Accuracy of ChatGPT on Medical Questions in the National Medical Licensing Examination in Japan: Evaluation Study
Bibliographic record
Abstract
BACKGROUND: ChatGPT (OpenAI) has gained considerable attention because of its natural and intuitive responses. ChatGPT sometimes writes plausible-sounding but incorrect or nonsensical answers, as stated by OpenAI as a limitation. However, considering that ChatGPT is an interactive AI that has been trained to reduce the output of unethical sentences, the reliability of the training data is high and the usefulness of the output content is promising. Fortunately, in March 2023, a new version of ChatGPT, GPT-4, was released, which, according to internal evaluations, was expected to increase the likelihood of producing factual responses by 40% compared with its predecessor, GPT-3.5. The usefulness of this version of ChatGPT in English is widely appreciated. It is also increasingly being evaluated as a system for obtaining medical information in languages other than English. Although it does not reach a passing score on the national medical examination in Chinese, its accuracy is expected to gradually improve. Evaluation of ChatGPT with Japanese input is limited, although there have been reports on the accuracy of ChatGPT's answers to clinical questions regarding the Japanese Society of Hypertension guidelines and on the performance of the National Nursing Examination. OBJECTIVE: The objective of this study is to evaluate whether ChatGPT can provide accurate diagnoses and medical knowledge for Japanese input. METHODS: Questions from the National Medical Licensing Examination (NMLE) in Japan, administered by the Japanese Ministry of Health, Labour and Welfare in 2022, were used. All 400 questions were included. Exclusion criteria were figures and tables that ChatGPT could not recognize; only text questions were extracted. We instructed GPT-3.5 and GPT-4 to input the Japanese questions as they were and to output the correct answers for each question. The output of ChatGPT was verified by 2 general practice physicians. In case of discrepancies, they were checked by another physician to make a final decision. The overall performance was evaluated by calculating the percentage of correct answers output by GPT-3.5 and GPT-4. RESULTS: Of the 400 questions, 292 were analyzed. Questions containing charts, which are not supported by ChatGPT, were excluded. The correct response rate for GPT-4 was 81.5% (237/292), which was significantly higher than the rate for GPT-3.5, 42.8% (125/292). Moreover, GPT-4 surpassed the passing standard (>72%) for the NMLE, indicating its potential as a diagnostic and therapeutic decision aid for physicians. CONCLUSIONS: GPT-4 reached the passing standard for the NMLE in Japan, entered in Japanese, although it is limited to written questions. As the accelerated progress in the past few months has shown, the performance of the AI will improve as the large language model continues to learn more, and it may well become a decision support system for medical professionals by providing more accurate 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 | 0.014 | 0.082 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.002 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 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".