Accuracy and reproducibility of <scp>ChatGPT</scp> responses to real‐world drug information questions
Bibliographic record
Abstract
Abstract Introduction The expanding use of Chat Generative Pre‐Trained Transformer (ChatGPT, OpenAI, San Francisco, CA) for drug information may enhance access to information. However, it is crucial to assess the accuracy and reproducibility of ChatGPT responses to drug information questions, examining its utility and limitations in clinical decision‐making. Objective To evaluate the accuracy and reproducibility of ChatGPT‐3.5 and ChatGPT‐4 in responding to clinician drug information questions compared with a commonly accepted resource, Lexicomp®(Wolters Kluwer Health, Philadelphia, PA). Methods A serial cross‐sectional study was conducted on ChatGPT from March 5 to 12, 2024 in the United States. ChatGPT‐3.5 is a free, artificial intelligence (AI) chatbot trained up to January 2022; ChatGPT‐4 is a paid‐subscription AI chatbot with internet access and more data. For trial 1 (day 0) we input 30 real‐world questions (10 drug information categories) into both ChatGPT‐3.5 and ChatGPT‐4. For trial 2 (day 1) and 3 (day 7), 10 randomly selected questions were re‐input into ChatGPT. The primary outcome evaluated the accuracy of ChatGPT‐3.5 responses versus (vs.) Lexicomp® using a 4‐point Likert scale. Secondary outcomes included assessing the accuracy of ChatGPT‐4 responses vs. Lexicomp, comparing the accuracy of both ChatGPT versions' responses, and comparing reproducibility of ChatGPT responses over time. Cohen's Kappa and Cochran's Q assessed reproducibility. Results ChatGPT‐3.5 demonstrated 30% accuracy (9/30), while ChatGPT‐4 had 40% (12/30) accuracy (p = 0.51). Neither ChatGPT versions accurately answered all the questions in any category. ChatGPT‐3.5's agreement between trials 1 vs. 2, 1 vs. 3, and 2 vs. 3 had fair (k = 0.21), moderate (k = 0.41), and substantial agreement (k = 0.62), respectively. ChatGPT‐4 trials 1 vs. 2, 1 vs. 3, and 2 vs. 3 had fair (k = 0.23), substantial (k = 0.80), and fair agreement (0.40). The accuracy of ChatGPT‐3.5 vs. ChatGPT‐4 for the 10 questions across the three trials was 30%, 20%, and 10% (p = 0.78), and 60%, 40%, and 50% (p = 0.82). Conclusions Both ChatGPT versions demonstrated limited accuracy and reproducibility in answering drug information questions, suggesting that health care professionals should exercise caution when using ChatGPT for drug information.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.055 | 0.203 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.004 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".