Calculated versus measured albumin-creatinine ratio to predict kidney failure and death in people with chronic kidney disease

Abstract

Albumin-to-creatinine ratio (ACR), the preferred method to quantify proteinuria, can be calculated from urine dipstick protein or protein-to-creatinine ratio (PCR). The performance of calculated vs. measured ACR in predicting kidney failure and death without kidney failure in people with chronic kidney disease (CKD) is unknown. Here, we used population-based data from Alberta, Canada, to identify adults with incident moderate-severe CKD (sustained for more than 90 days) from 2008-04-01 to 2017-03-31, who had same-day measures of ACR and urine dipstick (ACR-dipstick cohort) or PCR (ACR-PCR cohort) in the two years before cohort entry. We followed participants until 2019-03-31 and trained competing risk models of kidney failure and death without kidney failure including age, sex, estimated glomerular filtration rate, diabetes, cardiovascular disease, and either measured or calculated ACR. Model performance was tested in cohorts created using the same algorithm in Manitoba, Canada. The ACR-dipstick and ACR-PCR cohorts included 18,731 and 4,542 people (training cohorts) and 821 and 1,831 people (testing cohorts), respectively. In internal and external testing, there was closer agreement between predictions based on measured vs. PCR-calculated ACR than between those based on measured vs. dipstick-calculated ACR. The dipstick-calculated ACR had higher Brier scores than measured ACR from year three for both outcomes, indicating worsening calibration. Models including measured or calculated ACR had similar discrimination: year one-to-five area under the receiver operating characteristic curve of 83-89% for kidney failure and 69-75% for mortality. Thus, if confirmed in different ethnic groups, calculated ACR can be used for risk predictions when the measured ACR is not available. PCR-calculated ACR may have superior performance to dipstick-calculated ACR.