Table 2 Comparison of predictive performance for overall survival using TGI metrics and TDNODE-derived metrics.

From: Explainable deep learning for tumor dynamic modeling and overall survival prediction using Neural-ODE

Model

Input features

C-index evaluated via

5-fold cross-validation

(median±MAD)

C-index evaluated on the test set

TGI-OS.ML

TGI metrics only

0.73 ± 0.01

0.68

TGI metrics + 11 baseline covariates

0.78 ± 0.02

0.75

TDNODE-OS.ML

TDNODE metrics

0.84 ± 0.02

0.82

TDNODE metrics + 11 baseline covariates

0.86 ± 0.02

0.84

  1. Prediction of OS using TGI metrics compared with that of TDNODE encoder output metrics, both with and without eleven baseline covariates. The inclusion of these covariates significantly improved the prediction of OS in the TGI-OS model, whereas TDNODE metrics appear to capture the information provided by these covariates. In both cases, TDNODE-generated metrics are superior for the prediction of patients’ OS when compared to that of TGI-generated metrics. The variability of each metric is measured using median absolute deviation (MAD).
  2. The bolded row indicates the model that gave rise to superior predictive performance.
Back to article page