Examining real-world medication consultations and drug-herb interactions: ChatGPT performance evaluation
Open accessPublished version of a preprint
Abstract
BACKGROUND: Since OpenAI released ChatGPT, with its strong capability in handling natural tasks and its user-friendly interface, it has garnered significant attention. OBJECTIVE: A prospective analysis is required to evaluate the accuracy and appropriateness of medication consultation responses generated by ChatGPT. METHODS: A prospective cross-sectional study was conducted by the pharmacy department of a medical center in Taiwan. The test data set comprised retrospective medication consultation questions collected from February 1, 2023, to February 28, 2023, along with common questions about drug-herb interactions. Two distinct sets of questions were tested: real-world medication consultation questions and common questions about interactions between traditional Chinese and Western medicines. We used the conventional double-review mechanism. The appropriateness of each response from ChatGPT was assessed by 2 experienced pharmacists. In the event of a discrepancy between the assessments, a third pharmacist stepped in to make the final decision. RESULTS: Of 293 real-world medication consultation questions, a random selection of 80 was used to evaluate ChatGPT's performance. ChatGPT exhibited a higher appropriateness rate in responding to public medication consultation questions compared to those asked by health care providers in a hospital setting (31/51, 61% vs 20/51, 39%; P=.01). CONCLUSIONS: The findings from this study suggest that ChatGPT could potentially be used for answering basic medication consultation questions. Our analysis of the erroneous information allowed us to identify potential medical risks associated with certain questions; this problem deserves our close attention.
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
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?
Single studies reported high accuracy for specialized reasoning models (OpenAI o3, OpenAI o4-mini) in pharmacotherapy management and a customized ChatGPT-4 model for patient education and counseling.
“Furthermore, single studies reported high accuracy for specialized reasoning models, 100 OpenAI o3, 33 and OpenAI o4-mini, 60 in pharmacotherapy management and a customized ChatGPT-4 model for patient education and counseling.59”
From Large Language Models in Pharmacist-Led Direct Patient Care: A Rapid Review of Published Evaluations (2026) · cites this paper as [33]
The citation is attached to the customized ChatGPT-4 claim, but the Limitations section states that this study tested GPT-3.5. The Results section reports 51 of 80 medication-consultation responses (64%) as appropriate, not high accuracy for a customized GPT-4 patient-education model. No page numbers are available in our copy. (LLM, conf=0.89)
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measured
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
The citation lists haven’t been fetched yet
References 22 · Cited by 50
Research works printed but not linked: 2
- [7]Ouyang L Wu J Jiang X Almeida D Wainwright C Mishkin P Training language models to follow instructions with human feedback 2022 Advances in Neural Information Processing Systems 35 (NeurIPS 2022) Nov 28-Dec 9, 2022 New Orleans, LA 27730 27744
- [11]Vaswani A Shazeer N Parmar N Uszkoreit J Jones L Gomez A Attention is all you need 2017 Advances in Neural Information Processing Systems 30 (NIPS 2017) Dec 4-9, 2017 Long Beach, CA
- Authors
- Hsu, H.-Y., Hsu, K.-C., Hou, S.-Y., Wu, C.-L., Hsieh, Y.-W., & Cheng, Y.-D.
- Affiliations
- National Taiwan University (TW)
- China Medical University (TW)
- China Medical University Hospital (TW)
- Corresponding author
- Hsieh, Yow-Wen
- Venue
- JMIR Medical Education
- Publisher
- JMIR Publications Inc.
- Published
- 2023-08-21
- Volume
- 9
- Pages
- e48433
- ISSN
- 2369-3762 (ISSN-L)
- PMC
- PMC10477918
- OpenAlex
- W4385236544
- Cited by
- 50
- References
- 22
- Citations by year
- 2026: 9 · 2025: 15 · 2024: 23 · 2023: 3
- FWCI
- 1.99
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- cc-by
- Locations
- JMIR Medical Education (publishedVersion, cc-by)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Funders
- Agence Nationale de la Recherche (ANR‐19, ANR-19-P3IA-0001, ANR-19-CE38-0012)
- Nederlandse Organisatie voor Wetenschappelijk Onderzoek (406.DI.19.059)
- China Medical University Hospital
- Centre National de la Recherche Scientifique
- China Medical University
- Grand Équipement National De Calcul Intensif
- Topics
- Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
- Life Sciences › Pharmacology, Toxicology and Pharmaceutics › Pharmacology › Pharmacy and Medical Practices
- Health Sciences › Medicine › Geriatrics and Gerontology › Pharmaceutical Practices and Patient Outcomes
- Keywords
- ChatGPT, large language models, pharmaceutical consultation, herb-drug interactions, traditional Chinese medicine, pharmacist review, drug safety
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.
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
