Additional file 1 of Accelerating the pace and accuracy of systematic reviews using AI: a validation study

Abstract

Supplementary Material 1: Supplemental digital content: Table S1: STROBE Statement—checklist of items that should be included in reports of observational studies. Supplementary digital content: Table S2. Number of past human decisions of studies used for Review Copilot's performance metrics and validation. Supplementary digital content: Table S3. The performance of Review Copilot for all decisions in 4 datasets during the title/abstract screening and full-text screening stages. Supplementary digital content: Figure S1. Confusion matrices of the four validation datasets (A–D) for the title and abstract screening. Figure legend: This figure depicts the concordance and non-concordance between the gold-standard, human decision for abstract/title inclusion/exclusion (x-axis) compared to the AI/Review Copilot decision (y-axis) for each dataset. A is Cara et al. (Fiber) (2021), B is Cara et al. (Juice) (2021), C is Galaviz et al. (2022), and D is Meijboom et al. (2022). Supplementary digital content: Figure S2. Confusion matrices of two of the datasets (A,B) for full-text screening. Figure legend: This figure depicts the concordance and non-concordance between the gold-standard, human decision for full-text inclusion/exclusion (x-axis) compared to the AI/Review Copilot decision (y-axis) for each dataset. A is Cara et al. (Fiber) (2021), and B is Cara et al. (Juice) (2021). Datasets C (Galaviz et al. (2022)), and D (Meijboom et al. (2022)) could not be evaluated individually at the full-text level because of software restrictions used for these reviews.

Ähnliche Arbeiten