Identification of clinical and psychological factors associated with resilience in patients with schizophrenia using machine learning

This talk will focus on resilience in schizophrenia, conceptualized as the capacity to adapt one’s thinking when facing stress, and will shed light on which factors contribute to higher resilience levels in patients. Using data from participants in the Italian Network for Research on Psychoses (598 in a discovery sample and 298 in a validation sample), we analyzed a wide range of historical, clinical, psychological, and environmental variables. Based on these assessments, we trained a supervised machine learning algorithm (Support Vector Machine) on 85 features to distinguish individuals with high versus low resilience, applying a rigorous double-cycle, nested cross-validation approach and validating results in an unseen sample. The algorithm achieved strong classification performance, with balanced accuracy of 74.5% in the discovery sample and 80.2% in validation sample. Key factors driving the model’s predictions included greater self-esteem, broader social networks, and more adaptive coping strategies. Additionally, algolrithmic decisions were significantly associated with patients’ cognitive functioning and symptom severity. Overall, findings highlighted a robust and generalizable profile of potential residual resilience abilities in schizophrenia patients, pointing to specific psychological and clinical domains that could be targeted through non-pharmacological interventions to improve patient outcomes.