(P-426) Evaluating the Completeness of Electronic Health Records in Identifying Nonfatal Cardiovascular Events in Taiwan

Background: Electronic health record (EHR) databases contain rich data for conducting comparative effectiveness research. However, patients may receive care outside the study EHR system, and the EHR discontinuity could lead to substantial information bias on variables essential for the studies.

Objectives: To examine the completeness of EHR in identifying nonfatal cardiovascular events using a hospital database in Taiwan.

Methods: We first included patients who were diagnosed with hypertension, diabetes or dyslipidemia and had an encounter to the National Taiwan University Hospital (NTUH) between January 2009 and July 2017. The NTUH EHRs were linked to Taiwan’s National Health Insurance Research Database (NHIRD) to assess the EHR data completeness of three major cardiovascular events: hospitalization for myocardial infarction (MI), ischemic stroke (IS), and heart failure (HHF). We identified each patient’s first nonfatal MI, IS, and HHF in the NHIRD which occurred between their first and last NTUH encounters (at least 90 days apart). Patients with an inpatient or outpatient diagnosis of the study event prior to the hospitalization were excluded. For each event, capture rate by the NTUH EHR system was calculated. Sensitivity analyses were performed in patients who had regular encounters (with intervals < 90 or 120 days) and/or encounters in certain departments (internal medicine, family medicine, neurology, and geriatrics/gerontology).

Results: A total of 3,012 MIs, 6,536 ISs, and 3,891 HHFs were identified in the NHIRD. After excluding patients with a prior event diagnosis, there were 2,234 (74.2%) MIs, 4,460 (68.2%) ISs, and 898 (23.1%) HHFs. Of the 2,234 MIs identified in the claims, 759 (34.0%) occurred outside the NTUH, and only 189 patients revisited the NTUH with a record of MI diagnosis after the out-of-system event. Of the 4,460 patients with an IS, 1,598 (35.8%) occurred outside the NTUH; 342 revisited the NTUH with a record of IS diagnosis. Of the 898 patients with a HHF, 429 (47.8%) occurred outside the NTUH. Restricting patients to those with regular encounters and/or encounters in certain hospital departments decreased the sample sizes substantially; however, the EHR event capture rate increased from 66.0% to up to 81.9% for MI, from 64.2% to up to 82.3% for IS, and from 52.2% to up to 75.9% for HHF.

Conclusions: EHR data are limited in capturing major cardiovascular events such as MI, IS, and HHF. Employing requirements of patients’ encounter departments and encounter intervals may improve the EHR data completeness. Further studies are needed to examine how the EHR data incompleteness affects the validity of pharmacoepidemiologic studies.