Peachlab.org

ORIGINAL CONTRIBUTION Fluoxetine After Weight Restoration
in Anorexia Nervosa
A Randomized Controlled Trial
B. Timothy Walsh, MD Context Antidepressant medication is frequently prescribed for patients with an-
Allan S. Kaplan, MD, FRCPC orexia nervosa.
Objective To determine whether fluoxetine can promote recovery and prolong time-
Marion Olmsted, PhD to-relapse among patients with anorexia nervosa following weight restoration.
Michael Parides, PhD Design, Setting, and Participants Randomized, double-blind, placebo-
controlled trial. From January 2000 until May 2005, 93 patients with anorexia ner-
Jacqueline C. Carter, PhD vosa received intensive inpatient or day-program treatment at the New York State Psy- Kathleen M. Pike, PhD chiatric Institute or Toronto General Hospital. Participants regained weight to a minimumbody mass index (calculated as weight in kilograms divided by the square of height Michael J. Devlin, MD in meters) of 19.0 and were then eligible to participate in the randomized phase of Blake Woodside, MD, FRCPC Christina A. Roberto, BA Interventions Participants were randomly assigned to receive fluoxetine or pla-
Wendi Rockert, MEd cebo and were treated for up to 1 year as outpatients in double-blind fashion. All pa-tients also received individual cognitive behavioral therapy.
ous psychiatric illness with and the proportion of patients successfully completing 1 year of treatment.
substantial morbidity and a Results Forty-nine patients were assigned to fluoxetine and 44 to placebo. Similar
lifetime mortality arguably as percentages of patients assigned to fluoxetine and to placebo maintained a body mass high as that associated with any psy- index of at least 18.5 and remained in the study for 52 weeks (fluoxetine, 26.5%; chiatric illness.1 A major contributor to placebo, 31.5%; P=.57). In a Cox proportional hazards analysis, with prerandomiza- the poor prognosis of this illness is the tion body mass index, site, and diagnostic subtype as covariates, there was no signifi- high rate of relapse following initial cant difference between fluoxetine and placebo in time-to-relapse (hazard ratio, 1.12; treatment. Despite successful weight 95% CI, 0.65-2.01; P=.64).
restoration, 30% to 50% of patients re- Conclusions This study failed to demonstrate any benefit from fluoxetine in the treat-
quire rehospitalization within 1 year of ment of patients with anorexia nervosa following weight restoration. Future efforts discharge.2,3 This discouraging experi- should focus on developing new models to understand the persistence of this illness ence has prompted interest in inter- and on exploring new psychological and pharmacological treatment approaches.
ventions aimed at preventing deterio- Trial Registration clinicaltrials.gov Identifier: NCT00288574
ration following weight restoration.
Patients with anorexia nervosa often exhibit symptoms of other psychiatricdisorders,4 such as depression and ob- ingly, virtually all of the controlled trials Author Affiliations: New York State Psychiatric Insti-
tute/Columbia University Medical Center, New York,
sessive-compulsive disorder, which are of medication (most of which have been NY (Drs Walsh, Attia, Parides, Pike, and Devlin and responsive to antidepressant medica- conducted during the initial phase of Ms Roberto); and University of Toronto and TorontoGeneral Hospital, Toronto, Ontario (Drs Kaplan, Ol- tion, suggesting that pharmacological in- treatment when patients are under- msted, Carter, and Woodside and Ms Rockert).
terventions might be of use. Surpris- weight) have shown no benefit of medi- Corresponding Author: B. Timothy Walsh, MD, De-
cation compared with placebo.5,6 partment of Psychiatry, College of Physicians and Sur- Despite this lack of evidence of effec- geons of Columbia University/New York State Psy- See also p 2659 and Patient Page.
chiatric Institute, 1051 Riverside Dr, Unit 98, New York, tiveness, a substantial number of pa- NY 10032 ([email protected]).
2006 American Medical Association. All rights reserved.
(Reprinted) JAMA, June 14, 2006—Vol 295, No. 22 2605
FLUOXETINE AFTER WEIGHT RESTORATION IN ANOREXIA tients with anorexia nervosa receive an- cessfully completed treatment at 1 of the cebo, which were contained in identi- tidepressant medications. In our sites, study sites in an inpatient or day- cal bottles. A statistician performed approximately 60% of patients with an- program setting during which body mass primary analyses while the research staff orexia nervosa seeking treatment re- index (BMI [calculated as weight in ki- remained blind to medication assign- port having been treated with a selec- lograms divided by height in meters ment. To determine the effectiveness tive serotonin reuptake inhibitor (SSRI).
squared]) reached at least 19 and was of blinding procedures, patients, Since the ineffectiveness of medication maintained for 2 weeks; (4) was not at physicians, and therapists were asked at low weight might be attributable to imminent risk for suicide; (5) did not upon study termination to guess each the effects of starvation, researchers have have a serious medical illness aside from patient's assignment. Psychiatrists (␹2= more recently explored the utility of the eating disorder; and (6) was medi- 8.14, P = .004) and patients (␹2=5.49, medication after initial treatment has re- cation free (with the exception of occa- P = .02) guessed the correct medica- versed the acute effects of starvation, sional lorazepam, 0.5 mg, or zopiclone, tion assignment more frequently than with differing results.
7.5 mg, for anxiety or sleep disturbance).
chance, while therapists did not In a retrospective case-control analy- Recruitment methods included word sis, Strober et al7 found no difference of mouth, referrals, the Internet, and Medication was initiated in double- in the outcome of patients with an- advertisements. This study was de- blind fashion 1 week prior to dis- orexia nervosa taking fluoxetine fol- scribed either during the first evalua- charge from the hospital. The dose in- lowing hospitalization compared with tion of patients considering admission creased from 20 mg/d (1 pill) to 60 mg/d similar patients not taking fluoxetine.
to the eating disorders programs or fol- (3 pills) over 1 week under close su- However, in a double-blind, placebo- lowing admission. However, formal re- pervision for any adverse effects. Medi- controlled trial of 35 patients follow- cruitment and signing of the informed cation dose was maintained at 60 mg/d ing significant weight restoration, Kaye consent document did not occur until unless there were adverse effects, in et al8 found fluoxetine to be superior the end of the initial phase of treat- which case the dose was lowered at the to placebo in preventing relapse.
