Accuracy of LLMs to retrieve numeric data for meta-analysis in dentistry
Bibliographic record
Abstract
OBJECTIVES: Evidence-based dentistry relies heavily on systematic reviews and meta-analyses (SRMA), considered the most robust forms of evidence. Still, conducting SRMA is time- and resource-intensive, with high error rates in data extraction. Artificial intelligence (AI) and large language models (LLMs) offer the potential to automate and accelerate SRMA processes such as data extraction. However, assessing the reliability and accuracy of these new AI-based technologies for SRMA is crucial. This study evaluated the accuracy of four LLMs (DeepSeek v3 R1, Claude 3.5 Sonnet, ChatGPT-4o, and Gemini 2.0-flash) in extracting different primary numeric outcomes data in various dental topics. METHODS: LLMs were queried via APIs using default settings and a SMART-format prompt. Descriptive analysis was conducted at sub-outcome, outcome, and study levels. Errors were classified as hallucinations, missed, or omitted data. RESULTS: Overall extraction accuracy was exceptionally high at the sub-outcome level, with only 3 hallucinations (from Gemini 2.0-flash). Total errors increased at the outcome level and study level. Gemini 2.0-flash generally performed significantly worse than others (p < 0.01). Claude 3.5 Sonnet and DeepSeek-v3 R1 generally exhibited superior accuracy and lower omission rates in full-text extraction compared to Gemini 2.0-flash and ChatGPT-4o. CONCLUSIONS: This first comparative evaluation of multiple LLMs for data extraction in dental research from full-text PDFs highlights their significant potential but also limitations. Performance varied notably between models, with cost not directly correlating with superior performance. While single data point extraction was highly accurate, errors increased at higher aggregation levels. Standardized outcome reporting in studies could benefit future LLM extraction, and we offer a solid benchmark for future performance comparisons. CLINICAL SIGNIFICANCE: This study demonstrates that LLMs can achieve high accuracy in extracting single numeric outcomes, but omission errors in full-text analyses limit their independent use in SRMA. Improving outcome reporting standards and leveraging accurate, lower-cost models may enhance evidence synthesis efficiency in dentistry and beyond.
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 | 0.304 | 0.750 |
| Meta-epidemiology (narrow) | 0.003 | 0.002 |
| Meta-epidemiology (broad) | 0.005 | 0.016 |
| Bibliometrics | 0.017 | 0.013 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.008 | 0.006 |
| Open science | 0.003 | 0.006 |
| Research integrity | 0.002 | 0.002 |
| Insufficient payload (model declined to judge) | 0.008 | 0.002 |
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".