A laboratory-based algorithm to predict future kidney function decline in older adults with reduced estimated glomerular filtration rate

Abstract

Reduced estimated glomerular filtration rate (eGFR) in older adults is common and may reflect normal aging or significant kidney disease. Our objective was to develop a predictive model to better triage these individuals using routine laboratory data. Using a large US laboratory data set, we calculated individual eGFR regression slopes for 43,523 individuals aged 60-75 years with baseline eGFRs between 30 and 59 mL/min/1.73m2. We developed general linear models to predict the eGFR regression slope using urine protein measurements and other routinely available laboratory data as dependent variables. We validated these models on a similar data set comprised of 11,979 individuals. In a model utilizing log10 urine albumin/creatinine (UACR), the variables that significantly predicted the eGFR regression slope were log10 UACR, initial eGFR, serum albumin, chloride, glucose, and aspartate aminotransferase (AST). In an otherwise identical model substituting log10 urine protein/creatinine (UPCR) for UACR, results were similar except that serum calcium was significant and AST was not. We analyzed the correspondence between actual eGFR regression slopes and those predicted by our models using receiver operator characteristic (ROC) statistics to calculate areas under the curves (AUC) for four eGFR slope cut points: -2, -3, -4, and -5mL/min/year. AUCs using the UACR and UPCR models ranged from 0.716to0.900 and 0.751to0.868, respectively, for the training data set. Results were nearly identical for the validation data set. Use of a laboratory-based predictive model of eGFR decline for older adults with eGFR 30-59 mL/min/1.73m2 may help distinguish between individuals with and without risk for further decline in kidney function.