ment. The institutional review boards discretion of the psychiatrist; if the pa- The primary aim of the current study of the New York State Psychiatric In- tient's clinical status was deteriorat- was to determine, in a large sample, stitute and Toronto General Hospital ing, the dose could be raised to a maxi- whether fluoxetine, compared with pla- approved this study.
mum of 80 mg/d (4 pills). After the first cebo, reduced the rate of relapse and en- week receiving study medication, pa- hanced psychological and behavioral re- Treatment Prior to Randomization
tients were discharged to outpatient covery following initial treatment for The inpatient/intensive outpatient treat- care and treated for 12 months or un- anorexia nervosa. After successfully ment at both sites focused on nutri- til they met criteria for premature study completing a course of intensive treat- tional rehabilitation through the provi- withdrawal or voluntarily withdrew.
ment aimed at weight restoration, pa- sion of supervised meals and behavioral In addition to medication, all pa- tients were randomly assigned to re- and cognitive interventions. Both pro- tients received a form of manualized ceive fluoxetine or placebo and treated grams provided a weight maintenance cognitive behavioral therapy found to for up to 1 year in double-blind fash- period of several weeks following weight be useful in reducing relapse in an- ion. In addition to medication or pla- normalization, during which patients orexia nervosa.9 Experienced psycholo- cebo, patients received individual cog- spent increasing amounts of time out- gists and psychiatrists provided therapy, nitive behavioral therapy specifically side of the hospital.
which consisted of 50 individual ses- designed to prevent relapse.9 sions of 45 minutes' duration. Pa- tients were able to have up to 5 supple- After signing consent, patients were ran- mental sessions with family members domly assigned to receive either fluox- and/or significant others. A psychia- A patient was eligible to participate if she: etine or placebo. Randomization lists, trist saw patients to monitor the dose (1) was a female between the ages of 16 stratified by site and binge-purge sub- of medication, adverse effects, and gen- and 45 years; (2) met Diagnostic and Sta- type, were generated by a computer eral medical status. Therapy sessions tistical Manual of Mental Disorders, Fourth program utilizing a random number were audiotaped and reviewed by 2 of Edition (DSM-IV) criteria for anorexia generator seeded by time of day. Ran- the authors (K.M.P. and M.J.D.) to as- nervosa (except the requirement for domization assignments were kept in sess treatment fidelity and adherence.
amenorrhea; women who meet all DSM sealed envelopes and all clinical staff in- Blood samples were obtained every 3 criteria except amenorrhea do not ap- volved in the care of the patient, as well months for determination of plasma lev- pear to differ substantially from those as study coordinators, remained blind els of fluoxetine and nor-fluoxetine.
meeting all criteria10-12) prior to the ini- to medication assignment during the Study participation was terminated tial phase of treatment; (3) immedi- study. Program staff not involved in the and the patient was considered to have ately prior to randomization, had suc- study prepared medication and pla- relapsed if (1) the participant's BMI fell 2606 JAMA, June 14, 2006—Vol 295, No. 22 (Reprinted)
2006 American Medical Association. All rights reserved.
FLUOXETINE AFTER WEIGHT RESTORATION IN ANOREXIA to or below 16.5 for 2 consecutive weeks; Figure 1. Flow of Patients Through the Study
(2) severe medical complications (otherthan low weight) arose as a result of the 150 Patients Assessed for Eligibility eating disorder; (3) the participant wasjudged to be at imminent risk of sui- cide; or (4) the participant developed an- 23 Did Not Meet Inclusion Criteria other severe psychiatric disorder requir- 32 Refused to Participate ing treatment. Thus, the definition ofrelapse captured severe worsening of symptoms of anorexia nervosa and/orthe development of other major clini-cal problems.
49 Assigned to Receive Fluoxetine 44 Assigned to Receive Placebo 25 Terminated Trial Prematurely 28 Terminated Trial Prematurely To evaluate symptoms of disturbed eat- 10 Treatment Failure 10 Treatment Failure 6 Weight Loss to BMI ≤16.5 7 Weight Loss to BMI ≤16.5 ing behaviors and cognitions, depres- sion, anxiety, and self-esteem, the fol- 1 Suicide Attempt 1 Suicide Attempt 13 Patient-Initiated Withdrawal 14 Patient-Initiated Withdrawal lowing assessments were collected every 10 Dissatisfied With Treatment 9 Dissatisfied With Treatment 3 Unable to Keep Appointments 3 Unable to Keep Appointments 4 weeks: Eating Disorders Inventory 2 Staff-Initiated Withdrawals (EDI),13 the Beck Depression Inventory 1 Patient Missed Appointments 4 Staff-Initiated Withdrawals 1 Patient Stopped Medication (BDI),14 the Beck Anxiety Inventory 2 Patient Missed Appointments (BAI),15 and the Rosenberg Self-EsteemScale (RSE).16 Obsessions and quality of 49 Included in Primary Analysis 44 Included in Primary Analysis life were measured with the Yale BrownCornell Obsessive Compulsive Scale for Body mass index (BMI calculated as weight in kilograms divided by height in meters squared).
Eating Disorders (YBC-EDS)17 and theQuality of Life Enjoyment and Satisfac- mination were considered to have lapsed. Also, at study termination, pa- tion Questionnaire (QlesQ)18 and were relapsed; data from all other dropouts tient status was classified according to obtained before randomization, at 6 were censored. For the third analysis, modified Morgan-Russell criteria,19,20 months, and again at study termina- dropouts were considered to have and the percentage of patients in fluox- tion. Participants were weighed weekly.
relapsed if they met any of the follow- etine and placebo groups classified as ing criteria at the time they chose to end fair or better were compared using the participation in the study: (1) had a BMI ␹2 statistic. Additional secondary analy- The study was designed to randomize 90 less than 17.5; (2) were binge eating ses examined differences between patients with equal allocation to fluox- and/or purging 2 or more times per week groups in change in symptoms over etine and placebo. Ninety patients pro- for the 4 weeks prior to dropping out; time using random-effects regression vide 90% power to detect a relative 50% or (3) exhibited depressive and/or anxi- models. All outcome analyses adhered reduction in relapse for fluoxetine com- ety symptoms sufficient to interfere with to the intention-to-treat principle; a P pared with placebo based on a 2-sided functioning (eg, serious suicidal ide- value of .05 using 2-sided tests was the .05 level log-rank test. A relapse rate of ation). Cox proportional hazards regres- criterion for significance for all analy- 70% was assumed for placebo and 35% sion was used to test for differences ses except the Cox proportional haz- for fluoxetine corresponding to a haz- between randomization groups adjust- ards regressions, for which, because of ard ratio of approximately 0.34.
ing for site and diagnostic subtype (the an interim analysis, a P value of .04 was The primary outcome measure was 2 factors used to stratify the random- used. Statistical analysis was com- time-to-relapse. Patients who voluntar- ization) and by pretreatment BMI, which pleted using SAS version 9.1.3, SAS In- ily withdrew from treatment ("drop- was found to differ significantly between stitute Inc, Cary, NC, and SPSS ver- outs") were classified in 3 distinct ways.
randomization groups at the .05 level.
sion 13.0, SPSS Inc, Chicago, Ill.
