AI pre-consultation support in outpatient clinics

How to cite

Chitla, Sandeep; Cui, Yao; Li, Qi. 2025. "AI pre-consultation support in outpatient clinics." AEA RCT Registry. AEARCTR-0017494. https://doi.org/10.1257/rct.17494-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Sandeep Chitla* , Yao Cui , Qi Li

* Primary investigator

Registry ID
AEARCTR-0017494
DOI
https://doi.org/10.1257/rct.17494-1.0
First registered
Last updated
Status
On going
Trial dates
2025-12-15 – 2026-04-03
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/17494

Abstract

Artificial intelligence (AI) systems, particularly large language model (LLM)-based applications, are increasingly integrated into frontline healthcare workflows to facilitate documentation and clinical decision-making. However, empirical evidence regarding their impact on operational efficiency and the quality of medical decision-making in real-world environments remains scarce. This study examines the implementation of an AI-assisted pre-consultation tool in specialty outpatient clinics. The tool structures patient intake information specific to the clinician's specialty and generates content suitable for inclusion in electronic medical records (EMRs). In one experimental condition, the system produces structured summaries encompassing the chief complaint, history of present illness, and prior diagnostic tests. In a second condition, the AI additionally provides a proposed diagnosis and treatment plan by drafting a complete EMR note. Physicians may choose to accept, modify, or disregard the AI-generated output in favor of composing their own notes. The research employs a randomized controlled field experiment in China, comparing the two AI-assisted arms with a control group that uses standard intake procedures. Outcomes assessed include consultation duration, patient throughput, and the potential for AI-generated reports to induce behavioral distortions in physician decision-making.

Intervention

In both treatment arms, patients complete a structured pre-consultation intake on a tablet prior to seeing the physician. The AI tool generates electronic medical record (EMR) content that the physician may use as is, edit, or discard and document independently.
Treatment Arm 1 (AI triage and summarization): The system generates an EMR-ready structured summary of the chief complaint, history of present illness, relevant past history, and prior tests.
Treatment Arm 2 (AI summarization plus diagnostic): In addition to the summary, the system generates a preliminary diagnostic assessment and plan of action by drafting the full EMR note.
Control Arm (standard workflow): Patients follow the usual outpatient intake and consultation process without AI-generated pre-consultation content.

Experimental design

This study is conducted in specialty outpatient clinics in China. Patients complete the intake before seeing the physician. The study includes one control arm (standard workflow) and two treatment arms that introduce AI-generated EMR content from the same tablet-based pre-consultation intake: (1) triage and summarization support, and (2) summarization plus diagnostic support that drafts the complete EMR note. Physicians may use, edit, or discard the AI output.

Design details

Randomization method
Individual random assignment at the patient-visit level.
Randomization unit
Individual outpatient visit.
Clusters
Not clustered (individual randomization at the outpatient visit level).
Observations
Approximately 150 outpatient visits. We expect approximately 5 to 15 eligible patients per week and plan to run the experiment for 2 to 3 months. The experiment may be extended if hospital staffing and operational feasibility allow.
Arms
Approximately 50 outpatient visits control, approximately 50 outpatient visits Treatment Arm 1 (AI summarization), approximately 50 outpatient visits Treatment Arm 2 (AI summarization plus diagnostic).
Primary outcomes
Consultation time; diagnostic accuracy; physician reliance on AI.
Secondary outcomes
Patient throughput; downstream orders.

Cited by

No citation data available for this registration yet.

Keywords