Computational-Assisted Systematic Review and Meta-Analysis (CASMA): Effect of a Subclass of GnRH-a on Endometriosis Recurrence

Background: Evidence synthesis facilitates evidence-based medicine. Without information retrieval techniques, this task is impossible due to the vast and expanding literature. Objective: Building on prior work, this study evaluates an information retrieval-driven workflow to enhance the efficiency, transparency, and reproducibility of systematic reviews. We use endometriosis recurrence as an ideal case due to its complex and ambiguous literature. Methods: Our hybrid approach integrates PRISMA guidelines with computational techniques. We applied semi-automated deduplication to efficiently filter records before manual screening. This workflow synthesized evidence from randomised controlled trials on the efficacy of a subclass of gonadotropin-releasing hormone agonists (GnRH'as). A modified splitting method addressed unit-of-analysis errors in multi-arm trials. Results: Our workflow efficiently reduced the screening workload. It took only 11 days to fetch and filter 812 records. Seven RCTs were eligible, providing evidence from 841 patients in 4 countries. The pooled random-effects model yielded a Risk Ratio (RR) of 0.64 (95% CI (0.48 to 0.86)), with non-significant heterogeneity ($I^2=0.00\%$, $\tau=0.00$); i.e., a 36% reduction in endometriosis recurrence. Sensitivity analyses and bias assessments supported the robustness of our findings. Conclusion: This study demonstrates an information-retrieval-driven workflow for medical evidence synthesis. Our approach yields valuable clinical results while providing a framework for accelerating the systematic review process. It bridges the gap between clinical research and computer science and can be generalized to other complex systematic reviews.