In the most conservative analysis, all We also compared the proportion of patients who left the study prior to com- patients in each group who failed to pleting a full course of treatment (at least maintain normal weight (BMIⱖ18.5) 38 psychotherapy sessions over 50 to 60 for 52 weeks (ie, relapsed) using a ␹2 Patients were recruited between Janu- weeks) were considered to have relapsed.
test. Patients whose weight fell below ary 2000 and May 2004. The last pa- In a second least conservative analysis, this level for 4 consecutive weeks or tient completed the study in May 2005.
only patients who met the predefined cri- who withdrew from the study prior to One hundred fifty patients were evalu- teria described previously for study ter- 52 weeks were considered to have re- ated for study participation (FIGURE 1).
2006 American Medical Association. All rights reserved.
(Reprinted) JAMA, June 14, 2006—Vol 295, No. 22 2607
FLUOXETINE AFTER WEIGHT RESTORATION IN ANOREXIA Twenty-three did not meet criteria for 15.4 (1.8) at the time of entry into the However, there was a small but statis- eligibility, 32 declined participation, and initial phase of treatment. Forty-five pa- tically significant difference in BMI with 2 were excluded for other reasons. The tients (48.4%) met DSM-IV criteria for patients receiving placebo starting the remaining 93 patients entered the trial, binge-purge subtype. There were no sta- trial at higher BMI. Only patients who 45 in New York, 48 in Toronto. Forty- tistically significant differences in these had gained to a BMI of at least 19 were nine patients were assigned to fluox- parameters between the patients as- eligible to enter the randomized phase etine and 44 to placebo. On average, pa- signed to fluoxetine or placebo, or be- of the study. In practice, most patients tients were a mean (SD) of 23 (4.6) tween the New York or Toronto sites.
gained little additional weight, lead- years old, had been ill for 56.5 (44.7) At the time of randomization, pa- ing to very limited variance in BMI at months, had been previously hospital- tients in the fluoxetine and placebo the time of randomization. The lim- ized 2.1 (2.5) times, and had a BMI of groups were quite similar (TABLE 1).
ited variation contributed to the statis- Table 1. Demographic Information, Body Mass Index, and Psychological Assessments at Prerandomization and at Study Termination
Mean (SD)
Mean (SD)
Length of illness, y Rosenberg Self-Esteem Scale Beck Depression Inventory Beck Anxiety Inventory Yale Brown Cornell Obsessive Compulsive Scale for Eating Disorders Quality of Life Enjoyment and Satisfaction Questionnaire Termination assessments Rosenberg Self-Esteem Scale Beck Depression Inventory Beck Anxiety Inventory Yale Brown Cornell Obsessive Compulsive Scale for Eating Disorders Quality of Life Enjoyment and Satisfaction Questionnaire Eating Disorder Inventory at preassessment Drive for thinness Body dissatisfaction Interpersonal distrust Interoceptive awareness Eating Disorder Inventory at study termination Drive for thinness Body dissatisfaction Interpersonal distrust Interoceptive awareness *Calculated as weight in kilograms divided by height in meters squared.
†Significant difference between fluoxetine and placebo groups (t = 2.76, df = 91, P = .007).
2608 JAMA, June 14, 2006—Vol 295, No. 22 (Reprinted)
2006 American Medical Association. All rights reserved.
FLUOXETINE AFTER WEIGHT RESTORATION IN ANOREXIA tical significance of the small differ- Table 2. Reason for Study Termination and Status at Termination as Per Modified
ence in BMI between groups.
Morgan-Russell Criteria for Outcome Outcome
Forty of the 93 patients completed the
entire course of treatment and 53 ter- Reason for termination* minated prematurely (Figure 1 and Completed full study TABLE 2). The major reasons for pre-
Treatment failure† mature termination were meeting a Voluntarily withdrew priori criteria for relapse and patient dis- satisfaction with treatment. Patients were Outcome status at termination also withdrawn from the study if they failed to attend a minimum number of psychotherapy sessions in a quarter (12 during the first quarter, 9 during the sec- ond and third quarters, and 8 during the last quarter) or if they took less than 10 *Proportion of patients completing full study with fluoxetine vs plabebo: ␹2 = 0.001, P = .98.
†Withdrawn because of worsening of anorexia nervosa and/or other significant clinical deterioration (see "Methods").
mg of the medication for 2 weeks or ‡Withdrawn for nonadherence or potential drug adverse effects.
§Proportion of patients with outcome status of full recovery, good, or fair vs poor with fluoxetine vs placebo: ␹2 = 1.05, more. A 17-year-old patient assigned to P = .32.
fluoxetine made a suicide attempt dur- 㛳Dropped out early without meeting criteria for any of the other outcome categories.
ing the study; there were no other un-expected adverse events. There were no (fluoxetine: 26.5%; placebo: 31.5%; ence between fluoxetine and placebo significant differences between fluox- ␹2=1.31, P=.57). A greater percentage groups in time-to-relapse. At 52 weeks, etine and placebo groups in BMI or in of patients in Toronto (39.6%) met this using the least conservative analysis, in measures of psychological state at the outcome criterion compared with New which patients who voluntarily with- time of termination (Table 1). There York (17.8%; ␹2=5.36; P=.02). There drew were censored, 73% of the pla- were no significant differences be- was no statistically significant differ- cebo group and 71% of the fluoxetine tween the fluoxetine and placebo groups ence between binge-purge (22.2%) and group had not relapsed. Using the in the percentages of patients complet- restricting (35.4%) subtypes on this analysis in which patients who dropped ing the full study (␹2=0.001; P=.98), or measure (␹2=1.96; P=.16).
