AI-Based Measurement Tools for Intrinsic Capacity in Older Adults: Scoping Review.

چکیده اصلی

BACKGROUND: Intrinsic capacity (IC) has become a central concept in healthy aging because it emphasizes functional ability across the aging trajectory rather than disease alone. However, current IC assessment primarily relies on episodic clinical evaluations, which are insufficient for continuous monitoring and early identification of functional decline. Recent advances in AI and digital health technologies have created new opportunities for objective, continuous, and real-world assessment of IC. However, existing evidence remains fragmented across AI-enabled devices, digital biomarkers (DBs), AI techniques, and IC domains. OBJECTIVE: This scoping review aimed to systematically synthesize the current evidence on AI-based measurement tools for IC and to characterize the landscape of AI-enabled IC assessment using a 3D analytical framework integrating AI-enabled digital devices and systems, DBs, and AI techniques. METHODS: A comprehensive search of PubMed, Embase, CINAHL, PsycINFO, the Cochrane Library, SinoMed, and China National Knowledge Infrastructure (CNKI) was conducted from database inception to July 2025 and updated on May 31, 2026, in accordance with the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) guideline. Studies investigating AI-based measurement tools applicable to one or more IC domains were included. RESULTS: A total of 161 studies met the inclusion criteria. Research on AI-based measurement tools for IC has expanded rapidly since 2016, with studies conducted in 28 countries, predominantly the United States and China. Most studies focused on a single IC domain, with cognition accounting for the largest proportion. Eleven categories of AI-enabled digital devices and systems were identified, among which multimodal data acquisition devices, computer vision (CV) systems, and AI-driven health platforms were the most frequently reported. Twenty-one types of DBs were extracted and classified into 3 major categories, with gait parameters, digital task performance, physical activity features, speech and language features, and facial features representing the most commonly used biomarkers. Machine learning and deep learning were the predominant AI techniques, while CV and natural language processing played central roles in multimodal data interpretation. The distribution and maturity of evidence varied substantially across domains, with cognition and locomotor capacity representing the most developed areas, whereas vitality, hearing, and multidomain IC assessment remained comparatively underrepresented. CONCLUSIONS: This scoping review provides a 3D synthesis of AI-enabled digital devices and systems, DBs, and AI techniques across the 6 World Health Organization (WHO)-defined domains of IC. Unlike previous technology-, disease-, or domain-specific reviews, it compares evidence across the broader IC framework, identifying more developed areas, key evidence gaps, and priorities for standardization, external validation, and multidomain assessment. AI-based measurement tools may complement conventional assessment in community, primary care, and home settings, although their clinical translation will require robust validation, integration into care pathways, and implementation approaches that address the needs of older adults.

ترجمه فارسی چکیده

ظرفیت درونی (IC) مفهومی کلیدی در پیری سالم است که بر توانایی عملکردی در طول مسیر پیری تأکید دارد. ارزیابی فعلی IC عمدتاً بر اساس ارزیابی‌های بالینی پراکنده است که برای پایش مداوم و شناسایی زودهنگام افت عملکرد کافی نیستند. پیشرفت‌های اخیر در هوش مصنوعی و فناوری‌های سلامت دیجیتال فرصت‌های جدیدی برای ارزیابی هدفمند، مداوم و واقعی ارائه می‌دهند. این مرور جامع با استفاده از چارچوب تحلیلی سه‌بعدی، شواهد موجود در مورد ابزارهای اندازه‌گیری مبتنی بر هوش مصنوعی را در حوزه‌های دستگاه‌های دیجیتال، نشانه‌های دیجیتال و تکنیک‌های هوش مصنوعی جمع‌آوری و طبقه‌بندی می‌کند. نتایج نشان می‌دهد که ۱۶۱ مطالعه شامل شدند و اکثر آن‌ها بر یک حوزه IC تمرکز داشتند. تکنیک‌های یادگیری ماشین و یادگیری عمیق غالب بودند.

متن کامل اصلی