,within the company,due to the Xroblem:Prozac has encountered. Marketing (especially Rcerig) are very aware that apatient an ZOLor-involved La an indident of mass homicide could eevercLy affectthe image and commercial success of ZOLOFT. Patients with borderline per:cna.itydisorder, Psychopathic traits, aggressive impulse control disarders,Qt= _ =AYbethe very group who are likely to generate such an incident. Any przgassI tostudy these patients is likely to be vetoed,- therefore.I do notthink fluoxatinea_-eout of the woods' asregardstheir.asscciatianwith-vialance/suicidality.. Thee recent meta-analysis of controlled clinicaltrims(Seas-rayat al, HM.7 3C.: 683-692, 1991) was initially.fol2=wed byfavorihle comment but skeptical voices remain. A recent =a-analysis of =.he data.from this study using Mcnts-Carlo simulations damonstrates the conclusions at the8easlay pager to be invalid as this arigioa.meta-analysis had lour Pcws=-(LiSdanPeA_ Pha-macaegida~miolcgy lad Drug Safety 2: 7n.-84, 1994).iewever, same of the disorders you mention, such as pathological . gambling,•belongto the group of COD. spectrum disorder.. I am very interested is spcnsarin-' *wallopen studies inseveralcc=spec-.=s=disorder:(e.g.pathologicalam r= .lingst_ichctillomania, body-dysmarphic disorder, etc.). We era currently sponsoringan open. study in pedaphilia. This would support cur work in OCD and expand thenmarket beyond pure CC:2.Would this be a way for~rsrd to 3..nvolvei .:vastigatorsf experts such as, Linnclis and Cocca--=?Please do riot hesitate to discuss these issues with mefurther.Not ConfidentialDe-designated Confidentialby email datedJune25,200300NFIDEN T IALLnurastioc4 Fbw\"A'sCcoaF.Fs\"r I=23.S F...42nd ScrawcP!cWv ar;c,NT, USA 10017-57551

FDA epidemiologist, Dr. David Graham, also criticized Lilly’s meta-analysis, which had been submitted to FDA in September 1990. In a document obtained from the FDA through the Freedom of Information Act, Dr. Graham stated: Suicidality. The firm [Lilly] reviewed data from NDA studies, prefacing it with the acknowledgment that these trials were not designed for the prospective evaluation of suicidality. In these trials patients with current suicidal ideation were excluded. Suicidal ideation was studied in two ways. The first involved analysis of clinical comments ascertained through non-probing, open-ended questions during the trial. Also, at the beginning and end of the study, patients completed a self-administered questionnaire, the Hamilton Rating Scale for Depression, which included one question on suicide. This question, referred to as HAMD-3, rated suicidal ideation on an ordinal scale from 0 (absent) to 4 (severe ideation, usually with an attempt). The capacity of these trials to identify and describe the quality and intensity of suicidality was low. Dr. Graham made a number of other important points: 1. Dr. Graham criticized Lilly’s “meta-analysis,” which was being touted by Lilly as showing that there was no relationship between Prozac and suicidality. (“In the meta-analysis of suicidality from the IND trials, 76 fluoxetine cases were excluded from analysis because the patients were in studies or other trials lacking comparative controls.”) (Graham memo p. 4: ​VIEW FULL DOCUMENT​); 1. Dr. Graham questioned Lilly’s reliance on an abstract by Fava & Rosenbaum which Lilly asserted showed “no statistically significant differences among rates of treatment-emergent suicidal ideation associated with five classes of antidepressant therapy.” (When Dr. Graham re-analyzed Fava & Rosenbaum’s data he found that “Treatment-emergent suicidality was more frequent among ‘fluoxetine alone’ than ‘tricyclics with or without lithium’ patients. The relative risk of suicidality was 3.3. (95% CL 0.9, 12.2), p-0.07.”) (Graham memo p. 4); 2. Dr. Graham validated the report by Teicher, et al., which first discussed the relationship between Prozac and suicide (“Interestingly, the proportion of patients with treatment-emergent suicidality on fluoxetine in this study was similar to that reported by Teicher et al.”) (p. 6). 3. In conclusion, Dr. Graham stated: “The firm’s analysis of suicidality does not resolve the issue. The firm acknowledged that its clinical trials were not designed to study this and specificity of data to be gleaned from these trials to address suicidality were poor. . . . Because of apparent largescale underrreporting, the firm’s analysis cannot be considered as proving that fluoxetine and violent behavior are unrelated.”

MEMORANDUM DEPARTMENT OF HEALTH AND HUMANSERVICESPUBLIC HEALTHSERVICESFOOD AND DRUG ADMINISTRATION CENTER FOR DRUG EVALUATION AND RESEARCH 3BP\" 11990DATE: FROM:Chief,EpidemiologyBranch,HFD-733SUBJECT: -Meetingsonexperiencefromfluoxetinesurveillance:..September18(in-house)andSeptember25(withfirm)THROUGH:ActingDirector,CA.<*\\ioOfficeofEpidemiologyandBiostatistics,HFD-701TO: Director, Division of NeuropharmacologyDrug Products, HFD-120 Attached are three documents which have been prepared by Epidemiology Branch staff as backgroundfor the subject meetings: 1. A memorandumto you from Dr. David Graham, which reviewsthe sponsor's July 17, 1990 submission entitled\"Summary of Post MarketingSafetyExperience.\" 2. A report by Drs. Franz Rosa and Carlene Baum of findingsfromin-houseanalysis of Ohio Medicaid data, entitled\"MedicaidDiagnosesBefore and After Starting Fluoxetine.\" 3.An update by Dr. Robert Wise on \"Fluoxetine IncreasedFrequencyReport Submissions.\" I wish to expand here on the last paragraph of page four of Dr. Graham's memorandum, whichrefersto the sponsor's exclusion of 76 casesfrom the suicidalityanalyses presentedin the July 17, 1990 submission in Table VIII.2.(page 42276) and Table VIII.4.(page 42278): In the analyses of suicidality, 76 of the total of 97 cases were excluded becausethey occurred in compassionate use studies or other studies which did not have controls.It is Inappropriatein a safety analysisto excludesuch a large proportion of cases.Afluoxetinesuicidalityrate should be computed for the uncontrolledstudies and comparedto the rate for the controlled studies, whichis 21 cases/3333 users, or about 0.6 percent(representing 9 cases/1741 usersin the depression studies and 12 cases/1592 users in the non-depression studies, p- 0.39, two-sidedtest for equality of rates).If the 000522

cc : suicidalityratesdo notdiffersignificantlybetveentheuncontrolledstudiesandthe controlledstudies, an overallrateshouldbeusedinthecomparisonswithotherdrugs(andinthe estimationofsamplesizerequirementsforfutureresearch.)Iftheratesdodiffersignificantly,thegroupsof fluoxetineusersupon whichthey are basedshouldbestudiedfurtherina efforttoidentify pre-treatmentriskfactorsfortheemergenceof suicidalityduringfluoxetineuse.Finally,Irecommendthat suicidalitycase-control analysesnestedintheseandothercohortsoffluoxetineusers beperformedforinvestigationof pre-treatmentrisk factors. Bruce V. Stadel, MD, MPH HFD-120/Laughren/Brecher HFD-700/Anello HFD-733/Stadel/Graham/Rosa/Bauin/Vise HFD-735/Barash NDA 018,936DRU1.7fluoxetineChron 000523

