There is a new debate about long-term treatment with antipsychotics stimulated by data suggesting a dose-related brain volume loss, Reference Huhtaniska, Jaaskelainen, Hirvonen, Remes, Murray and Veijola1 supersensitivity effects of long-term treatment with antipsychotics Reference Chouinard and Jones2,Reference Moncrieff3 and some follow-up studies showing that patients who do not receive antipsychotics in the long-term have better outcomes than treated patients. Reference Harrow, Jobe and Faull4
In this context Takeuchi et al present an analysis of the symptom trajectories in relapse prevention studies over 1 year. Reference Takeuchi, Kantor, Sanches, Fervaha, Agid and Remington5 In the placebo-treated groups they find a continuous worsening of approximately 50% over baseline of the mean Positive and Negative Syndrome Scale (PANSS)/Brief Psychiatric Rating Scale (BPRS) scores at 1 year, compared with an only 10% worsening of these scores in the antipsychotic group. This finding is important because it means that antipsychotic efficacy is maintained over time and should not be discontinued. At first glance, this contrasts with our meta-analysis of relapse prevention studies where 64% of placebo-treated v. 27% of drug-treated patients experienced a relapse within 1 year. Reference Leucht, Tardy, Komossa, Heres, Kissling and Salanti6 When we summarised the relapse rates of only the 11 studies included by Takeuchi et al, the results were more similar 55% v. 22% relapsed. The drug-placebo difference became smaller over time in our analysis, but we discussed that this was likely because of statistical artefacts, among others because most studies did not use survival analyses (Fig. 3 in Leucht et al Reference Leucht, Tardy, Komossa, Heres, Kissling and Salanti6 ). To understand better how it is possible that the average PANSS/BPRS scores remained relatively stable over 1 year on the drug, despite some patients relapsing, one needs to know that in many studies the patients are not necessarily symptom free at baseline and therefore patients can improve, as well as get worse. For example, in Pigott et al 2003, Reference Pigott, Carson, Sana, Torbeyns, Stock and Ingenito7 the average PANSS at baseline was about 82 and at the end-point the drug-treated group improved by two points. Or, in our meta-analysis 30% of drug-treated patients improved at the end-point (Fig. 1 in Leucht et al Reference Leucht, Tardy, Komossa, Heres, Kissling and Salanti6 ). Some patients relapsed in the drug-treated group but this was balanced by those patients that improved.
The symptom trajectories presented by Takeuchi et al are also important in the context of rebound or supersensitivity psychosis, a possibility suggested in the 1970s by Chouinard and colleagues Reference Chouinard and Jones2 based on an observation of ten patients. This is pharmacologically plausible as dopamine supersensitivity is observed in animals. (For reviews, see Murray et al Reference Murray, Quattrone, Natesan, van Os, Nordentoft and Howes8 and Moncrieff. Reference Moncrieff3 ) Takeuchi et al found a gradual worsening of symptoms rather than an initial peak as would be expected for withdrawal symptoms. In our meta-analysis we examined rebound psychosis by comparing abrupt and gradual withdrawal of antipsychotics in the placebo groups, but we did not find any difference. However, the minimum duration of tapering was only 3 weeks, Reference Takeuchi, Kantor, Sanches, Fervaha, Agid and Remington5 which may have been too short. We also compared the relapse rate of drug v. placebo, in patients who were relapse free on either drug or placebo for 3 months, 6 months and 9 months, and then followed up until the end of the studies. The differences in relapse rates between drug and placebo were similar (Fig. 4 in Leucht et al Reference Leucht, Tardy, Komossa, Heres, Kissling and Salanti6 ). If the difference in relapse rates were simply explained by supersensitivity psychosis, the relapse rate should have been lower in those patients who had been relapse free for 9 months. A similar observation has been reported in a meta-analysis of antidepressants. Reference Geddes, Carney, Davies, Furukawa, Kupfer and Frank9 It is possible after some time on the drug that patients might have been ‘cured’, and would not need antipsychotics any longer. In a third analysis we therefore examined whether the drug–placebo difference in relapse rates became smaller in patients who had been stable on antipsychotics for up to 3–6 years before they were randomised to either stay on antipsychotics or to be switched to placebo. We found that even those patients who had been stable for 3–6 years and then were randomised to drug or placebo, those on placebo relapsed at a higher frequency than those who stayed on antipsychotics, and the drug-placebo difference was not influenced by the time patients were stable on antipsychotics before randomisation (Fig. 2 in Leucht et al Reference Leucht, Tardy, Komossa, Heres, Kissling and Salanti6 ). However, the limitation was, again, that drugs were withdrawn abruptly.
