Use of artificial intelligence to support the assessment of the methodological quality of systematic reviews
Bibliographic record
Abstract
OBJECTIVES: Published systematic reviews display a heterogeneous methodological quality, which can impact decision-making. Large language models (LLMs) can support and make the assessment of the methodological quality of systematic reviews more efficient, aiding in the incorporation of their evidence in guideline recommendations. We aimed to develop an LLM-based tool for supporting the assessment of the methodological quality of systematic reviews. METHODS: We assessed the performance of 8 LLMs in evaluating the methodological quality of systematic reviews. In particular, we provided 100 systematic reviews for eight LLMs (five base models and three fine-tuned models) to evaluate their methodological quality based on a 27-item validated tool (Reported Methodological Quality (ReMarQ)). The fine-tuned models had been trained with a different sample of 300 manually assessed systematic reviews. We compared the answers provided by LLMs with those independently provided by human reviewers, computing the accuracy, kappa coefficient and F1-score for this comparison. RESULTS: The best performing LLM was a fine-tuned GPT-3.5 model (mean accuracy = 96.5% [95% CI = 89.9%-100%]; mean kappa coefficient = 0.90 [95% CI = 0.71-1.00]; mean F1-score = 0.91 [95% CI = 0.83-1.00]). This model displayed an accuracy >80% and a kappa coefficient >0.60 for all individual items. When we made this LLM assess 60 times the same set of systematic reviews, answers to 18 of 27 items were always consistent (ie, were always the same) and only 11% of assessed systematic reviews showed inconsistency. CONCLUSION: Overall, LLMs have the potential to accurately support the assessment of the methodological quality of systematic reviews based on a validated tool comprising dichotomous items.
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 | 0.377 | 0.809 |
| Meta-epidemiology (narrow) | 0.004 | 0.003 |
| Meta-epidemiology (broad) | 0.010 | 0.009 |
| Bibliometrics | 0.031 | 0.015 |
| Science and technology studies | 0.002 | 0.004 |
| Scholarly communication | 0.016 | 0.007 |
| Open science | 0.005 | 0.010 |
| Research integrity | 0.004 | 0.006 |
| Insufficient payload (model declined to judge) | 0.003 | 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; the direct Gemma label and the distilled Codex classifier agree on what is shown here.
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".