Assessing the Limitations of Large Language Models in Clinical Practice Guideline–Concordant Treatment Decision-Making on Real-World Data: Retrospective Study
JMIRx Med · 2025-11-03 · 6 · e74899-e74899
DOI: 10.2196/74899 ↗At a glance
Leitlinie in JMIRx Med, published online on 2025-11-03. Written by Tobias Roeschl and 15 co-authors. Institution: Department of Cardiology, Angiology and Intensive Care Medicine, Deutsches Herzzentrum der Charité, Berlin, Germany. Cited 14 times so far (Crossref); 44 references. The full text is freely available in PubMed Central.
Abstract (original language)
Background
Studies have shown that large language models (LLMs) are promising in therapeutic decision-making, with findings comparable to those of medical experts, but these studies used highly curated patient data.
Objective
This study aimed to determine if LLMs can make guideline-concordant treatment decisions based on patient data as typically present in clinical practice (lengthy, unstructured medical text).
Methods
We conducted a retrospective study of 80 patients with severe aortic stenosis who were scheduled for either surgical (SAVR; n=24) or transcatheter aortic valve replacement (TAVR; n=56) by our institutional heart team in 2022. Various LLMs (BioGPT, GPT-3.5, GPT-4, GPT-4 Turbo, GPT-4o, LLaMA-2, Mistral, PaLM 2, and DeepSeek-R1) were queried using either anonymized original medical reports or manually generated case summaries to determine the most guideline-concordant treatment. We measured agreement with the heart team using Cohen κ coefficients, reliability using intraclass correlation coefficients (ICCs), and fairness using the frequency bias index (FBI; FBI >1 indicated bias toward TAVR).
Results
When presented with original medical reports, LLMs showed poor performance (Cohen κ coefficient: −0.47 to 0.22; ICC: 0.0‐1.0; FBI: 0.95‐1.51). The LLMs’ performance improved substantially when case summaries were used as input and additional guideline knowledge was added to the prompt (Cohen κ coefficient: −0.02 to 0.63; ICC: 0.01‐1.0; FBI: 0.46‐1.23). Qualitative analysis revealed instances of hallucinations in all LLMs tested.
Conclusions
Even advanced LLMs require extensively curated input for informed treatment decisions. Unreliable responses, bias, and hallucinations pose significant health risks and highlight the need for caution in applying LLMs to real-world clinical decision-making.
Subjects
Author keywords
Authors and institutions
- Department of Cardiology, Angiology and Intensive Care Medicine, Deutsches Herzzentrum der Charité, Berlin, Germany.
- Charité - Universitätsmedizin Berlin, corporate member of Freie Universität Berlin and Humboldt-Universität zu Berlin, Charitéplatz 1, Berlin, 10117, Germany, 49 17632864219.
- Berlin Institute of Health at Charité - Universitätsmedizin Berlin, BIH Biomedical Innovation Academy, BIH Charité Digital Clinician Scientist Program, Berlin, Germany.
- DZHK (German Centre for Cardiovascular Research), partner site Berlin, Berlin, Germany.
- Department of Cardiothoracic and Vascular Surgery, Deutsches Herzzentrum der Charité (DHZC), Berlin, Germany.
- Department of Infectious Diseases and Respiratory Medicine, Charité - Universitätsmedizin Berlin, Berlin, Germany.
Cite
APA
Roeschl, T., Hoffmann, M., Hashemi, D., Rarreck, F., Hinrichs, N., Trippel, T. D., Gröschel, M. I., Unbehaun, A., Klein, C., Kempfert, J., Dreger, H., O'Brien, B., Hindricks, G., Balzer, F., Falk, V., & Meyer, A. (2025). Assessing the Limitations of Large Language Models in Clinical Practice Guideline–Concordant Treatment Decision-Making on Real-World Data: Retrospective Study. JMIRx Med, 6, e74899-e74899. https://doi.org/10.2196/74899
Vancouver
Roeschl T, Hoffmann M, Hashemi D, Rarreck F, Hinrichs N, Trippel TD, et al. Assessing the Limitations of Large Language Models in Clinical Practice Guideline–Concordant Treatment Decision-Making on Real-World Data: Retrospective Study. JMIRx Med. 2025;6:e74899-e74899. doi:10.2196/74899
BibTeX
@article{Roeschl2025,
author = {Roeschl, Tobias and Hoffmann, Marie and Hashemi, Djawid and Rarreck, Felix and Hinrichs, Nils and Trippel, Tobias Daniel and Gröschel, Matthias I and Unbehaun, Axel and Klein, Christoph and Kempfert, Jörg and Dreger, Henryk and O'Brien, Benjamin and Hindricks, Gerhard and Balzer, Felix and Falk, Volkmar and Meyer, Alexander},
title = {Assessing the Limitations of Large Language Models in Clinical Practice Guideline–Concordant Treatment Decision-Making on Real-World Data: Retrospective Study},
journal = {JMIRx Med},
year = {2025},
volume = {6},
pages = {e74899--e74899},
doi = {10.2196/74899}
}
EUVIMED