Previously submitted to: Journal of Medical Internet Research (no longer under consideration since Feb 10, 2026)
Date Submitted: Oct 22, 2025
Open Peer Review Period: Oct 23, 2025 - Dec 18, 2025
(closed for review but you can still tweet)
NOTE: This is an unreviewed Preprint
Warning: This is a unreviewed preprint (What is a preprint?). Readers are warned that the document has not been peer-reviewed by expert/patient reviewers or an academic editor, may contain misleading claims, and is likely to undergo changes before final publication, if accepted, or may have been rejected/withdrawn (a note "no longer under consideration" will appear above).
Peer review me: Readers with interest and expertise are encouraged to sign up as peer-reviewer, if the paper is within an open peer-review period (in this case, a "Peer Review Me" button to sign up as reviewer is displayed above). All preprints currently open for review are listed here. Outside of the formal open peer-review period we encourage you to tweet about the preprint.
Citation: Please cite this preprint only for review purposes or for grant applications and CVs (if you are the author).
Final version: If our system detects a final peer-reviewed "version of record" (VoR) published in any journal, a link to that VoR will appear below. Readers are then encourage to cite the VoR instead of this preprint.
Settings: If you are the author, you can login and change the preprint display settings, but the preprint URL/DOI is supposed to be stable and citable, so it should not be removed once posted.
Submit: To post your own preprint, simply submit to any JMIR journal, and choose the appropriate settings to expose your submitted version as preprint.
Warning: This is an author submission that is not peer-reviewed or edited. Preprints - unless they show as "accepted" - should not be relied on to guide clinical practice or health-related behavior and should not be reported in news media as established information.
Decision Bias and Diagnostic Reasoning in ChatGPT-4o: Evidence from Orthodontic Extraction Decision
ABSTRACT
Background:
Large language models (LLMs) have demonstrated potential in clinical reasoning, yet their decision reliability in complex orthodontic planning remains unclear. Previous studies have primarily assessed diagnostic accuracy rather than authentic clinical judgment processes.
Objective:
This study aimed to examine the diagnostic reasoning and decision-layer behavior of the multimodal large language model ChatGPT-4o in orthodontic treatment planning, using extraction decisions as a representative task, and to evaluate whether information completeness and prompt-engineering strategies influence analytic reasoning, decision accuracy, and potential anchoring bias.
Methods:
30 standardized orthodontic cases with textual, quantitative, and radiographic inputs were evaluated under four controlled versions (V1–V4) using ChatGPT-4o. Outputs were independently assessed by four orthodontists across seven diagnostic factors defined in standard orthodontic references. Each factor was scored 0–2 for analytic quality (factor-level score), and the total across factors was summed as a comprehensive reasoning score. An instruction-only counterfactual experiment was conducted to examine potential sources of bias. Three prompting strategies—Chain-of-Thought (CoT), Knowledge Prompt (KP), and Few-Shot (FS)—were subsequently applied using identical scoring procedures.
Results:
Extraction cases were correctly identified in 95% (57/60, 95%; 95% CI 86%–99%), whereas all 15 non-extraction cases were misclassified (0/60, 0%; 95% CI 0%–6%). Decision accuracy did not differ significantly among the four input versions (Cochran Q(3)=4.714, P=.194). Factor-level scores varied, with the comprehensive input (V3) achieving the highest comprehensive score (mean 10.50, SD 1.31), exceeding V2 (mean 9.50, SD 2.03), V1 (mean 7.57, SD 1.65), and V4 (mean 7.20, SD 2.01). Significant version effects appeared for the vertical skeletal pattern (Friedman χ²(3) = 64.73, P<.001) and labial inclination of anterior teeth (Friedman χ²(3) = 73.282, P<.001). Among prompting strategies, FS raised non-extraction accuracy from 0% to 40%(6/15, 40%; 95% CI 16%–68%) (exact McNemar P=.031), while KP and CoT primarily enhanced factor-level reasoning.
Conclusions:
ChatGPT-4o produced more comprehensive analyses when supplied with richer clinical inputs, yet a persistent default-to-extraction bias remained. Prompt-engineering yielded limited but directionally positive effects—particularly with the Few-Shot strategy, which partially corrected non-extraction errors and improved factor-level reasoning. These findings indicate that while structured prompting can enhance analytic depth, decision reliability remains inadequate for autonomous clinical use. Continued integration of orthodontic expertise and multimodal evidence will be essential to translate large language models into clinically dependable diagnostic support systems.
Citation
Request queued. Please wait while the file is being generated. It may take some time.
Copyright
© The authors. All rights reserved. This is a privileged document currently under peer-review/community review (or an accepted/rejected manuscript). Authors have provided JMIR Publications with an exclusive license to publish this preprint on it's website for review and ahead-of-print citation purposes only. While the final peer-reviewed paper may be licensed under a cc-by license on publication, at this stage authors and publisher expressively prohibit redistribution of this draft paper other than for review purposes.