out were clinically classified, 51% of the in the average number of sessions of psy- placebo group and 49% of the fluox- chotherapy (fluoxetine, 38.1 [18.1]; pla- etine group had not relapsed at 52 cebo, 34.8 [16.8]; t = 0.48; df = 91; The most conservative analysis of time- P = .63). However, there were signifi- to-relapse, in which premature termi- As noted previously, the average pre- cant differences in the rates of full- nation for any reason was considered to randomization BMI of the placebo group study completion between sites (New signify relapse, found no significant dif- was slightly but significantly higher than York, 28.9%; Toronto, 56.3%; ␹2= ference between the fluoxetine and pla- that of the fluoxetine group. Because a 7.094; P=.008) and between subtypes cebo groups (␤ = −0.133; SE = 0.288; higher BMI was associated with a longer (binge-purge, 31.1%; restricting, 54.2%; df=1; P=.64) (FIGURE 2). At 52 weeks,
time-to-relapse, BMI was included in the 45% of the placebo group and 43% of analyses described above to control sta- At the time of termination, similar the fluoxetine group had not relapsed.
tistically for the difference between percentages of patients in the fluox- In this analysis, there were significant fluoxetine and placebo groups. Addi- etine and placebo groups met modi- effects of site (␤ = 0.814; SE = 0.287; tional secondary analyses of time-to- fied Morgan-Russell criteria for fair or df = 1; P = .005), of subtype (␤=0.777; relapse were conducted including only better outcome (fluoxetine, 65.31%; SE = 0.285; df = 1; P = .006), and of pre- the 55 patients whose prerandomiza- placebo, 56.82%; ␹2= 1.005; P = .32) randomization BMI (␤ = −0.761; tion BMIs were greater than 20.15.
(Table 2). By Morgan-Russell criteria SE = 0.323; df = 1; P = .02). Patients in Among these patients, there was no sta- at time of termination, there were no Toronto remained in the trial for a tistically significant difference between significant differences between sites or longer time prior to relapse than did the mean prerandomization BMIs of the between binge-purge and restricting those in New York, as did patients with fluoxetine (n=24; 20.51 [0.42]) and pla- subtypes (data not shown).
the restricting subtype and patients with cebo groups (n=31; 20.69 [0.41]; t(53) Similar percentages of patients as- a higher BMI at prerandomization. Re- =1.61; P=.11). In none of these analy- signed to fluoxetine and to placebo sults from the analyses using the other ses did the difference between fluoxetine maintained a BMI of at least 18.5 and methods of classifying dropouts were and placebo groups in time-to-relapse remained in the study for 52 weeks similar; neither indicated any differ- approach significance (P⬎.50 for all).
2006 American Medical Association. All rights reserved.
(Reprinted) JAMA, June 14, 2006—Vol 295, No. 22 2609
FLUOXETINE AFTER WEIGHT RESTORATION IN ANOREXIA Because the study by Kaye et al8 had Plasma samples for analysis of fluox- and the study by Kaye et al.8 Among the excluded patients with binge eating, etine and nor-fluoxetine levels were patients with detectable fluoxetine or these analyses were repeated using data available for 43 of the 49 patients as- nor-fluoxetine levels, there was no from only the 48 patients with the re- signed to receive fluoxetine. Plasma significant relationship between the stricting subtype of anorexia nervosa.
samples from 3 patients had no de- average total fluoxetine plus nor- There was no evidence of a significant tectible fluoxetine or nor-fluoxetine.
fluoxetine plasma concentration and difference between the fluoxetine and Exclusion of the data from these 3 time-to-relapse using any of the meth- placebo groups (data not shown).
patients did not substantially alter the ods for classifying dropouts (data not results of the analyses of time-to- shown). Plasma samples were avail- relapse. The average fluoxetine and nor- able for 38 of the 44 patients assigned At the last visit, the mean (SD) dose of fluoxetine levels for the remaining 40 to placebo; none had detectable levels medication was 63.5 [15.8] mg/d for pa- patients were 409 (225) and 297 (154) of fluoxetine or nor-fluoxetine.
tients on fluoxetine and the equiva- ng/mL, respectively; these levels are lent of 71.4 [13.2] mg/d for patients re- comparable with those reported in stud- Prior SSRI Treatment
ceiving placebo (t=2.6; df=91; P=.01).
ies of obsessive-compulsive disorder21 Twenty-four of the 93 patients had re-ceived 1 course of SSRI treatment (de-fined as 3 or more continuous weeks Figure 2. Patients Receiving Placebo and Fluoxetine Remaining in Treatment vs Number of
receiving an SSRI) prior to their par- Weeks After Treatment Inception ticipation in the current study, and 33 had received 2 or more previous coursesof SSRI treatment. There were no sig- nificant differences between the fluox- etine and placebo groups in time-to- relapse using any of the criteria for classifying dropouts among patients oportion Remaining with no prior SSRI treatment or among patients with no more than 1 course ofprior treatment (data not shown).
Random-effects regression models were used to compare randomization groups Circles indicate when patients left the trial after a full course of treatment (minimum 38 sessions, maximum 61 over time using all available data from sessions over 12 months). Twelve month/full course of treatment criterion was satisfied if 38-session mini- all patients (TABLE 3). The factors ex-
mum was met between 50 and 60 weeks after randomization.
amined were weight, several measuresof general (BDI, BAI, RSE) or specific(EDI-Drive for Thinness, Bulimia, Body Table 3. Random-Effects Regression Analysis of Change During Treatment
Dissatisfaction and Perfectionism, and Random Effects Regression
YBC-EDS) psychopathology, and a measure of quality of life (QlesQ). The only measure for which a significant Beck Depression Inventory fluoxetine-placebo difference was de- Beck Anxiety Inventory tected was the BAI; the mean (SD) Rosenberg Self-Esteem Scale change in BAI per month was −0.70 Quality of Life Enjoyment and (0.12) in the fluoxetine group vs −0.22 (0.13) in the placebo group (P = .007; Eating Disorders Inventory Drive for thinness Bonferroni corrected: P = .07).
Body dissatisfaction The current study was designed to Yale Brown Cornell Obsessive determine whether the antidepres- Compulsive Scale for sant medication fluoxetine reduced relapse following successful initial treat- *Coefficients and standard errors indicate change per month.
