Identifying Metabolic Syndrome Easily and Cost Effectively Using Non-Invasive Methods with Machine Learning Models

Figures & data

Table 1 Non-Invasive Factors of 10 Randomly Selected Subjects from Training Dataset

Samples	Training Dataset (2019 Group)
	Age, (Years)	Gender	BMI (kg/m2)	SBP, (mmHg)	DBP, (mmHg)	MetS
1	56	Male	24.57	138	86	No
2	33	Female	25.31	130	85	No
3	61	Male	25.06	132	88	Yes
4	53	Female	22.53	116	74	No
5	42	Female	21.45	136	90	No
6	63	Male	19.79	142	82	Yes
7	50	Female	18.07	104	64	No
8	45	Female	22.03	116	78	Yes
9	50	Male	23.67	160	110	No
10	62	Male	30.30	145	100	No

Abbreviations: BH, body height; BW, body weight; SBP, systolic blood pressure; DBP, diastolic blood pressure; MetS, metabolic syndrome.

Table 2 Non-Invasive Factors of 10 Randomly Selected Subjects from External Validation Dataset

Samples	External Validation Dataset (2019 Group)
	Age, (Years)	Gender	BMI (kg/m2)	SBP, (mmHg)	DBP, (mmHg)	MetS
1	56	Male	25.83	149	88	No
2	27	Male	26.53	161	92	Yes
3	58	Female	24.34	135	81	No
4	29	Male	24.77	140	84	No
5	46	Male	26.83	159	91	Yes
6	41	Female	22.92	123	72	No
7	40	Male	23.03	127	75	No
8	62	Female	24.95	136	82	No
9	33	Male	23.03	127	75	No
10	24	Female	20.19	104	52	No

Abbreviations: BH, body height; BW, body weight; SBP, systolic blood pressure; DBP, diastolic blood pressure; MetS, metabolic syndrome.

Table 3 General Characteristics of Subjects in the Training and Validation Datasets According to the Development of Metabolic Syndrome

Variables	Training Dataset (2009 Group)			External Validation Dataset (2019 Group)
	Non-MetS	MetS	p-value	Non-MetS	MetS	p-value
	N=4501 (92.9)	N=346 (7.1)		N=3728 (86.2)	N=596 (13.8)
Gender
Male, n (%)	2867 (63.70)	298 (86.13)	<0.001	1918 (51.45)	562 (94.30)	<0.001
Female, n (%)	1634 (36.30)	48 (13.87)	<0.001	1810 (48.55)	34 (5.70)	<0.001
Age, (years)	45.97 ± 13.74	54.73 ± 12.81	<0.001	44.34 ± 13.22	46.47 ± 14.84	0.0003
BMI (kg/m2)	23.35 ± 2.87	27.39 ± 2.27	<0.001	23.29 ± 1.82	27.05 ± 0.44	<0.001
SBP (mmHg)	120.35 ± 21.05	141.13 ± 15.98	<0.001	126.84 ± 13.07	167.96 ± 8.87	<0.001
DBP (mmHg)	79.90 ± 10.46	93.34 ± 9.67	<0.001	74.05 ± 9.27	92.52 ± 1.36	<0.001
FBG (mg/dL)	94.87 ± 18.42	125.76 ± 36.77	<0.001	87.65 ± 7.64	120.65 ± 8.46	<0.001
TG (mg/dL)	134.46 ± 106.74	266.23 ± 246.29	<0.001	123.18 ± 55.05	331.13 ± 34.20	<0.001
TC (mg/dL)	176.11 ± 33.70	187.95 ± 37.55	<0.001	167.96 ± 23.59	225.62 ± 5.18	<0.001
HDL-C (mg/dL)	48.57 ± 12.28	38.51 ± 9.96	<0.001	48.12 ± 10.25	74.58 ± 2.30	<0.001
LDL-C (mg/dL)	95.49 ± 25.51	99.09 ± 26.48	<0.001	93.19 ± 19.46	141.20 ± 4.52	<0.001
SUA (mg/dL)	5.22 ± 1.33	6.19 ± 1.36	<0.001	5.32 ± 1.17	7.90 ± 0.32	<0.001

Notes: Continuous variables were represented as mean ± standard deviation and categorical data were represented as a number (percentage). P-values of normal-distribution date derived from Student’s t-test; and P-values of categorical data derived from chi-squared test.

Abbreviations: MetS, metabolic syndrome; BMI, body mass index; SBP, systolic blood pressure; DBP, diastolic blood pressure; FBG, fasting blood glucose; TG, triglycerides; TC, total cholesterol; LDL-C, low-density lipoprotein cholesterol; HDL-C, high-density lipoprotein cholesterol; SUA, serum uric acid.

Table 4 The 10-Fold Cross-Validation Performance of the Machine Learning Models in the Training Dataset

Model	True Positive	False Positive	True Negative	False Negative	Sensitivity	Specificity	AUC	CA
Artificial Neural Network	58	60	4441	288	16.76%	98.67%	0.926	0.928
Logistic Regression	54	61	4440	292	15.61%	98.64%	0.916	0.927
Bayesian Network	212	421	4080	134	61.27%	90.65%	0.904	0.885
Random Forest	97	127	4374	249	28.03%	97.18%	0.896	0.922
K-NN	50	75	4426	296	14.45%	98.33%	0.811	0.924
Support Vector Machine	195	895	3606	151	56.36%	80.12%	0.652	0.652
Decision Trees	103	158	4343	243	29.77%	96.49%	0.591	0.921

Abbreviations: AUC, the area under the curve; CA, classification accuracy; Precision, positive predictive value; Recall, true positive rate.

Table 5 The External Validation Performance of the Machine Learning Models

Model	True Positive	False Positive	True Negative	False Negative	Sensitivity	Specificity	AUC	CA
Artificial Neural Network	429	23	3705	167	71.98%	99.38%	0.989	0.956
Bayesian Network	586	321	3407	10	98.32%	91.32%	0.976	0.923
Logistic Regression	18	0	3728	578	3.02%	100%	0.979	0.866

Abbreviations: AUC, the area under the curve; CA, classification accuracy; Precision, positive predictive value; Recall, true positive rate.

Figure 1 Area under the receiver-operating characteristic curves of three machine learning algorithms on the external validation set. (A) Artificial Neural Network; (B) Naive Bayesian; (C) Logistic Regression.

Data Sharing Statement

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.