Exploring the influencing factors of abdominal aortic calcification events in chronic kidney disease (CKD) and non-CKD patients based on interpretable machine learning methods

Abstract

Background: Calcification is prevalent in CKD patients, with abdominal aortic calcification (AAC) being a strong predictor of coronary calcification. We aimed to identify key calcification factors in CKD and non-CKD populations using machine learning models.

Methods: Data from the National Health and Nutrition Examination Survey (NHANES), including demographics, blood and urine tests, and AAC scores, were analyzed using machine learning models. The Shapley additive explanations (SHAP) analysis was applied to interpret the models.

Results: Among 505 CKD and 2,582 non-CKD participants, common key factors for calcification included age, estimated glomerular filtration rate (eGFR), smoking history, blood glucose levels (Glu), Ca*P and the urine albumin-to-creatinine ratio (UACR). Age, smoking history and eGFR were the top-ranking features in the model for both two groups. Inflammatory markers such as monocyte-to-lymphocyte ratio (MHR), monocyte-to-high-density lipoprotein ratio (MLR) and neutrophil-to-lymphocyte ratio (NLR) were more significant in CKD group. Trigger points for AAC events were identified: in CKD, eGFR of 90 mL/min/1.73 m², MHR values of 0.5 and 0.75, MLR values of 0.25, and SP of 120 mmHg; in non-CKD, eGFR of 105 mL/min/1.73 m², Ca*P values of 40, UACR values of 10, and TG of 200 mg/dL.

Conclusions: Regardless of CKD status, age, smoking history, and eGFR are key determinants of calcification. In the CKD population, inflammatory markers are more significant than in the non-CKD group.

Keywords: Abdominal aortic calcification; Chronic kidney disease; Machine learning; Prediction.