ARTIFICIAL INTELLIGENCE IN THE MANAGEMENT OF AGING-RELATED DISEASES: CURRENT MODELS, LIMITATIONS, AND FUTURE DIRECTIONS

Abstract

Populations worldwide are aging, with rapid growth in adults aged 65 years and older, particularly those aged 80 years and above. Aging is closely linked to multimorbidity, frailty and polypharmacy, which together create complex clinical profiles that traditional, single-disease models of care and conventional risk scores address poorly. At the same time, digital health infrastructures generate large, heterogeneous datasets (electronic health records, imaging, biosignals, wearable and ambient sensor data, and social determinants) that are well suited to artificial intelligence (AI), which is increasingly explored in geriatric care. We conducted a scoping review to map AI applications in the management of aging-related diseases and outcome prediction. MEDLINE (PubMed), Embase and Scopus were searched for peer-reviewed, English-language empirical studies using AI or machine learning in adults aged ≥60 years, or explicitly focused on older populations, to predict or classify clinically relevant outcomes. Studies limited to younger populations, purely simulated or technical work, and non–full-text reports were excluded. Two reviewers independently screened and extracted data on populations, data sources, model types, targets, performance and validation, followed by narrative synthesis. Most identified applications concerned risk prediction (mortality, hospitalisation, readmission, institutionalisation, frailty progression) using routinely collected clinical data, often enriched with geriatric assessments. Additional use cases included early detection of dementia, frailty and sarcopenia; prediction of treatment response and adverse drug events; remote monitoring and early warning systems; care pathway optimisation; and emerging large language model–based decision support. Across domains, many machine learning models outperformed traditional scores and captured more complex risk patterns, but methodological quality was variable, external validation was infrequent and very old, frail and institutionalised patients were under-represented. Concerns about interpretability, bias, equity, workflow integration and medico-legal responsibility remain prominent. Overall, AI has substantial potential to support more precise, person-centred care for older adults, but realising this promise will require multimorbidity-aware, transparent models, robust evaluation in diverse geriatric populations and governance frameworks that ensure fairness, privacy and meaningful human oversight.

Ähnliche Arbeiten