Evidence Prime contributed to the sixth Evidence-Based Research (EBR) Conference, held online on 13–16 October 2026. Under the theme “Evidence-Based Research for All,” the meeting examined how transparent, systematic use of existing evidence can inform research decisions across disciplines.
Can AI help turn eligibility criteria into screening instructions?
Our team’s oral contribution on 16 October, Can AI Generate Screening Instructions? Evaluation of Automated Decision Logic for Study Selection, was authored by Ewelina Sadowska, Monika Opałek, Ewa Borowiack and Artur Nowak. It formed part of the conference’s SR Efficiency session. The final programme places the presentation alongside other work on improving systematic-review methods.
The study addressed a step that can be easy to overlook when discussing AI screening: translating a review’s eligibility criteria into operational questions. Those questions determine how the system distinguishes an eligible study from an ineligible or unclear case.
We compared automatically generated and expert-developed decision logic across seven published systematic reviews. Both workflows used the same eligibility criteria and the same AI screening model, which answered the decision questions before those answers were mapped to screening decisions.
Evaluate the instructions as well as the model
In this experimental comparison, median sensitivity was 92.5% with automatically generated instructions and 96.4% with expert-developed instructions. Sensitivity ranged from 76% to 100% in both workflows, and neither approach had a consistent advantage across individual reviews.
These findings suggest that instruction generation deserves evaluation as a distinct part of the screening workflow. They do not establish that expert instruction design can be replaced generally. Performance varied by review, and the comparison was limited to the seven reviews studied.
The published abstract gives the methods and results. The conference overview provides the wider context for this discussion of efficient and trustworthy evidence-based research.
