Validation of an Algorithm to Identify End-Stage Kidney Disease in Electronic Health Records Data.

Abstract

RATIONALE & OBJECTIVES Accurate ascertainment of end-stage kidney disease (ESKD) in electronic health records (EHRs) data is important for much epidemiological research. This study aimed to develop and validate an algorithm using diagnosis and procedure codes to identify patients with ESKD (treated with maintenance dialysis or kidney transplantation) in EHRs data. STUDY DESIGN Study of diagnostic algorithms. SETTING & PARTICIPANTS The development cohort included 559,615 patients treated at the Geisinger Health System (January 1996-June 2018). The validation cohort included 767,186 patients treated at New York University Langone Health System (January 2018-December 2020). ALGORITHMS COMPARED An algorithm developed using diagnosis and procedure codes compared to a nominal gold standard designation within the United States Renal Data System (USRDS) data. The performance of the algorithm was characterized by sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). The dates of incident ESKD between the algorithm and USRDS were compared in a subset of cases. OUTCOMES ESKD (maintenance dialysis, prior recipient of a kidney transplant, or kidney transplantation surgery) cases. RESULTS In Geisinger, we developed the ESKD algorithm that identified 4,766 (0.85%) ESKD cases, while there were 5,155 (0.92%) ESKD cases reported by the USRDS. The sensitivity, specificity, PPV, and NPV of the algorithm were 73.9% (95% CI, 72.7-75.1%), 99.83% (99.82-99.84%), 79.9% (78.9-81.0%), and 99.76% (99.75-99.77%), respectively. When applying the algorithm to New York University Langone Health System data, the sensitivity, specificity, PPV, and NPV were 71.8% (95% CI, 70.7-73.0%), 99.95% (99.95-99.96%), 91.6% (90.8-92.4%), and 99.79 (99.78-99.80%), respectively. The median (interquartile range) difference between dates of incident ESKD (algorithms minus USRDS) were -3 (-21 to 83) days in Geisinger and 0 (-12 to 69) days in New York University Langone Health. LIMITATIONS Use of structured EHRs data only. CONCLUSIONS Algorithms combining diagnosis and procedure codes show high specificity and modest sensitivity for identifying patients with ESKD, providing a research tool to inform future EHR-based studies.