Table 6 Performance comparison of DNFCR models versus deepsurvcr and coxph models.

From: Development of a deep learning model for survival prediction in heart failure: competing risk and frailty model

Model performance comparison

Event

ΔC-index (95% CI)

Z-score

p-value

ΔIBS

ΔINBLL

DNFCR-PF vs. CoxPH

HF

+ 0.05 (0.01 to 0.09)

2.14

0.032

-0.02

+ 0.12

Other causes

+ 0.06 (0.02 to 0.10)

2.31

0.021

-0.02

+ 0.10

DNFCR-NF vs. CoxPH

HF

+ 0.04 (0.005 to 0.075)

2.01

0.044

-0.01

+ 0.09

Other causes

+ 0.05 (0.01 to 0.09)

2.18

0.029

-0.02

+ 0.08

DeepSurvCR vs. CoxPH

HF

+ 0.04 (0.001 to 0.079)

1.96

0.049

-0.02

+ 0.07

Other causes

+ 0.05 (0.01 to 0.09)

2.05

0.040

-0.02

+ 0.06

DNFCR-PF vs. DNFCR-NF

HF

+ 0.03 (0.005 to 0.055)

1.99

0.046

-0.01

+ 0.03

Other causes

+ 0.01 (-0.01 to 0.03)

0.87

0.384

0.00

+ 0.02

DNFCR-PF vs. DeepSurvCR

HF

+ 0.01 (0.001 to 0.019)

1.97

0.048

0.00

+ 0.05

Other causes

-0.01 (-0.03 to 0.01)

0.45

0.652

0.00

+ 0.04

  1. (I) All advanced models showed statistically significant improvements over CoxPH (p < 0.05 for all comparisons), (II) DNFCR-PF demonstrated superior performance for mortality from heart failure prediction (ΔC-index + 0.05, 95% CI: 0.01–0.09), (III) Between-model differences for mortality from other causes predictions were non-significant (p > 0.05 for DNFCR-PF vs. DNFCR-NF).