AI for Delirium Care: Promise and Implementation Gaps

Delirium affects up to one-third of hospitalized older adults and is associated with adverse outcomes. This review considers how artificial intelligence (AI) might address gaps in risk stratification, detection, documentation, intervention delivery and monitoring. It distinguishes model development and validation from real-world implementation and emerging concepts; the abstract does not report a systematic review method or a study sample size.

Prediction models using structured health-record data and clinical notes have reported good discrimination, but few have been integrated into practice. A live multimodal machine-learning model increased targeted delirium detection; however, a pre-post comparative study found longer hospital stays. Natural language processing, large language models, computer vision, ambient sensing and wearables may enable less labor-intensive assessment, but evidence is limited and concerns include false positives and poor generalizability. AI-supported management approaches such as reinforcement learning and digital twins remain largely conceptual.

The authors frame AI as an adjunct to, not a replacement for, clinical assessment. They call for prospective comparative studies measuring patient outcomes, resource use and balancing measures. Responsible implementation also requires external validation, clear action pathways, equitable performance, privacy protections, clinician oversight and post-deployment monitoring.