A Novel CGM Metric-Gradient and Combining Mean Sensor Glucose Enable to Improve the Prediction of Nocturnal Hypoglycemic Events in Patients with Diabetes

Table 1

The specificity and sensitivity of the prediction of nocturnal hypoglycemic events based on different metrics and algorithms (CGM).

pH	Method	LR		SVM		RF		LSTM
		SPE (%)	SEN (%)	SPE (%)	SEN (%)	SPE (%)	SEN (%)	SPE (%)	SEN (%)
15 min	Raw data	95.76	95.36	96.31	96.03	94.82	94.70	93.51	92.05
	MSG	85.89	90.07	86.06	90.07	80.21	90.07	86.06	90.07
	Gradient	52.00	90.73	54.51	89.40	49.02	90.07	51.32	89.40
	SD	24.24	91.39	26.83	89.40	13.53	89.40	22.17	91.39
	CV	69.75	76.82	89.42	54.97	40.84	90.07	39.94	92.72
	LAGE	21.37	91.39	21.37	91.39	16.15	90.73	21.37	91.39
	MAG	32.87	84.77	32.78	84.77	15.58	90.73	32.78	84.77
	LBGI	89.18	90.07	89.03	90.07	86.06	90.07	89.26	90.07
	LI	32.57	84.77	32.57	84.77	16.06	89.40	32.57	84.77
	MSG+gradient	97.12	98.01	97.40	98.01	96.07	96.03	96.88	97.35
	MSG+SD	93.61	94.70	93.82	94.04	92.42	92.72	94.13	94.04
	MSG+CV	87.81	90.07	88.43	90.07	93.02	93.38	87.61	90.73
	MSG+LAGE	94.26	94.70	94.67	94.70	93.00	94.04	94.18	95.36
	MSG+MAG	93.89	94.04	94.94	94.70	93.46	93.38	93.34	94.70
	MSG+LBGI	88.53	90.07	89.03	90.07	90.09	90.73	89.03	90.07
	MSG+LI	95.27	90.73	94.09	96.03	93.40	96.03	94.36	94.70
30 min	Raw data	83.70	90.07	85.43	90.07	83.99	90.07	85.50	90.07
	MSG	75.99	90.07	76.11	90.07	61.52	90.07	76.11	90.07
	Gradient	30.97	90.73	31.38	90.07	27.40	89.40	28.32	90.07
	SD	21.45	89.40	23.22	88.74	8.48	88.74	23.22	88.74
	CV	26.95	94.70	1.29	90.73	26.66	90.73	43.84	86.75
	LAGE	20.84	89.40	20.84	89.40	6.30	90.73	20.84	89.40
	MAG	31.75	82.12	31.76	82.12	17.17	90.07	43.91	69.54
	LBGI	82.92	90.07	83.12	90.07	81.38	89.40	83.34	90.07
	LI	42.70	70.20	16.84	92.05	10.08	88.74	31.49	82.12
	MSG+gradient	90.83	90.07	90.30	90.73	87.79	90.07	90.20	90.07
	MSG+SD	87.94	90.73	88.61	90.07	83.11	90.73	88.57	90.07
	MSG+CV	78.11	90.07	78.11	90.07	83.68	90.07	79.07	90.07
	MSG+LAGE	87.36	90.73	88.47	90.07	81.15	90.07	89.27	90.07
	MSG+MAG	86.02	90.07	88.15	90.07	79.79	90.07	87.53	90.07
	MSG+LBGI	79.26	90.73	77.00	90.07	81.24	90.07	81.34	90.07
	MSG+LI	87.99	90.73	88.48	90.07	78.93	90.73	79.58	90.07
45 min	Raw data	75.56	90.07	78.84	90.73	80.24	90.07	80.72	90.07
	MSG	65.66	90.07	65.46	90.07	54.54	90.07	65.11	90.07
	Gradient	12.05	90.73	11.09	91.39	21.34	90.07	11.43	90.73
	SD	20.58	89.40	22.42	88.08	7.03	90.07	19.91	89.40
	CV	75.75	57.62	10.82	60.93	26.56	90.73	53.56	78.81
	LAGE	19.95	89.40	19.94	89.40	7.77	89.40	19.95	89.40
	MAG	30.93	86.09	15.72	90.07	12.91	89.40	15.72	90.07
	LBGI	82.01	82.78	79.15	87.42	79.35	85.43	81.29	82.78
	LI	42.62	71.52	30.61	86.09	16.06	86.09	30.52	86.09
	MSG+gradient	82.42	90.07	81.15	90.73	79.79	90.07	82.84	90.07
	MSG+SD	81.79	91.39	82.14	90.07	74.84	90.07	82.19	90.07
	MSG+CV	65.84	90.07	66.98	90.07	73.32	90.07	67.02	90.07
	MSG+LAGE	81.55	90.73	81.95	90.73	72.27	91.39	81.32	90.73
	MSG+MAG	83.01	90.07	82.23	90.07	70.34	90.73	81.68	90.07
	MSG+LBGI	64.38	90.07	65.65	90.07	54.87	90.07	65.41	90.07
	MSG+LI	82.48	90.07	80.39	90.73	72.95	90.07	77.43	90.07
60 min	Raw data	71.68	90.07	70.65	90.73	64.49	90.73	70.19	90.07
	MSG	58.73	90.07	58.56	90.07	41.16	90.07	58.73	90.07
	Gradient	11.66	90.07	11.36	90.07	13.77	90.07	14.93	86.75
	SD	22.39	88.74	20.52	90.73	7.91	89.40	30.93	82.12
	CV	77.35	59.60	8.05	68.21	30.43	90.07	40.47	87.42
	LAGE	19.73	90.73	19.73	90.73	23.87	86.09	19.65	90.73
	MAG	30.62	84.11	30.63	84.11	21.83	87.42	30.64	84.11
	LBGI	81.43	80.13	82.32	78.15	79.45	79.47	80.71	80.79
	LI	49.75	67.55	30.33	84.11	18.29	86.09	41.69	73.51
	MSG+gradient	77.87	90.73	77.25	90.07	76.79	90.07	81.10	90.07
	MSG+SD	74.95	91.39	74.20	90.07	67.32	90.73	74.55	90.07
	MSG+CV	59.28	90.73	61.46	90.07	63.15	90.07	59.25	90.07
	MSG+LAGE	76.75	90.73	75.97	90.07	62.30	90.07	76.41	90.07
	MSG+MAG	77.75	90.73	78.85	90.07	69.74	90.07	76.89	90.73
	MSG+LBGI	57.46	90.07	58.71	90.07	45.14	90.07	58.10	90.07
	MSG+LI	76.51	90.73	76.67	90.73	70.87	90.07	78.90	90.07

PH: prediction horizon; SPE: specificity; SEN: sensitivity; LR: logistic regression; SVM: support vector machine; RF: random forest; LSTM: long short-term memory; MSG: mean sensor glucose; SD: standard deviations of sensor glucose; CV: coefficient of variation; LAGE: largest amplitude of glycemic excursion; MAG: mean absolute glucose; LBGI: low blood glucose index; LI: lability index. The CGM length which was used for the prediction of nocturnal hypoglycemic events was 30 minutes in this table.