Predicting future mortality risk in first-episode psychosis: External validation of the MIRACLE-FEP machine learning model

Le résumé fourni par la source

BACKGROUND: Identifying patients with first-episode psychosis (FEP) at high mortality risk may facilitate personalized treatment regimen development and reduce the mortality gap between individuals with psychotic disorders and the general population. Following validation studies in Sweden and Finland, we aimed to externally validate the recently developed mortality risk prediction machine learning model, MIRACLE-FEP, with a new Norwegian national cohort. METHODS: We analyzed a Norwegian national register-based cohort of patients with FEP (N = 4632), with follow-up extending to 8 years. The performance of MIRACLE-FEP was evaluated via the area under the receiver operating characteristic curve (AUROC) and calibration. In addition to all-cause mortality (primary outcome), we assessed the model's discrimination performance for specific causes of death. Finally, we examined whether the model demonstrates any bias toward gender, education level, or immigration status. RESULTS: MIRACLE-FEP demonstrated an AUROC of 0.71 (95% CI 0.63-0.79) for the prediction of 2-year all-cause mortality. Calibration was relatively good, with a calibration slope of 1.04 (95% CI 0.68-1.47) and a calibration-in-the-large value of 0.31 (-1.93-1.26). Among specific causes of death, the model showed the highest discrimination for deaths due to accidents (AUROC 0.86, 95% CI 0.79-0.93) and the lowest for suicide (AUROC 0.47, 95% CI 0.35-0.59). No evidence of bias was observed in discrimination accuracy by gender, education level, or immigration status. CONCLUSIONS: The performance metrics of this external validation study aligned with those reported in the development study. Efforts to enhance MIRACLE-FEP's performance in suicide prediction are needed.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.