M E M O R A N D U M DEPARTMENTOFHEALTHANDHUMANSERVICESPUBLICHEALTHSERVICEFOODANDDRUCADMINISTRATIONCENTERFORDRUCEVALUATIONANDRESEARCHDATE : H98 orrn 1990 TO: D i r e c t o r ,D i v i s i o nofNeuropharmacologyDrugProducts(HFD-120)THROUGH:ActingD i r e c t o r ,OfficeofEpidemiologyandB i o s t a t i s t i c s(HFD-700)Cft-M^l'?[SUBJECT:Sponsor'sADRsubmissiononfluoxetinedatedJ u l y17,1990[r FROM:. Section Chief, Epidemiology Branch(HFD-733)IThe sponsor wasasked bythe reviewing divisionto analyzeand discuss postmarketing data on fluoxetineforits firsttwo yearsof marketingrelatingto several differentpotentialreactions.Thereport submitted bythe firm addressedeightreaction entitlesand included a review of bothIND clinical trial experienceand domesticspontaneous adverse reaction reporting. Eos inophllla.Eosinophilia was notedIn 19 fluoxetineand14 placebopatientsduringIND studies.Two fluoxetineand one placebo patient developedrashinassociation with eosinophilia, and one other fluoxetinepatient developed associatedfever.The study groupsizes were 2044 fluoxetineand1397 placebo patients. Frompostmarketingdata, there were17 reports of eosinophilia.Eight had eosinophilia-myalgiasyndromeor someching resemblingIt.Three ofthese eight hadcoTicomitant L-tryptophanandthe remaining five did not.Thefirm concluded thatthere was no pattern suggestiveof eosinophilia-myalgiasyndromein this data.Whileitis truethat CDC has epidemiologicallylinked eosinophilia-myalgiasyndrometo a single Japanese manufacturerof L-tryptophanand has postulatedthatthe syndrome may be due to a contaminant,this does not explain the five cases reported witheosinophiliaand arthralgiaor myalgia, with or withoutfeveroccurringInthe absenc of L-tryptophan. :CulllalnBarre SyndromeCGBS).In its introductiontothis section,the sponsor notedthat zimelidine, a serotonin uptake inhibitor, was associated with GBS and thisledto its withdrawal.Becausefluoxetineis also a serotonin uptake inhibitor,, thefirm wasinterestedin pursuingthis.Thefirm reportedsevencases, of whichthey considerthreeto be probable or definite,two unlikelyandtwo uncertain becauseof Incompletefollow-upor data.In reviewingthematerialsubmitted,one ofthe caseslabeled unlikely bythe sponsormay be a true case.Case6 is compatible withthe diagnosis of GBSin thatrapidlyprogressiveexteraity weakness wasassociated with a demyellnativeEMG/NCV.000524

2 Thefirm cited backgroundratesfor CBS of 0.6to 1.9per100,000 per year and with an estimated2.1 millionfluoxetineexposed patients(method of this estimation not described),concludedthat there was notrendin the datatosuggest an association. Severalissues areimportantto consider becausethey may necessitate a change in this conclusion.First, underreportingof adversereactionsis not addressed bythe firm and may havesubstantial effect here.Second,the incidence rates cited bythe sponsorare based on100,000 person-yearsof observation.In its analysis,the firm hasimplicitlyassumedthatthe estimated2.1 million patientstreated withfluoxetineall receivedit for one year.Fromother work we havedone withantidepressantsin the past, thisis probably not a valid assumption.Finally, in reviewingthe case materialprovided, duration of therapyfor most was abouttwo monthsor less.The risk of drug-inducedCBSisusually confinedto someinitialperiod of exposure, after whichitfallstobackgroundlevels.Thisis consistent with whatis believedto betheunderlyingimmunologicbasisforthe reaction./•AveragePerson-ExpectedCasesDurationYearsBackgroundRatesTherapyAccrued0.61.91 month180,0001.13.43 months500,0003.09.56 months1,000,0006.019.012 months2,100,00012.038.0Thistableshowsthe \"expected\" number of CBS casesin a populationof2.1million people followedfor varying durations of time upto one yearif the backgroundrate for diseaseis 0.6or 1.9per 100,000 per year.Fromspontaneousreportswe have3-5casesbythe firm's estimateand 4-6by our estimate.Giventhat underreporting may be substantial, that most cases had GBS onset bytwo months oftherapy, andthat only someinitial period oftime on drugis importantto reaction onset, it seems possiblethat fluoxetine use might be associated with GBS occurrence. Hyponatremia.Thefirm statedthat one case of hyponatremia had beenreportedas an ADR duringIND studiesin 6630 patients, but that serum sodium was not routinelycheckedso that effects of drug on serum sodiumcould not beevaluatedfromthese patients.From spontaneous sources, 20 cases were reportedthroughSeptember1988(covering8 months of marketing).Thefirm also presented reportsfromthe scientificliteratureshowingthat both serotonin and fluoxetineIncrease ADHlevels in experimental animals.Thefirm mentionedthatpossibleSIADHisinthe product label. Monoamineoxidaseinhibitor(\"MAPI) interactions.IND studies werereviewedforpatients whotook fluoxetineand KAOI'sIn closetemporal proximityorconcurrently.ThisIncluded16 patients on phenylzine, 24 ontranylcypromineand17 on Isocarboxacid.Amongthese, there weretwo patients with myoclonus, two with somnolence, one with syncopeand one with orthostatic hypotension. Spontaneousreportsthrough November1989included 5 fatal and 1 non-fatal case of fluoxetine/HAOIinteraction.Studiesin rats havealso shown that 000525

3 hyperpyrexia canresultfromthis interaction.Thefirm statesthisinteractionis describedin productlabelingandthat because ofthe long half-lifeofmetabolites,that KAOI'sshould not be usedin patients until afterthey have been off fluoxetinefor at least five weeks. Pulmonaryevents.ThefirmrevieweditsIND and postmarketingexperiencethrough mid-June1989, forthe reportingof a varietyof pulmonaryreactionterms and notedthat a nuinber of cases suggestive ofan \"inflammatory\" or \"allergic\" mechanism had been reported.It commentedthat many ofthese cases were complex but some had occurredin otherwise healthy people.Forthemajority,the only presenting symptom was dyspnea.\"Many\" had symptom resolution with discontinuation of fluoxetineandthe addition of steroidsinsome.No discussion was directedat those who did not haveresolution of symptoms.Thefirm alsonotedthatthe estimatedreportingrate forpulmonaryevents had declinedovertime.It concluded by drawing a connection between immune-mediatedor vasculiticrashreactions and pulmonaryevents, suggesting a spectrumof hypersensitivityresponsestothe drug.rReportingrates do nottranslateintoincidencerates becausethereIs probably substantial underreportingof events.Thethree-fold declinein reportingratesby quarterof marketingseenin two years provides evidence of this.The firm's analysis does not separate seriousfrom non-serious pulmonaryeventsand does not discussthe presenceor absenceof fatal cases.Thefirm statedit has modifiedthe productlabelto includesome referenceto \"other allergic events.\" Dyspneais generallythe onlysymptom presentin patientstaking fluoxetine who develop drug-Inducedpulmonary disease.We believe dyspneais atleast as importantas rashas anindicationof an immune/hypersensitivityreaction.Thefirm hasincludedin productlabelingthe recommendationto discontinue fluoxetineuponthe appearence of rash.Dyspnea, asthe most important, and usuallyonly symptom of allergic pulmonarydiseaseis not specificallymentionedas an indication for discontinuation. Selected hematolc^i: events.Thefirm received 506 spontaneousreports of hematolgicevents possiblyrelatedto increased bleeding(4X of allreports), of which130(262) were serious.Concomitant drugs capable of potentially affecting bleeding were presentIn 112.Dose did not appearto bea factor, and reports seemedlesslikely withinthe first 2 weeks and more commonafter 8 weeksof therapy.Platelet studies were donein 7 patients.Theresultsshownin table 3 underthe column labeled\"Epi 2\" suggesttothis reviewerthat both aspirin and fluoxetine have plateletinhibiting properties.The sponsor has reachedthe opposite conclusionthat fluoxetine does not inhibit platelet function.AsIn other sections of the firm's submission, fatal events were not separatelyevaluated or commented upon.. Sulcldalitv.Thefirmreviewed datafromIND studies, prefacingit withtheacknowledgementthatthesetrials were not designedforthe prospective evaluationof suicidality.In thesetrials, patientswith currentsuicidalideation were excluded.SuicidalIdeation was studiedintwo ways.ThefirstInvolvedanalysisof clinical comments ascertainedthrough non-probing, open-ended questionsduringthetrial.Also, atthe beginningandend ofthe study, patients completeda self-administeredquestionnaire,the Hamilton Rating Scale for Depression, whichincluded one question on suicide.This question, referred 000526

