Assessing the Effectiveness and Scalability of Fast Healthcare Interoperability Resource-Based Granular Data Segmentation Technology.

Abstract

Background: Protecting sensitive health information, particularly substance use data, is essential to preserving patient trust, improving care engagement, and ensuring compliance with federal confidentiality laws such as 42 CFR Part 2. Manual segmentation of sensitive data remains burdensome and error-prone, creating a need for scalable, automated, and patient-centered data segmentation technologies.

Objectives: The objective of this study is to evaluate the technical effectiveness, scalability, and computational performance of SHARES, an open-source data segmentation platform based on HL7 Fast Healthcare Interoperability Resources (FHIR), designed to support patient-directed, granular control over sharing sensitive substance use-related health information.

Methods: We generated 11,519 synthetic FHIR patient records using Synthea and conducted physician-led classifications of 1,848 unique clinical terms contained in those patient records. These classifications were used to assign code-category-confidence rules applied by the SHARES segmentation engine. We deployed the SHARES platform in simulation environments to label and segment substance use-related data using varying confidentiality thresholds. We benchmarked throughput, scalability, and computational performance across large FHIR datasets, and measured segmentation accuracy via physician manual review.

Results: Physicians categorized 232 data items as substance use or "maybe" substance use data elements. The SHARES segmentation engine processed 197.6 million FHIR resources across 11 confidentiality thresholds. Overall, 17 million data items (8.64%) were flagged as sensitive. SHARES achieved a throughput of 10.36 patient bundles per second, or approximately 18,909 FHIR resources per second, on a commodity laptop. A full segmentation run on 11,519 patients had a cost of under one cent.

Conclusion: This study demonstrates the feasibility and scalability of automated, rule-based data segmentation within a FHIR-native architecture, using a physician-derived sensitivity construct for substance use as an applied test case. SHARES offers high-throughput, low-cost classification performance, and supports auditability and deterministic decision-making, which are key strengths for compliance with 42 CFR Part 2. Future enhancements will incorporate clinical quality language (CQL)-based logic to better reflect real-world clinical reasoning and contextual data interpretation.