Evaluation of artificial intelligence-based tool Covidence in literature screening for guideline updates: A prospective study
Bibliographic record
Abstract
Regularly updating literature evidence for clinical practice guidelines (CPGs) is necessary but time intensive. Artificial intelligence (AI) tools, like Covidence, may accelerate literature screening. This study prospectively evaluated the effectiveness and accuracy of two Covidence functions: the auto-marking filter (adopted from Cochrane’s RCT classifier) for identifying randomized controlled trials (RCTs) and the machine learning-assisted title and abstract (Stage I) screening in two cancer-related CPGs. We updated the literature searches for the breast and lung cancer CPGs at the Program in Evidence-Based Care (PEBC), Ontario Health (Cancer Care Ontario). Two methodologists trained the machine learning model by indicating which of the first 25 references were tagged by Covidence’s RCT filter. Main outcomes included workload and time savings, Work Saved over Sampling (WSS), and the impact on whether original guideline recommendations required changes. A total of 1,270 (breast cancer) and 1,734 (lung cancer) references were imported into Covidence. Among them, 633 breast cancer and 656 lung cancer references were tagged by Covidence’s RCT filter as possible RCTs and required review at Stage I. Covidence’s RCT filter excluded non-RCTs with high accuracy, reducing manual screening by 50.2% (breast cancer) and 62.2% (lung cancer). For Stage I screening, Covidence reduced workload by 48.3% and 55.2% at 95% sensitivity, and by 44.2% and 16.9% at 100% sensitivity, respectively. WSS values aligned with these reductions. Due to the small number of references screened and the time needed to become initially trained to use Covidence, time savings was minimal. At 95% sensitivity, one included reference per CPG was missed, but guideline recommendations remained unchanged. Covidence AI-assisted screening may effectively support updating literature reviews for oncology CPGs by reducing workload without compromising the integrity of final recommendations. A 95% detection threshold may provide a practical balance between efficiency and accuracy.
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 distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.015 | 0.048 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.001 | 0.002 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 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 teacher head, 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".