BACKGROUND: Antipsychotics reduce risk of relapse in schizophrenia, but longer-term use is associated with adverse effects, often prompting dose reduction or discontinuation. Although discontinuation increases relapse risk, the clinical course is heterogeneous. Rapid relapse following antipsychotic discontinuation has been observed in clinical trials and could reflect a direct pharmacological consequence of stopping medication, as opposed to re-emergence of underlying illness. We aimed to identify distinct trajectories of symptom change preceding relapse, to examine whether trajectory membership was associated with treatment arm and baseline clinical characteristics, and to compare symptom profiles at time of relapse between individuals whose illness relapsed following discontinuation and those whose illness relapsed during continued treatment. METHODS: For this meta-analysis, on May 4, 2025, we searched the Yale University Open Data Access project database for randomised, double-blind, discontinuation trials of antipsychotics in individuals with schizophrenia or schizoaffective disorder. No language restrictions were applied. We included double-blind, placebo-controlled, relapse-prevention, randomised trials of antipsychotic medications. Studies with available individual participant data were included if participants were first treated with antipsychotic drugs for more than 3 months and clinically stabilised before they were randomly assigned to placebo (discontinuation) or continued active treatment. We excluded studies that involved diagnoses other than schizophrenia or schizoaffective disorder. We analysed longitudinal symptom data (using the Positive and Negative Syndrome Scale [PANSS]) from individuals whose illness relapsed. Latent class mixed modelling identified distinct trajectories of symptom change preceding relapse. We first examined associations between trajectory membership and treatment discontinuation or continuation. We then compared symptom profiles at both baseline and point of relapse, between both trajectory groups and discontinuation status. The study was not preregistered; the Yale University Open Data Access project ID is 2025-0740. FINDINGS: We identified five randomised, double-blind, placebo-controlled, discontinuation trials of oral and long-acting injectable (LAI) paliperidone. In our analysis, we included 271 participants from the LAI trials (155 men [57%] and 116 women [43%], mean age 38·4 years [SD 11·2, range 18-65]) and 146 from the oral trials (72 men [49%] and 74 women [51%], mean age 34·8 years [SD 11·5, range 18-61]). Two latent classes of relapse were identified: rapid and delayed onset, with rapid relapse associated with more severe symptoms at point of relapse. The proportion with rapid relapse did not differ between continuation and discontinuation groups in either LAI (discontinuation 39 [20%] of 197, continuation eight [11%] of 74; p=0·12) or oral trials (discontinuation 29 [27%] of 108, continuation ten [26%] of 38, p=0·95). Symptom profiles at relapse did not differ by discontinuation status. Across both formulations, those with rapid relapse had significantly higher baseline PANSS scores than those with delayed relapse (p<0·0010). INTERPRETATION: Relapse following paliperidone discontinuation follows two distinct trajectories, rapid and delayed, the former being related to baseline severity. We found that rapid relapse was not over-represented among those who discontinued treatment, which is not the pattern expected if rapid relapse following paliperidone discontinuation were commonly a pharmacological discontinuation effect. Higher baseline severity in rapid relapse could reflect pre-existing susceptibility, or trial-related measurement artifact whereby symptom severity is minimised at screening to meet study entry thresholds. Our findings underscore the importance of risk stratification and individualised monitoring during antipsychotic de-prescribing. FUNDING: National Institute of Health and Care Research Oxford Biomedical Research Centre and the Wellcome Trust.