Table 2 Performance on synthetic benchmark data

From: Deep representation learning for clustering longitudinal survival data from electronic health records

Method

ACC

NMI

ARI

CI

k-means+Cox PH

0.334 ± 0.002*(0.008)

0.0003 ± 0.0001*(0.008)

-0.0003 ± 0.0003*(0.008)

0.505 ± 0.001*(0.008)

SSC

0.3334 ± 0.0004*(0.008)

0.0003 ± 0.0002*(0.008)

0.00005 ± 0.00050*(0.008)

 

SCA

0.336 ± 0.007*(0.012)

0.03 ± 0.02*(0.012)

0.006 ± 0.040*(0.012)

0.639 ± 0.006*(0.008)

DSM

0.3327 ± 0.0004*(0.008)

0.0005 ± 0.0005*(0.008)

-0.0005 ± 0.001*(0.008)

0.49 ± 0.01*(0.008)

RDSM

0.502 ± 0.002*(0.007)

0*(0.007)

0*(0.007)

0.500 ± 0.001*(0.008)

VaDeSC-MLP

0.57 ± 0.05*(0.008)

0.37 ± 0.04*(0.008)

0.44 ± 0.03*(0.008)

0.79 ± 0.02*(0.008)

VaDeSC-EHR

0.64 ± 0.04

0.72 ± 0.04

0.66 ± 0.07

0.77 ± 0.01

  1. Comparison between VaDeSC-EHR and other methods adapted for clustering longitudinal survival data, in terms of balanced accuracy (ACC), normalized mutual information (NMI), adjusted Rand index (ARI), and concordance index (CI). Reported ± is one standard deviation and the significance of the difference (p-value < 0.05) between VaDeSC-EHR and the other methods is indicated by an asterisk. Significance is based on a two-sided Mann–Whitney U test. Detailed p-values are shown in the brackets.
Back to article page