Evaluating a Customized Version of ChatGPT for Systematic Review Data Extraction in Health Research: Development and Usability Study
Bibliographic record
Abstract
Background: Systematic reviews are essential for synthesizing research in health sciences; however, they are resource-intensive and prone to human error. The data extraction phase, in which key details of studies are identified and recorded in a systematic manner, may benefit from the application of automation processes. Recent advancements in artificial intelligence, specifically in large language models (LLMs) such as ChatGPT, may streamline this process. Objective: This study aimed to develop and evaluate a custom Generative Pre-Training Transformer (GPT), named Systematic Review Extractor Pro, for automating the data extraction phase of systematic reviews in health research. Methods: OpenAI's GPT Builder was used to create a GPT tailored to extract information from academic manuscripts. The Role, Instruction, Steps, End goal, and Narrowing (RISEN) framework was used to inform prompt engineering for the GPT. A sample of 20 studies from two distinct systematic reviews was used to evaluate the GPT's performance in extraction. Agreement rates between the GPT outputs and human reviewers were calculated for each study subsection. Results: The mean time for human data extraction was 36 minutes per study, compared to 26.6 seconds for GPT generation, followed by 13 minutes of human review. The GPT demonstrated high overall agreement rates with human reviewers, achieving 91.45% for review 1 and 89.31% for review 2. It was particularly accurate in extracting study characteristics (review 1: 95.25%; review 2: 90.83%) and participant characteristics (review 1: 95.03%; review 2: 90.00%), with lower performance observed in more complex areas such as methodological characteristics (87.07%) and statistical results (77.50%). The GPT correctly extracted data in 14 instances (3.25% in review 1) and four instances (1.16% in review 2) when the human reviewer was incorrect. Conclusions: The custom GPT significantly reduced extraction time and shows evidence that it can extract data with high accuracy, particularly for participant and study characteristics. This tool may offer a viable option for researchers seeking to reduce resource demands during the extraction phase, although more research is needed to evaluate test-retest reliability, performance across broader review types, and accuracy in extracting statistical data. The tool developed in the current study has been made open access.
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 | 0.192 | 0.379 |
| Meta-epidemiology (narrow) | 0.002 | 0.003 |
| Meta-epidemiology (broad) | 0.002 | 0.005 |
| Bibliometrics | 0.006 | 0.005 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.003 | 0.004 |
| Open science | 0.003 | 0.005 |
| Research integrity | 0.001 | 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".