A new tool for predicting the probability of chronic kidney disease from a specific value of estimated GFR

Abstract

Objective. To demonstrate how patients’ probability of having chronic kidney disease (CKD) stage 3–5 (measured GFR <60 mL/min/1.73 m2) can be predicted from a specific value of estimated glomerular filtration rate (eGFR). Material and methods. The probability of CKD stage 3–5 was predicted from a logistic regression model (n = 850) using three different eGFR prediction equations: Lund-Malmö, MDRD and CKD-EPI. Population weighting was used to illustrate how this probability varies in three different populations: original sample (55% true prevalence of CKD stage 3–5), a screening (6.7% prevalence) and a CKD population (84% prevalence). Results. All three eGFR-equations had high classification ability (area under the receiver-operating-characteristic curve = 97%). The probability of CKD stage 3–5 increased with decreasing eGFR, varied substantially among the populations studied and to some extent between the eGFR-equations. Using the Lund-Malmö equation as illustration, the probability of CKD stage 3–5 is > 90% only when eGFR is <38 mL/min/1.73 m2 in a screening population, whereas it is > 90% already when eGFR is <51 mL/min/1.73 m2 in a CKD population. Conversely, the probability of CKD stage 3–5 is <10% if eGFR > 59 mL/min/1.73 m2 in a screening population, whereas it is <10% only when eGFR is > 88 mL/min/1.73 m2 in a CKD population. Conclusion. Instead of reporting diagnostic accuracy as sensitivity, specificity, and predictive values, actual eGFR supplemented with the probability that it represents a true GFR <60 mL/min/1.73 m2 may be more valuable for physicians. Clinical (pre-test) probability in the population must be considered when predicting this probability.