Quantifying interoperability: An analysis of oncology practice electronic health record data variability.

Abstract

e18080 Background: Implementation of electronic health records (EHRs) has engendered a large quantity of machine-readable data. However, different practices choose different EHR vendors and the same vendor product may be implemented differently at each practice. Motivated by the desire to facilitate appropriate integration of data, our goal was to describe and quantify the consistency and variation of structured data within EHRs. Methods: De-identified and aggregated CancerLinQ data from 47 practices regarding the standards and variability of structured data including race, diagnoses, encounters, cancer staging, selected cancer-relevant medications, lab values and biomarkers were analyzed. EHR represented included ARIA, MOSAIQ, Allscripts, Centricity, Epic, Intellidose, NextGen, and OncoEMR. Results: De-identified EHR implementations included 23 A, 12 B, C and 5 other vendors. Only 6 practices (13%) used non-standard race representation. All practices used ICD-9/10 for diagnoses. There was variability in coding of encounters. Sixteen practices always used CPT, 5 practices always used SNOMED CT and 26 practices used multiple standards. Multiple staging systems were used. An average of 48% (range 11%-; including patients staged more than once) of patient records included coded staging information. Only one practice used a standard (LOINC) for laboratory data. No standards were used for medications ordered/administered or biomarkers. The table shows the number of distinct names for selected lab tests, medications and biomarkers across systems. Conclusions: In this cross-sectional sample, standards are used consistently for diagnoses and encounter data, often for race and rarely for medications, laboratory tests or biomarkers.[Table: see text]