Prediction of cardiac arrest in critically ill patients presenting to the emergency department using a machine learning score incorporating heart rate variability compared with the modified early warning score

Table 4 Discriminatory values of the machine learning score and the modified early warning score

Variable	ML score (95% CI)^a	MEWS (95% CI)^b	Difference (95% CI for difference)^c	P value
Cardiac arrest within 72 hours after presentation
Area under ROC curve	0.781	0.680	0.101 (0.006 to 0.197)	0.037
Sensitivity	81.4	74.4	7.0 (-11.1 to 21.9)	0.581
Specificity	72.3	54.2	18.1 (14.3 to 22.0)	< 0.001
Positive predictive value	12.5 (9.0 to 17.1)	7.4 (5.3 to 10.3)		< 0.001
Negative predicting value	98.8 (97.5 to 99.4)	97.8 (95.9 to 98.8)		0.133
Likelihood ratio (+)^d	2.94 (2.46 to 3.52)	1.62 (1.34 to 1.96)
Death after admission^e
Area under ROC curve	0.741	0.693	0.048 (-0.023 to 0.119)	0.185
Sensitivity	69.8	74.4	-4.7 (-16.7 to 7.4)	0.572
Specificity	73.9	55.7	18.2 (14.3 to 22.2)	< 0.001
Positive predictive value	21.5 (16.9 to 26.9)	14.7 (11.5 to 18.4)		< 0.001
Negative predicting value	96.0 (94.1 to 97.3)	95.5 (93.2 to 97.1)		0.608
Likelihood ratio (+)^d	2.67 (2.23 to 3.20)	1.68 (1.45 to 1.94)

CI, confidence interval; MEWS, modified early warning score; ML, machine learning; ROC, receiver operating characteristic. ^aA cutoff value of 60 and above was used for the ML score. ^bA cutoff value of 3 and above was used for the MEWS. ^cThe 95% confidence interval for the difference between the ML score and the MEWS for each diagnostic statistic was calculated, except for positive predictive value, negative predictive value, and likelihood ratio (+) that are not well established. ^dLikelihood ratio of a positive test. ^eIn-hospital death during current admission. 95% confidence interval or statistical test not computed because the method is not well established for the diagnostic statistics concerned.

ISSN: 1364-8535