Predictive Model for High Coronary Artery Calcium Score in Young Patients with Non-Dialysis Chronic Kidney Disease

Figures

• 

  Figure 1. Flow chart representing the workflow of the machine learning analysis. We utilized the derivation cohort to construct the predictive models and external validation cohort to assess the universality of final predictive model. Abbreviations: AUROC, area under the receiver operator curve; CAC, coronary artery calcium; KNOW-CKD, Korean Cohort Study for Outcomes in Patients with Chronic Kidney Disease; SHAP, Shapley additive explanations.

• 

  Figure 2. Comparing area under the receiver operator curve of prediction models. The RandomForest model showed the highest area under the receiver operator curve of the coronary artery calcification score prediction. Abbreviations: AUC, area under the receiver operator curve; SVM, support vector machine; XGBoost, extreme gradient boost.

• 

  Figure 3. Shapley additive explanations (SHAP) value for RandomForest model. The top 20 features showed the highest SHAP value in the RandomForest model. Abbreviations: SHAP, Shapley additive explanations; FBS, fasting blood glucose; WHR, waist–hip ratio; HDL, high-density lipid; UPCR, urine protein to creatinine ratio; ALP, alkaline phosphatase; LDL, low-density lipid; SBP, systolic blood pressure; TC, total cholesterol; CRP, C-reactive protein; eGFR, estimated glomerular filtration rate; TG, triglyceride; DBP, diastolic blood pressure.

• 

  Figure 4. Area under the receiver operator curve of final predictive RandomForest model with test dataset. The area under the receiver operator curve of the final predictive model showed a reliable result. Abbreviation: AUC, area under the receiver operator curve.