Table 1 Summary of model mortality detection performance

From: A deep learning model for real-time mortality prediction in critically ill children

	Development cohort				Validation cohort
Lead time window	AUROC	95% CI	AUPRC	95% CI	AUROC	95% CI	AUPRC	95% CI
PROMPT
6 h	0.965	± 0.006	0.831	± 0.018	0.922	± 0.004	0.716	± 0.016
12 h	0.948	± 0.009	0.745	± 0.029	0.945	± 0.004	0.701	± 0.023
24 h	0.933	± 0.009	0.733	± 0.027	0.946	± 0.005	0.605	± 0.024
48 h	0.899	± 0.013	0.570	± 0.041	0.849	± 0.007	0.360	± 0.023
60 h	0.887	± 0.018	0.565	± 0.052	0.881	± 0.011	0.445	± 0.031
GBDT
6 h	0.944	± 0.008	0.767	± 0.022	0.877	± 0.005	0.499	± 0.032
12 h	0.927	± 0.008	0.684	± 0.028	0.915	± 0.005	0.605	± 0.022
24 h	0.908	± 0.014	0.612	± 0.032	0.897	± 0.007	0.442	± 0.021
48 h	0.853	± 0.014	0.452	± 0.031	0.805	± 0.009	0.342	± 0.025
60 h	0.831	± 0.022	0.419	± 0.051	0.790	± 0.012	0.403	± 0.035
LSTM
6 h	0.945	± 0.010	0.808	± 0.019	0.875	± 0.006	0.547	± 0.039
12 h	0.915	± 0.016	0.703	± 0.031	0.870	± 0.012	0.520	± 0.034
24 h	0.889	± 0.013	0.644	± 0.032	0.837	± 0.012	0.348	± 0.032
48 h	0.844	± 0.014	0.530	± 0.029	0.770	± 0.013	0.348	± 0.027
60 h	0.814	± 0.025	0.429	± 0.050	0.759	± 0.019	0.353	± 0.034
PIM 3
Total	0.767	–	0.509	–	0.881	–	0.500	–
Subset 1^*	0.787	–	0.315	–	0.876	–	0.462	–
Subset 2^**	0.785	–	0.298	–	0.876	–	0.462	–

AUROC area under the receiver operating characteristic curve, CI confidence interval, AUPRC area under the precision-recall curve, PROMPT pediatric risk of mortality prediction tool, GBDT Gradient Boosting Decision Trees, LSTM Long Short-Term Memory, PIM 3 Pediatric Index of Mortality 3
^*Subset of the cohort with data of at least 48 h
^**Subset of the cohort with data of at least 60 h

Back to article page

ISSN: 1364-8535

Contact us

Submission enquiries: journalsubmissions@springernature.com