Unsupervised k-means clustering identified routine clinical biomarkers for classifying chronic kidney disease in patients with chronic heart failure.

INTRODUCTION

We sought to identify a diagnostic panel based on routine clinical biomarkers that can distinguish chronic kidney disease (CKD) from non-CKD among patients who have chronic heart failure (HF) and predict changes in kidney function in this patient population.

METHODS

A total of 432 patients with chronic HF were enrolled at Ningde Municipal Hospital of Ningde Normal University in China. The k-means clustering method was applied to identify a diagnostic panel capable of distinguishing CKD from non-CKD and predicting changes in kidney function after 1 year in patients with chronic HF.

RESULTS

The k-means clustering method identified 2 distinct subgroups among patients with chronic HF, demonstrating 71.59% concordance with actual CKD diagnostic labels. Five biomarkers showed a statistically significant difference between identified subgroups: albumin, lymphocyte percentage, hemoglobin, cholesterol, and apolipoprotein A. The 5-biomarker panel can distinguish CKD from non-CKD in patients with chronic HF at an area under the curve (AUC) of 0.81. Furthermore, this panel demonstrated predictive value for kidney function changes, with an AUC of 0.87 for identifying improved kidney function and 0.75 for predicting worsening kidney function.

DISCUSSION: Our findings suggested that a panel of routine clinical biomarkers can enhance the detection of CKD in patients with chronic HF and may provide additional prognostic value beyond traditional assessments, such as estimated glomerular filtration rate.