Table 2.

Comparison with state-of-the-art AI models for sepsis prediction.

	Model	Performance metrics					Data source
		${\text{t}}_{\text{onset}}$ (h)	Accuracy (%)	Sensitivity	Specificity	AUROC	Vitals	Lab${}^{\mathrm{a}}$	Dem.${}^{\mathrm{b}}$
17CinC 2019	LSTM	4	84.5	−	0.66	0.8	8	26	6
18CIBM 2016	Ensemble	3	82.7	0.81	0.9	0.83	9	0	0
19CinC 2019	Random forest	6	74.6	−	−	0.63	8	26	6
20CinC 2019	Regression	6	86.4	0.30	0.97	0.87	8	26	6
21CCM 2018	Survival model	4	67	0.85	0.67	0.85	16	30	19
22JAMIA 2020	RNN	4	−	0.84	0.80	0.94	9	39	36
23Nature Comm. 2021	Random forest	4	−	0.87	0.89	0.92	5	6	4
15EMBC 2020	GRU	6	99.8	0.94	0.98	0.97	8	26	0
24J.Elect.cardiology 2017	Regression	4	61	0.55	0.85	0.78	8	0	7
25CIBM 2017	LSTM	3	93	0.94	0.91	0.93	8	0	1
16ICHI 2018	LSTM+CNN	3	91.5	0.97	0.86	0.92	6	37	35
This work	Late fusion	4	90${}^{\mathrm{c}}$	0.90${}^{\mathrm{c}}$	0.90${}^{\mathrm{c}}$	0.90${}^{\mathrm{c}}$	1	0	4
			93${}^{\mathrm{d}}$	0.90${}^{\mathrm{d}}$	0.95${}^{\mathrm{d}}$	0.99${}^{\mathrm{d}}$	1	0	5

${}^{\mathrm{a}}$ Includes laboratory test results and culture results.

${}^{\mathrm{b}}$ includes demographics and co-morbidities.

${}^{\mathrm{c}}$ late fusion of demographics and ECG.

${}^{\mathrm{d}}$ late fusion of demographics, co-morbidity and ECG.