Fd-scope helps clinicians question cancer trial data with verified ai answers
Answering clinicians' questions over trial evidence tables with verifiable, feedback-driven language models
Artificial Intelligence
Summary
Clinicians often use tables summarizing many clinical trials to find useful information, but these tables can miss some important details they want to ask about. The authors created FD-SCoPE, a system that uses language models to answer both simple and complex questions by explaining how it got each answer and learning from expert corrections. When tested on cancer trial data, FD-SCoPE answered nearly all clinician-like questions accurately and improved over time with feedback. This approach makes it easier for clinicians to get trustworthy, detailed answers about trials.
What this means in practice
- •For clinical informatics teams: Enable clinicians to query clinical trial evidence tables with explanations and improved answers through expert feedback.
- •For medical ai developers: Build AI tools that combine natural language understanding with executable queries to answer complex medical trial questions transparently.$Commercial implications: Allows creation of verifiable clinical decision support products that explain trial evidence answers for healthcare providers.
Authors
Manan Roy Choudhury, Suparno Roy Chowdhury, Swastik Sahoo, Muhammad Ali Khan, Kaneez Zahra Rubab Khakwani, Mohamad Bassam Sonbol, Irbaz Bin Riaz, Vivek Gupta
Abstract
Systematic reviews condense clinical trials into evidence tables, yet clinicians can interrogate these tables only through database queries, and many questions concern attributes that the table does not record, such as a drug's target class or a harmonised endpoint. Here we introduce FD-SCoPE, a language-model framework that answers both kinds of question, exposes the query, the selected trials and the derivation rule behind every answer, and learns from expert corrections. On an oncology evidence table of 159 immune checkpoint inhibitor trial records, FD-SCoPE completed all 140 clinician-style tasks (alternatives, 90.7-97.9%). For questions needing derived attributes it retrieved 99.3% of relevant trial records at a positive predictive value of 89.8% and outperformed four alternative approaches (derived-value F1 77.7% versus 64.8-73.4%). Corrections on 299 questions, simulated from reference answers, raised F1 on 1,201 unseen questions from 77.9% to 84.9%. Language models coupled with executable queries, verified programs and expert feedback can give clinicians auditable access to trial evidence.