Antipsychotic drugs versus placebo for relapse prevention in schizophrenia: a systematic review and meta-analysis

Articles

Antipsychotic drugs versus placebo for relapse prevention in schizophrenia: a systematic review and meta-analysis Stefan Leucht, Magdolna Tardy, Katja Komossa, Stephan Heres, Werner Kissling, Georgia Salanti, John M Davis

Summary Background Relapse prevention with antipsychotic drugs compared with placebo in patients with schizophrenia has not been sufficiently addressed by previous systematic reviews. We aimed to assess the association between such drugs and various outcomes in patients with schizophrenia to resolve controversial issues. Methods We searched the Cochrane Schizophrenia Group’s specialised register for reports published before Nov 11, 2008; and PubMed, Embase, and ClinicalTrials.gov for those before June 8, 2011. We also contacted pharmaceutical companies and searched the reference lists of included studies and previous reviews. Randomised trials of patients with schizophrenia continued on or withdrawn from any antipsychotic drug regimen after stabilisation were eligible. Our primary outcome was relapse between 7 and 12 months. We also examined safety and various functional outcomes. We used the random effects model and verified results for the primary outcome with a fixed effects model. Heterogeneity was investigated with subgroup and meta-regression analyses. Findings We identified 116 suitable reports from 65 trials, with data for 6493 patients. Antipsychotic drugs significantly reduced relapse rates at 1 year (drugs 27% vs placebo 64%; risk ratio [RR] 0·40, 95% CI 0·33–0·49; number needed to treat to benefit [NNTB] 3, 95% CI 2–3). Fewer patients given antipsychotic drugs than placebo were readmitted (10% vs 26%; RR 0·38, 95% CI 0·27–0·55; NNTB 5, 4–9), but less than a third of relapsed patients had to be admitted. Limited evidence suggested better quality of life (standardised mean difference –0·62, 95% CI –1·15 to –0·09) and fewer aggressive acts (2% vs 12%; RR 0·27, 95% CI 0·15–0·52; NNTB 11, 6–100) with antipsychotic drugs than with placebo. Employment data were scarce and too few deaths were reported to allow significant differences to be identified. More patients given antipsychotic drugs than placebo gained weight (10% vs 6%; RR 2·07, 95% CI 2·31–3·25), had movement disorders (16% vs 9%; 1·55, 1·25–1·93), and experienced sedation (13% vs 9%; 1·50, 1·22–1·84). Substantial heterogeneity in size of effect was recorded. In subgroup analyses, number of episodes, whether patients were in remission, abrupt or gradual withdrawal of treatment, length of stability before trial entry, first-generation or secondgeneration drugs, and allocation concealment method did not significantly affect relapse risk. Depot preparations reduced relapse (RR 0·31, 95% CI 0·21–0·41) more than did oral drugs (0·46, 0·37–0·57; p=0·03); depot haloperidol (RR 0·14, 95% CI 0·04–0·55) and fluphenazine (0·23, 0·14–0·39) had the greatest effects. The effects of antipsychotic drugs were greater in two unblinded trials (0·26, 0·17–0·39) than in most blinded studies (0·42, 0·35–0·51; p= 0·03). In a meta-regression, the difference between drug and placebo decreased with study length.

Lancet 2012; 379: 2063–71 Published Online May 3, 2012 DOI:10.1016/S01406736(12)60239-6 See Comment page 2030 Department of Psychiatry and Psychotherapy, Technische Universität München, Klinikum rechts der Isar, Munich, Germany (Prof S Leucht MD, M Tardy MSC, K Komossa MD, S Heres MD, W Kissling MD); Department of Hygiene and Epidemiology, University of Ioannina School of Medicine, University Campus Ioannina, Ioannina, Greece (G Salanti PhD); Psychiatric Institute, University of Illinois at Chicago, Chicago, IL, USA (Prof J M Davis MD); and Maryland Psychiatric Research Center, Baltimore, MD, USA (Prof J M Davis) Correspondence to: Prof Stefan Leucht, Department of Psychiatry and Psychotherapy, Technische Universität München, Klinikum rechts der Isar, 81675 Munich, Germany [email protected]

Interpretation Maintenance treatment with antipsychotic drugs benefits patients with schizophrenia. The advantages of these drugs must be weighed against their side-effects. Future studies should focus on outcomes of social participation and clarify the long-term morbidity and mortality of these drugs. Funding German Ministry of Education and Research.

Introduction Schizophrenia is a debilitating, often lifelong disease. Naturalistic studies have shown that about 80% of patients relapse within 5 years.1,2 Although the half-life of many oral antipsychotic drugs is roughly 24 h, relapses often occur only months or years after last treatment. In 1995, Gilbert and co-workers3 reported that antipsychotic maintenance treatment reduces relapse rates, but many issues remain unresolved and guidelines are not consistent.4 Should patients who have had one episode of acute psychosis—20% of whom will not have another2,5— receive maintenance treatment and for how long? Previous reviews3,6 grouped remitted and symptomatic patients together, but remitted individuals might www.thelancet.com Vol 379 June 2, 2012

relapse at a decreased rate. Long-term treatment with antipsychotics has been associated with increased mortality,7 but in another study8 mortality was reduced, possibly because suicide is prevented. The acquisition expense of antipsychotic drugs is substantial (estimated US$18·5 billion worldwide in 2010),9 but schizophrenia’s main cost is admission to hospital. Therefore, policy makers need to know by how much antipsychotic drugs reduce this outcome. Whether depot drugs are better than oral forms because of improved compliance is unclear.10 Additionally, previous reviews3,6,11,12 have not assessed side-effects (eg, weight gain, sedation, or socially disfiguring tardive dyskinesia), or addressed whether antipsychotic drugs improve functional outcomes (eg, employment or quality of life). Finally, the 2063

Articles

evidence for so-called supersensitivity psychosis13 needs to be examined. The theory suggests that long-term use of antipsychotic drugs increases dopamine receptor sensitivity; abrupt rather than gradual withdrawal of antipsychotics could then cause rebound psychoses. In the last review of all maintenance antipsychotic drugs, Gilbert and colleagues3 did not undertake a full meta-analysis or subgroup analyses. We aimed to obtain information about antipsychotic drugs compared with palcebo as maintenance treatment in patients with schizophrenia to update guidelines and practice.

