Relapse in clinically stable adult patients with schizophrenia or schizoaffective disorder: evidence-based criteria derived by equipercentile linking and diagnostic test accuracy meta-analysis

Background: There is no consensus on defining relapse in schizophrenia, and scale-derived criteria with unclear clinical relevance are widely used. We aimed to develop an evidence-based scale-derived set of criteria to define relapse in patients with schizophrenia or schizoaffective disorder. Methods: We searched the Yale University Open Data Access (YODA) for randomised controlled trials (RCTs) in clinically stable adults with schizophrenia or schizoaffective disorder, and obtained individual participant data on Positive and Negative Syndrome Scale (PANSS), Clinical Global Impression Severity (CGI-S), Personal and Social Performance (PSP), and Social and Occupational Functioning Assessment Scale (SOFAS). Our main outcomes were PANSS-derived criteria based on worsening in PANSS total score. We examined their relevance using equipercentile linking with CGI-S and functioning scales, and their test-performance in defining relapse with diagnostic test accuracy meta-analysis against CGI-S worsening (≥1-point increase together with a score ≥4 points) and psychiatric hospitalisation. Findings: Based on data from seven RCTs (2354 participants; 1348 men [57·3%] and 1006 women [42·7%], mean age of 39·5 years [SD 12·0, range 17–89]; 303 Asian [12.9%], 255 Black [10.8%], 1665 White [70.7%], and other or unspecified 131 [5.6%]), an increase of 12 points or more in PANSS total (range 30–210 points) corresponded to clinically important deterioration in global severity of illness (≥1 point increase in CGI-S, range 1–7) and functioning (≥10 points decline in PSP or SOFAS, range 1–100). The interpretation of percentage changes varied importantly across different baseline scores. An increase of 12 points or more in PANSS total had good sensitivity and specificity using CGI-S as reference standard (sensitivity 82·1% [95% CI 77·1–86·4], specificity 86·9% [82·9–90·3]), as well as good sensitivity but lower specificity compared to hospitalisation (sensitivity 81·7% [74·1–87·7], specificity 69·2% [60·5–76·9]). Requiring either an increase in PANSS total or in specific items for positive and disorganization symptoms further improved test-performance. Cutoffs for situations where high sensitivity or specificity is needed are presented. Interpretation: An increase of either 12 points or more in the PANSS total score, or worsening of specific positive and disorganisation symptom items could be a reasonable evidence-based definition of relapse in schizophrenia, potentially linking symptoms used to define remission and relapse. Percentage changes should not be used to define relapse because their interpretation depends on baseline scores. Funding: German Research Foundation (grant number: 428509362).