Table 4.

Comparison of Accuracy of In‐Hospital Cardiac Arrest Prediction Model With Same Sensitivity Point

TTS	Sensitivity	Specificity	PPV	NPV	F‐Measure	MACHP	NRI (95% CI)
MEWS ≥3	63.0	79.9	0.5	99.9	1.0	0.293
DEWS ≥7.1	63.0	87.0	0.8	99.9	1.5	0.199	0.071 (0.061–0.082)
MEWS ≥4	49.3	86.8	0.6	99.9	1.2	0.198
DEWS ≥18.2	49.3	94.6	1.4	99.9	2.8	0.084	0.078 (0.067–0.089)
MEWS ≥5	37.3	90.6	0.6	99.9	1.3	0.143
DEWS ≥52.8	37.3	98.4	3.7	99.9	7.1	0.025	0.079 (0.068–0.090)
SPTTS	60.7	77.0	0.4	99.9	0.8	0.334
DEWS ≥8.0	60.7	88.3	0.8	99.9	1.6	0.180	0.151 (0.138–0.163)
Random forest	75.3	69.9	0.4	99.9	0.8	0.409
DEWS ≥3.0	75.3	77.0	0.5	99.9	1.0	0.354	0.071 (0.060–0.082)
Logistic regression	76.3	34.6	0.2	99.9	0.4	0.622
DEWS ≥2.9	75.7	76.5	0.5	99.9	1.0	0.360	0.413 (0.399–0.427)

CI indicates confidence interval; DEWS, deep learning–based early warning system score; MACHP, mean alarm count per hour per patient; MEWS, modified early warning score; NPV, negative predictive value; NRI, net reclassification index; PPV, positive predictive value; SPTTS, single‐parameter track‐and‐trigger system; TTS, track‐and‐trigger system.