Automatically inferring treatment effects from reports of Randomized Controlled Trials (RCTs)

This material is based upon work supported by the National Science Foundation (NSF) under Grant No 1750978: “CAREER: Structured Scientific Evidence Extraction: Models and Corpora”.

Disclaimer: Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

Project duration (expected): 7/1/2018 – 6/30/2023.

Project PI: Byron Wallace.

Project goals We aim to design natural language processing (NLP) models that can “read” through clinical trial reports extract reported findings from these (i.e., which drugs actually work?). Such models, if successful, would allow domain experts (physicians, in this case) to harness the entirety of the published evidence base to inform treatment decisions, something not currently possible due to published evidence being predominantly unstructured.

Research challenges Designing and training models to make sense of the findings reported in clinical trials requires core technical innovations to realize models that can jointly extract entities and infer relationships between them over lengthy technical articles, and new corpora with which to train and evaluate them. Further, this problem necessitates models that provide explicit support for predictions, i.e., “rationales”. In this project we seek to address both the application (models to extract structured evidence from the trials literature) and the underlying methodological challenges (more transparent neural NLP models that can rationalize outputs).

This page last updated: 5/25/2020; please contact Byron Wallace with any questions.