Methods Search strategy and selection criteria See Online for appendix

2064

For a full version of our methods see the published protocol14 and appendix. Briefly, we searched the Cochrane Schizophrenia Group’s specialised register (compiled by regular systematic searches; appendix p 12) for reports published before Nov 18, 2008; and PubMed, Embase, and ClinicalTrials.gov for those published before June 8, 2011. We used the search term “[cessation* or withdr?w* or discontinu* or halt* or stop* or drop?out* or dropout* or rehospitalis* or relaps* or maintain* or maintenance* or recur* in title, abstract, index terms of reference] or [withdrawal* in interventions of study]”. Additionally, we contacted pharmaceutical companies who produce second-generation antipsychotic drugs, and searched the reference lists of included studies and previous reviews.3,6 Trials with adequate randomisation and allocation concealment methods, or those described as randomised without further description were eligible. We excluded quasi-randomised studies (eg, allocation according to day of the week). We assumed randomisation in three trials described as double-blind, but excluded them from a sensitivity analysis. Quality of sequence generation, allocation concealment, blinding, missing outcomes, selective reporting, and other biases was further assessed with the Cochrane Collaboration’s risk of bias method.15 We did not use this method for study selection. Trials were eligible when antipsychotic drugs (any compound or route of administration and dose, but available in at least one country) were maintained or withdrawn from patients with schizophrenia, or with schizoaffective, schizophreniform, or delusional disorders, who had stabilised during treatment. We used no language restrictions. We did not set a minimum study duration, because we wanted to establish how soon differences between drug and placebo are reported. We included studies that did not apply standardised diagnostic criteria, because these criteria are not meticulously applied in clinical routine; however, we did exclude them in a sensitivity analysis. Studies of intermittent treatment16 and those that included follow-up of acute-phase responders without further randomisation were excluded, because this design compromises randomisation.

Data extraction Two authors (SL and KK) independently selected trials. Three (SL, MT, and KK) independently extracted information about study design, treatment, and outcomes into standard forms. Disagreements were resolved by discussion with other team members (SH, WK, and JMD) or contact with original investigators, who were all sent data extraction sheets with requests for correction. We used intention-to-treat data when available. When results only of patients who completed follow-up were presented, we assumed that the same percentage of those who did not complete would have had an event. The primary outcome was relapse between 7 and 12 months, as defined in the original studies. Other outcomes were readmission, dropout, improvement of disease, death, violence or aggressive behaviour, adverse events, quality of life, satisfaction with care, and employment. All outcomes were analysed for different lengths of follow-up (up to 3 months, 6 months, 1 year, or more than 1 year; appendix).

Statistical analyses For binary outcomes, we calculated risk ratios (RRs), absolute risk differences (RDs), and the number needed to treat to benefit or harm (NNTB/H) as the inverse of RDs. For continuous outcomes, we calculated standardised mean differences (Hedges’ g). Because we expected heterogeneity, we used DerSimonian and Laird’s random effects model17 (weights calculated by Mantel-Haenszel method). The degree of heterogeneity was assessed by visual inspection, and by a χ²-test combined with the I² method (I² >50% was regarded as substantial heterogeneity).18 In case of heterogeneity, we checked data extraction from outlier studies, investigated reasons for their different findings, and explored the effects of study exclusion in sensitivity analyses. We investigated potential sources of heterogeneity with subgroup analyses and meta-regressions. We assessed whether inclusion only of patients who had had one episode affected results, because these patients might have a better outlook than do others. Similarly, patients in remission might have more favourable outlooks. Additionally, we analysed whether duration of stability before study entry affected relapse risk. Posthoc analyses of single antipsychotic drugs, depot versus oral medication, first-generation versus secondgeneration drugs, open-label versus blinded trials, and allocation concealment methods were also undertaken. We assessed duration of stability before study entry and duration of drug withdrawal in placebo groups with unrestricted maximum likelihood random-effect metaregressions. Gradual withdrawal was defined as a reduction in dose for 3 weeks, or when depot treatment was given before study initiation (as in Viguera and co-workers’ study19). Other post-hoc meta-regressions addressed mean dose of chlorpromazine equivalents20 and study duration. www.thelancet.com Vol 379 June 2, 2012

Articles

Number Drug group of studies included

Control group

Mean study duration* (months)

Risk ratio (95% CI) Absolute difference (95% CI)

0·40 (0·33 to 0·49) –0·39 (–0·46 to –0·32)

NNTB/H (95% CI)

3 (2 to 3)

Relapse 7–12 months

24

392/1465 (27%)

773/1204 (64%)

11

Relapse independent of duration

62

744/3395 (22%)

1718/2997 (57%)

9

0·35 (0·29 to 0·41)

–0·38 (–0·43 to –0·33)

3 (2 to 3)

Participants readmitted to hospital 16

112/1132 (10%)

245/958 (26%)

13

0·38 (0·27 to 0·55)

–0·19 (–0·27 to –0·11)

5 (4 to 9)

Dropout for any reason

57

802/2642 (30%)

1130/2076 (54%)

9

0·53 (0·46 to 0·61)

–0·24 (–0·30 to –0·17)

4 (3 to 6)

Dropout because of inefficacy

46

412/2539 (16%)

830/2007 (41%)

8

0·37 (0·31 to 0·44)

–0·27 (–0·34 to –0·19)

4 (3 to 5)

Participants unimproved/worse

14

614/880 (70%)

569/644 (88%)

5

0·73 (0·64 to 0·84) –0·25 (0·35 to 0·14)

5

9/403 (2%)

34/277 (12%)

8

0·27 (0·15 to 0·52)

–0·09 (–0·17 to –0·01)

11 (6 to 100)

2

63/130 (48%)

50 (H7 to B10)†

Violent/aggressive behaviour Participants employed Death (any) Suicide

14 8

4 (3 to 7)

65/129 (50%)

11

0·96 (0·75 to 1·23)

