Artificial intelligence applications in delirium prediction, diagnosis, and management: a systematic review

Delirium is a prevalent, acute, and reversible neuropsychiatric syndrome in elderly populations, notable for its high incidence and mortality rates, both of which impose a substantial burden on patient outcomes and healthcare systems. Currently, the detection of delirium primarily depends on clinical assessments performed by physicians, which poses challenges for less experienced clinicians in early identification of the condition. The application of artificial intelligence (AI) technologies to analyze large-scale data from delirium patients enables the identification and quantification of relevant delirium markers, thereby effectively assisting clinicians in the diagnosis, prediction, and monitoring of patient status. However, existing reviews on AI in the field of delirium exhibit several limitations, particularly regarding the comprehensive classification of existing studies. Current reviews often focus on a single type of data and often lack a systematic analysis of studies by data type. Most clinical models for delirium rely on electronic medical records, though physiological time-series data and imaging features also offer crucial biomarkers for the identification of delirium. This paper offers an overview of the medical foundation and recent technological advancements in delirium, aiming to establish a theoretical framework for novices. It systematically reviews the prediction, diagnosis, and management of delirium from a multi-source data perspective, enumerates relevant data sources and public databases, and examines various AI models used in this field, along with their advantages and limitations. Finally, this review addresses the challenges and emerging trends of AI in delirium, with the aim of elucidating key directions for future research and development.