A to as HAMD-3, rated suicidalideation on an ordinal scalefrom 0(absent)to 4 (severeideation, usuallywith an attempt).The capacityof thesetrialstoidentify and describethe quality andintensityof suicidality waslow.Thefirm's review coveredIND studiesthroughlate December 1989.There were 97 reportsof suicidality withfluoxetine(21 whilein INDtrials and76 during compassionateor open-label use), 9 with placeboand 2 withtricyclic controls. The76 fluoxetine casesfrom studies other than double-blindand controlled were excludedfromthe firm's meta-analysis.Combining all studies(table 8.2), the suicidalityrate was 0.517Z withfluoxetine, 0.1781 withplaceboand 0.273Z with tricyclic controls.Thefirm reportedthese differences were notstatisticallysignificant. InitsIntroductorydiscussion,the firm called attentionto an abstract by Fava and Rosenbaua which\"concludedthatthere were no statisticallysignificantdifferencesamong rates of treatment-emergent suicidalideation associated with five classesof antidepressanttherapy.\"Whilethis istechnically correct, the actual datafromthis retrospectivechart-review studydo raisesomepotentialquestions.The data are shown in thetable below. Fluox +HAOI or FluoxTCATGA+/-L1OtherTotaltreated29473458192Pre-existing suicidality651375?Treatment-emergent suicidality6(2.9Z)2 (3.3Z)3 (0.8Z)1 (0. Z) Treatment-emergentsuicidality was morefrequent among\"fluoxetinealone\" than \"tricyclics withor withoutlithium\" patients.The relativerisk of suicidality was3.3(95Z CL 0.9, 12.2),p - 0.07. There are many problems withthis studythat cannot be assessed.Thedistribution of pre-existing suicidality betweenthe fluoxetineandtricyclicgroups was different(p - 0.05).Thisraisesthe questionthattheremainingpatientsnot suicidal at baselinein the fluoxetine group may have been more severelydepressedthan thosein thetricyclic group, butthisis purely speculative.Wealso don't know if patient groups were similar or dissimilar withrespectto other factorsimportantto suicidal ideation. Overall,the analysis presented bythe firm had several shortcomingswhichshould be noted.In the meta-analysisof suicidalityfromIND trials, 76 fluoxetinecases were excludedfrom analysis becausethe patients wereinstudiesor othertrialslacking comparative controls.It can be arguedthattheseexclusionsare not justifiedor appropriatein a meta-analysiswhere data contributingto boththe numeratorand denominatorof fluoxetinewerecollectedand are analyzable.Werethese casesIncluded, substantial differencesinsuicidality between drugscould have been observed.A related problemis that suicidalIdeation was probably viewedas a component ofthe underlying 00052?

5 depressivedisorderand hencefrequentlynot commenceduponor noted byresearchphysicians andnurse monitorsintheIND studies.This possibilityis mentioned bythe firm. Other problemsrelatetothe analysis of HAMD-3 scores.The analyses compared onlythe startand finish scores, ignoringthe possibilityof intercurrent suicidalicy whichresolved bythe completion of the study.Also,to be counted as a case of suicidality,the patient hadto have a HAMD-3score of 3 or 4, requiring\"suggestive behavior\"indicative of suicidalityor a serious attempt. This may betoo stringent a requirement, especiallyif the goalisto detect increasesin or characterizethe nature of suicidalideation.Violent behavior.Thefirm beganthis section with an overview of the prevalence of violent behaviorinthe United Statesand juxtaposedthis with mentionof\"fewerthan 10\" spontaneousreports of violenceamongfluoxetineusers.'This cannot beinterpretedco meanthat fluoxetinereducestheoccurrence of this behavior asimplied bythe firm.Rather,it demonstrates how great underreporting is. The analysis of clinicaltrials data was reported bythe firmto show a statisticallysignificantlower occurrenceof violent behavior as defined bythe\"aggression cluster\" ofterms amongfluoxetine patients comparedto placebo. The data forthis comparison werederivedfrom spontaneouslyreported events during clinicaltrials, not intentionally ascertained.As a result, these data do not permit any conclusion regardingthe comparative occurrenceof violent behavior. Discussion Thefirm presented a review of eight selectedadverseevents.Our assessment differs somewhatfromthe sponsor'sin several areas.One comment applicable to the entire submissionis that fatal reports were not separately analyzed or described. For eosinophilia,therearefive spontaneousreportsof reactionssuggestive of eosinophilia-myalgiasyndromein the absence of L-tryptophanuse.The firm does not viewthis as a problem. For CBS, we believethe existing data raisethe possibilitythat fluoxetine confers an increasedrisk of occurrence.The firm reached an opposite conclusion but failedto account for underreporting of adverse events.In other situations which have been documented, fewer than 10-202 of fatal or potentially fatal adverseevents have beenreported.Thefirm also did not account forthedifference between number of personsexposedto a drugandthe cumulative person-timeof exposure.In the situation of GBS, one mustalso account for the probableimmunologic basisforthe disorder.Foreign antigen(s) capable of triggeringthis reactiontypicallydo so over a shorter ratherthan longer periodof exposure.Inthe case with fluoxetine, itis possiblethat a patient on fluoxetinefor morethan one ortwo months ceasesto beat riskfor \"drug-induced\" CBS.Ifthe majorityof patients usedthe drugforlonger periodsof000528

6 time,che relativerisk could be substantiallyunderestimatedif the concept of \"periodat risk\"is not adjusted. For pulmonaryreactions, dyspneais the only symptomin most patients subsequentlyfound or suspectedto haveimmunologicallybased(hypersensitivity)lung disease.Thefirmcurrentlyrecommends discontinuationof fluoxetineifrash appears.We believedyspnea,asa symptomof possibleallergicpulmonarydiseaseis at least as importantas rash, but dyspneais not currentlyspecifiedas an indicationfor discontinuation. Thefirm's analysisof suicidalitydoes not resolvetheissue.Thefirmacknowledgedthatits clinicaltrials were not designedto studythis andthatthe quality and specificityof datato be gleanedfromthesetrialsto address suicidality were poor.Thedata presentedin sometables shoved higher percentages of suicidalityamongfluoxetine patientsthan amongtricyclicorplacebo patients, butthesedifferencesdid not reachstatisticalsignificance.The discussionofthe reportby Teicheret al. pointedoutthe difficultproblemof studyingthis question.However,the firm's strongestargumentagainstthefindingsof Teicher werechoseit presentedfrom Fava and Rosenbaum.As shown above,the summary providedbythefirm whiletechnicallycorrectdid not expressthe overall appearenceof the data.The actual datashowed a higher percentageoftreatment-emergentsuicidalityamong fluoxetine(2.9Z) than tricyclic(0.8X) patients with borderlinestatisticalsignificance.Interestingly,the proportion of patients withtreatment-emergentsuicidalityonfluoxetinein this study wassimilarto that reported by Teicher et al. Because of apparentlargescale underreporting,the firm'sanalysis cannot be consideredas provingthatfluoxetineand violent behaviorare unrelated. David J.Graham, MD, MPH Concur:/, Chief, Epidemiology Branch cc: Laughren/Brecher(HFD-120)Anello(HFD-700)Stadel/Craham/Rosa/Baum/Vise(HFD-733)Barash(HFD-735)NDA tf 18,936 • DRU1.7fluoxetineChron 000529

