DQueST

DQueST transforms unstructured ClinicalTrials.gov eligibility criteria into dynamic, user-specific questionnaires to filter and identify potentially eligible clinical trials.

Key Features:

• Criteria library curation: Curates a comprehensive library from ClinicalTrials.gov by extracting narrative eligibility criteria and storing them in an organized format.
• Narrative criteria extraction: Applies information extraction techniques to convert free-text eligibility criteria into structured representations.
• Clinical entity normalization: Normalizes clinical entities to standard concepts.
• Criteria clustering: Clusters related criteria to group semantically similar eligibility conditions.
• Dynamic question generation algorithm: Generates real-time questions by selecting criteria from the library one at a time based on relevance scores.
• Relevance scoring for exclusion power: Computes relevance scores that reflect each criterion's ability to exclude ineligible trials.
• Iterative trial refinement: Iteratively updates and filters the set of potential trials by eliminating those deemed ineligible according to user responses.

Scientific Applications:

• Information overload reduction: Simulation experiments across 10 diseases filtered 60%–80% of initial trial listings after approximately 50 questions.
• Question relevance evaluation: An evaluation found that on average 79.7% of generated questions were relevant to the queried conditions.
• Exclusion accuracy assessment: Random samples of trials excluded by DQueST yielded an estimated exclusion accuracy rate of 63.7%.
• Comparative retrieval performance: In a study with five mock patient profiles, DQueST retrieved trials with 1.465 times higher density of eligible trials compared to existing search engines.
• Free-text to interactive-question framework: Transforms free-text eligibility criteria into interactive questionnaires to improve the precision of clinical trial searches.

Methodology:

The approach comprises two phases: criteria library curation—extracting narrative criteria from ClinicalTrials.gov, normalizing clinical entities to standard concepts, clustering related criteria, and storing them—and real-time dynamic question generation that selects criteria one at a time based on relevance scores reflecting exclusion power and iteratively refines the candidate trial set as users answer.