Prediction of hyperkalemia in ESRD patients by identification of multiple leads and multiple features on ECG

Figures & data

Figure 1. A preprocessing is required for the original ECG to become analyzable. (A) The original ECG. (B) ECGs after correction for mild baseline drift by polynomial fitting. (C) ECG after the second baseline correction. (D) ECG after wavelet transform. (E) ECG after bandpass filters.

Figure 2. The flowchart of this study.

Table 1. Characteristics of 80 patients.

Characteristic	All patients (N = 80)
Gender, n (%)
Male	50 (62.5)
Female	30 (37.5)
Age (years), mean (SD)	53.7 (13.6)
Height (cm), mean (SD)	168.9 (8.2)
Weight (kg), mean (SD)	66.9 (12.4)
BMI (kg/m^2), mean (SD)	23.4 (3.3)
Average time of dialysis (year), mean (SD)	3.0 (2.7)
Scr (μmol/L), mean (SD)	992.3 (273)
Bun (mmol/L), mean (SD)	27.3 (7.3)
kt/v, mean (SD)	1.5 (0.1)
HCO3^- (mmol/L), mean (SD)	20.5 (2.8)
Comorbidities, n (%)
Diabetes mellitus	11 (13.8)
Coronary artery disease	12 (15)
Hypertension	69 (86.3)
Cerebral infarction	4 (5)

Figure 3. Distribution of 1024 blood potassium concentration results.

Table 2. Performance of different models at different serum potassium concentration thresholds.

	Threshold of hyperkalemia (mmol/L)	Accuracy	Precision	Sensitivity	Specificity	F1-score	AUC
CNN	5	0.815	0.769	0.851	0.784	0.808	0.900
	5.5	0.717	0.495	0.930	0.635	0.646	0.824
	6	0.834	0.395	0.577	0.872	0.469	0.829
	6.5	0.741	0.155	0.692	0.745	0.254	0.754
SVM	5	0.820	0.777	0.851	0.793	0.812	0.892
	5.5	0.829	0.696	0.684	0.885	0.690	0.861
	6	0.810	0.362	0.654	0.832	0.466	0.784
	6.5	0.883	0.261	0.462	0.911	0.333	0.781
XGB	5	0.863	0.806	0.926	0.811	0.861	0.931
	5.5	0.824	0.672	0.719	0.865	0.695	0.872
	6	0.883	0.542	0.500	0.939	0.520	0.806
	6.5	0.922	0.364	0.308	0.964	0.333	0.751
AdaBoost	5	0.863	0.884	0.809	0.910	0.844	0.923
	5.5	0.771	0.563	0.789	0.764	0.657	0.831
	6	0.878	0.526	0.385	0.950	0.444	0.809
	6.5	0.922	0.364	0.308	0.964	0.333	0.740
LR	5	0.795	0.765	0.798	0.793	0.781	0.858
	5.5	0.815	0.702	0.579	0.905	0.635	0.824
	6	0.873	0.500	0.538	0.922	0.519	0.823
	6.5	0.898	0.300	0.462	0.927	0.364	0.744

Figure 4. ROC of different machine learning models for different degrees of hyperkalemia.

Figure 5. Comparison of AUC of machine learning models at different hyperkalemia concentration thresholds.