Screening Automation for Systematic Reviews: A 5-Tier Prompting Approach Meeting Cochrane’s Sensitivity Requirement
Bibliographic record
Abstract
Systematic Reviews are essential for synthesizing evidence from multiple studies, but the process, particularly the title and abstract screening phase, is time-consuming and labour-intensive. Traditional machine learning methods for automating this phase often fall short of the sensitivity required by Cochrane, which is set at greater than 0.99. This paper introduces a novel 5-tier prompting approach leveraging a foundational Large Language Model to automate the screening process. First, each study is assigned to one of five classes based on its likelihood of meeting predefined inclusion and exclusion criteria. Using a specified threshold, these classifications are then converted into binary decisions. This approach minimizes the risk of excluding relevant papers while automatically excluding the majority of irrelevant ones.Evaluation conducted on 5,643 records from four published systematic reviews resulted in zero wrong excludes when compared to human full-text screening decisions. The executed experiments resulted in a 68% average reduction in human workload, which enables a 50% decrease in the time needed to complete the screening process, all without compromising the accuracy of the results. These findings suggest that the 5-tier prompting approach offers a promising solution for enhancing the efficiency of systematic reviews.
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 | 0.141 | 0.393 |
| Meta-epidemiology (narrow) | 0.003 | 0.003 |
| Meta-epidemiology (broad) | 0.004 | 0.005 |
| Bibliometrics | 0.014 | 0.008 |
| Science and technology studies | 0.002 | 0.002 |
| Scholarly communication | 0.008 | 0.008 |
| Open science | 0.004 | 0.008 |
| Research integrity | 0.004 | 0.003 |
| Insufficient payload (model declined to judge) | 0.010 | 0.005 |
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".