Clinical Trials are not designed to adequately test for side effect of suicidalityThe hypothesis of whether antidepressants cause suicidality has ​never ​been prospectively studied. Clinical trials are useful to prove that a drug has an intended effect, however, they were not designed to determine questions such as whether a drug is causing suicidality. 1. Clinical trial data is not a good measure to test rare events. Suicidality, particularly completed suicides, are rare events. Patients who are suicidal are excluded from most clinical trials, and a significant percentage of patients quit clinical trials due to side effects. Epidemiologists Gunnell and Ashby (1995 ​BMJ ​article) wrote: “Suicide is rare, even among people with depression. [Cite omitted.] Thus, most clinical trials have insufficient power to provide clear evidence on the effect of antidepressants on suicide.” As the editor of the New England Journal of Medicine thoughtfully explained: First “a drug is approved because it is more effective than placebo.” Then, “worries emerge about its safety.” However, “few or no adequately powered controlled trials [conducted by drug companies] are conducted to address these issues.” Thus, “The health care system has a hard time performing drug safety analyses, in large part because it relies on the pharmaceutical industry to conduct most research on the risks and benefits of medications. It is naive to expect companies to voluntarily fund studies that could sink lucrative products” and further complicating matters is the fact that “the FDA lacks regulatory clout to require them.” (N N Engl J Med. 2006 Nov 23; 355(21):2169-71: ​VIEW FULL DOCUMENT​)











Epidemiologists Gunnell and Ashby (1995 ​BMJ ​article) wrote:“ Suicide is rare, even among people with depression. [Cite omitted.] Thus, most clinical trials have insufficient power to provide clear evidence on the effect of antidepressants on suicide.” Despite the unlikelihood that a statistically significant increased risk will be found from clinical trials conducted by drug companies to obtain FDA approval, meta-analyses of both child/adolescent and adult clinical trials have revealed a risk. Dr. Thomas Newman, an epidemiologist from the University of San Francisco and an advisor to the FDA on the issue of child/adolescent suicidality (following recommendations that antidepressants carry black box warnings) noted that, the fact that higher rates of suicidality in those taking antidepressants have emerged from the clinical trials is “striking” and “such a dramatic result would be expected to occur by chance only 1 time in 20,000.” Dr. Newman observed that “some FDA staff and committee members expressed reservations about the data used for this analysis,” but as he explained, “these concerns only made the results more compelling.” He further stated: “The fact that an association emerged from the meta-analysis … for an outcome that the sponsors of the trials were not looking for, and presumably did not wish to find, was quite convincing.” (​VIEW FULL DOCUMENT​) The fact that a risk has been detected in clinical trials not intended to answer this question for an event that is rare, even in depressed patients, makes the evidence even stronger.

n engl j med 351;16www.nejm.orgoctober 14, 20041595PERSPECTIVEreports about resistance of the balloon to withdraw- terms of the rate of adverse events. Our responsi-al attempts after the stent is deployed — a problem bility to protect the public health, however, neces-sometimes described as “stickiness” of the balloon sitates that we consider early notification about anyduring withdrawal. This problem does not appear legitimate risk, weighing the risk posed by neglect-to be related to the mechanical defect observed in ing to notify practitioners about a clinically signifi-the investigation of the balloon-deflation failures; cant problem against the consequences of “cryingrather, it seems to result from a combination of wolf” on the basis of inconclusive information. Inpatient-related and device-related factors. These attempting to strike the right balance, we may en-factors are still under review but are currently pre- gage clinical and other stakeholders for review andsumed to include friction between the stent-delivery comment on our draft notifications before we re-balloon and the drug-polymer coating on the stent, lease them. Such a dialogue is particularly impor-as well as certain characteristics of the coronary ves- tant in the case of breakthrough technologies, insel. The association, if there is any, between these which our decisions can have an immediate andreports of resistance to withdrawal and adverse clin- profound effect on people’s lives.ical events is currently being assessed.The Boston Scientific case illustrates the dilem-ma we face in determining the appropriate thresh-old for product recall and notification when therisk is of a severe but rare adverse event. In thesecases, the potential benefit to be accrued from re-calling the product must be balanced against thepossible negative effects this action might have onpublic health. Denying a potentially beneficial ther-apy to all patients, as in a complete recall of a prod-uct, might have greater adverse consequences thanallowing the device to remain in use.These cases exemplify the challenges that weface in regulating breakthrough technologies. In al-most all instances, as in the Cypher case, we do nothave a complete accounting of all cases or informa-tion on the number of devices actually in use, mak-ing it impossible to compare similar products inFrom the Center for Devices and Radiological Health, Food andDrug Administration, Rockville, Md.1.CYPHER sirolimus-eluting coronary stent on RAPTOR over-the-wire delivery system or RAPTORRAIL rapid exchange deliverysystem — P020026. Rockville, Md.: Food and Drug Administra-tion, April 24, 2003. (Accessed September 29, 2004, at http://www.fda.gov/cdrh/pdf2/p020026.html.)2.FDA public health Web notification: information for physi-cians on sub-acute thromboses (SAT) and hypersensitivity reac-tions with use of the Cordis CYPHER coronary stent. October 29,2003. (Accessed September 24, 2004, at http://www.fda.gov/cdrh/safety/cypher.pdf.)3.FDA public health Web notification: updated information forphysicians on sub-acute thromboses (SAT) and hypersensitivityreactions with use of the Cordis CYPHER sirolimus-eluting coro-nary stent. November 25, 2003. (Accessed September 28, 2004,at http://www.fda.gov/cdrh/safety/cypher2.pdf.)4.TAXUS Express 2 paclitaxel-eluting coronary stent system(monorail and over-the-wire) — P030025. Rockville, Md.: Foodand Drug Administration, March 4, 2004. (Accessed September28, 2004, at http://www.fda.gov/cdrh/pdf3/p030025.html.)On September 14, 2004, a Food and Drug Ad- and behavior (“suicidality”) in pediatric patients. Al-ministration (FDA) joint advisory committee voted though, as an epidemiologist and general pediatri-15 to 8 to recommend that a “black-box” warning cian, I do not have clinical experience caring forlabel be required for antidepressant drugs, indicat- depressed patients, after reviewing the evidence, Iing that they increase the risk of suicidal thinking strongly favored the black-box warning.treating depression in childrenThe editors asked two members of the Psychopharmacologic Drugs and Pediatric Advisory Committees of the Food and Drug Administrationto comment on the committees’ recent recommendations regarding the use of antidepressant medications in children and adolescents. Their responses follow.A Black-Box Warning for Antidepressants in Children?Thomas B. Newman, M.D., M.P.H.Problems with Drug-Eluting Coronary Stents — The FDA PerspectiveCopyright © 2004 Massachusetts Medical Society. All rights reserved. Downloaded from www.nejm.org by on July 8, 2005 .

