Mapping personal recovery in schizophrenia spectrum disorders: an exploratory machine learning study of self-reported stage classifications

Le résumé fourni par la source

Recovery in schizophrenia is a complex and multidimensional process involving functional, clinical, and psychological dimensions. This study aims to use machine learning (ML) to explore multivariate associations with classifications of personal recovery assessed with the Stages of Recovery Instrument (STORI). We analyzed cross-sectional data from 1,361 individuals with schizophrenia-spectrum disorders enrolled in the French REHABase cohort, a national psychosocial rehabilitation network. Random forest models were developed using internal cross-validation on a training set and subsequently evaluated on an independent test set (70/30 split). Model performance was high in the training data (AUC = 0.997; accuracy = 94.3%) and more modest in the test set (AUC = 0.724; accuracy = 49.8%). SHAP analyses (explainable ML) were used to describe stage-associated recovery profiles based on multivariate patterns across sociodemographic, clinical, and psychological variables. Overall, the modest performance observed in the independent test set underscores the complexity of personal recovery and is consistent with partial conceptual overlap across STORI stage classifications. The Moratorium stage was characterized by higher negative self-esteem and internalized stigma, alongside lower wellbeing and positive self-esteem. Awareness was associated with mental wellbeing and negative self-esteem, with additional contributions from positive self-esteem and relational satisfaction. The Preparation stage showed less distinct profiles, with autonomy emerging as a relevant feature. Rebuilding was associated with intermediate-to-high levels of wellbeing and resilience. The Growth stage was characterized by high wellbeing, positive self-esteem, and relational satisfaction. These findings provide an integrated description of stage-associated recovery profiles and illustrate how explainable ML can be used to explore the multidimensional organization of personal recovery.

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