Artificial Intelligence (AI)-Initiated Clinical Decision Support in Hospitalist Practice

This is a three-arm randomized feasibility trial evaluating AI-initiated clinical decision support in routine inpatient hospital medicine practice. Eligible patient cases cared for by participating hospitalist physicians will be randomized to one of three AI prompting strategies: automated electronic health record (EHR) data retrieval, automated EHR data retrieval with explicit inclusion of the admission history and physical (H\&P) note, or the admission H\&P note alone. OpenEvidence will generate a clinical decision support recommendation for each randomized case.

The recommendation will be delivered to the attending hospitalist, who will then complete a brief electronic survey assessing the clinical utility and interpretation of the recommendation and whether it was implemented or planned for implementation. The study will assess the feasibility of generating and delivering AI-initiated recommendations during routine clinical care, physician survey completion, timely survey completion, and recommendation adoption. Findings from this feasibility trial will inform the design of a future larger randomized trial.

• AI Clinical Decision Support: Automated EHR Data Retrieval (Other): OpenEvidence generates an AI-based clinical decision support recommendation using automated retrieval of available EHR data.
• AI Clinical Decision Support: Admission H&P Note (Other): The admission H\&P note is explicitly provided to OpenEvidence as clinical context for generation of the AI-based clinical decision support recommendation.

Primary outcomes

• Successful AI Recommendation Delivery Rate [Within 24 hours of hospital admission]: Proportion of randomized patient cases in which an AI-generated recommendation is successfully delivered to the attending hospitalist.
• Physician Survey Completion Rate [Within 48 hours after AI recommendation delivery]: Proportion of successfully delivered AI-generated recommendations for which the attending hospitalist completes the electronic physician survey.
• Timely Physician Survey Completion Rate [Within 12 hours of AI recommendation delivery]: Proportion of successfully delivered AI-generated recommendations for which the attending hospitalist completes the electronic physician research survey within 12 hours of recommendation delivery.
• Recommendation Adoption Rate [Within 48 hours after AI recommendation delivery]: Proportion of AI-generated recommendations that the attending hospitalist reports as implemented or planned for implementation. Responses will be recorded separately as implemented, planned but not yet implemented, or not adopted.

Secondary outcomes

• Physician-Perceived Clinical Utility [Within 48 hours after AI recommendation delivery]: Physician-perceived clinical utility of the AI-generated recommendation, assessed at the patient-case level using a single-item 0-10 numeric rating scale, with anchors at 0 ("Not at all useful"), 5 ("Moderately useful"), and 10 ("Extremely useful").
• Interpretation of AI-Generated Recommendations [Within 48 hours after AI recommendation delivery]: Physician interpretation of the AI-generated recommendation, assessed using six predefined response categories in the physician research survey.
• Physician-Reported Recommendation Adoption Status [Within 48 hours after AI recommendation delivery]: Physician-reported adoption of the AI-generated recommendation, categorized as implemented, planned but not yet implemented, or not adopted.

Sex: All, ages: 18 Years to any, accepts healthy volunteers

Inclusion Criteria:

* Age 18 years or older.
* Attending physician providing inpatient hospital medicine care at a participating Cedars-Sinai Health System site.
* Expected to provide clinical care during the study period.
* Willing and able to provide informed consent.

Exclusion Criteria:

* Declines to provide or withdraws informed consent.
* Not actively providing attending-level inpatient care on a participating hospitalist service during the study period.
* Unable to complete study procedures.

• Joshua M Pevnick, MD, Principal investigator, Cedars-Sinai Medical Center