n engl j med 351;16www.nejm.orgoctober 14, 20041596PERSPECTIVEDuring consideration of the proposed labeling not wish to find, was quite convincing.change, the committee heard a number of presen-tations summarizing evidence that suicidality in ported by public testimony from people who be-children and adolescents may be increased by the lieved that antidepressant drugs had caused theirnewer antidepressant drugs, primarily selective se- loved ones to commit suicide (or, in some cases,rotonin-reuptake inhibitors. The most convincing homicide). Several of these cases involved patientsevidence came from an FDA analysis of random- who had shown no hint of suicidality before begin-ized trials. Most of these trials had been conducted ning treatment with the drugs and who had beenby the drug manufacturers under the Best Pharma- given these drugs for indications other than depres-ceuticals for Children Act, which provides compa- sion, including migraine headaches, nail biting,nies an additional six months of patent protection anxiety, and insomnia.for their product if they do pediatric studies. Thesestudies need not be published and need not be of the antidepressants, citing either their own clini-high quality. In fact, we heard that because these cal experience or the Treatment for Adolescentsmedications are already widely prescribed “off-la- with Depression Study (TADS), a recently pub-bel” and patents may be close to expiration, spon- lished randomized, double-blind study of fluoxe-sors may have more incentive to do the studies tine for major depression in adolescents, whichquickly than to do them well. To facilitate analysis the committee reviewed in detail. However, othersof the patchwork of pediatric studies, the FDA ob- and I found the evidence of efficacy much less con-tained narratives of adverse-event reports from vincing than the evidence of harm. In reviewingthe trials and contracted with experts on suicide at TADS, we were struck by the small size of the dif-Columbia University to review them. The Colum- ference between fluoxetine and placebo as com-bia staff members, who were unaware of the treat- pared with the effect of placebo alone. For example,ment-group assignments, were asked to determine after 12 weeks, the average decrease in the Child-whether the adverse events represented suicidali- hood Depression Scale–Revised (scores on whichty. FDA staff members then combined the results were around 60 of a possible 113 at baseline in bothinto a meta-analysis.The results were striking. When all the pediatric depression) was 19.4 points with placebo, as com-trials were pooled, the rate of definite or possible pared with 22.6 points with fluoxetine (see Figure).suicidality among children assigned to receive an-tidepressants was twice that in the placebo group.(The summary risk ratio was 2.19; 95 percent con-fidence interval, 1.50 to 3.19.) Although the FDAstaff did not provide this information to the commit-tee, according to my own calculations, such a dra-matic result would be expected to occur by chanceonly 1 time in 20,000 (P=0.00005).Nonetheless, some FDA staff and committeemembers expressed reservations about the dataused for this analysis. For example, there was a rel-atively small number of events, the trials had notbeen designed to evaluate suicidality, and the meth-ods of ascertainment and classification of the eventsin the various trials were not uniform. To me, how-ever, these concerns only made the results morecompelling. Inadequate sample size and misclas-sification of outcomes make it more — not less —difficult to detect differences between groups inrandomized, blinded trials. The fact that an associ-ation emerged from the meta-analysis with a P val-ue of 0.00005, for an outcome that the sponsors ofthe trials were not looking for, and presumably didInferences from the randomized trials were sup-Several committee members spoke in favor of1groups, with higher scores indicating more severeFigure. Mean Changes in Score on the Childhood De-pression Scale–Revised in the Treatment for Adolescents with Depression Study.Data are from March et al. The average scores at base-1line, out of a possible 113, were 58.9 in the fluoxetine group and 61.2 in the placebo group. Higher scores indi-cate more severe depression.Mean Change in ChildhoodDepression Scale – Revised¡10¡5¡15¡20 ¡250612Weeks of Treatment0FluoxetinePlaceboA Black-Box Warning for Antidepressants in Children?Copyright © 2004 Massachusetts Medical Society. All rights reserved. Downloaded from www.nejm.org by on July 8, 2005 .

n engl j med 351;16www.nejm.orgoctober 14, 20041597PERSPECTIVEA Black-Box Warning for Antidepressants in Children?It is easy to see why the personal experience of cli- cacy of antidepressants in children was minimal andnicians and patients would lead them to believe the likely to have been overestimated, because publisheddrug to be effective, since they would have no way studies have much more favorable results than un-of knowing that more than 85 percent of the bene- published studies. Thus, both clinical experiencefit they observed would also have occurred with and published trials are likely to lead to inflated es-placebo.Randomized trials other than TADS have hadless favorable results. The FDA indicated that only wide gaps in our knowledge about antidepres-3 of 15 trials of antidepressant use in children sants. Perhaps the most important relate to theirwith depression had found a statistically signifi- medium-term and long-term safety and efficacy.cant benefit. The agency also provided us with a The FDA’s meta-analysis suggested that the newmeta-analysis that showed that the estimated effi- antidepressants double the risk of suicidality, from2timates of the efficacy of these drugs.The committee members agreed that there areFDA Statement on Recommendations of the PsychopharmacologicDrugs and Pediatric Advisory Committees, September 16, 2004.*he Food and DrugAdministration (FDA)generallysupportsthe recommendations thatwere recently made to theagency by the Psychophar-macologic Drugs and Pedi-atric Advisory Committeesregarding reports of an in-creased risk of suicidality (sui-cidal thoughts and actions)associated with the use of cer-tain antidepressants in pe-diatric patients. FDA hasbegun working expeditious-ly to adopt new labeling toenhance the warnings asso-ciated with the use of antide-pressants and to bolster theinformation provided to pa-tients when these drugs aredispensed.In summary, the members ofthe advisory committees:•endorsed FDA’s approachto classifying and analyz-ing the suicidal events andbehaviors observed in con-trolled clinical trials and ex-pressed their view that thenew analyses increased theirconfidence in the results• concluded that the find-ing of an increased risk ofsuicidality in pediatric pa-tients applied to all the drugsstudied (Prozac, Zoloft, Re-meron, Paxil, Effexor, CelexaWellbutrin, Luvox and Ser-zone) in controlled clinicaltrials• recommended that anywarning related to an in-creased risk of suicidality inpediatric patients should beapplied to all antidepressantdrugs, including those thathave not been studied incontrolled clinical trials inpediatric patients, since theavailable data are not ade-quate to exclude any singlemedication from an in-creased risk• reached a split decision(15-yes, 8-no) regarding rec-ommending a “black-box”warning related to an in-creased risk for suicidality inpediatric patients for all anti-depressant drugs• endorsed a patient infor-mation sheet (“MedicationGuide”) for this class ofdrugs to be provided to thepatient or their caregiverwith every prescription• recommended that theproducts not be contraindi-cated in this country becausethe Committees thought ac-cess to these therapies wasimportant for those whocould benefit• recommended that theresults of controlled pediat-ric trials of depression beincluded in the labeling forantidepressant drugs.*From the Food and Drug Adminis-tration, www.fda.gov/bbs/topics/news/2004/new01116.html.tCopyright © 2004 Massachusetts Medical Society. All rights reserved. Downloaded from www.nejm.org by on July 8, 2005 .

n engl j med 351;16www.nejm.orgoctober 14, 20041598PERSPECTIVEabout 2.5 percent to 5 percent, in trials lasting two ed exclusivity should be granted only if the studiesor three months. But what happens if you take them conducted in order to receive it are judged to be offor a year? Does a 5 percent risk over the course of high quality by independent peer review and if theirthree months become a 20 percent risk over the results are disseminated in a timely manner.course of a year, or does the benefit–risk balanceimprove over time? Does the increase in the rate pressants in children illustrates the need for a moreof suicidality translate into an increase in the rate sophisticated approach to evaluating harms andof completed suicide? Do additional adverse effects efficacy than simply seeing whether they are statis-occur after treatment with the medications is tically significant at a P value of less than 0.05. Re-stopped? Are there important differences among gardless of the P value, no psychotropic drug is freethe drugs in this class? What age groups are at risk? of potential negative effects. The best available es-It seems unlikely that the drugs suddenly become timates of the magnitude and nature of the effectssafe after the patient’s 18th birthday. Currently, no of the drugs must be discussed with patients andone knows the answer to any of these questions.There was also agreement that the experience sion about treatment. My hope is that the FDA willwith antidepressants illustrates serious deficien- follow the recommendation of the advisory com-cies in the implementation of the Best Pharmaceu- mittee and require a black-box warning — and thatticals for Children Act U.S. consumers, through doing so will make these discussions more likelyhigher drug prices buoyed by extended market ex- to take place.clusivity, are paying a high price for pediatric stud-ies that may be poorly conducted and selectively dis-seminated. The public is the loser, both becausethe availability of generic versions of the drugs isthen delayed and because poor science and selec-tive publication can lead to false conclusions aboutpediatric safety and efficacy. If these studies are toprovide value, the requirement to do them must bechanged to a requirement to do them well. Extend-Finally, this controversy over the use of antide-families, so that they can make an informed deci-From the Departments of Epidemiology and Biostatistics and Pe-diatrics, University of California San Francisco, San Francisco.1.March J, Silva S, Petrycki S, et al. Fluoxetine, cognitive-behav-ioral therapy, and their combination for adolescents with depres-sion: Treatment for Adolescents With Depression Study (TADS)randomized controlled trial. JAMA 2004;292:807-20.2.Whittington CJ, Kendall T, Fonagy P, Cottrell D, Cotgrove A,Boddington E. Selective serotonin reuptake inhibitors in childhooddepression: systematic review of published versus unpublisheddata. Lancet 2004;363:1341-5.There is great concern that antidepressants used in too much stress. There was debate about whetherchildren and adolescents may paradoxically in- depression could occur in children, and the prevail-crease their risk of suicidal thoughts and behavior. ing view was that moodiness was normal in ado-Is this concern valid, and if so, how should it mod- lescents. Furthermore, even if we could have diag-ify our clinical approach to pediatric depression?Twenty-five years ago, long before the intro- who were at risk for suicide, there were no empiri-duction of selective serotonin-reuptake inhibitors cally validated treatments to offer.(SSRIs), the adolescent suicide rate was increasingrapidly, having tripled over the previous two de- deed affect children and adolescents. Through ret-cades, but the risk factors involved were unknown. rospective interviews with family members andAdolescents who committed suicide were regarded friends, this disorder emerged as the single mostas misunderstood teenagers who had been under important risk factor for adolescent suicide, al-nosed depression and recognized young peopleEventually, we learned that depression did in-1Antidepressants and Pediatric Depression — The Risk of Doing NothingDavid A. Brent, M.D.A Black-Box Warning for Antidepressants in Children?Copyright © 2004 Massachusetts Medical Society. All rights reserved. Downloaded from www.nejm.org by on July 8, 2005 .

