The AI Consultation Report as a Format for Documenting Clinical AI Failure Modes
DOI:
https://doi.org/10.5281/zenodo.22740633Keywords:
ai consultation report, clinical ai failure modes, large language models, generative artificial intelligence, clinical decision making, signal detection, medical publishingAbstract
Controlled evaluations of large language models (LLMs) in medicine measure how often a system answers a clinical question correctly, but they cannot describe what happens when a clinician consults one about a specific patient. No current reporting instrument lets an individual clinician document a consultation, and isolated published accounts cannot be compared or pooled because no format defines them. The artificial intelligence (AI) consultation report is proposed as a defined article type: a first-person account specifying the clinical impasse, the tool, version, and approximate date, the question as posed, the output as returned, the clinician's decision, and the outcome. Accumulated, such reports offer a mechanism for identifying clinical AI failure modes before they can be measured, on the model of spontaneous reporting in pharmacovigilance.
References
1. Gallifant J, Afshar M, Ameen S, Aphinyanaphongs Y, Chen S, Cacciamani G, et al. The TRIPOD-LLM reporting guideline for studies using large language models. Nat Med. 2025;31: 60–69. doi:10.1038/s41591-024-03425-5
2. Perlis RH, Fihn SD. Evaluating the Application of Large Language Models in Clinical Research Contexts. JAMA Netw Open. 2023;6: e2335924. doi:10.1001/jamanetworkopen.2023.35924
3. Wong EYT, Verlingue L, Aldea M, Franzoi MA, Umeton R, Halabi S, et al. ESMO guidance on the use of Large Language Models in Clinical Practice (ELCAP). Ann Oncol. 2025;36:1447–1457. doi:10.1016/j.annonc.2025.09.001
4. Heston TF, Lewis LM. ChatGPT provides inconsistent risk-stratification of patients with atraumatic chest pain. PloS One. 2024;19: e0301854. doi:10.1371/journal.pone.0301854
5. Li Z, Yan C, Cao Y, Gong A, Li F, Zeng R. Evaluating performance of large language models for atrial fibrillation management using different prompting strategies and languages. Sci Rep. 2025;15: 19028. doi:10.1038/s41598-025-04309-5
6. Heston TF, Gillette J. Large language models demonstrate distinct personality profiles. Cureus. 2025;17: e84706. doi:10.7759/cureus.84706
7. Ye Y, Sarkar S, Bhaskar A, Tomlinson B, Monteiro O. Using ChatGPT in a clinical setting: A case report. MedComm – Future Med. 2023;2: e51. doi:10.1002/mef2.51
8. Heston TF, Khun C. Prompt Engineering in Medical Education. Int Med Educ. 2023;2:198–205. doi:10.3390/ime2030019
9. Meyboom RHB, Egberts ACG, Edwards IR, Hekster YA, De Koning FHP, Gribnau FWJ. Principles of Signal Detection in Pharmacovigilance: Drug Saf. 1997;16: 355–365. doi:10.2165/00002018-199716060-00002
10. Pacurariu AC, Straus SM, Trifirò G, Schuemie MJ, Gini R, Herings R, et al. Useful Interplay Between Spontaneous ADR Reports and Electronic Healthcare Records in Signal Detection. Drug Saf. 2015;38: 1201–1210. doi:10.1007/s40264-015-0341-5
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2026 Thomas F. Heston
This work is licensed under a Creative Commons Attribution 4.0 International License.
