Global variations in artificial intelligence-generated information on juvenile idiopathic arthritis
Bibliographic record
Abstract
OBJECTIVES: We aimed to evaluate similarities and variations of information provided by Large Language Models (LLMs) across diverse world regions by analysing responses to validated questions on oligoarticular juvenile idiopathic arthritis (oJIA). METHODS: The 10 PICOs related to the oJIA treatment on the 2021 American College of Rheumatology recommendations were simultaneously prompted in English to ChatGPT 4o from five different countries (Canada, India, Italy, Kenya and Türkiye). Readability was assessed through the Flesch Reading Ease Score (FRES), distinctiveness of terms through the Term Frequency-Inverse Document Frequency (TF-IDF) analysis. Co-occurrence networks (CONs) detailed the relationships between terms. Three experts rated the adherence of responses to recommendations using a Likert-like scale. RESULTS: All the responses were difficult or very difficult to read, with a median FRES of 30 (IQR 24-34). Depending on the expert, 52-84% of responses were mostly or fully adherent to the recommendations, with similar adherence rates across countries. No response was not adherent at all. Inter-rater agreement on the adherence of LLM-generated responses was generally weak (Kappa values mostly below 0.40), highlighting the challenges of consistently evaluating AI-generated medical information. The TF-IDF analysis showed that the distinctiveness of terminology in LLM-generated responses varied across countries, with scores ranging from 0.60 to 0.85. CONs detailed a strong focus on intra-articular corticosteroid treatments in Italy and emphasis on short- and long-term outcomes in Kenya. CONCLUSION: LLM-generated content should be critically evaluated in clinical practice, especially in the context of regional differences.
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 | 0.009 | 0.051 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.004 | 0.003 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.000 | 0.002 |
| Research integrity | 0.001 | 0.000 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
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".