AutoForest: Automatically Generating Forest Plots from Biomedical Studies with End-to-End Evidence Extraction and Synthesis

Forest plots are the standard visual output of meta-analyses, summarizing effect sizes across multiple clinical trials in a single chart. The bottleneck AutoForest targets is not just extraction but harmonization: different studies report outcomes in incompatible formats, and reconciling them before computation has historically required a statistician, not just a careful reader.

This sits squarely in a cluster of clinical NLP work Modelwire has been tracking. The Llama-3 fine-tuning paper on hospital stay summarization (covered the same day) addresses a structurally similar problem: aggregating fragmented, multi-source clinical text into coherent structured output. Both projects treat domain-specific LLM adaptation as the mechanism for collapsing expert workflows. The self-harm surveillance paper from Australian emergency departments adds another data point, showing that LLM generalization across institutional contexts is now a design requirement, not a bonus. AutoForest extends this pattern upstream into the research pipeline itself, rather than the point-of-care record.

The credibility test here is whether AutoForest's extracted effect sizes and confidence intervals match those in published meta-analyses on a held-out benchmark. If the authors release that validation dataset publicly, independent replication will settle whether this is a reliable tool or a promising prototype.