–0·02 (–0·14 to 0·10)

5/1240 (<1%)

7/1116 (1%)

7

0·77 (0·28 to 2·11)

0·00 (–0·01 to 0·00)

∞

0/1021

2/920 (<1%)

6

0·34 (0·04 to 3·28)

0·00 (–0·01 to 0·00)

∞ ∞

Death from natural causes

14

5/1272 (1%)

3/1129 (<1%)

7

1·24 (0·39 to 3·97)

0·00 (0·00 to 0·01)

Dropout because of AE

43

129/2437 (5%)

78/1896 (4%)

8

1·16 (0·70 to 1·91)

0·00 (–0·01 to 0·02)

At least one AE

10

575/1188 (48%)

450/996 (45%)

7

1·01 (0·87 to 1·18)

0·01 (–0·06 to 0·08)

At least one MD

22

304/1901 (16%)

134/1510 (9%)

7

1·55 (1·25 to 1·93)

0·06 (0·03 to 0·10)

Dyskinesia

13

18/1051 (2%)

37/769 (5%)

9

0·52 (0·28 to 0·97)

–0·01 (–0·02 to 0·01)

Use of antiparkinsonian medication 7

182/748 (24%)

90/569 (16%)

7

1·40 (1·03 to 1·89)

0·09 (0·02 to 0·16)

11 (6 to 50)

Sedation

10

158/1174 (13%)

85/972 (9%)

6

1·50 (1·22 to 1·84)

0·05 (0·00 to 0·10)

20 (B=∞ to H10)†

Weight gain

10

128/1231 (10%)

61/1090 (6%)

7

2·07 (2·31 to 3·25)

0·05 (0·03 to 0·07)

20 (14 to 33)

0·1

1·0 Favours drug

∞ 100 (H17 to B13)† 17 (10 to 33) 100 (H50 to B100)†

10 Favours placebo

Figure 1: Summary of pooled results Data are n/N (%) unless otherwise stated. The random effects model by DerSimonian and Laird17 was used throughout, with weights calculated by the Mantel-Haenszel method. NNTB/H=number needed to treat to benefit or harm. H=harm. B=benefit. AE=adverse event. MD=movement disorder. *Weighted by sample size of individual trials. †Because of space limitations, we did not use the display suggested by Altman.31

To establish the robustness of the primary outcome by sensitivity analyses, we applied a fixed effects model; and excluded studies that were not double blind, studies in which randomisation was implied by descriptions of blinding, trials with unclear randomisation or allocation-concealment methods (post hoc), and studies without standardised diagnostic criteria (post hoc). We used the original studies’ assumptions for dropouts. To test the theory that any beneficial effect of antipsychotic drugs in relapse prevention merely resulted from rebound psychosis after abrupt withdrawal,19 we did post-hoc analyses for subsequent relapse risk of participants who had not relapsed for 3–9 months after randomisation. Publication bias was assessed with contour-enhanced funnel-plots21 and the trim-and-fill method.22 We used GRADE Profiler 3.2 to produce a summary table of outcomes defined a priori. All meta-analytic calculations were made with RevMan (version 5.1), STATA (version 10.1), and Comprehensive Meta-analysis (version 2). All analyses were two-tailed, with an α of 0·05.

Role of the funding source The sponsors of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. www.thelancet.com Vol 379 June 2, 2012

Results We identified 1923 records, of which 116 reports of 65 randomised controlled trials (63 double-blind, placebo-controlled; two open with no treatment as comparator) with a total of 6493 participants were eligible (appendix pp 12, 16–30). However, one of these reports could not be included in the meta-analyses (appendix p 31). We obtained one unpublished study from a pharmaceutical company (Vanderburg D, Pfizer, personal communication), and additional information about seven trials.23–29 Investigators of one study30 communicated that data were no longer available. Astellas, Bristol-Myers Squibb, Eli Lilly, Lundbeck, Pfizer, Johnson and Johnson, and Sanofi-Aventis declared that no other randomised controlled trials of their second-generation antipsychotic drugs exist. Median number of participants was 47 (IQR 30–121) and study duration was 26 weeks (1·75–156). Reports had been published between 1959 and 2011. 31 studies were done with hospital inpatients, 27 with outpatients, and three with both; in four, the setting was unclear (appendix). In 11 trials, all participants were in remission at baseline, and eight included results from first-episode groups or subgroups (appendix). In 33 trials, participants were diagnosed by various standardised diagnostic criteria and in 32, the criteria used were not mentioned. Mean age was 40·8 years (SD 9·3), duration of illness was 13·6 years (8·8), and length of stability was 36 weeks (56). 2065

Articles

Number of randomised controlled trials

Drug group

Control group

Risk ratio (95% CI)

First vs several episodes First episode Other

8

67/255 (26%)

167/273 (61%)

0·47 (0·38–0·58)

19

325/1210 (27%)

606/931 (65%)

0·39 (0·31–0·49)

Test for subgroup differences: Q=1·36, df=1, p=0·24, I2=26·5% Remission vs non-remission Remitted Other

8

92/304 (30%)

134/212 (63%)

0·38 (0·24–0·61)

16

300/1161 (26%)

639/992 (64%)

0·40 (0·33–0·49)

Test for subgroup differences: Q=0·03, df=1, p=0·87, I2=0% Abrupt withdrawal vs tapering Abrupt

16

313/1108 (28%)

540/838 (64%)

0·43 (0·34–0·54)

Tapered

8

79/357 (22%)

233/366 (64%)

0·34 (0·23–0·50)

Test for subgroup differences: Q=1·05, df=1, p=0·31, I2=4·9% Time stable before the study* ≥1 month

5

49/232 (21%)

140/196 (71%)

0·27 (0·15–0·46)

≥3 months

5

71/413 (17%)

227/393 (58%)

0·30 (0·21–0·44)

≥6 months

1

1/10 (10%)

3/10 (30%)

0·33 (0·04–2·69)

≥12 months

5

35/165 (21%)

97/161 (60%)

0·31 (0·17–0·57)

≥3–6 years

2

6/27 (22%)

17/27 (63%)

0·38 (0·18–0·78)

