Use of Conventional Artificial Intelligence Methods in the Identification of Frailty: A Scoping Review

Early identification and management of frailty are crucial, yet its detection in early stages remains difficult for clinicians. Artificial intelligence (AI) has emerged as a promising tool in healthcare. However, the absence of a standard frailty definition and diversity of AI methods create a need for a comprehensive review. This study examines the clinical tools and conceptual frameworks used as reference standards in training AI algorithms for frailty identification and management, describes current AI methods, and explores the engagement of knowledge users in developing and evaluating these technologies.

A scoping review was conducted following the Arksey and O'Malley framework, enhanced by Levac et al. and the Joanna Briggs Institute. Eight academic databases-Medline, Embase, PsycInfo, Cumulative Index to Nursing and Allied Health Literature, Ageline, Web of Science, Scopus, and Institute of Electrical and Electronics Engineers Xplore-and one gray literature source-ProQuest Dissertations & Theses Global-were searched. Abstracts and full-text screening and data charting were performed in duplicate. Results were summarized through text and graphical representations.

The review included 33 publications, predominantly emerging after 2020. Twenty-three different AI techniques were presented, with standard modeling approaches such as logistic regression and decision trees being most common. Among the 21 distinct reference standards used to train AI models, the Physical Frailty Phenotype was cited most frequently (n = 7). Most AI methods (n = 27) prioritized frailty identification, one addressed frailty management, and five focused on both. None of the papers engaged knowledge users in defining or validating AI tools, and only three studies explored algorithmic biases that could lead to inequities.

Like the broader frailty literature, emerging AI tools lack a consistent definition of frailty, leading to design and implementation inconsistencies. The absence of knowledge user involvement may further limit the clinical relevance and equity of these technologies.

OSF Registries [https://doi.org/10.17605/OSF.IO/T54G8].