Using artificial intelligence to predict adverse outcomes in emergency department patients with hyperglycemic crises in real time

Hsu, Chin-Chuan; Kao, Yuan; Hsu, Chien-Chin; Chen, Chia-Jung; Hsu, Shu-Lien; Liu, Tzu-Lan; Lin, Hung-Jung; Wang, Jhi-Joung; Liu, Chung-Feng; Huang, Chien-Cheng

doi:10.1186/s12902-023-01437-9

Table 2 Comparison of performance among the random forest, logistic regression, SVM, KNN, LightGBM, and MLP algorithms for adverse outcomes in ED patients with hyperglycemic crises

From: Using artificial intelligence to predict adverse outcomes in emergency department patients with hyperglycemic crises in real time

Outcomes and algorithms	Accuracy		Sensitivity		Specificity		PPV		NPV		F1		AUC (95%CI)		p-value*
Outcomes and algorithms	Train	Test	Train	Test	Train	Test	Train	Test	Train	Test	Train	Test	Train	Test	p-value*
Sepsis or septic shock
MLP	0.789	0.790	0.828	0.791	0.750	0.789	0.768	0.636	0.814	0.890	0.797	0.705	0.854 (0.839–0.869)	0.852 (0.825–0.880)	0.910
Random forest	0.920	0.779	0.948	0.780	0.892	0.778	0.898	0.621	0.945	0.884	0.922	0.691	0.984 (0.980–0.987)	0.848 (0.821–0.875)	< 0.001
LightGBM	0.868	0.764	0.896	0.803	0.840	0.745	0.848	0.595	0.889	0.891	0.871	0.683	0.948 (0.940–0.956)	0.842 (0.815–0.870)	< 0.001
SVM	0.855	0.765	0.907	0.764	0.803	0.766	0.821	0.602	0.896	0.874	0.862	0.674	0.945 (0.937–0.953)	0.818 (0.787–0.849)	< 0.001
KNN	0.800	0.738	0.889	0.744	0.711	0.734	0.755	0.566	0.865	0.861	0.817	0.643	0.890 (0.877–0.902)	0.816 (0.786–0.846)	< 0.001
Logistic regression	0.718	0.720	0.690	0.720	0.746	0.720	0.731	0.545	0.706	0.847	0.710	0.620	0.789 (0.771–0.806)	0.802 (0.770–0.833)	0.487
ICU admission
MLP	0.692	0.680	0.714	0.688	0.670	0.680	0.684	0.120	0.700	0.971	0.698	0.205	0.744 (0.728–0.760)	0.743 (0.663–0.822)	0.973
LightGBM	0.960	0.676	0.924	0.667	0.997	0.677	0.997	0.116	0.929	0.970	0.959	0.198	0.985 (0.981–0.989)	0.737 (0.671–0.803)	< 0.001
Random forest	0.969	0.668	0.958	0.667	0.980	0.668	0.980	0.113	0.959	0.969	0.969	0.194	0.996 (0.995–0.997)	0.730 (0.661–0.799)	< 0.001
Logistic regression	0.728	0.654	0.727	0.646	0.730	0.654	0.729	0.107	0.728	0.967	0.728	0.183	0.801 (0.786–0.815)	0.706 (0.626–0.786)	0.024
SVM	0.689	0.611	0.770	0.604	0.607	0.612	0.662	0.090	0.725	0.960	0.712	0.157	0.766 (0.751–0.782)	0.682 (0.598–0.765)	0.052
KNN	0.791	0.601	0.973	0.604	0.610	0.601	0.714	0.088	0.957	0.960	0.823	0.154	0.949 (0.942–0.955)	0.667 (0.585–0.749)	< 0.001
All-cause mortality
MLP	0.770	0.741	0.816	0.716	0.724	0.745	0.747	0.291	0.797	0.947	0.780	0.414	0.836 (0.823–0.850)	0.796 (0.755–0.837)	0.065
Random forest	0.952	0.740	0.940	0.716	0.965	0.744	0.964	0.290	0.941	0.947	0.952	0.412	0.990 (0.988–0.993)	0.790 (0.750–0.831)	< 0.001
LightGBM	0.924	0.690	0.884	0.716	0.964	0.686	0.961	0.250	0.892	0.943	0.921	0.371	0.977 (0.972–0.981)	0.771 (0.725–0.816)	< 0.001
SVM	0.925	0.709	0.896	0.706	0.953	0.709	0.950	0.262	0.902	0.943	0.923	0.382	0.982 (0.978–0.985)	0.761 (0.714–0.808)	< 0.001
KNN	0.780	0.715	0.932	0.716	0.629	0.716	0.715	0.268	0.902	0.945	0.809	0.390	0.907 (0.897–0.917)	0.761 (0.713–0.808)	< 0.001
Logistic regression	0.751	0.666	0.770	0.667	0.731	0.666	0.741	0.226	0.761	0.932	0.755	0.337	0.812 (0.797–0.827)	0.760 (0.716–0.805)	0.031

MLP Multilayer perceptron, LightGBM Light Gradient Boosting Machine, SVM Support vector machine, KNN K-nearest neighbors, ED Emergency department, PPV Positive predictive value, NPV Negative predictive value, F1, 2 × (precision × recall/precision + recall), AUC Area under the curve, CI Confidence interval, ICU intensive care unit
^*The DeLong test was used to compare the AUC between train and test models [27]

Back to article page

ISSN: 1472-6823

Contact us

General enquiries: journalsubmissions@springernature.com

BMC Endocrine Disorders

Contact us