Can we predict mortality of older patients with advanced chronic kidney disease?

Abstract

Background: Studies have shown that patients with advanced chronic kidney disease (CKD) desire to know their prognosis. Although many predictive tools exist, they are not widely used in clinical practice as their applicability is not clear, particularly for older patients.

Aim: To evaluate the validity of existing prognostic tools in older patients with advanced CKD.

Methods: This was a single-centre retrospective validation cohort study on patients aged ≥65 years with grade 4 or 5 CKD between 2009 and 2018. The Ivory, Schmidt and Cohen models were tested by comparing predicted and actual mortality outcomes. Discriminatory ability was measured by area under the receiver operating characteristic curve (AUC), and calibration was assessed by the Hosmer-Lemeshow statistic. The global performance of the predictive regression models was assessed using the measure of goodness of fit with the coefficient of determination Nagalkerke R².

Results: A total of 387 patients with a median age of 80 years (IQR 74-85) were included in the study. For the Ivory and Schmidt models, the c-statistic was 0.617 (95% CI 0.47-0.74) and 0.60 (95% CI 0.47-0.71) respectively, indicating poor discrimination. The Hosmer-Lemeshow statistic was 1.42 (P = 0.22) and 1.59 (P = 0.15) respectively, indicating reasonable calibration. The Cohen model had an overall poor predictive value.

Conclusion: Existing prognostication tools demonstrate overall suboptimal performance in our validation cohort of older patients. Further research is needed for the development of a prognostic tool specific to the older advanced CKD population.