†Z scores are for the difference between the coefficients for fluoxetine and placebo.
ment of anorexia nervosa. Ninety- ‡P values are for the difference between the coefficients for fluoxetine and placebo.
three patients entered the double- 2610 JAMA, June 14, 2006—Vol 295, No. 22 (Reprinted)
2006 American Medical Association. All rights reserved.
FLUOXETINE AFTER WEIGHT RESTORATION IN ANOREXIA blind phase, making this trial, to our the treatment of anorexia nervosa, we for example, in patients who have had knowledge, the largest controlled medi- anticipated that many patients would anorexia nervosa for only a short time cation trial conducted to date in an- not complete a full year of outpatient or who maintained a normal weight for orexia nervosa. All patients had reached care and designed this study to exam- a longer period of time before receiving a minimum BMI of 19.0, so that the ine the course of relapse during this antidepressant treatment. Similarly, it is acute effects of undernutrition were time. The results of the survival analy- conceivable that fluoxetine might pro- minimized. The average dose of fluox- ses help clarify the reasons for prema- vide benefit if it were the sole treatment etine was 63.5 mg/d similar to the dose ture termination and the impact of provided or given in association with routinely used in the treatment of bu- medication. Twenty of the 53 patients some other intervention. However, the limia nervosa. The study was con- who did not complete a full course of results of the current study are consis- ducted at 2 sites experienced in the treatment met our stringent, a priori cri- tent with most previous controlled trials treatment of anorexia nervosa, lend- teria for full relapse and were there- of medication in anorexia nervosa, which ing support to the generalizability of the fore withdrawn from the study and re- have failed to show significant benefit findings. Despite these strengths, the ferred for other treatment. An additional from antidepressant medication com- current study found no convincing evi- 24 of the 53 noncompleters had not de- pared with placebo.5,22 dence that fluoxetine provided signifi- teriorated so severely as to require with- Unlike Kaye et al,8 we were unable to cant benefit compared to placebo. On drawal but were clinically classified, demonstrate a difference between fluox- virtually all measures, including time- blind to medication assignment, as do- etine and placebo on time-to-relapse. A to-relapse, rate of study completion, ing sufficiently poorly to be consid- direct comparison between that study BMI, and clinical measures at time of ered as having relapsed at the time they and the current one is confounded by a termination, including depression, the left the study; many of these patients number of differences. In the current group receiving fluoxetine was statis- voluntarily withdrew from the study be- study, all patients received a structured tically indistinguishable from that re- cause of their lack of progress. There- psychotherapy specifically aimed to re- ceiving placebo. There was no indica- fore, 44 (83%) of the 53 patients who duce relapse, whereas in the study by tion that the effect of fluoxetine varied did not complete treatment had dete- Kaye et al,8 the psychosocial treatment significantly by site, by subtype of an- riorated substantially, and this deterio- received was not standardized; this dif- orexia nervosa, or by history of prior ration was a major contributor to their ference may explain the high relapse rate exposure to SSRIs. A single secondary withdrawal. We had hoped that fluox- in the placebo group in the study by analysis examining change in the BAI etine would alter the course of illness Kaye et al.8 In addition, criteria for re- over time found a significant differ- and thereby reduce such deterioration lapse were less specific in that study and ence favoring the fluoxetine group; and the rate of withdrawal, as sug- patients were encouraged to withdraw however, since the fluoxetine and pla- gested by Kaye et al.8 Unfortunately, if they felt they were doing poorly. Be- cebo groups did not differ in similar medication assignment had no discern- cause of limited information available analyses of 9 other measures, we can- ible impact on time-to-relapse whether about the treatment response of indi- not be confident that this difference be- all dropouts were considered to have viduals with the restricting subtype of tween groups was not due to chance.
relapsed, whether no dropouts were anorexia nervosa and of indications that Both patients and their psychia- considered to have relapsed, or whether such patients might respond more fa- trists, but not their therapists, were dropouts were classified as having re- vorably to fluoxetine,25 Kaye et al8 fo- more likely than chance to guess lapsed based on their status at time of cused their analysis on 35 patients who whether patients were receiving ac- study termination. In other words, be- denied binge eating. Additional pa- tive medication. We speculate that this cause discontinuation of treatment by tients with binge eating were random- was due to the presence of minor ad- patients with anorexia nervosa is a likely ized according to personal communi- verse effects. However, since virtually sign of impending relapse, systematic cation with Walter H. Kaye, MD, no differences in outcome were de- examination of such withdrawal can Western Psychiatric Institute and Clinic, tected between the fluoxetine and medi- provide information about treatment ef- University of Pittsburgh School of Medi- cation groups, it seems unlikely that this ficacy. In this study, fluoxetine was no cine, and the restriction of the analysis information substantially affected the more efficacious than placebo.
to a subgroup may have inflated the pos- assessment of clinical status.
A limitation on the conclusions of the sibility of a false-positive result.27 The interpretation of treatment trials current study is that we examined the The current study has implications for such as this one, in which a substan- utility of fluoxetine at a particular stage both clinical practice and research. The tial number of patients did not com- of illness and in conjunction with a par- present findings, coupled with those of plete the full course of treatment, can ticular form of structured psychologi- previously published studies, indicate be problematic. However, since re- cal treatment. We cannot exclude the that the common practice of prescrib- lapse soon after successful weight gain possibility that antidepressants might ing antidepressant medication is un- is a well-recognized clinical problem in have an effect at other stages of illness, likely to provide substantial benefit for 2006 American Medical Association. All rights reserved.
(Reprinted) JAMA, June 14, 2006—Vol 295, No. 22 2611
FLUOXETINE AFTER WEIGHT RESTORATION IN ANOREXIA most patients with anorexia nervosa, were presented at the meeting of the Eating Disor- 9. Pike KM, Walsh BT, Vitousek K, Wilson GT, Bauer
ders Research Society; September 29, 2005; Toronto, J. Cognitive behavior therapy in the posthospitaliza- either when they are underweight or im- tion treatment of anorexia nervosa. Am J Psychiatry.
mediately upon weight restoration.
Acknowledgment: We thank the staffs of the Gen-
These data imply that therapeutic ef- eral Clinical Research Unit at New York State Psychi- 10. Cachelin FM, Maher BA. Is amenorrhea a critical
atric Institute/Columbia University Medical Center, criterion for anorexia nervosa? J Psychosom Res. 1998; forts would be better devoted to psy- New York, NY and of Toronto General Hospital, chological and behavioral interven- Toronto, Ontario. The members of the data and 11. Garfinkel PE, Lin E, Goering P, et al. Should amen-
safety monitoring board: Sidney Kennedy, MD, Uni- orrhoea be necessary for the diagnosis of anorexia ner- tions for which there is some, albeit versity of Toronto, Toronto, Ontario; Andrew Leon, vosa? evidence from a Canadian community sample.
modest, evidence of efficacy.9,23 Future PhD, Weill Medical College of Cornell University, Br J Psychiatry. 1996;168:500-506.
