Accuracy of Identification of Cardiovascular Events with International Classification of Diseases (ICD) Diagnosis Codes vs. Physician Adjudication in CKD and ESKD

Background: Cardiovascular disease (CVD) in chronic kidney disease (CKD) is an important end-point for research studies. Adjudication by a central committee is considered the most rigorous approach of ascertaining CVD outcomes, but it is resource intensive. We aimed to compare the accuracy of unadjudicated ICD codes vs. physician adjudication for hospitalizations for heart failure (HF), stroke, atrial fibrillation (AF), and myocardial infarction (MI) in those with CKD and ESKD. Methods: Using the Chronic Renal Insufficiency Cohort (CRIC), we evaluated 31,521 hospitalization events. We determined the positive predictive value (PPV), negative predictive value, sensitivity and specificity of primary code position as well as when using all diagnosis codes from the hospitalization. We calculated hazard ratios for the CVD outcomes determined by each type of ascertainment (ICD-10 codes vs. physician adjudicated) for known CVD risk factors. Results: In those with CKD, comparing primary ICD diagnosis codes with the adjudicated outcomes, we found the PPV to be 82.5% for HF, 80.8% for ischemic stroke, and 87.6% for AF (Table 1). The PPVs were generally similar for participants who were and those who were not on dialysis. MI events by ICD code had much lower PPV rates, 38.4% in non-dialysis dependent CKD, 47.3% in dialysis dependent, and 62.5% in transplant participants. HR of outcomes for those with CVD risk factors diabetes, body mass index, hypertension, age, and eGFR were similar whether ascertained through ICD codes or physician adjudication. Conclusion: When using ICD codes to determine CVD outcomes in CKD, PPV was near 80% for HF, AF, and ischemic stroke, and was similar amongst those with ESKD. These data may inform the approach to CVD outcome ascertainment in future studies of kidney disease patients.