Adoption of Machine Learning in US Hospital Electronic Health Record Systems: Retrospective Observational Study

Comprehensive Summary

Huang et al. conducted a retrospective observational study to assess the landscape of machine learning (ML) adoption in electronic health records (EHRs) across US hospitals and determine characteristics associated with adoption using the Technology-Organization-Environment (TOE) framework. Data were collected from 2,562 unique hospitals through the 2022‐2023 American Hospital Association (AHA) Annual Survey and the 2023‐2024 AHA IT Supplement Survey. The surveys sampled US acute care and general hospitals across all 50 states and the District of Columbia. Respondents were asked to select uses of ML. Adoption was then categorized as no ML adoption, clinical adoption, operational adoption, or adoption of both. Bed size, hospital ownership, teaching hospital status, and critical access hospital (CAH) status provided organizational context. Metropolitan location, contract with leading EHR vendor, and health system affiliation provided environmental context. Due to the lack of technical content in the survey, technological context was omitted. Among the 2023-2024 sample, 73-76% of hospitals reported ML adoption in any capacity, with the majority of US hospitals implementing both clinical and operational ML. Adoption rates were higher among large hospitals (94.4% vs 65.0% for small), not-for-profit hospitals (82.7% vs 69.8% for for-profit), teaching hospitals (92.1% vs 56.7% for non-teaching), non-CAHs (82.2% vs 57.4% for CAHs), metropolitan hospitals (83.8% vs 61.3% of non-metropolitan), those contracted with leading EHR vendors (92.1% vs 56.7% for non-leading EHR), and those affiliated with health systems (87.9% vs 40.1% for non-health-system hospitals), with all comparisons reaching statistical significance (p<0.001). The most common ML uses were predicting inpatient risk or health trajectories and identifying high-risk patients for follow-up care.

Outcomes and Implications

Previous literature has focused on the adoption of ML in the hospital setting, which presents this technology as a homogenous feature and obscures variations in implementation. This study provides the organizational and environmental context surrounding ML adoption, which is needed to promote equitable and effective ML application across diverse health care settings. The results suggest that the unequal distribution of resources in rural or underserved areas poses a significant barrier to ML adoption, which may be mitigated by external support. The federal government could introduce similar incentives from the 2009 Health Information Technology for Economic and Clinical Health Act, which accelerated meaningful EHR use in rural and underserved hospitals. Important limitations include self-report bias due to reliance on secondary data and potential non-response bias, as 44.7% of the AHA hospitals did not respond to the IT Supplement survey.