OBJECTIVES: Manual chart review is a traditional method of data collection in which trained individuals abstract clinical data from electronic or paper charts into an electronic case-report form. This can be a laborious and costly method. We assessed use cases of artificial intelligence (AI) to support chart review studies.

METHODS: A scoping review was performed of literature between 01 January 2020 and 17 June 2024. We aimed to identify original research describing the use of AI to support elements of chart review. A search was conducted in PubMed to find potentially relevant citations. Both authors reviewed an equal number of abstracts. Inter-rater reliability between the two authors was not formally assessed; however, they discussed their assessments until a consensus was reached.

RESULTS: 363 citations were identified, of which 85 described original research of use cases of AI to support chart review research. These abstracts reported using AI to automate (a) data collection from patient charts (41%, n=35), (b) case identification (39%, n=33), and (c) case ascertainment (20%, n=17). The majority (76%, 65/85) reported using a natural language processing (NLP) model, either alone or in combination with other AI approaches such as machine learning (ML) or optical character recognition (OCR); the remaining 24% (20/85) reported using an ML method. The most common therapeutic area was oncology with 24% of abstracts (20/85), followed by neurology with 11% of abstracts (9/85).

CONCLUSIONS: AI models are being applied to support different elements of chart review research. While still in early stages, the use of AI has potential to support certain chart review activities. The results of the review align with the overall trend of AI usage in healthcare.