Background: Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia that is associated with increased risk of stroke, heart failure and mortality. The increasing availability of electronic medical records (EMRs) for clinical research offers opportunities to generate timely real-world evidence to enhance patient care and facilitate research. Previous EMR-derived AF patient cohorts that had relied on International Classification of Diseases (ICD) codes have limited AF case identification. We have generated and assessed a novel method to enhance EMR-derived AF by using electronic prescribing records.
Methods: This was a population-based cohort study of EMR data that anonymously used a patient-specific identifier, the Community Health Index (CHI), to link the community-dispensed prescriptions maintained by the Health Informatics Centre (HIC) in Scotland. These records contain detailed prescribing information on all residents of Tayside region (population >400,000) since 1993. This dataset was linked to other clinical datasets that included echocardiography data, hospital admissions (Scottish Morbidity Record) and mortality data (General Registry Office). Electronic prescription records were filtered for patients prescribed a direct oral anticoagulant (DOAC) or warfarin. Merging with Scottish Morbidity Records (SMR), cases were excluded if an ICD-10 code for deep vein thrombosis (DVT), pulmonary embolism (PE) or mechanical heart valve was present. Following Caldicott guardian approval, we conducted a validation and concordance exercise to assess the validity of this prescribing-enhanced EMR-derived AF.
Results: From electronic prescribing records, we identified 2,798 patients who were prescribed a DOAC or warfarin. After excluding based on ICD-10 code for DVT, PE or mechanical heart valve, an AF subset of 2,573 patients was established. AF cases from the SMRs using ICD-10 codes only to phenotype for AF yielded 854 patients. Combining these two methodologies yielded 2,839 unique patients after accounting for duplicates. An open-chart validation and concordance exercise of 34 patients showed a positive predictive value of 97% and a false positivity rate of 3%.
Conclusions: Identifying AF cases by prescribing record yielded superior sample sizes with high positive predictive value. Identification of AF through ICD codes alone may not identify a representative sample for EMR research. ❑
