Are routinely collected clinical and sociodemographic characteristics associated with social functioning and activities of daily living in schizophrenia? A machine learning approach descriptive of a schizophrenia cohort

Patients with schizophrenia often experience substantial impairments in social functioning and activities of daily living (ADLs). Previous studies have highlighted links between sociodemographic and clinical factors and daily life functioning. However, routinely collected, non-scale-based clinical and sociodemographic characteristics have rarely been systematically evaluated as a standalone predictive framework. Understanding whether such variables alone can inform functional outcomes in schizophrenia may have important implications for public health. Using the French multicentric psychosocial rehabilitation database REHABase, we predicted five dimensions of the Social Autonomy Scale (a validated, clinician-administered questionnaire widely used in France): personal care, basic ADLs, financial autonomy, complex ADLs, and social and affective relationships. We used a SuperLearner ensemble machine learning method with a large set of routinely collected socio-demographic and basic clinical predictors descriptive of schizophrenia patients. Our sample comprised 948 participants. Averaged R2 on hold-out testing sets were higher for basic ADLs (mean R2: 0.35) than for social and affective relationships (0.16), financial autonomy (0.14), complex ADLs (0.13), and personal care (0.01). Factors associated with improved functioning included: being in a relationship, higher education, lower Clinical Global Impression scores, higher Global Assessment of Functioning scores, living in personal housing, being employed, mid-range illness duration, a history of suicide attempts and psychiatric comorbidities. Our findings indicate an association between socio-demographic and standard clinical variables routinely assessed in practice and outcomes in social functioning and ADLs. However, these variables account for only a limited proportion of the observed variance. This underscores the need for more specialized and precise assessments, e.g., based on cognitive abilities, to better understand and address patient functioning. We recommend targeted interventions focused on improving clinical symptoms, housing conditions, and supporting employment. Finally, clinicians should not assume that patients with seemingly protective factors, such as shorter illness duration, or absence of comorbidities, do not require further support.