Table 3 Comparison of prediction performance between traditional statistics and machine learning models using both non-invasive and invasive variables.

From: Invasive and non-invasive variables prediction models for cardiovascular disease-specific mortality between machine learning vs. traditional statistics

Prediction models

Sex

Time dependent mean AUC

(95% CI)

C-index

(95% CI)

Integrated Brier Score

(95% CI)

Traditional statistics

Cox regression

All

0.833 (0.815–0.854)

0.829 (0.815–0.854)

0.048 (0.044–0.051)

Male

0.814 (0.792–0.844)

0.809 (0.792–0.844)

0.058 (0.052–0.062)

Female

0.853 (0.802–0.897)

0.858 (0.802–0.897)

0.027 (0.023–0.030)

Cox regression with elastic net penalty

All

0.813 (0.788–0.834)

0.810 (0.788–0.834)

0.049 (0.047–0.053)

Male

0.806 (0.777–0.832)

0.802 (0.777–0.832)

0.058 (0.054–0.065)

Female

0.846 (0.787–0.895)

0.852 (0.787–0.895)

0.027 (0.023–0.031)

Machine learning

Random Survival Forest

All

0.844 (0.825–0.863)

0.840 (0.825–0.863)

0.048 (0.045–0.051)

Male

0.819 (0.792–0.845)

0.812 (0.792–0.845)

0.058 (0.053–0.065)

Female

0.862 (0.819–0.906)

0.862 (0.819–0.906)

0.027 (0.023–0.032)

Gradient Boosting Survival

All

0.841 (0.819–0.858)

0.838 (0.819–0.858)

0.058 (0.053–0.062)

Male

0.818 (0.792–0.838)

0.810 (0.792–0.838)

0.071 (0.067–0.077)

Female

0.903 (0.883–0.920)

0.861 (0.821–0.901)

0.032 (0.028–0.037)

Survival Tree

All

0.819 (0.799–0.838)

0.817 (0.799–0.838)

0.049 (0.046–0.053)

Male

0.793 (0.762–0.815)

0.786 (0.762–0.815)

0.060 (0.055–0.067)

Female

0.819 (0.760–0.872)

0.823 (0.760–0.872)

0.029 (0.023–0.033)

  1. Models adjusted for age, sex (for all models only), waist to height ratio, hypertension status, diabetes status, physical activity level, fasting glucose, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, and triglyceride.
  2. Abbreviations: AUC, the area under the receiver operating characteristic curve; CI, confidence intervals.
Back to article page