CoChat
Sign Up
Cited by 29Pharmacovigilance and Adverse Drug ReactionsArtificial Intelligence in Healthcare and EducationPharmaceutical Practices and Patient Outcomes

Can large language models detect drug–drug interactions leading to adverse drug reactions?

Open access
Sicard, Justine, Montastruc, François, Achalme, Coline, Jonville-Bera, Annie Pierre, Songue, Paul, Babin, Marina, Soeiro, Thomas, Schiro, Pauline +2 more·2025·Therapeutic Advances in Drug Safety
DOI 10.1177/20420986251339358PMC PMC12084699
Open this paper in CoChat — free account

Abstract

Background: Drug–drug interactions (DDI) are an important cause of adverse drug reactions (ADRs). Could large language models (LLMs) serve as valuable tools for pharmacovigilance specialists in detecting DDIs that lead to ADR notifications? Objective: To compare the performance of three LLMs (ChatGPT, Gemini, and Claude) in detecting and explaining clinically significant DDIs that have led to an ADR. Design: Observational cross-sectional study. Methods: We used the French National Pharmacovigilance Database to randomly extract Individual Case Safety Reports (ICSRs) of ADRs with DDI (positive controls) and ICSRs of ADRs without DDI (negative controls) registered in 2022. Interaction cases were classified by difficulty level (level-1 DDI being the easiest and level-2 DDI being the most difficult). We give each LLM (ChatGPT, Gemini, and Claude) the same prompt and case summary. Sensitivity, specificity, and F -measure were calculated for each LLM in detecting DDIs in the case summaries. Results: We assessed 82 ICSRs with DDIs and 22 ICSRs without DDIs. Among ICSRs with DDIs, 37 involved level-1 DDIs, and 45 involved level-2 DDIs. Correct responses were more frequent for level-1 DDIs than for level-2 DDIs. Regardless of difficulty level, ChatGPT detected 99% of DDI cases, and Claude and Gemini detected 95%. The percentage of correct answers to all DDI-related questions was 66% for ChatGPT, 68% for Claude, and 33% for Gemini. ChatGPT and Claude produced comparable results and outperformed Gemini ( F -measure between 0.83 and 0.85 for ChatGPT and Claude and 0.63–0.68 for Gemini) to detect drugs involved in DDI. All exhibited low specificity (ChatGPT 0.68, Claude 0.64, and Gemini 0.36) and reported nonexistent DDIs for negative controls. Conclusion: LLMs can detect DDIs leading to pharmacovigilance cases, but cannot reliably exclude DDIs in cases without interactions. Pharmacologists are crucial for assessing whether a DDI is implicated in an ADR.

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.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1177/20420986251339358

Cites

2 of 2 references verified0 citation errors · 0 not checkable

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.

A few free checks a day · papers already verified open instantly · results are public

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?

  • Large language models such as ChatGPT, Gemini and DeepSeek could rapidly analyze medication lists and cross-reference known interactions, providing comprehensive and standardized DDI screening at high speed and potentially streamlining preoperative prescribing workflows.

    “Large language models (LLMs) such as Chat-GPT, Gemini and specialized systems like DeepSeek could rapidly analyze medication lists and cross-reference known interactions, providing comprehensive and standardized DDI screening at high speed and potentially streamlining preoperative prescribing workflows [29][30][31][32].”

    From Comparing Artificial Intelligence (ChatGPT, Gemini, DeepSeek) and Oral Surgeons in Detecting Clinically Relevant Drug–Drug Interactions in Dental Therapy (2025) · cites this paper as [32]

    Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

    Experimental — accuracy not yet measured

  • Earlier evaluations reported that general-purpose LLMs such as ChatGPT and Claude achieved very high sensitivity (up to 95-99%) but poor specificity due to hallucinated or duplicated DDIs.

    “Earlier evaluations reported that general-purpose LLMs such as ChatGPT and Claude achieved very high sensitivity (up to 95-99%) but poor specificity due to hallucinated or duplicated DDIs, as shown by Sicard et al. (2025) [32] and Al-Ashwal et al. (2023) [33].”