New York, NY; Steven Roose, MD, New York State research on the utility of novel psycho- 12. Watson TL, Andersen AE. A critical examination
Psychiatric Institute/Columbia University Medical of the amenorrhea and weight criteria for diagnosing logical treatments24,25 and innovative Center; Jane Pearson, PhD, program officer, National anorexia nervosa. Acta Psychiatr Scand. 2003;108:175- psychotropic (eg, olanzapine)26,27 and Institute of Mental Health, Bethesda, Md. The mem- bers of the Eating Disorders Research Unit at New 13. Garner DM, Olmsted MP. The Eating Disorder In-
nonpsychotropic (eg, cycloserine)28 York State Psychiatric Institute/Columbia University ventory Manual. Odessa, Fla: Psychological Assess- medications is obviously needed. In ad- Medical Center: Lauren Escott, MA, and Dana Satir, ment Resources; 1984.
BA, Boston University, Boston, Mass, for data collec- 14. Beck AT, Ward CH, Mendelson M, Mock J, Er-
dition, it may be important to examine tion; Juli Goldfein, PhD, Diane Klein, MD, Lisa Kotler, baugh J. An inventory for measuring depression. Arch the fundamental behavioral and cogni- MD, Laurel Mayer, MD, Pamela Raizman, PhD, Gen Psychiatry. 1961;4:561-571.
Joanna Steinglass, MD, and Sara Wolk, PhD, New tive disturbances of anorexia nervosa 15. Beck AT, Epstein N, Brown G, Steer RA. An in-
York State Psychiatric Institute/Columbia University ventory for measuring clinical anxiety: psychometric from fresh perspectives. Investigators Medical Center, for clinical care of study participants; properties. J Consult Clin Psychol. 1988;56:893-897.
Katharine Loeb, PhD, Mt Sinai School of Medicine, have recently suggested that patients 16. Rosenberg M. Conceiving the Self. New York, NY:
New York, NY; Sally Woodring, RN, New York State Basic Books; 1979.
with anorexia nervosa might be viewed Psychiatric Institute/Columbia University Medical 17. Mazure CM, Halmi KA, Sunday SR, Romano SJ,
as having disturbances in extinguish- Center, for medication supply management; Einhorn AM. The Yale-Brown-Cornell eating disor- Jonathan Cohen, MS, New York State Psychiatric ing conditioned fears29 or in changing der scale: development, use, reliability and validity.
Institute/Columbia University Medical Center, for J Psychiatr Res. 1994;28:425-445.
cognitive set.30,31 New frameworks for database creation and support. The members of the 18. Endicott J, Nee J, Harrison W, Blumenthal R. Qual-
Toronto General Hospital research team: Traci understanding the impressive persis- ity of life enjoyment and satisfaction questionnaire: a McFarlane, PhD, and Randy Staab, MD, FRCP(C), new measure. Psychopharmacol Bull. 1993;29:321- tence of anorexia nervosa may, eventu- Toronto General Hospital/University of Toronto; Pavla Reznicek, PhD, private practice, Toronto, for ally, lead to more effective treatments, 19. Eckert ED, Halmi KA, Marchi P, Grove W, Crosby
clinical care of study participants; Anita Federici, MSc, R. Ten-year follow-up of anorexia nervosa: clinical which are sorely needed.
York University, Toronto, for data collection; and course and outcome. Psychol Med. 1995;25:143-156.
Nancy Lipson, MSW, RSW, Concurrent Disorders Author Contributions: Dr Walsh had full access to all
20. Russell GFM, Szmukler GI, Dare C, Eisler I. An
Service, Center for Addiction and Mental Health, of the data in the study and takes responsibility for evaluation of family therapy in anorexia nervosa and Toronto, for medication supply management.
the integrity of the data and the accuracy of the data bulimia nervosa. Arch Gen Psychiatry. 1987;44:1047- The following individuals included in the acknowl- edgments reported having received compensation Study concept and design: Walsh, Kaplan, Attia, 21. Koran LM, Cain JW, Dominguez RA, Rush AJ, Thi-
from the NIH grant for their involvement in the Olmsted, Parides, Carter, Pike, Devlin, Woodside, emann S. Are fluoxetine plasma levels related to out- study: Jonathan Cohen, MS; Lauren Escott, MA; come in obsessive-compulsive disorder? Am J Anita Federici, MSc; Juli Goldfein, PhD; Lisa Kotler, Acquisition of data: Walsh, Kaplan, Attia, Olmsted, MD; Nancy Lipson, MSW; Pamela Raizman, PhD; Carter, Pike, Devlin, Woodside, Roberto, Rockert.
22. Holtkamp K, Konrad R, Kaiser N, et al. A retro-
Pavla Reznicek, PhD; Randy Staab, MD, FRCP(C); Analysis and interpretation of data: Walsh, Kaplan, Dana Satir, BA; and Sara Wolk, PhD.
spective study of SSRI treatment in adolescent an- Attia, Olmsted, Parides, Roberto, Rockert.
orexia nervosa: insufficient evidence for efficacy. J Psy- Drafting of the manuscript: Walsh, Kaplan, Parides, chiatr Res. 2005;39:303-310.
23. Lock J, le Grange D. Family-based treatment of
Critical revision of the manuscript for important in- eating disorders. Int J Eat Disord. 2005;37:S64-S67.
tellectual content: Attia, Olmsted, Carter, Pike, Devlin, 1. Sullivan PF. Mortality in anorexia nervosa. Am J
24. McIntosh VV, Jordan J, Carter FA, et al. Three psy-
Woodside, Rockert.
chotherapies for anorexia nervosa: a randomized, con- Statistical analysis: Walsh, Parides, Roberto.
2. Pike KM. Long-term course of anorexia nervosa:
trolled trial. Am J Psychiatry. 2005;162:741-747.
Obtained funding: Walsh, Kaplan.
response, relapse, remission, and recovery. Clin Psy- 25. Kaye WH, Weltzin TE, Hsu LK, Bulik CM. An open
Administrative, technical, or material support: Walsh, chol Rev. 1998;18:447-475.
trial of fluoxetine in patients with anorexia nervosa.