2. HAMD an insensitive measure of treatment-emergentsuicidalityRather than legitimately studying the link between their drugs andsuicidality, SSRI manufacturers have conducted analyses of theirclinical trial databases using a scale called the “Hamilton DepressionScale” (HAMD), a scale used to assess changes in the degree ofdepression of patients enrolled in the clinical trials. The HAMDcontains one question concerning suicidality. It is an insensitivemeasure of treatment-emergent suicidality. Nevertheless, theseanalyses, inaccurately described as “thorough,” have been used bythe manufacturers to claim their drugs have been “exonerated”against claims of increased suicidality.Drs. Healy and Creaney criticized Lilly’s analysis in 1991 andLilly’s use of HAMD Item 3 to analyze the risk (1991BritishMedical Journalarticle):Lilly’s “analysis of whether there is an association betweenfluoxetine and suicidality does not entirely settle the questionraised by Teicher/Cole of whether treatment with fluoxetine mayin certain instances lead to suicidal ideation.” There were severalreasons for this, the first being that “item 3 of the Hamilton scalefor depression, ratings on which provide the data for the analysis,is an insensitive measure of suicidality … the capacity of thesetrials to identify and describe the quality and intensity ofsuicidality was low.”Senior FDA epidemiologist, Dr. David Graham was likewiseskeptical of Lilly’s use of the HAMD Item 3 measure:Lilly’s “analysis of suicidality does not resolve the issue.”(VIEW FULL DOCUMENT)

MEMORANDUM DEPARTMENT OF HEALTH AND HUMANSERVICESPUBLIC HEALTHSERVICESFOOD AND DRUG ADMINISTRATION CENTER FOR DRUG EVALUATION AND RESEARCH 3BP\" 11990DATE: FROM:Chief,EpidemiologyBranch,HFD-733SUBJECT: -Meetingsonexperiencefromfluoxetinesurveillance:..September18(in-house)andSeptember25(withfirm)THROUGH:ActingDirector,CA.<*\\ioOfficeofEpidemiologyandBiostatistics,HFD-701TO: Director, Division of NeuropharmacologyDrug Products, HFD-120 Attached are three documents which have been prepared by Epidemiology Branch staff as backgroundfor the subject meetings: 1. A memorandumto you from Dr. David Graham, which reviewsthe sponsor's July 17, 1990 submission entitled\"Summary of Post MarketingSafetyExperience.\" 2. A report by Drs. Franz Rosa and Carlene Baum of findingsfromin-houseanalysis of Ohio Medicaid data, entitled\"MedicaidDiagnosesBefore and After Starting Fluoxetine.\" 3.An update by Dr. Robert Wise on \"Fluoxetine IncreasedFrequencyReport Submissions.\" I wish to expand here on the last paragraph of page four of Dr. Graham's memorandum, whichrefersto the sponsor's exclusion of 76 casesfrom the suicidalityanalyses presentedin the July 17, 1990 submission in Table VIII.2.(page 42276) and Table VIII.4.(page 42278): In the analyses of suicidality, 76 of the total of 97 cases were excluded becausethey occurred in compassionate use studies or other studies which did not have controls.It is Inappropriatein a safety analysisto excludesuch a large proportion of cases.Afluoxetinesuicidalityrate should be computed for the uncontrolledstudies and comparedto the rate for the controlled studies, whichis 21 cases/3333 users, or about 0.6 percent(representing 9 cases/1741 usersin the depression studies and 12 cases/1592 users in the non-depression studies, p- 0.39, two-sidedtest for equality of rates).If the 000522

cc : suicidalityratesdo notdiffersignificantlybetveentheuncontrolledstudiesandthe controlledstudies, an overallrateshouldbeusedinthecomparisonswithotherdrugs(andinthe estimationofsamplesizerequirementsforfutureresearch.)Iftheratesdodiffersignificantly,thegroupsof fluoxetineusersupon whichthey are basedshouldbestudiedfurtherina efforttoidentify pre-treatmentriskfactorsfortheemergenceof suicidalityduringfluoxetineuse.Finally,Irecommendthat suicidalitycase-control analysesnestedintheseandothercohortsoffluoxetineusers beperformedforinvestigationof pre-treatmentrisk factors. Bruce V. Stadel, MD, MPH HFD-120/Laughren/Brecher HFD-700/Anello HFD-733/Stadel/Graham/Rosa/Bauin/Vise HFD-735/Barash NDA 018,936DRU1.7fluoxetineChron 000523

M E M O R A N D U M DEPARTMENTOFHEALTHANDHUMANSERVICESPUBLICHEALTHSERVICEFOODANDDRUCADMINISTRATIONCENTERFORDRUCEVALUATIONANDRESEARCHDATE : H98 orrn 1990 TO: D i r e c t o r ,D i v i s i o nofNeuropharmacologyDrugProducts(HFD-120)THROUGH:ActingD i r e c t o r ,OfficeofEpidemiologyandB i o s t a t i s t i c s(HFD-700)Cft-M^l'?[SUBJECT:Sponsor'sADRsubmissiononfluoxetinedatedJ u l y17,1990[r FROM:. Section Chief, Epidemiology Branch(HFD-733)IThe sponsor wasasked bythe reviewing divisionto analyzeand discuss postmarketing data on fluoxetineforits firsttwo yearsof marketingrelatingto several differentpotentialreactions.Thereport submitted bythe firm addressedeightreaction entitlesand included a review of bothIND clinical trial experienceand domesticspontaneous adverse reaction reporting. Eos inophllla.Eosinophilia was notedIn 19 fluoxetineand14 placebopatientsduringIND studies.Two fluoxetineand one placebo patient developedrashinassociation with eosinophilia, and one other fluoxetinepatient developed associatedfever.The study groupsizes were 2044 fluoxetineand1397 placebo patients. Frompostmarketingdata, there were17 reports of eosinophilia.Eight had eosinophilia-myalgiasyndromeor someching resemblingIt.Three ofthese eight hadcoTicomitant L-tryptophanandthe remaining five did not.Thefirm concluded thatthere was no pattern suggestiveof eosinophilia-myalgiasyndromein this data.Whileitis truethat CDC has epidemiologicallylinked eosinophilia-myalgiasyndrometo a single Japanese manufacturerof L-tryptophanand has postulatedthatthe syndrome may be due to a contaminant,this does not explain the five cases reported witheosinophiliaand arthralgiaor myalgia, with or withoutfeveroccurringInthe absenc of L-tryptophan. :CulllalnBarre SyndromeCGBS).In its introductiontothis section,the sponsor notedthat zimelidine, a serotonin uptake inhibitor, was associated with GBS and thisledto its withdrawal.Becausefluoxetineis also a serotonin uptake inhibitor,, thefirm wasinterestedin pursuingthis.Thefirm reportedsevencases, of whichthey considerthreeto be probable or definite,two unlikelyandtwo uncertain becauseof Incompletefollow-upor data.In reviewingthematerialsubmitted,one ofthe caseslabeled unlikely bythe sponsormay be a true case.Case6 is compatible withthe diagnosis of GBSin thatrapidlyprogressiveexteraity weakness wasassociated with a demyellnativeEMG/NCV.000524

