Assessing the accuracy of ChatGPT responses to guideline-based inquiries: A cross-sectional study
Bibliographic record
Abstract
Introduction: Physicians treating multisystem diseases face challenges in consulting expanding, complex clinical guidelines. Large language models like ChatGPT may help consolidate this information, providing quick access to guideline recommendations. The objective of this study was to assess the accuracy of ChatGPT 3.5 and 4o responses to questions based on specialist-level guideline recommendations. Methods: A framework was developed for authors to pose questions, based on a guideline recommendation, to ChatGPT. A validation tool graded responses as concordant, partially concordant, or discordant to the guideline recommendation. A total of 581 recommendations from three guidelines were analyzed. The primary outcome was overall accuracy. Subgroup analyses assessed accuracy based on number of criteria, strength of evidence, and type of recommendation. Results: For ChatGPT 3.5, 347 recommendations were concordant (59.72%), 128 partially concordant (22.03%), and 106 discordant (18.24%). Questions seeking a single response (Z = 5.289, p < .001) and questions based on recommendations with strong levels of evidence (OR 2.23, p = .001) generated higher levels of concordance. For ChatGPT 4o, 474 recommendations were concordant (81.6%), 82 partially concordant (14.1%), and 25 discordant (4.3%). Mean concordance ratings for single questions were significantly higher compared to multipart questions (Z = 3.08, p = .002). Mean concordance ratings for ChatGPT 4o were substantially higher compared to ChatGPT 3.5 (Z = 8.66, p < .00001). Discussion: ChatGPT 3.5 had a moderate level of accuracy. There remain weaknesses in its ability to answer multi-part questions or those backed by weaker evidence. ChatGPT 4o performed substantially better than ChatGPT 3.5, though both models were vulnerable to hallucination.
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 | 0.060 | 0.192 |
| Meta-epidemiology (narrow) | 0.000 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.003 | 0.003 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
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".