Diagnosis and Management of Acute and Chronic Lithium-Associated Nephrotoxicity.

Addiction – encompassing substance use and behavioral disorders – remains a major global health challenge characterized by chronic relapse, premature mortality, and substantial social and economic burden [, ]. Although pharmacological and psychosocial interventions have advanced in recent decades, treatment retention, adherence, and relapse prevention continue to be suboptimal across settings []. A large body of evidence shows that outcomes are strongly shaped by psychosocial and structural determinants, including housing instability, employment status, psychiatric comorbidity, legal involvement, and neighborhood disadvantage [, ]. Artificial intelligence (AI) and machine learning (ML) have been increasingly applied in addiction research with the aim of improving risk stratification, anticipating relapse, and supporting more individualized treatment planning [, ]. Studies using electronic health records (EHRs), administrative databases, digital phenotyping, and ecological momentary assessment (EMA) have demonstrated that ML models can outperform traditional statistical methods for predicting treatment dropout, discontinuation of medications for opioid use disorder (MOUD), or relapse events [–]. Importantly, several recent investigations show that predictive performance improves when models incorporate variables reflecting patients’ everyday lives – care continuity, symptom fluctuations, stress, craving, sleep disruption, or neighborhood-level disadvantage – rather than relying exclusively on medication-related, diagnostic, or routinely available clinical variables [–]. These psychosocial dimensions – motivation to change, coping capacity, stability of social support, and the quality of the therapeutic relationship – are well-established drivers of engagement and recovery across addiction care settings []. Decades of evidence show that strong therapeutic alliance, higher motivation, and robust coping skills predict treatment retention and lower relapse risk [, ]. Although rarely captured in structured clinical data, these factors substantially influence whether individuals remain in treatment or disengage after relapse. Incorporating such dimensions into AI models is therefore essential not only for predictive accuracy but also for ensuring clinical relevance and alignment with person-centered care [].