Automatically Summarizing Evidence from Clinical Trials: A Prototype Highlighting Current Challenges
This work addresses the challenge of efficiently summarizing clinical trial evidence for healthcare professionals, but it is incremental as it builds on prior work and highlights current limitations rather than achieving a breakthrough.
The authors tackled the problem of automatically summarizing evidence from clinical trials by developing TrialsSummarizer, a system that retrieves and ranks trials based on sample size and quality, then generates summaries using neural models. The result showed that while the models produced fluent and relevant summaries, they often introduced unsupported statements, making them currently unsuitable for clinical use.
We present TrialsSummarizer, a system that aims to automatically summarize evidence presented in the set of randomized controlled trials most relevant to a given query. Building on prior work, the system retrieves trial publications matching a query specifying a combination of condition, intervention(s), and outcome(s), and ranks these according to sample size and estimated study quality. The top-k such studies are passed through a neural multi-document summarization system, yielding a synopsis of these trials. We consider two architectures: A standard sequence-to-sequence model based on BART and a multi-headed architecture intended to provide greater transparency to end-users. Both models produce fluent and relevant summaries of evidence retrieved for queries, but their tendency to introduce unsupported statements render them inappropriate for use in this domain at present. The proposed architecture may help users verify outputs allowing users to trace generated tokens back to inputs.