Table 2 Comparison of model’s accuracy for mortality prediction

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

	Development cohort					Validation cohort
Lead time window	Sensitivity	Specificity	PPV	NPV	Accuracy	Sensitivity	Specificity	PPV	NPV	Accuracy
PROMPT
6 h	0.846	0.963	0.663	0.986	0.953	0.800	0.850	0.288	0.982	0.846
12 h	0.800	0.946	0.555	0.983	0.935	0.800	0.890	0.336	0.985	0.884
24 h	0.800	0.931	0.454	0.985	0.922	0.849	0.887	0.334	0.989	0.884
48 h	0.800	0.834	0.224	0.986	0.832	0.800	0.752	0.177	0.983	0.755
60 h	0.800	0.882	0.268	0.988	0.878	0.800	0.772	0.190	0.983	0.773
GBDT
6 h	0.800	0.933	0.509	0.982	0.922	0.800	0.805	0.238	0.981	0.805
12 h	0.801	0.898	0.398	0.982	0.891	0.800	0.854	0.276	0.984	0.850
24 h	0.800	0.854	0.283	0.983	0.850	0.800	0.818	0.227	0.984	0.817
48 h	0.800	0.769	0.172	0.985	0.771	0.800	0.629	0.126	0.979	0.640
60 h	0.800	0.693	0.123	0.985	0.698	0.800	0.551	0.107	0.976	0.567
LSTM
6 h	0.800	0.951	0.588	0.982	0.939	0.800	0.770	0.209	0.981	0.772
12 h	0.800	0.888	0.374	0.981	0.881	0.800	0.782	0.204	0.982	0.783
24 h	0.800	0.828	0.251	0.983	0.826	0.800	0.740	0.170	0.982	0.743
48 h	0.800	0.729	0.150	0.984	0.733	0.800	0.537	0.104	0.976	0.554
60 h	0.800	0.626	0.103	0.983	0.635	0.800	0.505	0.098	0.974	0.524
PIM 3
Total	0.800	0.617	0.392	0.909	0.661	0.800	0.799	0.298	0.974	0.799
Subset 1^*	0.806	0.643	0.218	0.964	0.661	0.818	0.754	0.182	0.984	0.758
Subset 2^**	0.800	0.643	0.200	0.966	0.659	0.818	0.754	0.182	0.984	0.758

PPV positive predictive value, NPV negative predictive value, 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