Evaluating machine learning-based polypharmacy risk prediction in multigenerational households under family medicine, internal and pediatric care interfaces

Polypharmacy commonly defined as the concurrent use of five or more medications presents significant clinical risks, especially in multigenerational households where pediatric, adult, and geriatric care intersect. With increasing medication burdens and comorbidities, traditional methods of medication review and reconciliation are insufficient for timely and accurate risk stratification. This study evaluates the utility of machine learning (ML) algorithms in predicting polypharmacy-associated risks across diverse patient cohorts within shared household contexts. Drawing on anonymized electronic health records (EHRs) from family medicine, internal medicine, and pediatric care units, we developed and validated ensemble-based models that integrated medication histories, diagnostic codes, socioeconomic indicators, and household composition data. Our models achieved strong predictive performance, with area under the ROC curve (AUC) values exceeding 0.87 across age-stratified subgroups. We specifically examined the performance of random forests, gradient boosting machines, and neural networks in identifying medication interaction risks, hospitalization likelihoods, and early warning signs of adverse drug events (ADEs). Multigenerational dynamics such as caregiver stress, medication sharing, and uncoordinated prescribing were found to significantly influence risk scores. Pediatric risks were often underestimated in conventional screening tools, while elder populations showed higher susceptibility to anticholinergic burden and cumulative sedative effects. Our results highlight the importance of incorporating familial and generational context into predictive healthcare models. ML-based polypharmacy risk stratification can augment care coordination across departments and improve anticipatory interventions, especially in under-resourced.