Nomogram for Predicting Cardiovascular Mortality in Incident Peritoneal Dialysis Patients: An Observational Study

Abstract

Cardiovascular mortality risk is high for peritoneal dialysis (PD) patients but it varies considerably among individuals. There is no clinical tool to predict cardiovascular mortality for PD patients yet. Therefore, we developed a cardiovascular mortality risk nomogram in a PD patient cohort. We derived and internally validated the nomogram in incident adult PD patients randomly assigned to a training (N = 918) or a validation (N = 460) dataset. The nomogram was built using the LASSO Cox regression model. Increasing age, history of cardiovascular disease or diabetes were consistent predictors of cardiovascular mortality. Low hemoglobin and serum albumin, high hypersensitive C-reactive protein and decreasing 24 hours urine output were identified as non-traditional cardiovascular risk predictors. In the validation dataset, the above nomogram performed good discrimination (1 year c-statistic = 0.83; 3 year c-statistic = 0.78) and calibration. This tool can classify patients between those at high risk of cardiovascular mortality (high-risk group) and those of low risk (low-risk group). Cardiovascular mortality was significantly different in the internal validation set of patients for the high-risk group compared to the low-risk group (HR 3.77, 2.14-6.64; p < 0.001). This novel nomogram can accurately predict cardiovascular mortality risk in incident PD patients.

Figure 1

Enrollment and outcomes of the cohort. Abbreviations: CVD; cardiovascular disease.

Figure 2

Nomogram to predict risk of cardiovascular mortality in peritoneal dialysis patients.

Figure 3

(A) Time-dependent ROC curves and (B) Kaplan-Meier survival curves in the training sets on the basis of the nomogram. Abbreviations: ROC; receiver operator characteristic, AUC; area under the curve.

Figure 4

Plots depict the calibration of the nomogram in terms of agreement between predicted and observed 3-year outcomes in the training sets. Model performance is shown by the plot, relative to the 45-degree line, which represents perfect prediction.

Figure 5

(A) Time-dependent ROC curves and (B) Kaplan-Meier survival curves in the internal testing sets on the basis of the nomogram. Abbreviations: ROC; receiver operator characteristic, AUC; area under the curve.

Figure 6

Plots depict the calibration of the nomogram in terms of agreement between predicted and observed 3-year outcomes in the internal testing sets. Model performance is shown by the plot, relative to the 45-degree line, which represents perfect prediction.