One argument against ever reducing antipsychotics is that stopping causes a distinct, rapid, pharmacologically driven relapse — supersensitivity psychosis — different in kind from the illness coming back. If true, that would mean discontinuation carries a risk over and above the underlying condition. This individual participant data meta-analysis tested it directly, using five randomised double-blind placebo-controlled discontinuation trials of oral and long-acting injectable paliperidone from the Yale Open Data Access database: 271 participants from the injectable trials and 146 from the oral trials, all stabilised on treatment for more than three months before randomisation.
Latent class modelling of symptom trajectories before relapse identified two distinct patterns: rapid and delayed onset, with rapid relapse carrying more severe symptoms at the point of relapse. But rapid relapse was not more common among those who stopped. In the injectable trials it occurred in 39 of 197 (20%) who discontinued against 8 of 74 (11%) who continued, p=0.12; in the oral trials, 29 of 108 (27%) against 10 of 38 (26%), p=0.95. Symptom profiles at relapse did not differ by discontinuation status either.
What did predict rapid relapse was baseline severity: higher PANSS scores at baseline in both formulations, p<0.001.
This reframes the de-prescribing conversation in a useful way. Discontinuation does increase relapse risk — that is not in question and these were relapse-prevention trials. What it apparently does not do is cause a separate, drug-specific kind of rapid relapse. So the honest framing for a patient asking about reducing medication is that stopping raises the chance of the illness returning, and that if it returns quickly, that reflects their illness rather than a withdrawal phenomenon.
The practical action is risk stratification. Someone with higher baseline symptom severity is the person who relapses fast, and is therefore the person for whom a reduction attempt needs closer monitoring, a slower taper, and a clear relapse plan agreed in advance with the family. Someone who has been stable at low symptom levels for a long period is a different proposition. Two caveats: this covers paliperidone only, and the authors themselves note that low baseline scores may partly be an artefact of trials minimising severity at screening to meet entry criteria.
- Stratify de-prescribing by baseline symptom severity — that predicted rapid relapse, discontinuation status did not.
- Tell patients honestly that stopping raises relapse risk, but not that it causes a distinct withdrawal psychosis.
- Agree a written relapse plan with the patient and family before any reduction, with defined early signs.
- Monitor most closely in the first weeks after a reduction, when rapid relapses occurred.
- This is paliperidone data; do not assume it generalises to clozapine or to abrupt cessation of other agents.
The statistics, in plain English
This is an analysis of relapse patterns among people who relapsed, not of who relapses — so it says nothing about whether discontinuation increases relapse overall, which the parent trials established that it does. Latent class modelling is a statistical technique that sorts trajectories into groups without being told what the groups are; how many classes emerge depends partly on modelling choices, so 'two trajectories' is a description of these data rather than a discovery of two biological types. The comparisons rest on small numbers — 8 rapid relapses among 74 continuing patients in the injectable trials — so a p value of 0.12 means no difference was detected, not that none exists.
Read the rest in the app
You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for psychiatry, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free