Artificial intelligence in pain assessment and management for older adults: A scoping review

Conventional pain management strategies often fall short in addressing the complex needs of older adults with pain. Artificial intelligence (AI) represents a significant potential for advancing personalized, data-driven approaches. This review explores the literature on AI in pain management for older adults, showing trends and existing gaps. Following the PRISMA-ScR guidelines, five databases were searched using terms related to older adults, pain management, and artificial intelligence in peer-reviewed articles published between 2014 and 2025. Included studies were synthesized and reported using descriptive and narrative analyses. A total of 96 studies were included. Machine learning was the most common AI method used (71, 74%), followed by deep learning (14, 14.6%), robotics (7, 7.3%), natural language processing (3, 3%), and rule-based systems (1, 1%). AI was primarily used for pain prediction (40, 41.7%) and classification (33, 34.4%), with fewer studies focusing on pain detection (6, 6.3%) or treatment optimisation (17, 17.7%). Input data types included clinical records (29, 30.2%), facial analysis (18, 18.8%), imaging (7, 7.3%), and video sequences (4, 4.2%). The most frequently studied pain types were chronic secondary musculoskeletal pain (30, 31%), chronic primary pain (13, 13.8%), and visceral pain (11, 11.7%). AI applications for pain management in older adults are growing rapidly, with machine learning dominating prediction and classification tasks. Despite technical progress, most tools remain early-stage, rely on limited datasets, and lack aging-specific dynamics. Future research should prioritize diverse cohorts and inclusive algorithms to ensure these innovations translate into targeted pain care for older adults. • AI in pain management for older adults is expanding, but it remains in an early-stage and under-validated phase. • Machine learning dominates prediction and classification, with few tools reaching clinical readiness. • Factors like frailty, polypharmacy, and biopsychosocial influences, which are specific to older adults, are rarely integrated into models. • Ethical gaps persist with limited co-design, cultural validation, and bias mitigation strategies. • Small datasets, workflow barriers, and digital exclusion hinder real-world adoption.