. 2021 Sep 9;4:134. doi: 10.1038/s41746-021-00504-6

Table 2.

Summary of performance of COMPOSER on the (A) ICU and (B) ED cohorts when evaluated in a sequential prediction setting.

	SEN	SPC	PPV	NPV	DOR	AUC
(A) ICU cohorts
Hospital-A ICU test N^a=3,406; S^a=767 (22.5%)	91.6^a	93.0^b	38.0^b	98.8^b	53.5^b (1.088)	0.953
N^b=157,527; S^b=6499 (4.1%)	80.1^b	93.0^b	38.0^b	98.8^b	53.5^b (1.088)	0.953
Hospital-A Temporal ICU N^a=3,596; S^a=733 (20.4%)	92.3^a	93.6^b	35.5^b	99.0^b	57.2^b (1.090)	0.953
N^b=171,242; S^b=6156 (3.6%)	79.3^b	93.6^b	35.5^b	99.0^b	57.2^b (1.090)	0.953
Hospital-B ICU N^a=45,812; S^a=7,913 (17.3%)	91.3^a	90.7^b	24.3^b	99.1^b	35.9^b (1.027)	0.925
N^b=1,884,383; S^b=56,637(2.9%)	78.9^b	90.7^b	24.3^b	99.1^b	35.9^b (1.027)	0.925
(B) ED cohorts
Hospital-A ED test N^a=19,807; S^a=1,624 (8.2%)	95.6^a	93.5^b	20.1^b	99.5^b	51.8^b (1.109)	0.945
N^b=162,504 S^b=7691 (4.7%)	78.1^b	93.5^b	20.1^b	99.5^b	51.8^b (1.109)	0.945
Hospital-A Temporal ED N^a=19,945; S^a=1,795 (9.0%)	96.0^a	92.4^b	20.8^b	99.5^b	49.2^b (1.107)	0.940
N^b=158,045; S^b=8766 (5.5%)	80.0^b	92.4^b	20.8^b	99.5^b	49.2^b (1.107)	0.940
Hospital-B ED N^a=330,299; S^a=14,454 (4.4%)	90.5^a	94.7^b	13.4^b	99.6^b	51.2^b (1.031)	0.938
N^b=2,362,426; S^b=64032(2.7%)	70.5^b	94.7^b	13.4^b	99.6^b	51.2^b (1.031)	0.938

Decision threshold corresponding to 80% sensitivity on the Hospital-A training cohort (0.561 and 0.414 for the ICU and ED cohorts respectively).

N Number of patients within the cohort, S Number of septic patients.

^aPatient-wise.

^bhourly window-wise.

SEN Sensitivity, SPC Specificity, PPV Positive predictive value, NPV Negative predictive value, DOR Diagnostic odds ratio with 95% confidence interval.