Development of prediction models for cardiovascular disease mortality risk in maintenance hemodialysis patients based on nomogram and CART algorithm.

BACKGROUND

Patients on maintenance hemodialysis (MHD) face a dramatically elevated risk of cardiovascular death, which is 10 - 20 times higher than in the general population. To address this high risk, we developed and validated a prediction model to accurately estimate cardiovascular disease (CVD) mortality and guide preemptive clinical management.

MATERIALS AND

METHODS

This study retrospectively collected data from MHD patients at the First Affiliated Hospital of Chengdu Medical College from 2016 to 2021 (Approval No. CYFYEC-C-005), including demographic characteristics, medical history, biochemical indicators, and echocardiogram indices. Variables were screened using univariate logistic regression and stepwise regression to construct a nomogram model. The dataset was randomly divided (6 : 4) into training and validation sets, and a classification and regression tree (CART) decision tree model was also constructed. Both models' discrimination, calibration, and clinical utility were evaluated.

RESULTS

The nomogram identified systolic blood pressure, uric acid, total cholesterol, diabetes, myoglobin, serum albumin, and procalcitonin as predictors, with an AUC of 0.947 (95% CI: 0.903 - 0.991) and good clinical applicability. The CART model identified serum albumin, procalcitonin, and myoglobin as predictors, categorizing the population into four groups. AUC values were 0.933 (95% CI: 0.851 - 1.000) in the training set and 0.774 (95% CI: 0.612 - 0.936) in the validation set.

CONCLUSION

In conclusion, this study consistently identified serum albumin, procalcitonin, and myoglobin as key factors associated with CVD mortality risk in MHD patients. Both models demonstrated promising predictive performance in our cohort. These findings suggest the potential of such models to inform clinical risk assessment. However, external validation in larger, multi-center studies is necessary before these tools can be considered for direct clinical decision-making.