Test for subgroup differences: Q=0·58, df=4, p=0·96, I2=0% Antipsychotic drugs† Chlorpromazine

2

63/207 (30%)

137/199 (69%)

0·44 (0·36–0·55)

Fluphenazine depot

6

18/160 (11%)

74/136 (54%)

0·23 (0·14–0·39)

Haloperidol depot

1

2/20 (10%)

16/23 (70%)

0·14 (0·04–0·55)

Quetiapine

1

27/89 (30%)

56/89 (63%)

0·48 (0·34–0·69)

Paliperidone

2

78/311 (25%)

212/306 (69%)

0·36 (0·30–0·45)

10

107/400 (27%)

185/305 (61%)

0·42 (0·27–0·65)

2

97/278 (35%)

93/146 (64%)

0·52 (0·43–0·64)

Mixed antipsychotics Ziprasidone

Test for subgroup differences: Q=14·68, df=6, p=0·02, I2=59·1% Depot vs oral drugs Depot Oral

7

60/332 (18%)

202/331 (61%)

0·31 (0·23–0·41)

14

307/1017 (30%)

507/768 (66%)

0·46 (0·37–0·57)

Test for subgroup differences: Q=4·52, df=1, p=0·03, I2=77·9% First-generation vs second-generation drugs First Second

18

188/778 (24%)

402/652 (62%)

0·35 (0·25–0·48)

6

204/687 (30%)

371/552 (67%)

0·44 (0·37–0·53)

Test for subgroup differences: Q=1·44, df=1, p=0·23, I2=30·4% Blinded vs unblinded studies Blinded Unblinded

22

370/1334 (28%)

691/1078 (64%)

0·42 (0·35–0·51)

2

22/131 (17%)

82/126 (65%)

0·26 (0·17–0·39)

Test for subgroup differences: Q=4·50, df=1, p=0·03, I2=77·8% Appropriate vs unclear allocation concealment Appropriate

9

214/778 (28%)

416/632 (66%)

0·41 (0·33–0·52)

Unclear AC

15

178/687 (26%)

357/572 (62%)

0·39 (0·28–0·53)

Test for subgroup differences: Q=0·09, df=1, p=0·76, I2=0% 0·01

0·10

1·0 Favours drug

Favours placebo

Figure 2: Summary of subgroup analyses of relapse at 7–12 months Data are n/N (%) unless otherwise stated. The random effects model by DerSimonian and Laird17 was used throughout, with weights calculated by the Mantel-Haenszel method. *No trials of patients stable for the predefined duration of at least 24 months were available. †Excluding the mixed group does not change the result.

60% of participants were men. 15 studies were industry sponsored; the others had public sponsors or the sponsorship was not mentioned. Trial definitions of relapse varied: 26 used clinical judgment, 17 need of medication, 15 rating-scale-based 2066

definitions, three admission to hospital, two dropout because symptoms worsened, two unclear methods. 19 studies used mixed groups of antipsychotic drugs; others used single drugs (appendix). Doses were usually flexible, with a mean chlorpromazine equivalent20 of www.thelancet.com Vol 379 June 2, 2012

Articles

www.thelancet.com Vol 379 June 2, 2012

0·40 –0·04 –0·48 –0·92 Log risk ratio

459 mg (SD 247; range 180–1425). Antipsychotics were mostly withdrawn abruptly, but 11 studies included gradual withdrawal (mean period 28 days; SD 20·82). In terms of potential bias, 16 studies had adequate randomisation and 18 adequate allocation concealment (appendix). The others were randomised, but no details were provided. All but two studies were double-blind and only three reports confirmed the success of blinding. Most protocols defined relapse as an endpoint after which patients were excluded from the study, leading to high drop-out rates. Most selective reporting was of adverse events, which some investigators reported only when at least 5–10% of participants were affected. Sources of other bias were early termination after interim analysis, high or low doses, baseline imbalance, loss of blinding after relapse, and an altered randomisation scheme. Results for relapse, readmission, and dropout were of high quality, but data for suicide and quality or life were poor, and for employment very poor (appendix pp 90–92). No data were available for satisfaction of care. Fewer participants given antipsychotic drugs relapsed than did those given placebo (figure 1). This difference was recorded at all time intervals assessed and was irrespective of study length (figure 1, appendix). Similarly, fewer individuals in experimental groups of outpatient studies were readmitted than were those in control groups (figure 1, appendix). Notably, less than a third of patients had such severe relapses that admission was necessary. When only trials that included reports of both relapse and readmission were used in analyses, 25% of patients given drugs relapsed compared with 69% given placebo (RR 0·35, 95% CI 0·27–0·45; RD 0·40, 95% CI 0·31–0·49). 10% of patients given drugs were readmitted compared with 25% given placebo (RR 0·39, 95% CI 0·27–0·56; RD 0·18, 95% CI 0·09–0·26). Participants in drug groups dropped out for any reason and because of treatment inefficacy less often than did controls (figure 1). The main cause of dropout was relapse, because most protocols stipulated that relapsed patients had to leave the study. Dropout because of adverse events did not differ significantly (figure 1, appendix pp 42–47). At trial conclusion, the number of patients with unimproved or worse disease than at entry was lower in drug groups than in controls (figure 1, appendix p 48). Fewer participants given antipsychotic drugs behaved aggressively than did those given placebo (figure 1, appendix p 49). Furthermore, findings from three studies (n=527) showed that quality of life was better in experimental groups than in placebo groups (standardised mean difference –0·62, 95% CI –1·15 to –0·09; appendix p 50). Only two studies reported whether participants were employed, and groups did not differ significantly (figure 1, appendix p 50). No significant difference in mortality due to any cause, natural causes, and suicide or suicide attempts was recorded (figure 1, appendix pp 51–54), although

–1·36 –1·80 –2·24 –2·68 –3·12

Slope=0·007 (95% CI 0·002–0·010), p<0·0001

–3·56 –4·00 0

5

23

41

60 79 97 Study duration (weeks)

116

134

153

171

Figure 3: Meta-regression with study duration as a moderator Outcome was relapse. Circle size represents weight each study was given in the analysis.

