Machine learning analysis of serum biomarkers for cardiovascular risk assessment in chronic kidney disease.

BACKGROUND: Chronic kidney disease (CKD) patients show an increased burden of atherosclerosis and high risk of cardiovascular events (CVEs). There are several biomarkers described as being associated with CVEs, but their combined effectiveness in cardiovascular risk stratification in CKD has not been tested. The objective of this work is to analyse the combined ability of 19 biomarkers associated with atheromatous disease in predicting CVEs after 4 years of follow-up in a subcohort of the NEFRONA study in individuals with different stages of CKD without previous CVEs. METHODS: Nineteen putative biomarkers were quantified in 1366 patients (73 CVEs) and their ability to predict CVEs was ranked by random survival forest (RSF) analysis. The factors associated with CVEs were tested in Fine and Gray (FG) regression models, with non-cardiovascular death and kidney transplant as competing events. RESULTS: RSF analysis detected several biomarkers as relevant for predicting CVEs. Inclusion of those biomarkers in an FG model showed that high levels of osteopontin, osteoprotegerin, matrix metalloproteinase-9 and vascular endothelial growth factor increased the risk for CVEs, but only marginally improved the discrimination obtained with classical clinical parameters: concordance index 0.744 (95% confidence interval 0.609-0.878) versus 0.723 (0.592-0.854), respectively. However, in individuals with diabetes treated with antihypertensives and lipid-lowering drugs, the determination of these biomarkers could help to improve cardiovascular risk estimates. CONCLUSIONS: We conclude that the determination of four biomarkers in the serum of CKD patients could improve cardiovascular risk prediction in high-risk individuals.