Table 6 30-day readmission prediction on MIMIC-III. U, T, S represents unstructured data, temporal signals, and static information respectively

Baseline models

LR

\(T + S\)

0.144 (0.136, 0.152)

0.649 (0.646, 0.652)

0.071 (0.062, 0.080)

1

LR

U

0.142 (0.133, 0.151)

0.638 (0.634, 0.642)

0.070 (0.056, 0.084)

< 0.001

LR

\(U + T + S\)

0.144 (0.137, 0.151)

0.660 (0.657, 0.663)

0.072 (0.059, 0.085)

< 0.001

RF

\(T + S\)

0.123 (0.113, 0.133)

0.575 (0.559, 0.591)

0.060 (0.054, 0.066)

< 0.001

RF

U

0.117 (0.105, 0.129)

0.557 (0.539, 0.575)

0.059 (0.056, 0.062)

< 0.001

RF

\(U + T + S\)

0.118 (0.111, 0.125)

0.560 (0.543, 0.577)

0.059 (0.056, 0.062)

< 0.001

Deep models

Fusion-CNN

\(T + S\)

0.155 (0.146, 0.164)

0.657 (0.650, 0.664)

0.077 (0.073, 0.081)

0.0208

Fusion-CNN

U

0.163 (0.160, 0.166)

0.663 (0.660, 0.666)

0.078 (0.077, 0.079)

< 0.001

Fusion-CNN

\(U + T + S\)

0.164 (0.161, 0.167)

0.671 (0.668, 0.674)

0.080 (0.076, 0.084)

< 0.001

Fusion-LSTM

\(T + S\)

0.149 (0.146, 0.152)

0.653 (0.651, 0.655)

0.074 (0.071, 0.077)

0.0158

Fusion-LSTM

U

0.158 (0.154, 0.162)

0.641 (0.635, 0.647)

0.075 (0.072, 0.078)

0.0076

Fusion-LSTM

\(U + T + S\)

0.160 (0.151, 0.169)

0.674 (0.672, 0.676)

0.079 (0.076, 0.082)

< 0.001