the total number of each event was small (five deaths in the drug groups and seven in the placebo groups). However, patients given antipsychotic drugs had more adverse events, such as movement disorders, akathisia (after exclusion of an outlier), dystonia, use of antiparkinsonian drugs, sedation, and weight gain, than did those given placebo (figure 1, appendix). Dyskinesia was more frequent in the placebo groups than in the drug ones (figure 1). No significant differences were recorded for number of patients with at least one adverse effect (figure 1), akinesia, rigidity, or tremors (appendix pp 55–65). Various results were heterogeneous, but heterogeneity was usually a result of variation in degree of difference rather than in direction of effect. Exploratory exclusion of outliers did not substantially change the results (appendix pp 35–65). In subgroup analyses, risk ratios in studies or subgroups of patients who had had one episode and of participants in remission (for definitions see appendix p 67) did not differ from those in other studies or subgroups (figure 2, appendix pp 66, 67). Depot formulations of haloperidol and fluphenazine had larger effects than did other drugs, and this finding was confirmed by subgroup analysis of depot versus oral preparations (figure 2). The effects of first-generation and second-generation drugs were not different (figure 2, appendix pp 68–70). In the two openlabel studies, the antipsychotic drugs’ effects on relapse were larger than in blinded trials (figure 2). Quality of allocation concealment did not alter effect size (figure 2, appendix pp 73, 74). Additionally, the effect size did not differ with length of time participants had been stable before trial entry. Even in studies in which participants had been stable for 3–5 years32 and 6 years,33 more relapses were noted in the placebo groups than in experimental ones (figure 2, 2067

Articles

Number of Drug group randomised controlled trials

Risk ratio (95% CI)

Control group

0·40 (0·30–0·55)*

3 months 20

243/1678 (14%) 445/1264 (35%)

6 months 13

133/862 (15%)

199/520 (38%)

0·40 (0·26–0·61)*

9 months 10

100/556 (18%)

115/275 (42%)

0·46 (0·29–0·73)† 0·1 Favours drug

1·0 10 Favours placebo

Figure 4: Sensitivity analysis of subsequent relapse risk after specific lengths of time without a patient relapsing Data are n/N (%) unless otherwise stated. The random effects model by DerSimonian and Laird17 was used throughout, with weights calculated by the Mantel-Haenszel method. *p<0·0001. †p=0·0009.

Risk ratio Number Total number of (95% CI) of studies participants

p value

Fixed effects model

24

2669

0·40 (0·36–0·44) <0·0001

Exclusion of the two open trials

22

2412

0·42 (0·35–0·51)

Appropriately randomised trials only

8

1546

0·41 (0·34–0·50) <0·0001

Appropriate allocation concealment only

<0·0001

9

1410

0·41 (0·33–0·52)

Exclusion of two studies with assumed randomisation 23

2654

0·40 (0·33–0·49) <0·0001

<0·0001

Original studies’ assumptions for patients who did not complete follow-up

24

2669

0·40 (0·33–0·49) <0·0001

Excluding eight studies without standardised diagnostic criteria

16

2325

0·42 (0·34–0·49) <0·0001

Table: Sensitivity analyses of the primary outcome

appendix p 71). Abrupt or gradual withdrawal of prestudy antipsychotic drugs did not change relapse risk (figure 2, appendix p 72). The only statistically significant moderator in meta-regressions was that the difference between drug and placebo decreased with study length (figure 3, appendix pp 75–78). Even when only participants who had not relapsed until 3 months, 6 months, or 9 months were included in the analysis, antipsychotic drugs were still more effective than was placebo (figure 4, appendix). The other sensitivity analyses showed that findings did not change with exclusion of specific studies (table, appendix pp 79–86). The contour-enhanced funnel-plot to assess publication bias was asymmetrical (appendix pp 87–89). Nevertheless, the trim-and-fill adjusted effect size was only slightly reduced (RR 0·41, 95% CI 0·34–0·50 vs 0·46, 0·38–0·56). Even when only large trials (>200 participants) were included, the effect size did not greatly change (RR 0·39, 95% CI 0·32–0·48; appendix). Overall, inclusion of small, unpublished trials would not have substantially changed the findings.

Discussion We have established that antipsychotic maintenance treatment substantially reduces relapse risk in all patients with schizophrenia for up to 2 years of follow-up. The effect was robust in important subgroups such as 2068

patients who had had only one episode and those in remission, but seemed to decrease in size with time. Moreoever, we present novel results for outcomes other than relapse, particularly social participation. Clinicians and guideline developers need to know whether patients who have been stable for a specific number of years still benefit from treatment or whether they are cured and no longer have any need of drug treatment. Relapse reduction was not different in patients who had been stable for up to 3–6 years. However, this finding is based on two studies,32,33 and more data are needed for patients who have had different numbers of episodes. In studies of patients who had had only one episode, the longest duration of stability before inclusion in investigations was 1 year,34,35 which could serve as the minimum recommended duration for maintenance treatment. The meta-regression suggested that antipsychotic drugs might lose their effectiveness with time. However, in studies lasting 1 year or less, the decreasing difference between drugs and placebo was not apparent with visual inspection of survival curves23–25,27,29,36–41 or a meta-regression (appendix). Of the five studies lasting longer than 1 year, two35,42 of three that showed such a decrease were small (n≤30). Most (100%42 and 90%35) participants in the placebo groups had already relapsed after 1 year and so the difference could only decrease in size after this point because only patients given drugs were yet to relapse. The findings from the other three studies34,43,44 were not consistent. The reduction in effect size could be due to many factors, such as differences in patient characteristics between short-term and long-term studies, or falling adherence in the drug group with time. Few investigators have systematically examined non-compliance—as high as 50% in a large study45— reducing drug effectiveness. Large studies that last longer than 2 years and monitor compliance or use depot drugs are urgently needed for clarification. Our results do not support the suggestion that beneficial effects of antipsychotic drugs could be merely because of supersensitivity psychosis.13 We did not record a significant difference in relapse reduction between studies in which antipsychotics were withdrawn abruptly or gradually, nor was the duration of gradual withdrawal an important factor. Even when only patients who had not relapsed 9 months after study initiation were analysed, the subsequent relapse rate was still significantly higher in placebo groups than in drug groups. Similarly, Viguera and co-workers’19 primary metaanalysis showed that abrupt or gradual withdrawal of treatment did not alter the effect, but in an individual patient analysis of three trials, investigators reported significantly higher relapse rates after abrupt withdrawal than after gradual reduction. However, one of the studies was unpublished and another had a flawed design,46 because the experimental participants were gradually withdrawn during 1 year and the control participants www.thelancet.com Vol 379 June 2, 2012

