Gen AI in Systematic Literature Reviews: First Case Study on Gen AI in a NICE Submission

This session explores the transition of systematic literature reviews (SLRs) from traditional dual-review models to hybrid human–AI workflows, where AI tools conduct screening and evidence extraction in parallel with human reviewers, followed by quality control by a human reviewer. The discussion will consider how this approach can maintain or improve compliance with evolving regulatory standards, enhance efficiency, and reduce bias in evidence generation. (5 min) Barinder Singh will introduce the panel with a historical overview of systematic review methodologies and highlight the increasing pressure on evidence teams to deliver faster, scalable outputs without compromising on methodological rigor. He will also present the current regulatory landscape, reviewing NICE and CDA position papers on automation in SLRs. (5 min) Sven Klijn will compare legacy dual-human review with AI-augmented parallel review and outline automated SLR specs: source ingestion, transformer-based screening (human-in-the-loop), and PRISMA-grade audit trails. He’ll cover model governance and reproducibility with implications for compliance, transparency, and reproducibility. (10 min) Dr. Dilip Makhija will demonstrate how AI tools are operationalized for literature screening, data extraction, and QC, sharing real-world implementation data, accuracy benchmarks, and reviewer concordance. He will present a case study of an automated SLR used in an HTA submission, highlighting HTA feedback, regulatory alignment, and key lessons learned. (5 min) Dr. Raphael Sonabend-Friend will provide the HTA agency perspective, discussing how automation could be evaluated for HTA submissions, what evidence can support validation, and how agencies might distinguish between appropriate automation vs. insufficient oversight. (5 min) Audience Q&A and Panel Discussion.

Sponsored by Pharmacoevidence