This research introduces a two-step retrieval-augmented system that automatically translates clinical inclusion/exclusion criteria into SQL queries for patient cohort generation from electronic health records.

• Combines criteria parsing, two-level retrieval augmentation, and medical concept standardization
• Achieves 0.75 F1-score in cohort identification
• Creates patient funnels to visualize selection process
• Eliminates manual query writing while maintaining clinical accuracy

This innovation significantly accelerates medical research workflows by automating a previously manual, error-prone process, enabling faster patient recruitment and more efficient observational studies in healthcare settings.

Generating patient cohorts from electronic health records using two-step retrieval-augmented text-to-SQL generation