Articles

continued to receive antipsychotic drugs. The control participants were abruptly withdrawn only after 1 year. In the third study,39 more relapses were noted when oral rather than depot fluphenazine was discontinued abruptly (withdrawal of depot drugs is similar to gradual reduction of oral medication because of their long halflife). This study lasted only 15 weeks, which is perhaps too short for all depot fluphenazine to have been removed. Nevertheless, we recommend gradual withdrawal, because meta-analytic methods might not be sensitive enough to detect differences, and supersensitivity psychosis could also explain a certain proportion of the decreasing effect size with time. Because hospital treatment is the main cost factor in many countries, that its relative risk reduction was similar to that of relapse is important. Nevertheless, the absolute risk reduction was lower than for relapse. In trials that reported both relapse and admission, the difference in relapse risk between experimental and control groups was much larger than was reduction in admission to hospital. Cost calculations should take into account that less than a third of relapsed patients need to be admitted. Although how easily patients are admitted to hospital depends on the setting, this fairly hard and intuitively interpretable outcome should be reported in all future studies. Reduced rate of aggression is an important outcome, because aggressive acts of acutely ill people with schizophrenia contribute to stigma and can have dire effects on the patient and society. Improvement in quality of life—albeit on the basis of only three studies—is also notable, because this measure mirrors patients’ perspectives and combines effectiveness and side-effects. However, our finding (from only two studies) that rates of employment did not differ between groups suggests that function might not be affected by maintenance treatment. More evidence is needed to clarify whether antipsychotic drugs only suppress symptoms or also improve social participation. Few researchers reported death rates, so the debate about whether antipsychotics prevent suicides or increase mortality due to adverse events is unresolved.7,8 Any findings were not significant and were underpowered. Death should always be reported, because meta-analytic pooling might be the only way to obtain sufficient randomised data for rare events. As a class, antipsychotic drugs produced several side-effects. Weight gain is a concern because it can lead to diabetes and cardiovascular events. That dyskinesia occurred more frequently in the placebo groups is probably a result of so-called rebound dyskinesias induced by abrupt withdrawal of treatment rather than persisting tardive dyskinesia, but the disorder was not assessed by appropriate criteria.47 Future studies should clearly describe whether adverse events had newly emerged or were already present before randomisation, and whether they persisted. We believe that accurate quantification of long-term morbidity and mortality associated with antipsychotics should be prioritised. www.thelancet.com Vol 379 June 2, 2012

Pooling of trials that have been done in the past 50 years and that used different relapse definitions and drugs led to inevitable heterogeneity. However, the recorded variation was mostly a matter of differences in degree of effect, which is much less problematic than are differences in direction of effect.15 Universally accepted definitions of relapse and stability of symptoms are urgently needed.48 We investigated potential sources of bias, mainly because adherence to the CONSORT guidelines was poor,49 but we recorded no difference in subgroup and sensitivity analyses for randomisation and allocation concealment. Highly sensitive time-to-relapse data derived from survival analyses were not available for most studies; therefore, we used number of relapsed participants, which underestimates the difference. We believe that our assumption that missing participants had the same percentage of events as patients who completed the study is more realistic than are best-case or worst-case scenarios that are likely to underestimate or overestimate risk. The model assumes missing at random specific to group. Because imputations of events for missing participants were not needed for relapse and were applied only to a few other outcomes and participants in 14 of 65 trials, the anticipated effect is small. Although the funnel-plot was asymmetrical, it might not represent publication bias, but rather that small trials often overestimate drug effects (so-called small-study effects15). Indeed, many trials were small but neither the trim-and-fill test nor inclusion of only large trials substantially changed the effect size. Nevertheless, more than one trial from the past 50 years is likely to have remained unpublished. Almost all studies included relapse, but other outcomes were inconsistently reported. Because original study protocols were not available, we cannot accurately distinguish between outcomes that were not measured or not reported. Finally, pooling of different antipsychotic drugs as a class is appropriate for efficacy because differences between compounds are small,50 with the possible exception of clozapine (the most efficacious antipsychotic drug); however, there were no clozapine studies. Additionally, antipsychotic drugs that were more effective in trials of acute-phase schizophrenia than were others (amisulpride, olanzapine, and risperidone)50 did not contribute to the effects on the primary outcome. Indeed, no significant differences in effectiveness were noted, apart from an increased effect of the depot forms of haloperidol and fluphenazine. Whether depot drugs are better than oral formulations as suggested by epidemiological10 and mirrorimage studies51 can only be clarified by direct comparisons. A recent update (unpublished) of a meta-analysis52 did not show a difference. Finally, first-generation and secondgeneration drugs did not differ in effectiveness, failing to support such a classification.50 However, antipsychotic drugs do produce different side-effects—eg, many old antipsychotics produce movement disorders and newer ones weight gain.50 2069

