Dominic Oliver, Maite Arribas, Yanakan Logeswaran, Laura Fusar-Poli, Matthew J Kempton, Emily P Hedges, Clara Albinaña, Evangelos Vassos, Amir Sariaslan, William Pettersson-Yeo, Lucia Valmaggia, Mark van der Gaag, Lieuwe de Haan, Barnaby Nelson, Patrick McGorry, Anita Riecher-Rössler, Erich Studerus, Rodrigo Bressan, Neus Barrantes-Vidal, Marie-Odile Krebs, Merete Nordentoft, Stephan Ruhrmann, Gabriele Sachs, Bart Rutten, Jim van Os, Daniel Stahl, Philip McGuire, Paolo Fusar-Poli, EU-GEI High Risk Study Group
CHR-P detection may be improved by using the PPS, though evidence is needed from more representative detection settings. However, we did not find evidence that baseline clinical, environmental and/or genetic data enhanced the prediction of psychosis onset.
BACKGROUND AND HYPOTHESIS: Psychosis prevention can be supported by tools that facilitate the early detection of individuals at clinical high risk (CHR-P) and predicting their clinical outcomes. We aimed to use clinical, genetic, and environmental data to: (1) predict case-control status as a proof-of-concept for CHR-P detection, and (2) predict transition to psychosis.
STUDY DESIGN: We used data from the European Network of National Schizophrenia Networks Studying Gene-Environment Interactions: a multicenter cohort study comprising 344 CHR-P individuals and 67 healthy controls. To predict CHR-P status, we used environmental (Psychosis Polyrisk Score [PPS]) and genetic (polygenic risk score for schizophrenia [PRS]) measures with logistic regression (LR) and random forest (RF). To predict transition to psychosis, we used clinical, environmental, and genetic measures with Cox proportional hazards model and random survival forest. Primary outcomes were discrimination (C-index) and calibration (intercept and slope) in repeated nested cross-validation. Clinical utility was assessed with decision curve analysis.
STUDY RESULTS: For detection of CHR-P, both PPS (LR:C = 0.91, 95% CI, 0.87-0.94, intercept = 1.46, slope = 0.73; RF:C = 0.77, 95% CI, 0.61-0.90, intercept = -0.53, slope = 0.42) and PPS + PRS (LR:C = 0.89, 95% CI, 0.85-0.92, intercept = 1.45, slope = 0.73; RF:C = 0.78, 95% CI, 0.65-0.89, intercept = -0.24, slope = 0.44) had excellent discrimination performance, whereas PRS performed substantially worse (LR:C = 0.60, 95% CI, 0.52-0.67, intercept = 1.63, slope = 0.81; RF:C = 0.57, 95% CI, 0.41-0.72, intercept = -0.84, slope = 0.10). All models over-estimated risk in individuals with low observed risk. Prognosis model performance was poor (C ≤ 0.65).
CONCLUSIONS: CHR-P detection may be improved by using the PPS, though evidence is needed from more representative detection settings. However, we did not find evidence that baseline clinical, environmental and/or genetic data enhanced the prediction of psychosis onset.