Very long-term follow-up
Takeuchi et als data were restricted to 1 year and ours to 2 years, but the very long-term outcome of schizophrenia is complex. Some examples: according to a methodologically sound systematic review on average only 13.5% of patients achieve a full recovery. Reference Jaaskelainen, Juola, Hirvonen, McGrath, Saha and Isohanni10 In contrast, in both studies from the pre-antipsychotic era Reference Aastrup, Fossum and Holmboe11 and in recent studies a considerable proportion of patients with a first psychotic episode were in remission at follow-up, with wide variability in diagnostic criteria, remission criteria and frequency (for example 32% in Aastrup et al, Reference Aastrup, Fossum and Holmboe11 20% in Robinson et al Reference Robinson, Woerner, Alvir, Bilder, Goldman and Geisler12 and 40% in the AESOP study Reference Morgan, Lappin, Heslin, Donoghue, Lomas and Reininghaus13 ). But we cannot identify them in advance and full (social and symptomatic) recovery over a long period may be rare. Reference Jaaskelainen, Juola, Hirvonen, McGrath, Saha and Isohanni10 Harrow et al Reference Harrow, Jobe and Faull4 reported a 20-year follow-up of a Chicago cohort of 70 patients with schizophrenia, drawn in large part from a psychoanalytic hospital, where affluent, mainly first- or second-episode patients, were admitted to hospital for long-term psychoanalysis. In total, 50% (approximately 7 out of 15) of untreated patients compared with about 15% of treated patients were recovered at 20 years. Patients with a better prognosis might not have needed antipsychotics, thus not taking antipsychotics was not necessarily causal for the better outcome (which epidemiologists call ‘confounding by indication’). As an analogy, people who survived a cancer without treatment might also do better than those who underwent chemotherapy. In contrast, at the 14-year follow-up of a well-designed epidemiological study from a defined catchment area in China more untreated patients had died, had higher unemployment rates and fewer were in remission. Reference Ran, Weng, Chan, Chen, Tang and Lin14 Although there are some predictors of non-relapsing patients, they are not accurate enough to be fully confident of correct identification. We would hope that more data on imaging, cognition and genetics will provide robust criteria in the future.
Brain volume loss
The evidence suggesting that antipsychotics cause brain volume loss in a dose-related fashion is worrying, but equivocal, because it is difficult to disentangle whether this is because of the illness or the medication. Higher doses of medication are required for more severely ill patients. In six macaque monkeys olanzapine and haloperidol led to more brain volume loss than placebo. Reference Dorph-Petersen, Pierri, Perel, Sun, Sampson and Lewis15 In an observational 15-year follow-up study higher antipsychotic doses were associated with more brain volume loss, Reference Ho, Andreasen, Ziebell, Pierson and Magnotta16 but in the same sample there was more brain volume loss in those patients who spent most time in exacerbation of psychosis, which would mean that relapses need to be prevented. Reference Andreasen, Liu, Ziebell, Vora and Ho17 Moreover, the clinical relevance is unclear, because in one study treated patients performed better on cognitive tests than untreated patients despite more brain volume loss. Reference Lesh, Tanase, Geib, Niendam, Yoon and Minzenberg18
The final line of evidence is the open randomised controlled trial by Wunderink and colleagues Reference Wunderink, Nienhuis, Sytema, Slooff, Knegtering and Wiersma19 in which patients in a first-episode in remission were either continued on antipsychotics or were gradually withdrawn. The major difference, in comparison with the classical relapse intervention studies, was that if withdrawal was not possible, medication was reinstituted or the dose re-increased. This strategy is actually quite similar to what many physicians do when following the wishes of their patients. Gradual dose reduction did not prevent relapse in that there were 22% more relapses in the discontinuation group (43%) compared with the continuation group (21%). But most of the 13 patients who could be successfully discontinued initially remained relapse free, and, in the 7-year follow-up patients in the gradual discontinuation group had a significantly better functional outcome. The major limitation was that after the initial 2 years the study was no longer randomised, meaning that patients did not follow the treatment that they had been initially assigned to any longer. Much can have happened in 5 years. This makes it impossible to derive a causal relationship between the initial attempt to withdraw antipsychotics and better functional outcome at 7 years. We find a causal relationship unlikely, in particular because the doses in both groups were similarly low at the end of 2 years (in the attempted withdrawal group 2.8 mg/day haloperidol equivalent and in the maintenance group 2.9 mg/day).
There have been many attempts to stop medication, closely monitor early warning symptoms of relapse and reinstitute treatment rapidly, but these have failed to prevent relapse. Reference Sampson, Mansour, Maayan, Soares-weiser and Adams20 Attempts to identify the minimum effective doses for relapse prevention with dose-response studies have also proven to be difficult. Reference Kane, Davis, Schooler, Marder, Casey and Brauzer21,Reference Wang, Xiang, Cai, Weng, Bo and Zhao22 But the side-effects of antipsychotic drugs such as movement disorders, sedation, hormonal changes and particularly weight gain can produce significant physical comorbidities that can contribute to the documented excess mortality of schizophrenia. Reference McGrath, Saha, Chant and Welham23 Therefore, an approach to identify the patients with a first-episode who do not need maintenance treatment would solve a major problem. But replications are needed, because the stakes are high when doses are down-titrated with relapse as the lower limit. In the meantime, Takeuchi et al's results speak in favour of continuing antipsychotics at the same dose, at least for chronic illness, because the superiority of antipsychotics compared with placebo was large and increased even further over time.