Articles

Contributors SL, MT, KK, SH, WK, and JD designed the review and extracted data. SL, MT, and KK selected studies to be included. SL, MT, GS, and JD did statistical analyses. All authors wrote the report. Conflicts of interest SL has received honoraria as a consultant or a member of advisory boards from Alkermes, Bristol-Myers Squibb, Eli Lilly, Janssen, Johnson and Johnson, Medavante, Roche; he has received lecture honoraria from AstraZeneca, Bristol-Myers Squibb, Eli Lilly, EssexPharma, Janssen, Johnson and Johnson, Lundbeck Institute, Pfizer, and Sanofi-Aventis; and is primary investigator of a trial for which Eli Lilly has provided medication. SH has received honoraria from Janssen-Cilag, Sanofi-Aventis, and Johnson and Johnson; and has accepted travel or hospitality payment from Janssen-Cilag, Sanofi-Aventis, Johnson and Johnson, Pfizer, Bristol-Myers Squibb, AstraZeneca, Lundbeck, Novartis, and Eli Lilly. WK has received honoraria for board memberships, consulting, and lectures from Janssen and Eli Lilly; honoraria for development of educational materials from Janssen; grant support from Janssen and AstraZeneca; and travel and accommodation expenses from AstraZeneca, Eli Lilly, and Janssen. The other authors declare that they have no conflicts of interest. Acknowledgments Financial support was mainly provided by a grant from the German Ministry for Education and Research (Bundesministerium für Bildung und Forschung), number 01KG0816 88166528. GS was supported by a grant from the European Research Council (IMMA 260559). The authors thank the Cochrane Schizophrenia Group for its support for this review; Michael Borenstein for his statistical advice on publication bias; Richard Skodnek for his work on a preliminary version of this review; and Astellas, Bristol-Myers Squibb, Eli Lilly, Lundbeck, Pfizer, Johnson and Johnson, Sanofi-Aventis, Eric Chen, George Gardos, Julian Leff, and Erik Denys for additional information. This Article is a copublication with a Cochrane Review (Cochrane Database Syst Rev 2012; 5: CD008016). References 1 Robinson D, Woerner MG, Alvir JM, et al. Predictors of relapse following response from a first episode of schizophrenia or schizoaffective disorder. Arch Gen Psychiatry 1999; 56: 241–47. 2 Shepherd M, Watt D, Falloon I, Smeeton N. The natural history of schizophrenia: a five-year follow-up study of outcome and prediction in a representative sample of schizophrenics. Psychol Med Suppl 1989; 15: 1–46. 3 Gilbert P, Harris MJ, McAdams LA, et al. Neuroleptic withdrawal in schizophrenic patients: a review of the literature. Arch Gen Psychiatry 1995; 52: 173–88. 4 Kane JM, Leucht S, Carpenter D, Docherty JP. Optimising pharmacologic treatment of psychotic disorders. J Clin Psychiatry 2003; 64 (suppl 12): 1–100. 5 Robinson DG, Woerner MG, Alvir JMJ, et al. Predictors of treatment response from a first episode of schizophrenia or schizoaffective disorder. Am J Psychiatry 1999; 156: 544–49. 6 Davis JM. Overview: maintenance therapy in psychiatry: I Schizophrenia. Am J Psychiatry 1975; 132: 1237–45. 7 Ray WA, Chung CP, Murray KT, Hall K, Stein CM. Atypical antipsychotic drugs and the risk of sudden cardiac death. N Engl J Med 2009; 360: 225–35. 8 Tiihonen J, Lonnqvist J, Wahlbeck K, et al. 11-year follow-up of mortality in patients with schizophrenia: a population-based cohort study (FIN11 study). Lancet 2009; 374: 620–27. 9 BCC Research. Antipsychotic drugs: technologies and global markets. Wellesley, MA: BCC Research, 2010. 10 Tiihonen J, Haukka J, Taylor M, et al. A nationwide cohort study of oral and depot antipsychotics after first hospitalization for schizophrenia. Am J Psychiatry 2011; 168: 603–09. 11 National Collaborating Centre for Mental Health. Schizophrenia: core interventions in the treatment and management of schizophrenia in primary and secondary care (update). London: National Institute for Health and Clinical Excellence, 2009. 12 Leucht S, Barnes TRE, Kissling W, et al. Relapse prevention in schizophrenia with new-generation antipsychotics: a systematic review and exploratory meta-analysis of randomized, controlled trials. Am J Psychiatry 2003; 160: 1209–22.

