Development and validation of identification algorithms for five autoimmune diseases using electronic health records: a retrospective cohort study in China

TL;DR

This study developed and validated algorithms to identify five autoimmune diseases using electronic health records in China, showing good accuracy for most conditions.

Contribution

The study presents the first validated algorithms for identifying autoimmune diseases using EHR data in China.

Findings

The algorithm for Hashimoto’s thyroiditis achieved high accuracy with 97.44% sensitivity and 98.28% PPV.

Combining data sources improved performance for IBD and ITP, achieving PPV above 70%.

The T1D algorithm using admission and outpatient records had 84.09% sensitivity and 74.00% PPV.

Abstract

This study aims to assess the identification algorithms for five autoimmune diseases—Hashimoto’s thyroiditis, inflammatory bowel disease (IBD), primary immune thrombocytopenia (ITP), rheumatoid arthritis (RA), and type 1 diabetes (T1D)—using the Yinzhou Regional Health Information Platform (YRHIP) in China.

Diagnostic data was extracted from YRHIP’s population registry (2010-2021), combining ICD-10 codes and Chinese medical terminology from outpatient, inpatient, and discharge records. Algorithms were validated through chart reviews, adhering to global clinical guidelines. Cases were adjudicated using electronic case report forms. We evaluated algorithm performance based on sensitivity and positive predictive value (PPV), with a 70% PPV threshold for optimization.

Among all reviewed cases, we identified 136 cases for Hashimoto’s thyroiditis, 65 for IBD, 76 for ITP, 130 for RA, and 43 for T1D. Algorithm performance varied across diseases: the final algorithm for Hashimoto’s thyroiditis achieved optimal accuracy (sensitivity 97.44%, PPV 98.28%), followed by RA (sensitivity 100.00%, PPV 76.92%). Algorithms for IBD and ITP required synthesis of multiple data sources to achieve acceptable performance (IBD: sensitivity 79.66%, PPV 70.15%; ITP: sensitivity 62.50%, PPV 70.00%). For T1D, the final algorithm utilizing both admission and outpatient records yielded satisfactory results (sensitivity 84.09%, PPV 74.00%).

This study presents the first validated algorithms for identifying autoimmune diseases using EHR data in China, demonstrating satisfactory performance (PPV >70%) across all diseases. Our findings demonstrate that a combination of data sources is crucial for accurate case identification in complex autoimmune conditions, providing an important methodological foundation for future real-world studies in Chinese populations.