OBJECTIVES: The development of efficient, portable and connected sensors allows for the remote and potentially real-time monitoring of patients, in the context of clinical trials or routine practice. AI methods applied on those data have the potential to detect abnormalities and recognize personalized patterns. Our objective is to review existing AI solutions and their usefulness.

METHODS: A search strategy was implemented in MEDLINE and EMBASE via OVID, based on a previously developed search filter for AI, covering the period since 2020. Titles and abstracts were screened. It was completed by a grey literature search on specialized media, relevant conferences websites and research reports.

RESULTS: The search yields 524 hits and 268 articles were selected after screening. Additionally, 45 publications from grey literature were reviewed. AI-assisted monitoring solutions span over signal processing technics, smart display and consolidation of narratives, and prediction of events. In most published studies in theoretical settings, AI methods outperformed classical approaches, but the applications in real-life are less accurate than expected. They offer the advantages of providing good results in remote settings, with long-term and personalized follow-up, even with off-the-shelf devices, but are limited by the currently modest size of the training datasets and the complexity of medical situations. Accuracy is expected to improve as the quantity, quality and diversity of available data increases.

CONCLUSIONS: Refinement of AI and hardware can help improving clinical measurements and outcomes with comprehensive and real-time monitoring. Several solutions are already giving satisfying results, and there is a large pipeline promising solutions developed in academia still to be further assessed.