Frailty — the syndrome of accumulated physiological decline that predicts hospitalization, disability, and death in older adults — has long resisted early clinical detection. If AI can reliably flag frailty before it becomes clinically obvious, it could shift geriatric care from reactive to preventive, potentially altering outcomes for millions of aging adults worldwide. That promise, however, hinges on methodological coherence that this scoping review reveals is still largely absent.
Drawing on 33 publications identified across eight major academic databases, this review maps the landscape of conventional AI applied to frailty identification. The field is young — most relevant publications emerged after 2020 — yet already fractured: researchers deployed 23 distinct AI techniques, ranging from classical approaches like logistic regression and decision trees to more sophisticated ensemble and machine learning models. Equally problematic is the training foundation: 21 different reference standards were used across studies, meaning each algorithm was essentially learning a different version of what frailty is. Without consensus on the ground-truth definition — whether phenotypic frailty, accumulation-of-deficits index, or clinical judgment — cross-study comparisons become nearly meaningless.
This fragmentation reflects a deeper unresolved tension in geriatrics: frailty itself lacks a single universally accepted definition. The review's finding that knowledge users — clinicians, patients, caregivers — were minimally engaged in developing and evaluating these AI tools compounds the concern. Algorithmic frailty detection trained without clinical input risks optimizing for measurable proxies rather than meaningful patient outcomes. For the field to advance, standardization on both the outcome definition and AI methodology is essential. This review is best understood as an incremental but necessary cartographic exercise — useful for identifying where the field is, and more importantly, where critical infrastructure is still missing before AI frailty tools could responsibly enter clinical practice.