Kaplan, Attia, Olmsted, Roberto, Rockert.
3. Jones LM, Halford WK, Dooley RT. Long term out-
J Clin Psychiatry. 1991;11:464-471.
Study supervision: Walsh, Kaplan, Pike.
come of anorexia nervosa. Behav Change. 1993;10:93- 26. Powers PS, Santana CA, Bannon YS. Olanzapine
Financial Disclosures: Dr Walsh has reported receiv-
in the treatment of anorexia nervosa: an open label ing research support from Eli Lilly and Co, Abbott Labo- 4. Halmi KA, Eckert E, Marchi P, Sampugnaro V, Apple
trial. Int J Eat Disord. 2002;32:146-154.
ratories, Ortho-McNeil Pharmaceuticals, and R, Cohen J. Comorbidity of psychiatric diagnoses in 27. Lagakos SW. The challenge of subgroup analyses–
GlaxoSmithKline. Drs Kaplan and Woodside have re- anorexia nervosa. Arch Gen Psychiatry. 1991;48:712- reporting without distorting. N Engl J Med. 2006;354: ported receiving research support from Eli Lilly and Co.
Dr Devlin has reported receiving research support from 5. Zhu AJ, Walsh BT. Pharmacologic treatment of eat-
28. Ressler KJ, Rothbaum BO, Tannenbaum L, et al.
Ortho-McNeil Pharmaceuticals and Eli Lilly and Co.
ing disorders. Can J Psychiatry. 2002;47:227-234.
Cognitive enhancers as adjuncts to psychotherapy: use Dr Attia has reported receiving research support from 6. Attia E, Haiman C, Walsh BT, Flater SR. Does fluox-
of D-cycloserine in phobic individuals to facilitate ex- Eli Lilly and Co and Pfizer Inc. No other authors re- etine augment the inpatient treatment of anorexia tinction of fear. Arch Gen Psychiatry. 2004;61:1136- ported disclosures.
nervosa? Am J Psychiatry. 1998;155:548-551.
Funding/Support: This study was supported in part
7. Strober M, Freeman R, DeAntonio M, Lampert C,
29. Strober M. Pathologic fear conditioning and an-
by grants MH060271 and MH60336 from National Diamond J. Does adjunctive fluoxetine influence the orexia nervosa: on the search for novel paradigms. Int Institutes of Health. Eli Lilly supplied fluoxetine and post-hospital course of anorexia nervosa? a 24- J Eat Disord. 2004;35:504-508.
month prospective, longitudinal followup and com- 30. Tchanturia K, Morris RG, Surguladze S, Treasure
Role of the Sponsor: The National Institutes of Health
parison with historical controls. Psychopharmacol Bull.
J. An examination of perceptual and cognitive set shift- and Eli Lilly had no role in the design and conduct of ing tasks in acute anorexia nervosa and following the study; the collection, management, analysis, and 8. Kaye WH, Nagata T, Weltzin TE, et al. Double-
recovery. Eat Weight Disord. 2002;7:312-315.
interpretation of the data; or the preparation, re- blind placebo controlled administration of fluoxetine 31. Steinglass JE, Walsh BT, Stern Y. Set shifting defi-
view, or approval of the manuscript.
in restricting- and restricting-purging type anorexia cit in anorexia nervosa. J Int Neuropsychol Soc. In Previous Presentation: Preliminary results of this study
nervosa. Biol Psychiatry. 2001;49:644-652.
2612 JAMA, June 14, 2006—Vol 295, No. 22 (Reprinted)
2006 American Medical Association. All rights reserved.
Financial Disclosures: None reported.
10. Koreith K. Medical writing market appreciation. The Centerwatch Monthly.
Funding/Support: No external funding was used for this study.
Previous Presentation: An abstract of this research was presented at the Fifth In-
.pdf. Accessed May 9, 2006.
ternational Congress on Peer Review and Biomedical Publication; September 16- 11. Hamilton CW, Mallia-Hughes M, Mitrany D, Foote MA. Comments on "The
18, 2005; Chicago, Ill.
corporate author" [letter]. J Gen Intern Med. 2005;20:972.
Acknowledgment: We thank Duncan Purvis, PhD, and Christine Wichems, PhD,
employees of ProScribe Medical Communications, for their critical review of the
manuscript, and John Wlodarczyk, PhD, for calculating the confidence intervals.
Drs Purvis, Wichems, and Wlodarczyk received no financial compensation for theirservices.
Incorrect Data: In the Original Contribution entitled "Fluoxetine After Weight Res-
toration in Anorexia Nervosa" published in the June 14, 2006, issue of JAMA (2006;
1. Camacho LH, Bacik J, Cheung A, Spriggs DR. Presentation and subsequent pub-
295:2605-2612), the boxes in Figure 1 that stated the reasons for premature termi- lication rates of phase I oncology clinical trials. Cancer. 2005;104:1497-1504.
nation for participants taking fluoxetine and placebo were switched. In the box de- 2. Wise P, Drury M. Pharmaceutical trials in general practice: the first 100 pro-
scribing reasons for termination for participants taking the placebo, "Suicide Attempt" tocols: an audit by the clinical research ethics committee of the Royal College of was incorrectly substituted for "Clinical Deterioration." In Tables 1, 3, and the Mea- General Practitioners. BMJ. 1996;313:1245-1248.
sures section, The Yale Brown Cornell Obsessive Compulsive Scale for Eating Disor- 3. International Committee of Medical Journal Editors. Uniform requirements for
ders should have been termed the Yale-Brown-Cornell Eating Disorder Scale and the manuscripts submitted to biomedical journals: writing and editing for biomedical Eating Disorders Inventory should have been termed the Eating Disorder Inventory.
publication: ethical considerations in the conduct and reporting of research. http: In the Table 2 footnotes, the ␹2 value after "§Proportion of patients with outcome sta- //www.icmje.org/index.html#ethic. Accessed November 1, 2005.
tus of full recovery, good, or fair vs poor with fluoxetine vs placebo" should have been 4. Wager E, Field EA, Grossman L. Good publication practice for pharmaceutical
␹2=1.005. In Table 3, the random-effects regression for the fluoxetine group for the companies. Curr Med Res Opin. 2003;19:149-154.
Eating Disorder Inventory bulimia subscale should have been 0.11 rather than −0.11.
