Skip to main content
. 2021 Sep 30;9(9):e31311. doi: 10.2196/31311

Table 4.

Performance of each algorithm in the internal validation cohort.

Algorithm AUPRCa (95% CI) AUROCb (95% CI) Sensitivityc (95% CI) Precisionc (95% CI)



0.6 0.7 0.8 0.6 0.7 0.8
STEP-OPd 0.716
(0.708-0.723)		 0.961
(0.959-0.962)		 0.600
(0.591-0.609)		 0.700
(0.692-0.708)		 0.800
(0.793-0.806)		 0.742
(0.733-0.751)		 0.647
(0.640-0.655)		 0.502
(0.495-0.509)
Convolutional neural network 0.698
(0.690-0.705)		 0.955
(0.953-0.957)		 0.600
(0.591-0.608)		 0.700
(0.692-0.708)		 0.800
(0.793-0.806)		 0.717
(0.709-0.726)		 0.615
(0.606-0.622)		 0.466
(0.459-0.472)
Recurrent neural network 0.706
(0.698-0.715)		 0.958
(0.956-0.959)		 0.600
(0.591-0.608)		 0.700
(0.692-0.708)		 0.800
(0.793-0.806)		 0.738
(0.729-0.746)		 0.639
(0.631-0.647)		 0.488
(0.481-0.495)
Logistic regression 0.673
(0.665-0.682)		 0.948
(0.946-0.950)		 0.600
(0.592-0.609)		 0.700
(0.691-0.708)		 0.800
(0.793-0.807)		 0.711
(0.703-0.720)		 0.622
(0.614–0.630)		 0.481
(0.474-0.487)

aAUPRC: area under the precision-recall curve.

bAUROC: area under the receiver operating characteristic curve.

cSensitivity and precision values were evaluated at the thresholds for sensitivity of 0.6, 0.7, and 0.8.

dSTEP-OP: short-term event prediction in the operating room.