From many to one: Development and evaluation of a unified patient match algorithm

Background To integrate disease surveillance registries, the NYC DOHMH developed the Electronic Disease Reporting Infrastructure (EDRI). Previously, each disease program, including Sexually Transmitted Diseases, Tuberculosis, Communicable Diseases, Vaccine Preventable Diseases, and HIV Epidemiology Program had utilized different methods and technologies to match patients with no automated method to link patients across registries.

Methods To develop and evaluate the EDRI probabilistic patient match algorithm, each program submitted a sample gold-standard dataset of matches and non-matches as defined by pre-existing de-duplication methods. These records were de-duplicated using EDRI's algorithm. The outcome of the EDRI match was compared to the gold-standard results to determine the sensitivity, specificity, and positive predictive value (PPV) of the algorithm by program and in aggregate. After reviewing discordant results for trends within and across programs, the algorithm was adjusted to achieve a priori defined goals of at least 90% sensitivity and 95% specificity overall.

Results A total of 433,240 records, with 80,379 sets, were submitted from all programs, including 72,671 matches, and 7,708 non-matches. Across all programs, the EDRI patient match algorithm correctly identified 94.9% of the matches (range by program: 83.2%-95.0%) and 98.2% of the non-matches (range by program: 91.7%-98.4%). The overall PPV was 99.8% (range by program: 80.5%-99.9%).

Conclusion This analysis demonstrates that a unified match algorithm can be successfully achieved from disparate data sources. The resulting unified data system allows programs to leverage data from other sources and facilitates the analysis of co-morbidities.