5. Jacobs A, Wager E. European Medical Writers Association (EMWA) guidelines
In the last paragraph of the article, the references 24,25 after "psychological treatments" on the role of medical writers in developing peer-reviewed publications. Curr Med should have been numbered 24 and references 26,27 after "olanzapine" should have Res Opin. 2005;21:317-322.
been numbered 26. None of these corrections affect the conclusions in the article.
6. Hamilton CW, Royer MG; AMWA 2002 Task Force on the Contributions of
Medical Writers to Scientific Publications. AMWA position statement on the
Error in Wording: In the Editorial entitled "Radiosurgery and Whole-Brain Radia-
contributions of medical writers to scientific publications. AMWA Journal.
tion Therapy for Brain Metastases: Either or Both as the Optimal Treatment" pub- lished in the June 7, 2006, issue of JAMA (2006;295:2535-2536), an error oc- 7. Flanagin A, Carey LA, Fontanarosa PB, et al. Prevalence of articles with hon-
curred in wording. In the final paragraph on page 2536, the term "stereotactic orary authors and ghost authors in peer-reviewed medical journals. JAMA. 1998; radiosurgery" should have been "whole-brain radiation therapy" in both in- stances. The sentence should have read "Aoyama et al10 have prospectively shown 8. Phillips SG, Carey LA, Biedermann G. Attitudes toward writing and writing as-
that withholding whole-brain radiation therapy does not affect survival for pa- sistance in peer-reviewed articles. AMWA Journal. 2001;16:10-16.
tients who have 4 or fewer brain metastases; these patients have a higher rate of 9. Healy DT. Transparency and trust: figure for ghost written articles was mis-
local brain failure, but apparently withholding whole-brain radiation therapy does quoted [letter]. BMJ. 2004;329:1345.
not influence how patients die of their disease." 934 JAMA, August 23/30, 2006—Vol 296, No. 8 (Reprinted)
2006 American Medical Association. All rights reserved.
ity, specificity, positive predictive power, negative predic- we believe that the use of single point estimates as a thresh- tive power, and the C statistic) that we and others have old for recommending genetic testing should be discour- used to compare mutation prediction models,1,4 the abil- aged in favor of a clinical judgment–grounded approach. This ity to examine multiple predictors in logistic regression is consistent with the updated American Society of Clinical can add depth of understanding to statistical results. We Oncology guidelines5 and is critically important to indi- agree that the corrective probability index we created was vidual women contemplating the benefits of risk reduc- not based on a theory of inheritance but, rather, was a temporizing measure to improve the sensitivity of the Jeffrey N. Weitzel, MD
model, hopefully to be replaced by a model wherein the parameters are adjusted for accuracy for single-case indi- Veronica I. Lagos, MS
cators in clinical settings such as ours. We intend to vali- Deborah J. MacDonald, PhD
date the effect of family structure in additional clinical City of Hope
and population-based data sets.
BRCAPRO certainly has great value in estimating the Financial Disclosures: None reported.
proportion of a population with a BRCA mutation and, in Additional Contributions: Gwen Uman, PhD, Vital Research LLC, Los Angeles,
California, provided assistance with the statistical methods. Dr Uman received com-
particular, in assigning likely carrier status when test pensation via a subcontract for her assistance.
results are uninformative. However, we maintain that 1. James PA, Doherty R, Harris M, et al. Optimal selection of individuals for BRCA
from the clinician's perspective, neither BRCAPRO nor mutation testing. J Clin Oncol. 2006;24(4):707-715.
any of the other models tested is adequately accurate to 2. Domchek SM, Eisen A, Calzone K, Stopfer J, Blackwood A, Weber BL. Appli-
cation of breast cancer risk prediction models in clinical practice. J Clin Oncol. 2003;
determine whether to offer BRCA testing for a single case of breast cancer, especially in the setting of limited family 3. Fewell Z, Smith GD, Sterne JA. The impact of residual and unmeasured con-
founding in epidemiological studies. Am J Epidemiol. 2007;166(6):646-655.
4. Berry DA, Iversen ES Jr, Gudbjartsson DF, et al. BRCAPRO validation, sensitiv-
While model-driven estimates have an important role as ity of genetic testing of BRCA1/BRCA2, and prevalence of other breast cancersusceptibility genes. J Clin Oncol. 2002;20(11):2701-2712.
a component of the genetic cancer risk assessment process, 5. American Society of Clinical Oncology. American Society of Clinical Oncology
policy statement update: genetic testing for cancer susceptibility. J Clin Oncol.
2003;21(12):2397-2406.
Table. BRCAPRO Prior Probability for Documented BRCA Carriers
BRCAPRO Prior Probability, No. (%)
⬍10%a (n = 16)b
ⱖ10% (n = 13)
Ashkenazi Jewish ancestry Data Processing Error: In the Original Contribution entitled "Fluoxetine After Weight
Restoration in Anorexia Nervosa: A Randomized Controlled Trial" published in the June 14, 2006, issue of JAMA (2006;295(22):2605-2612), a data processing er-
ror resulted in data errors in TABLE 3. The corrected analyses do not change the
conclusions of the study. The data processing error also resulted in the reporting of a statistically significant difference between the New York and Toronto sites in the fraction of patients who remained in the study for 1 full year and whose body mass indexes (calculated as weight in kilograms divided by height in meters squared) A 10% probability threshold was chosen because it was the commonly accepted thresh- old at the time the study was conducted and has been used in other published studies.
did not fall to less than 18.5 for more than 4 weeks, which lost statistical signifi- b All 16 patients had a prior probability of less than 5%.
cance (P=.10) in the corrected data set.
Table 3. Random-Effects Regression Analysis of Change During Treatment
Random Effects Regression
Beck Depression Inventory Beck Anxiety Inventory Rosenberg Self-Esteem Scale Quality of Life Enjoyment and Satisfaction Questionnaire Eating Disorder Inventory Drive for thinness Body dissatisfaction Yale-Brown-Cornell Eating Disorder Scale a Coefficients and standard errors indicate change per month.
b Z scores are for the difference between the coefficients for fluoxetine and placebo.
c P values are for the difference between the coefficients for fluoxetine and placebo.
2008 JAMA, November 7, 2007—Vol 298, No. 17 (Reprinted)
2007 American Medical Association. All rights reserved.

Source: http://www.peachlab.org/wp-content/uploads/2013/05/Walsh_JAMA-Fluox.pdf