Table 3 Comparison of the ROC-AUC of the used ML strategies in different patient groups.

From: Machine learning for fast identification of bacteraemia in SIRS patients treated on standard care wards: a cohort study

 

All patients

AB naïvety

Patients with 2 SIRS criteria

Patients with 3 SIRS criteria

Patients with 4 SIRS criteria

n

466

380

182

213

71

Bacteraemia rate

28.8%

30.5%

28.6%

27.2%

33.8%

PCT

0.729 (0.679–0.779)

0.734 (0.680–0.787)

0.679 (0.598–0.762)

0.756 (0.678–0.833)

0.751 (0.633–0.869)

rf

0.738 (0.606–0.870)

0.727 (0.548–0.905)

0.698 (0.349–0.999)

0.781 (0.573–0.988)

0.585 (0.188–0.981)

nn

0.698 (0.549–0.857)

0.688 (0.499–0.876)

0.640 (0.355–0.925)

0.714 (0.497–0.930)

0.583 (0.181–0.985)

en

0.654 (0.493–0.815)

0.627 (0.396–0.858)

0.594 (0.334–0.854)

0.690 (0.466–0.914)

0.612 (0.214–0.999)

  1. PCT = procalcitonin, rf = random forest, nn = neural network, en = elastic net.