2 Thefirm cited backgroundratesfor CBS of 0.6to 1.9per100,000 per year and with an estimated2.1 millionfluoxetineexposed patients(method of this estimation not described),concludedthat there was notrendin the datatosuggest an association. Severalissues areimportantto consider becausethey may necessitate a change in this conclusion.First, underreportingof adversereactionsis not addressed bythe firm and may havesubstantial effect here.Second,the incidence rates cited bythe sponsorare based on100,000 person-yearsof observation.In its analysis,the firm hasimplicitlyassumedthatthe estimated2.1 million patientstreated withfluoxetineall receivedit for one year.Fromother work we havedone withantidepressantsin the past, thisis probably not a valid assumption.Finally, in reviewingthe case materialprovided, duration of therapyfor most was abouttwo monthsor less.The risk of drug-inducedCBSisusually confinedto someinitialperiod of exposure, after whichitfallstobackgroundlevels.Thisis consistent with whatis believedto betheunderlyingimmunologicbasisforthe reaction./•AveragePerson-ExpectedCasesDurationYearsBackgroundRatesTherapyAccrued0.61.91 month180,0001.13.43 months500,0003.09.56 months1,000,0006.019.012 months2,100,00012.038.0Thistableshowsthe \"expected\" number of CBS casesin a populationof2.1million people followedfor varying durations of time upto one yearif the backgroundrate for diseaseis 0.6or 1.9per 100,000 per year.Fromspontaneousreportswe have3-5casesbythe firm's estimateand 4-6by our estimate.Giventhat underreporting may be substantial, that most cases had GBS onset bytwo months oftherapy, andthat only someinitial period oftime on drugis importantto reaction onset, it seems possiblethat fluoxetine use might be associated with GBS occurrence. Hyponatremia.Thefirm statedthat one case of hyponatremia had beenreportedas an ADR duringIND studiesin 6630 patients, but that serum sodium was not routinelycheckedso that effects of drug on serum sodiumcould not beevaluatedfromthese patients.From spontaneous sources, 20 cases were reportedthroughSeptember1988(covering8 months of marketing).Thefirm also presented reportsfromthe scientificliteratureshowingthat both serotonin and fluoxetineIncrease ADHlevels in experimental animals.Thefirm mentionedthatpossibleSIADHisinthe product label. Monoamineoxidaseinhibitor(\"MAPI) interactions.IND studies werereviewedforpatients whotook fluoxetineand KAOI'sIn closetemporal proximityorconcurrently.ThisIncluded16 patients on phenylzine, 24 ontranylcypromineand17 on Isocarboxacid.Amongthese, there weretwo patients with myoclonus, two with somnolence, one with syncopeand one with orthostatic hypotension. Spontaneousreportsthrough November1989included 5 fatal and 1 non-fatal case of fluoxetine/HAOIinteraction.Studiesin rats havealso shown that 000525

3 hyperpyrexia canresultfromthis interaction.Thefirm statesthisinteractionis describedin productlabelingandthat because ofthe long half-lifeofmetabolites,that KAOI'sshould not be usedin patients until afterthey have been off fluoxetinefor at least five weeks. Pulmonaryevents.ThefirmrevieweditsIND and postmarketingexperiencethrough mid-June1989, forthe reportingof a varietyof pulmonaryreactionterms and notedthat a nuinber of cases suggestive ofan \"inflammatory\" or \"allergic\" mechanism had been reported.It commentedthat many ofthese cases were complex but some had occurredin otherwise healthy people.Forthemajority,the only presenting symptom was dyspnea.\"Many\" had symptom resolution with discontinuation of fluoxetineandthe addition of steroidsinsome.No discussion was directedat those who did not haveresolution of symptoms.Thefirm alsonotedthatthe estimatedreportingrate forpulmonaryevents had declinedovertime.It concluded by drawing a connection between immune-mediatedor vasculiticrashreactions and pulmonaryevents, suggesting a spectrumof hypersensitivityresponsestothe drug.rReportingrates do nottranslateintoincidencerates becausethereIs probably substantial underreportingof events.Thethree-fold declinein reportingratesby quarterof marketingseenin two years provides evidence of this.The firm's analysis does not separate seriousfrom non-serious pulmonaryeventsand does not discussthe presenceor absenceof fatal cases.Thefirm statedit has modifiedthe productlabelto includesome referenceto \"other allergic events.\" Dyspneais generallythe onlysymptom presentin patientstaking fluoxetine who develop drug-Inducedpulmonary disease.We believe dyspneais atleast as importantas rashas anindicationof an immune/hypersensitivityreaction.Thefirm hasincludedin productlabelingthe recommendationto discontinue fluoxetineuponthe appearence of rash.Dyspnea, asthe most important, and usuallyonly symptom of allergic pulmonarydiseaseis not specificallymentionedas an indication for discontinuation. Selected hematolc^i: events.Thefirm received 506 spontaneousreports of hematolgicevents possiblyrelatedto increased bleeding(4X of allreports), of which130(262) were serious.Concomitant drugs capable of potentially affecting bleeding were presentIn 112.Dose did not appearto bea factor, and reports seemedlesslikely withinthe first 2 weeks and more commonafter 8 weeksof therapy.Platelet studies were donein 7 patients.Theresultsshownin table 3 underthe column labeled\"Epi 2\" suggesttothis reviewerthat both aspirin and fluoxetine have plateletinhibiting properties.The sponsor has reachedthe opposite conclusionthat fluoxetine does not inhibit platelet function.AsIn other sections of the firm's submission, fatal events were not separatelyevaluated or commented upon.. Sulcldalitv.Thefirmreviewed datafromIND studies, prefacingit withtheacknowledgementthatthesetrials were not designedforthe prospective evaluationof suicidality.In thesetrials, patientswith currentsuicidalideation were excluded.SuicidalIdeation was studiedintwo ways.ThefirstInvolvedanalysisof clinical comments ascertainedthrough non-probing, open-ended questionsduringthetrial.Also, atthe beginningandend ofthe study, patients completeda self-administeredquestionnaire,the Hamilton Rating Scale for Depression, whichincluded one question on suicide.This question, referred 000526

