CoChat

The DOI matches its Crossref record; 54 of the 54 references it cites were found in a registry or an index; 22 of the 26 sources its claims cite were checked for support.

Mitigating Cognitive Biases in Clinical Decision-Making Through Multi-Agent Conversations Using Large Language Models: Simulation Study

Open access

Published version of a preprint

Abstract

Background: Cognitive biases in clinical decision-making significantly contribute to errors in diagnosis and suboptimal patient outcomes. Addressing these biases presents a formidable challenge in the medical field. Objective: This study aimed to explore the role of large language models (LLMs) in mitigating these biases through the use of the multi-agent framework. We simulate the clinical decision-making processes through multi-agent conversation and evaluate its efficacy in improving diagnostic accuracy compared with humans. Methods: A total of 16 published and unpublished case reports where cognitive biases have resulted in misdiagnoses were identified from the literature. In the multi-agent framework, we leveraged GPT-4 (OpenAI) to facilitate interactions among different simulated agents to replicate clinical team dynamics. Each agent was assigned a distinct role: (1) making the final diagnosis after considering the discussions, (2) acting as a devil’s advocate to correct confirmation and anchoring biases, (3) serving as a field expert in the required medical subspecialty, (4) facilitating discussions to mitigate premature closure bias, and (5) recording and summarizing findings. We tested varying combinations of these agents within the framework to determine which configuration yielded the highest rate of correct final diagnoses. Each scenario was repeated 5 times for consistency. The accuracy of the initial diagnoses and the final differential diagnoses were evaluated, and comparisons with human-generated answers were made using the Fisher exact test. Results: A total of 240 responses were evaluated (3 different multi-agent frameworks). The initial diagnosis had an accuracy of 0% (0/80). However, following multi-agent discussions, the accuracy for the top 2 differential diagnoses increased to 76% (61/80) for the best-performing multi-agent framework (Framework 4-C). This was significantly higher compared with the accuracy achieved by human evaluators (odds ratio 3.49; P=.002). Conclusions: The multi-agent framework demonstrated an ability to re-evaluate and correct misconceptions, even in scenarios with misleading initial investigations. In addition, the LLM-driven, multi-agent conversation framework shows promise in enhancing diagnostic accuracy in diagnostically challenging medical scenarios.

Overview

CoChat Summary

TL;DR

This simulation study used GPT-4 Turbo and an AutoGen multi-agent framework to re-evaluate 16 clinical case reports in which cognitive biases had caused misdiagnosis. The best-performing 4-agent framework, which explicitly discussed cognitive biases, improved final top-2 differential diagnostic accuracy to 76% (61/80), significantly better than human evaluators (odds ratio 3.49; P=.002).

Key findings

  • Initial diagnoses by the first-responder agent were correct in 0% (0/80) of responses across all three multi-agent frameworks, whereas human-generated initial diagnoses were correct in 27% (13/48).
  • After multi-agent discussion, final diagnosis accuracy was 64% (51/80) for Framework 3, 69% (55/80) for Framework 4, and 76% (61/80) for Framework 4-C; only Framework 4 (odds ratio 2.39; P=.03) and Framework 4-C (odds ratio 3.49; P=.002) were significantly better than human answers.
  • Human final diagnoses were correct in 48% (23/48).
  • Across all three frameworks, multi-agent discussions led to the correct final diagnosis in 13 of 16 scenarios (81%), and identified anchoring bias, confirmation bias, availability bias, and premature closure. In case 12, Framework 4-C reached the correct diagnosis of pneumothorax in 100% (5/5) of simulations after an initially misleading chest radiograph interpretation.

Methods

  • PubMed was searched for case reports of misdiagnosis attributed to cognitive bias; 162 reports were found, 37 were eligible, and 15 published cases were selected, plus 1 unpublished personal critical care case, for 16 scenarios.
  • GPT-4 Turbo was accessed via API and used within the AutoGen multi-agent conversation framework. Three configurations were tested: Framework 3 (Junior Resident I, Junior Resident II, Recorder), Framework 4 (plus Professional Expert), and Framework 4-C (plus Senior Doctor tasked with discussing cognitive biases). Each scenario was repeated 5 times.
  • Diagnostic accuracy was assessed for initial and final diagnoses. Two physicians graded answers, with consensus for discrepancies. Three doctors with at least 5 years of clinical experience provided human comparator differential diagnoses. Multi-agent and human answers were compared using the Fisher exact test.

Limitations

  • The paper states that reliance on published case reports limits the breadth of clinical scenarios and may affect generalizability.
  • The paper states that excluding visual data such as medical imaging confines the model to text-based information, omitting a critical component of clinical diagnosis.
  • The paper states that inherent LLM biases from pre-training data and technical limitations in understanding complex medical terminology and nuances may limit applicability compared with experienced clinicians. Evident from the text: the study used only 16 simulated scenarios, did not test real-time clinical outcomes, and the human comparator involved only 3 doctors.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.2196/59439

Cites

54 of 54 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Different roles in the multi-agent conversation framework.
Figure 1 Different roles in the multi-agent conversation framework.CC BY · PubMed Central
Figure 2 Schematic illustration of multi-agent (Framework 4-C) discussion dynamics leading to accurate differential diagnosis. ED: emergency department; LLM: large language model.
Figure 2 Schematic illustration of multi-agent (Framework 4-C) discussion dynamics leading to accurate differential diagnosis. ED: emergency department; LLM: large language model.CC BY · PubMed Central
Figure 3 Flow diagram for identification of case reports with cognitive bias.
Figure 3 Flow diagram for identification of case reports with cognitive bias.CC BY · PubMed Central

Verification

Claims the paper makes

15 Valid · 1 Not valid · 3 Not checked · 46 Own results and interpretations

24 more claims

Create a free account to see all 65 claims and every source verdict.

See all claims — free account
  • Probably not a substantial claim. 29
  • It may state two separate claims as one. 13

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Citations

Cites

Cited by

See all 66

Metadata

Authors
Ke, Y., Yang, R., Lie, S. A., Lim, T. X. Y., Ning, Y., Li, I., Abdullah, H. R., Ting, D. S. W., & Liu, N.
Affiliations
  • Singapore General Hospital (SG)
  • Duke-NUS Medical School (SG)
  • The University of Tokyo (JP)
  • Singapore National Eye Center (SG)
  • National University of Singapore (SG)
Corresponding author
Liu, Nan
Venue
Journal of Medical Internet Research
Published
2024-11-19
Volume
26
Pages
e59439
ISSN
1438-8871 (ISSN-L)
OpenAlex
W4404518047
Cited by
64
References
52
Citations by year
2026: 35 · 2025: 27 · 2024: 2
FWCI
5.95
Type
journal article
Language
en
Open access
Licence
cc-by
Locations
  • Journal of Medical Internet Research (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
Funders
Duke-NUS Medical School
Topics
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Health Sciences › Medicine › Family Practice › Clinical Reasoning and Diagnostic Skills
  • Health Sciences › Health Professions › Health Information Management › Artificial Intelligence in Healthcare
Keywords
cognitive biases, clinical decision-making, large language models, multi-agent systems, GPT-4, diagnostic accuracy, misdiagnosis, confirmation bias, anchoring effect, devil's advocate, differential diagnosis, AI-assisted diagnosis
MeSH terms
Clinical Decision-Making, Cognition, Communication, Computer Simulation, Diagnostic Errors, Humans, Language, Bias

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-07, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-07 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.