Abstract

I read with genuine admiration the article by Reason et al, which presents an innovative exploration of generative artificial intelligence as an “artificial intelligence statistician” capable of autonomously selecting appropriate models and executing network meta-analyses (NMAs). The study is particularly compelling in demonstrating that a large language model (LLM)-based automated process could consistently identify the correct statistical framework across all 16 National Institute for Health and Care Excellence TSD2 case studies, generate appropriate model specifications, and reproduce network diagrams and executable code with high fidelity. The finding that the artificial intelligence (AI)-generated incremental treatment effects closely replicated published estimates—for example, yielding d₁₂ = −0.26 (95% CrI −0.36, −0.16) in the fixed-effects blocker example, identical to the National Institute for Health and Care Excellence TSD2 results—is both technically impressive and highly encouraging for the future automation of evidence-synthesis workflows.