Large language models encode medical oncology knowledge: Performance on the ASCO and ESMO examination questions.
Bibliographic record
Abstract
511 Background: Chatbots based on large language models (LLM) recently developed an unprecedented ability to answer questions across a broad range of applications. Whether LLMs encode sufficient knowledge to answer questions about medical oncology, a highly specialized domain requiring rapid integration of new evidence, is unknown. Methods: We presented ChatGPT (GPT-3.5 and GPT-4) with the American Society of Oncology (ASCO) Self Assessment Program and the European Society of Medical Oncology (ESMO) Examination Trial questions, excluding those that included images or required knowledge unavailable before the algorithm’s training cutoff date. The proportion of correct answers was compared against random chance. ChatGPT was prompted again for a different answer if the previous was incorrect. The reasoning provided by ChatGPT was qualitatively evaluated by two medical oncologists. Results: ChatGPT (GPT-4) correctly answered 84.4% (38/45, 95% confidence interval [CI] 70.5-93.5%, P<0.0001 versus random answering) of ASCO and 86.7% (65/75, 95% CI 76.8-93.4%, P<0.0001) of the ESMO examination questions. GPT-4 outperformed GPT-3.5 (57.8% [26/45, 95% CI 42.2%-72.3%, P=0.001] for ASCO and 65.3% [49/75, 95% CI 53.5%-76.0%, P=0.004] for ESMO). Including second attempts, GPT-4 correctly answered 93.3% (42/45, 95% CI 81.7-98.6%) of ASCO and 93.3% (70/75, 95% CI 85.1-97.8%) of the ESMO examination questions. Incorrect responses for ASCO questions were more common in questions whose answers referenced papers published after 2018 (22.2% [4/18] versus 11.1%, [3/27], P=0.03). Oncologists rated the reasoning behind correct answers by GPT-3.5 as complete for 93.3% of questions (70/75, CI 85.1-97.8%). Conclusions: LLMs can answer examination questions designed for medical oncology fellows with impressive and improving accuracy, alongside correct reasoning. These results imply broad potential applications of LLMs during cancer care to improve the patient and provider experience.
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 distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.005 | 0.008 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 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 teacher head, 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".