Abstract P415: Validation of an Algorithm to Identify End-Stage Kidney Disease Cases in Electronic Health Record Data

Abstract

Introduction: End-stage kidney disease (ESKD) is an important clinical outcome. However, there is limited work on validating algorithms to detect ESKD cases within electronic health record (EHR) data. Hypothesis: Algorithms using diagnostic and procedure codes in EHR data can accurately identify ESKD cases. Methods: Using data from Geisinger linked to the United States Renal Data System (USRDS), we developed algorithms to identify individuals with ESKD (either dialysis or kidney transplant, whichever occurred first) and validated them against ESKD cases ascertained from the USRDS data from January 1, 1996, through June 28, 2018 (gold standard). We calculated sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). Results: To identify dialysis cases, we required 1) diagnosis codes of stage 5 chronic kidney disease or ESKD with dialysis codes within seven days or 2) three dialysis codes lasted longer than a month with any two consecutive codes occurring within 3 months. To identify kidney transplant cases, we required 1) one procedure code of kidney transplant or 2) one diagnosis code of kidney transplant in inpatients or problem list. Among 572,574 individuals (mean age, 41.6 years; female, 55.4%; White, 92.9%), the algorithms identified 4,187 ESKD cases (3,875 dialysis and 792 kidney transplant cases), while there were 4,529 ESKD cases (4,368 dialysis and 772 kidney transplant cases) by USRDS data. Our ESKD identifying algorithms’ sensitivity, specificity, PPV, and NPV were 71.1%, 99.8%, 76.9%, and 99.8%, respectively. Median (interquartile range) duration between incident dates by the algorithm and by the USRDS was -2 (-21, 86) days ( Table ). Conclusions: The algorithms developed by diagnostic and procedure codes accurately identified ESKD cases with high specificity in EHR data. Further validation is required in different health systems. Table The validity of the algorithms to identify ESKD cases IQR: interquartile range