Towards cross-application model-agnostic federated cohort discovery

Area of Science:

• Biomedical Informatics
• Health Data Interoperability
• Clinical Trial Recruitment

Background:

• Federated data networks are crucial for large-scale clinical research.
• Cohort discovery tools facilitate patient identification for trials.
• Interoperability challenges hinder data sharing across diverse healthcare systems.

Purpose of the Study:

• To demonstrate the interoperability of Leaf and Shared Health Research Information Network (SHRINE) cohort discovery tools.
• To adapt Leaf to function within a SHRINE federated network and translate queries for heterogeneous data models.

Main Methods:

• Leaf was modified to act as a node in a SHRINE network.
• SHRINE queries were dynamically translated to other data models using a Python script.
• 500 ENACT network queries were used for evaluation, with 100 for refinement.

Main Results:

• 91.1% of translated queries yielded results within 5% of the i2b2 data model.
• 91.3% query recall was achieved.
• Most discrepancies were attributed to script/ETL errors, with Leaf's translation function later corrected.

Conclusions:

• Leaf and SHRINE can effectively interoperate within federated data networks.
• The study validates the dynamic translation of cohort discovery queries across heterogeneous data models.
• This approach enhances the potential for large-scale, multi-institutional cohort identification.