A to as HAMD-3, rated suicidalideation on an ordinal scalefrom 0(absent)to 4 (severeideation, usuallywith an attempt).The capacityof thesetrialstoidentify and describethe quality andintensityof suicidality waslow.Thefirm's review coveredIND studiesthroughlate December 1989.There were 97 reportsof suicidality withfluoxetine(21 whilein INDtrials and76 during compassionateor open-label use), 9 with placeboand 2 withtricyclic controls. The76 fluoxetine casesfrom studies other than double-blindand controlled were excludedfromthe firm's meta-analysis.Combining all studies(table 8.2), the suicidalityrate was 0.517Z withfluoxetine, 0.1781 withplaceboand 0.273Z with tricyclic controls.Thefirm reportedthese differences were notstatisticallysignificant. InitsIntroductorydiscussion,the firm called attentionto an abstract by Fava and Rosenbaua which\"concludedthatthere were no statisticallysignificantdifferencesamong rates of treatment-emergent suicidalideation associated with five classesof antidepressanttherapy.\"Whilethis istechnically correct, the actual datafromthis retrospectivechart-review studydo raisesomepotentialquestions.The data are shown in thetable below. Fluox +HAOI or FluoxTCATGA+/-L1OtherTotaltreated29473458192Pre-existing suicidality651375?Treatment-emergent suicidality6(2.9Z)2 (3.3Z)3 (0.8Z)1 (0. Z) Treatment-emergentsuicidality was morefrequent among\"fluoxetinealone\" than \"tricyclics withor withoutlithium\" patients.The relativerisk of suicidality was3.3(95Z CL 0.9, 12.2),p - 0.07. There are many problems withthis studythat cannot be assessed.Thedistribution of pre-existing suicidality betweenthe fluoxetineandtricyclicgroups was different(p - 0.05).Thisraisesthe questionthattheremainingpatientsnot suicidal at baselinein the fluoxetine group may have been more severelydepressedthan thosein thetricyclic group, butthisis purely speculative.Wealso don't know if patient groups were similar or dissimilar withrespectto other factorsimportantto suicidal ideation. Overall,the analysis presented bythe firm had several shortcomingswhichshould be noted.In the meta-analysisof suicidalityfromIND trials, 76 fluoxetinecases were excludedfrom analysis becausethe patients wereinstudiesor othertrialslacking comparative controls.It can be arguedthattheseexclusionsare not justifiedor appropriatein a meta-analysiswhere data contributingto boththe numeratorand denominatorof fluoxetinewerecollectedand are analyzable.Werethese casesIncluded, substantial differencesinsuicidality between drugscould have been observed.A related problemis that suicidalIdeation was probably viewedas a component ofthe underlying 00052?

5 depressivedisorderand hencefrequentlynot commenceduponor noted byresearchphysicians andnurse monitorsintheIND studies.This possibilityis mentioned bythe firm. Other problemsrelatetothe analysis of HAMD-3 scores.The analyses compared onlythe startand finish scores, ignoringthe possibilityof intercurrent suicidalicy whichresolved bythe completion of the study.Also,to be counted as a case of suicidality,the patient hadto have a HAMD-3score of 3 or 4, requiring\"suggestive behavior\"indicative of suicidalityor a serious attempt. This may betoo stringent a requirement, especiallyif the goalisto detect increasesin or characterizethe nature of suicidalideation.Violent behavior.Thefirm beganthis section with an overview of the prevalence of violent behaviorinthe United Statesand juxtaposedthis with mentionof\"fewerthan 10\" spontaneousreports of violenceamongfluoxetineusers.'This cannot beinterpretedco meanthat fluoxetinereducestheoccurrence of this behavior asimplied bythe firm.Rather,it demonstrates how great underreporting is. The analysis of clinicaltrials data was reported bythe firmto show a statisticallysignificantlower occurrenceof violent behavior as defined bythe\"aggression cluster\" ofterms amongfluoxetine patients comparedto placebo. The data forthis comparison werederivedfrom spontaneouslyreported events during clinicaltrials, not intentionally ascertained.As a result, these data do not permit any conclusion regardingthe comparative occurrenceof violent behavior. Discussion Thefirm presented a review of eight selectedadverseevents.Our assessment differs somewhatfromthe sponsor'sin several areas.One comment applicable to the entire submissionis that fatal reports were not separately analyzed or described. For eosinophilia,therearefive spontaneousreportsof reactionssuggestive of eosinophilia-myalgiasyndromein the absence of L-tryptophanuse.The firm does not viewthis as a problem. For CBS, we believethe existing data raisethe possibilitythat fluoxetine confers an increasedrisk of occurrence.The firm reached an opposite conclusion but failedto account for underreporting of adverse events.In other situations which have been documented, fewer than 10-202 of fatal or potentially fatal adverseevents have beenreported.Thefirm also did not account forthedifference between number of personsexposedto a drugandthe cumulative person-timeof exposure.In the situation of GBS, one mustalso account for the probableimmunologic basisforthe disorder.Foreign antigen(s) capable of triggeringthis reactiontypicallydo so over a shorter ratherthan longer periodof exposure.Inthe case with fluoxetine, itis possiblethat a patient on fluoxetinefor morethan one ortwo months ceasesto beat riskfor \"drug-induced\" CBS.Ifthe majorityof patients usedthe drugforlonger periodsof000528

6 time,che relativerisk could be substantiallyunderestimatedif the concept of \"periodat risk\"is not adjusted. For pulmonaryreactions, dyspneais the only symptomin most patients subsequentlyfound or suspectedto haveimmunologicallybased(hypersensitivity)lung disease.Thefirmcurrentlyrecommends discontinuationof fluoxetineifrash appears.We believedyspnea,asa symptomof possibleallergicpulmonarydiseaseis at least as importantas rash, but dyspneais not currentlyspecifiedas an indicationfor discontinuation. Thefirm's analysisof suicidalitydoes not resolvetheissue.Thefirmacknowledgedthatits clinicaltrials were not designedto studythis andthatthe quality and specificityof datato be gleanedfromthesetrialsto address suicidality were poor.Thedata presentedin sometables shoved higher percentages of suicidalityamongfluoxetine patientsthan amongtricyclicorplacebo patients, butthesedifferencesdid not reachstatisticalsignificance.The discussionofthe reportby Teicheret al. pointedoutthe difficultproblemof studyingthis question.However,the firm's strongestargumentagainstthefindingsof Teicher werechoseit presentedfrom Fava and Rosenbaum.As shown above,the summary providedbythefirm whiletechnicallycorrectdid not expressthe overall appearenceof the data.The actual datashowed a higher percentageoftreatment-emergentsuicidalityamong fluoxetine(2.9Z) than tricyclic(0.8X) patients with borderlinestatisticalsignificance.Interestingly,the proportion of patients withtreatment-emergentsuicidalityonfluoxetinein this study wassimilarto that reported by Teicher et al. Because of apparentlargescale underreporting,the firm'sanalysis cannot be consideredas provingthatfluoxetineand violent behaviorare unrelated. David J.Graham, MD, MPH Concur:/, Chief, Epidemiology Branch cc: Laughren/Brecher(HFD-120)Anello(HFD-700)Stadel/Craham/Rosa/Baum/Vise(HFD-733)Barash(HFD-735)NDA tf 18,936 • DRU1.7fluoxetineChron 000529

3. FDA finally acknowledge the inadequacy of its review of SSRI suicide data Unfortunately, it took FDA officials over a decade to figure this out. Senior FDA officials recently acknowledged at the Congressional hearings and the Advisory Committee hearings in 2004 that its analysis of the SSRI-induced suicidality was inadequate. In testimony before Congress resulting from investigations of the FDA’s failure to protect consumers related to the antidepressant suicidality issue, the FDA’s Dr. Robert Temple defended the agency’s failure, stating that although the FDA “had been systematically looking at the adult data for almost that entire decade” and had “not seen a signal in that data,” Dr. Temple admitted that the FDA’s analyses could have been far “better, more structured, [and] more careful, … but we didn’t know to do that.” (​VIEW FULL DOCUMENT​)





At the February 2, 2004, FDA advisory committee meeting concerning the risk of suicidality in children and adolescents taking antidepressants, the FDA’s Dr. Thomas Laughren similarly explained: “Just one follow up on a suggestion that has come up from several committee members now about looking at items from the rating scales. That was actually done here, and it turned out not to be very helpful. Now, this was a similar analysis that had been done with the adult data years ago. … ” He explained that this method “did not detect a signal in these trials … ” and admitted that the method was “was not particularly productive.” (​VIEW FULL DOCUMENT​)