Multivariate longitudinal data for survival analysis of cardiovascular event prediction in young adults: insights from a comparative explainable study

Table 2 Predictive performance of all models on 35 variables (mean and 95% empirical bootstrap interval)

Strategy	Model	iAUC	C-index	Last AUC
Time-series (TS) massive feature extraction	RSF on TS-extracted features	0.808 (0.790, 0.826)	0.778 (0.757, 0.801)	0.758 (0.733, 0.784)
Time-series (TS) massive feature extraction	LASSO-Cox on TS-extracted features	0.744 (0.711, 0.781)	0.713 (0.686, 0.739)	0.701 (0.674, 0.727)
Recurrent neural network	Dynamic-DeepHit	0.794 (0.764, 0.825)	0.767 (0.745, 0.789)	0.762 (0.733, 0.792)
Trajectory clustering	RSF on trajectory clustering data	0.793 (0.772, 0.816)	0.741 (0.721, 0.76)	0.725 (0.705, 0.744)
Data concatenation	RSF on concatenated data	0.797 (0.778, 0.817)	0.766 (0.745, 0.788)	0.751 (0.725, 0.779)
Joint modeling	JMBayes	Did not converge
Last observed values	RSF on Y15 data	0.793 (0.773, 0.812)	0.750 (0.729, 0.77)	0.731 (0.705, 0.76)
	Cox on Y15 data	0.778 (0.758, 0.804)	0.75 (0.733, 0.769)	0.728 (0.705, 0.752)
	Cox on Y15 data	0.793 (0.772, 0.818)	0.748 (0.73, 0.763)	0.727 (0.707, 0.745)
Reference (Y0 data)	RSF on Y0 data	0.754 (0.73, 0.777)	0.721 (0.698, 0.743)	0.699 (0.672, 0.726)
	Cox on Y0 data	0.748 (0.724, 0.773)	0.709 (0.686, 0.73)	0.685 (0.654, 0.716)
	LASSO-Cox on Y0 data	0.739 (0.713, 0.768)	0.698 (0.678, 0.717)	0.678 (0.645, 0.711)

The best scores are bolded. iAUC: integrated AUC, LASSO-Cox: Cox Proportional Hazards penalized by LeAst Shrinkage and Selection Operator. JMBayes Joint modeling with Bayesian approach, RSF Random Survival Forest

ISSN: 1471-2288