    From Comparing Artificial Intelligence (ChatGPT, Gemini, DeepSeek) and Oral Surgeons in Detecting Clinically Relevant Drug–Drug Interactions in Dental Therapy (2025) · cites this paper as Sicard et al. (2025) [32]

    The Results report that ChatGPT detected 99% of DDI cases and Claude 95%, while both had low specificity (0.68 and 0.64). The Abstract Results say the models reported nonexistent DDIs for negative controls, supporting the false-positive aspect. The paper does not attribute poor specificity to duplicated DDIs; the redundancy mentioned in Results concerns cases excluded from the dataset, not model outputs. (LLM, conf=0.88)

    Experimental — accuracy not yet measured

  • Sicard et al. (2025) showed that general-purpose LLMs reached very high detection rates of DDIs but remained unable to reliably exclude non-interactions.

    “As shown by Sicard et al. (2025) [32], general-purpose LLMs reached very high detection rates of DDIs but remained unable to reliably exclude non-interactions; the moderate AUC values and relatively low positive predictive values (16-20%) observed here align with that limitation, indicating that most alerts generated by even the best-performing models do not correspond to truly actionable DDIs.”

    From Comparing Artificial Intelligence (ChatGPT, Gemini, DeepSeek) and Oral Surgeons in Detecting Clinically Relevant Drug–Drug Interactions in Dental Therapy (2025) · cites this paper as Sicard et al. (2025) [32]

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Cites

Flagged first, then by citations

Research works printed but not linked: 20

  1. [2]Wei J Tay Y Bommasani R, et al. Emergent abilities of large language models. arXiv:2206.07682. 2022. 10.48550/arXiv.2206.07682. · DOI · Paper page
  2. [4]Matheny ME Yang J Smith JC, et al. Enhancing postmarketing surveillance of medical products with large language models. JAMA Netw Open 2024; 7: e2428276. 39150707
  3. [5]Li D Wu L Zhang M, et al. Assessing the performance of large language models in literature screening for pharmacovigilance: a comparative study. Front Drug Saf Regul 2024; 4: 1379260. PMC12443093 40979379
  4. [6]Dong G Bate A Haguinet F, et al. Optimizing signal management in a vaccine adverse event reporting system: a proof-of-concept with COVID-19 vaccines using signs, symptoms, and natural language processing. Drug Saf 2024; 47: 173– 182. 38062261 PMC10821983
  5. [7]Pandya S Patel C Sojitra B, et al. Causality assessment of adverse drug reaction toxic epidermal necrolysis with the aid of ChatGPT: a case report. Cureus 2024; 16: e60638. PMC11187461 38903274
  6. [8]Montastruc F Storck W de Canecaude C, et al. Will artificial intelligence chatbots replace clinical pharmacologists? An exploratory study in clinical practice. Eur J Clin Pharmacol 2023; 79: 1375– 1384. 37566133
  7. [9]Pariente A Salvo F Bres V, et al. Can we ask chatgpt about drug safety? Appropriateness of ChatGPT responses to questions about drug use and adverse reactions received by pharmacovigilance centers. Drug Saf 2024; 47: 921– 923. 38717670
  8. [10]Juhi A Pipil N Santra S, et al. The capability of ChatGPT in predicting and explaining common drug-drug interactions. Cureus 2023; 15: e36272. PMC10105894 37073184
  9. [11]Al-Ashwal FY Zawiah M Gharaibeh L, et al. Evaluating the sensitivity, specificity, and accuracy of ChatGPT-3.5, ChatGPT-4, Bing AI, and Bard against conventional drug-drug interactions clinical tools. Drug Healthc Patient Saf 2023; 15: 137– 147. 37750052 PMC10518176
  10. [12]Radha Krishnan RP Hung EH Ashford M, et al. Evaluating the capability of ChatGPT in predicting drug–drug interactions: real-world evidence using hospitalized patient data. Brit J Clin Pharm 2024; 90: 3361– 3366. PMC11602951 39359001
  11. [13]Bedi S Liu Y Orr-Ewing L, et al. Testing and evaluation of health care applications of large language models: a systematic review. JAMA 2025; 333: 319– 328. 39405325 PMC11480901
  12. [14]Wang H Ding YJ Luo Y. Future of ChatGPT in pharmacovigilance. Drug Saf 2023; 46: 711– 713. 37306853 PMC11168724
  13. [15]Bihan K Lipszyc L Lemaitre F, et al. Nirmatrelvir/ritonavir (Paxlovid®): French pharmacovigilance survey 2022. Therapies 2023; 78: 531– 547. PMC9990891 37012153
  14. [16]Théophile H Dutertre J-P Gérardin M, et al. Validation and reproducibility of the updated French Causality Assessment Method: an evaluation by pharmacovigilance centres & pharmaceutical companies. Therapie 2015; 70: 465– 476. 26423144
  15. [17]Arimone Y Bidault I Dutertre J-P, et al. Updating the french method for the causality assessment of adverse drug reactions. Therapies 2013; 68: 69– 76. 23773347
  16. [21]He N Yan Y Wu Z, et al. Chat GPT-4 significantly surpasses GPT-3.5 in drug information queries. J Telemed Telecare 2025; 31: 306– 308. 37350055
  17. [22]Sonoda Y Kurokawa R Nakamura Y, et al. Diagnostic performances of GPT-4o, Claude 3 Opus, and Gemini 1.5 Pro in “Diagnosis Please” cases. Jpn J Radiol 202; 42: 1231– 1235. PMC11522128 38954192
  18. [24]Andrikyan W Sametinger SM Kosfeld F, et al. Artificial intelligence-powered chatbots in search engines: a cross-sectional study on the quality and risks of drug information for patients. BMJ Qual Saf 2025; 34: 100– 109. PMC11874309 39353736
  19. [25]Kim H Hwang H Lee J, et al. Small language models learn enhanced reasoning skills from medical textbooks. arXiv:2404.00376, 2024. PMC12048634 40316765 · arXiv
  20. [28]Wang L Chen X Deng X, et al. Prompt engineering in consistency and reliability with the evidence-based guideline for LLMs. npj Digital Med 2024; 7: 1– 9. PMC10879172 38378899