2070

13

14

15 16

17 18 19

20

21

22

23

24

25

26

27

28 29

30

31 32 33

34

35

36

Chouinard G, Annable L, Jones BD, Collu R. Lack of tolerance to long-term neuroleptic treatment in dopamine tuberoinfundibular system. Acta Psychiatr Scand 1980; 64: 353–62. Komossa K, Depping AM, Heres S, et al. Maintenance treatment with antipsychotic drugs for schizophrenia. http://onlinelibrary.wiley. com/doi/10.1002/14651858.CD008016/full (accessed March 8, 2012). Higgins JPT, Green S. Cochrane Handbook for Systematic Reviews of Interventions. Chichester, UK: Wiley and Sons, 2008. Carpenter WT, Hanlon TE, Heinrichs DW, et al. Continuous versus targeted medication in schizophrenic outpatients: outcome results. Am J Psychiatry 1990; 147: 1138–63. DerSimonian R, Laird N. Meta-analysis in clinical trials. Contr Clin Trials 1986; 7: 177–88. Higgins JPT, Thompson SG, Deeks JJ, Altman DG. Measuring inconsistency in meta-analyses. BMJ 2003; 327: 557–60. Viguera AC, Baldessarini RJ, Hegarty JD, van Kammen DP, Tohen M. Clinical risk following abrupt and gradual withdrawal of maintenance neuroleptic treatment. Arch Gen Psychiatry 1997; 54: 49–55. Gardner DM, Murphy AL, O’Donnell H, Centorrino F, Baldessarini RJ. International consensus study of antipsychotic dosing. Am J Psychiatry 2010; 167: 686–93. Peters JL, Sutton AJ, Jones DR, Abrams KR, Rushton L. Contour-enhanced meta-analysis funnel plots help distinguish publication bias from other causes of asymmetry. J Clin Epidemiol 2008; 61: 991–96. Duval S, Tweedie R. Trim and fill: a simple funnel-plot-based method of testing and adjusting for publication bias in meta-analysis. Biometrics 2000; 56: 455–63. Arato M, O’Connor R, Meltzer HY. A 1-year, double-blind, placebo-controlled trial of ziprasidone 40, 80 and 160 mg/day in chronic schizophrenia: the Ziprasidone Extended Use in Schizophrenia (ZEUS) study. Int Clin Psychopharmacol 2002; 17: 207–15. Beasley CM, Sutton VK, Hamilton SH, et al. A double-blind, randomized, placebo-controlled trial of olanzapine in the prevention of psychotic relapse. J Clin Psychopharmacol 2003; 23: 582–94. Cooper SJ, Butler A, Tweed J, Welch C, Raniwalla J. Zotepine in the prevention of recurrence: a randomised, double-blind, placebo-controlled study for chronic schizophrenia. Psychopharmacology 2000; 150: 237–43. Gardos G, Cole JO, Rapkin RM, et al. Anticholinergic challenge and neuroleptic withdrawal: changes in dyskinesia and symptom measures. Arch Gen Psychiatry 1984; 41: 1030–35. Kramer M, Simpson G, Maciulis V, et al. Paliperidone extended-release tablets for prevention of symptom recurrence in patients with schizophrenia: a randomized, double-blind, placebo-controlled study. J Clin Psychopharmacol 2007; 27: 6–14. Leff JP, Wing JK. Trial of maintenance therapy in schizophrenia. BMJ 1971; 3: 599–604. Chen EYH, Hui CLM, Lam M, et al. Maintenance treatment with quetiapine versus discontinuation after one year of treatment in patients with remitted first episode psychosis: randomised controlled trial. BMJ 2010; 341: c4024. Denijs EL, Vereeken JL. Pimozide (OrapR, R 6238) in residual schizophrenia: a clinical evaluation with long-term double-blind follow-up. Psychiatr Neurol Neurochir 1973; 76: 47–59. Altman DG. Confidence intervals for the number needed to treat. BMJ 1998; 317: 1309–12. Cheung HK. Schizophrenics fully remitted on neuroleptics for 3–5 years—to stop or continue drugs? Br J Psychiatry 1981; 138: 490–94. Sampath G, Shah A, Krska J, Soni SD. Neuroleptic discontinuation in the very stable schizophrenic patient—relapse rates and serum neuroleptic levels. Hum Psychopharmacol 1992; 7: 255–64. Crow TJ, MacMillan JF, Johnson AL, Johnstone EC. A randomised controlled trial of prophylactic neuroleptic treatment. Br J Psychiatry 1986; 148: 120–27. Boonstra G, Burger H, Grobbee DE, Kahn RS. Antipsychotic prophylaxis is needed after remission from a first psychotic episode in schizophrenia patients: results from an aborted randomised trial. Int J Psychiatry Clin Pract 2011; 15: 128–34. Caffey EM, Diamond LS, Frank TV, et al. Discontinuation or reduction of chemotherapy in chronic schizophrenics. J Chronic Dis 1964; 17: 347–58.

www.thelancet.com Vol 379 June 2, 2012

Articles

37

38

39

40

41

42

43

44

Hough D, Gopal S, Vijapurkar U, et al. Paliperidone palmitate maintenance treatment in delaying the time-to-relapse in patients with schizophrenia: a randomized, double-blind, placebo-controlled study. Schizophr Res 2010; 116: 107–17. Kane JM, Mackle M, Snow-Adami L, et al. A randomized placebo-controlled trial of asenapine for the prevention of relapse of schizophrenia after long-term treatment. J Clin Psychiatry 2011; 72: 349–55. Levine J, Schooler NR, Severe J, et al. Discontinuation of oral and depot fluphenazine in schizophrenic patients after one year of continuous medication: a controlled study. Adv Biochem Psychopharmacol 1980; 24: 483–93. Peuskens J, Trivedi J, Malyarov S, et al. Prevention of schizophrenia relapse with extended release quetiapine fumarate dosed once daily: a randomized, placebo-controlled trial in clinically stable patients. Psychiatry 2007; 4: 34–50. Pigott TA, Carson WH, Saha AR, et al. Aripiprazole for the prevention of relapse in stabilized patients with chronic schizophrenia: a placebo-controlled 26-week study. J Clin Psychiatry 2003; 64: 1048–56. Nishikawa T, Tsuda A, Tanaka M, Koga I, Uchida Y. Prophylactic effect of neuroleptics in symptom-free schizophrenia. Psychopharmacology 1982; 77: 301–04. Pietzcker A, Gaebel W, Koepcke W, et al. Intermittent versus maintenance neuroleptic long-term treatment in schizophrenia— 2-year results of a German multicenter study. J Psychiatr Res 1993; 27: 321–39. Hogarty GE, Goldberg SC, Schooler NR, Ulrich RF. Drug and sociotherapy in the aftercare of schizophrenic patients: II two-year relapse rates. Arch Gen Psychiatry 1974; 31: 603–08.

www.thelancet.com Vol 379 June 2, 2012

45

46

47 48

49

50

51 52

Hogarty GE, Goldberg SC. Drug and sociotherapy in the aftercare of schizophrenic patients: one-year relapse rates. Arch Gen Psychiatry 1973; 28: 54–64. Greenberg LM, Roth S. Differential effects of abrupt versus gradual withdrawal of chlorpromazine in hospitalized chronic schizophrenic patients. Am J Psychiatry 1966; 123: 221–26. Schooler NR, Kane JM. Research diagnosis for tardive dyskinesia. Arch Gen Psychiatry 1982; 39: 468–69. Gleeson JF, varez-Jimenez M, Cotton SM, Parker AG, Hetrick S. A systematic review of relapse measurement in randomized controlled trials of relapse prevention in first-episode psychosis. Schizophr Res 2010; 119: 79–88. Schulz KF, Altman DG, Moher D, CONSORT Group. CONSORT 2010 statement: updated guidelines for reporting parallel group randomised trials. BMJ 2010; 340: c332. Leucht S, Corves C, Arbter D, et al. Second-generation versus first-generation antipsychotic drugs for schizophrenia: a meta-analysis. Lancet 2009; 373: 31–41. Davis JM, Mentalon L, Watanabe MD. Depot antipsychotic drugs: place in therapy. Drugs 1994; 47: 741–73. Leucht C, Heres S, Kane JM, Kissling W, Davis JM, Leucht S. Oral versus depot antipsychotic drugs for schizophrenia—a critical systematic review and meta-analysis of randomised long-term trials. Schizophr Res 2011; 127: 83–92.

2071