Cited by

Sorted by citations

Authors
Sicard, J., Montastruc, F., Achalme, C., Jonville-Bera, A. P., Songue, P., Babin, M., Soeiro, T., Schiro, P., de Canecaude, C., & Barus, R.
Affiliations
  • Centre Hospitalier Universitaire de Toulouse (FR)
  • Department of Medical and Clinical Pharmacology, Tours University Hospital (CHU), Tours, France
  • Centre Hospitalier Universitaire d'Angers (FR)
  • Assistance Publique Hôpitaux de Marseille (FR)
  • Department of Medical and Clinical Pharmacology, Centre of PharmacoVigilance and Pharmacoepidemiology, Faculty of Medicine, Toulouse University Hospital (CHU), Toulouse, France
Corresponding author
Barus, Romain
ORCID
Venue
Therapeutic Advances in Drug Safety
Publisher
SAGE Publications
Published
2025-05-16
Volume
16
Pages
20420986251339358-20420986251339358
ISSN
  • 2042-0986 (ISSN-L)
  • 2042-0994
DOI
10.1177/20420986251339358
PMC
PMC12084699
OpenAlex
W4410440973
Cited by
29
References
2
Citations by year
2026: 20 · 2025: 9
FWCI
14.78
Type
journal article
Language
en
Open access
Licence
Locations
  • Therapeutic Advances in Drug Safety (publishedVersion, cc-by-nc)
  • PubMed Central (publishedVersion, cc-by-nc)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Topics
  • Life Sciences › Pharmacology, Toxicology and Pharmaceutics › Toxicology › Pharmacovigilance and Adverse Drug Reactions
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Health Sciences › Medicine › Geriatrics and Gerontology › Pharmaceutical Practices and Patient Outcomes
Keywords
drug-drug interactions, adverse drug reactions, large language models, pharmacovigilance, individual case safety reports, ChatGPT, Gemini, Claude

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