The Construction of Primary Screening Model and Discriminant Model for Chronic Obstructive Pulmonary Disease in Northeast China

Figures & data

Table 1 The Questionnaire

Factor	Question	Option
x1	Which is your resident type now?	a. Bungalow b. Building
x2	Gender	a. Male b. Female
x3	Age	Year of birth to 2019
x4	Which is your education level?	a. Undergraduate and above b. High school c. Junior high school d. Primary school e. Uneducated
x5	Do you have any dyspnea without strenuous exercise recently?	a. Never b. <1 Time/week c. 1–2 Times/week d. 3–6 Times/week e. Daily
x6	Which of the following was the cooking fuel you often use?	a. No b. Electricity c. Natural gas/liquefied gas/biogas d. Coal e. Firewood
x7	How many packs do you smoke per year?
	How many years have you smoked?
x8	How many days are you exposed to secondhand smoke each year?
	How many years have you been exposed to secondhand smoke?
x9	Do your parents or siblings have respiratory problems?	a. No b. Yes
x10	Have you ever had the following diseases during childhood? (Multiple choice)	a. None b. Pneumonia c. Tuberculosis d. Bronchiectasis
x11	BMI (kg/m^2)
x12	Will the weather change cause you to cough?	a. Never b. <1 Time/week c. 1–2 Times/week d. >3 Times/week
	Do you often cough when you do not have a cold?	a. Never b. <1 Time/week c. 1–2 Times/week d. >3 Times/week
x13	Do you often cough with sputum from your chest in the morning?	a. Never b. <1 Time/week c. 1–2 Times/week d. >3 Times/week
x14	Do you have any wheezing without strenuous exercise recently?	a. No b. Yes
x15	Which of the following is your birth quarter	a. 4–9 months b. 10–3 months
x16	Are you a farmer?	a. No b. Yes
x17	Has your mother ever been exposed to smoke during pregnancy?	a. Never b. Yes, “She smoked at that time” or “She was exposed to secondhand smoke at that time” c. Yes, “She smoked at that time” and “She was exposed to secondhand smoke at that time”
x18	Which of the following is your kitchen ventilation?	a. Range hood b. Ventilation fan c. Chimney d. No
x19	What did you use to keep warm in childhood?	a. Central heating b. Electricity or air conditioning c. Coal stove d. Firewood/brazier/fire
x20	What do you use for heating at present?	a. Central heating b. Electricity or air conditioning c. Coal stove d. Firewood/brazier/fire

Abbreviation: BMI, body mass index.

Table 2 The Variable Assignment

Factor	Variable	Quantification
x1	Resident type	Building = 0, Bungalow = 1
x2	Gender	Female=0, Male=1
x3	Age	Year of birth to 2019
x4	Education level	Undergraduate and above=0, High school=1, Junior high school=2, Primary school=3, Uneducated=4
x5	Dyspnea	Never = 0, <1 Time/week = 1, 1–2 Times/week = 2, 3–6 Times/week = 3, Daily = 4
x6	Cooking fuel grade	No = 0, Electricity = 1, Natural gas/liquefied gas/biogas = 2, Coal = 3, Firewood = 4
x7	Smoking index	pack*year
x8	Second-hand smoking index	Contact day/year* contact year
x9	Family history	No = 0, Yes =1
x10	Infectious history at child age	None = 0, One type= 1, Two types= 2, Three types= 3
x11	BMI	kg/m^2
x12	Cough	None = 0, Level 1 = 1, Level 2 = 2, Level 3 = 3, Level 4 = 4, Level 5 = 5, Level 6 = 6
x13	Cough with sputum	None=0, Level 1=1, Level 2=2, Level 3=3
x14	Wheezing	No = 0, Yes=1
x15	Birth quarter	4–9 months=0, 10–3 months=1
x16	Farmers	No=0, Yes=1
x17	Mother’s smoking exposure history during pregnancy	No = 0, level 1 = 1, level 2 = 2
x18	Kitchen ventilation	Range hood=0, Ventilation fan=1, Chimney=2, No=3
x19	Childhood heating	Central heating = 0, Electricity or air conditioning = 1, Coal stove = 2, Firewood/brazier/fire = 3
x20	Current heating	Central heating=0, Electricity or air conditioning=1, Coal stove=2, Firewood/brazier/fire=3
y1	FEV1%	Actual value
y2	Binary classification	Normal person: y2=0, COPD patients: y2=1
y3	Three classification	Normal person: y3=1, Mild/moderate COPD patient: y3=2, Severe/very severe COPD patient: y3=3

Note: The asterisks (*) in columns X₇ and X₈ in the table stand for multiplication.

Abbreviations: BMI, body mass index; FEV₁, forced expiratory volume in one second; COPD, chronic obstructive pulmonary disease.

Table 3 The Comparison of Basic Information Among Three Groups

Test	COPD			Control (n=218)	P
	GOLD I+II (n=124)	GOLD III+IVⅣ (n=108)	P
Resident type			0.02245		<0.001
Building	56	33		124
Bungalow	68	75		94
Gender			0.7097		0.5415
Male (%)	67 (54%)	61 (56%)		114 (52.2%)
Age (years)			0.5525		0.9503
40–59	44	37		73
60–80	76	70		144
Means	61.30	62.03		61.58
Education level			<0.001		<0.001
Undergraduate and above	10	5		10
High school	16	7		85
Junior high school	58	28		98
Primary school	26	52		23
Uneducated	14	16		2
Dyspnea			<0.001		<0.001
Never	4	0		100
<1 Time/week	5	4		92
1–2 Times/week	33	6		23
3–6 Times/week	28	23		3
Daily	54	75		0
Cooking fuel grade			<0.001		<0.001
No	0	0		0
Electricity	6	2		96
Liquefied gas	58	35		92
Coal	46	40		10
Firewood	14	31		20
Smoking (pack*years)	20.1±25.8	25.6±28.8	0.1296	1.2±3.0	<0.001
Second-hand smoking index	4703.99±5799.85	4521.30±5576.68	0.8081	1923.62±3078.06	<0.001
Family history			0.07935		<0.001
No	66	45		169
Yes	58	63		49
Infectious history at child age			0.06133		0.7544
None	87	61		142
One	22	28		46
Two	11	14		22
Three	4	5		8
BMI (kg/m^2)	23±4	21.50±3.48	<0.001	25±3	<0.001
Cough			<0.001		<0.001
None	7	3		14
Level 1	3	4		2
Level 2	16	4		143
Level 3	18	5		47
Level 4	25	21		11
Level 5	21	22		0
Level 6	34	49		1
Cough with sputum			0.00607		<0.001
None	12	9		16
Level 1	23	9		166
Level 2	40	23		35
Level 3	49	67		1
Wheeze			0.06013		<0.001
No	4	0		188
Yes	120	108		30
Birth quarter			0.7445		<0.001
First and fourth quarters (%)	72 (58%)	65 (60%)		87 (40%)
Farmers			<0.001		<0.001
No	63	30		205
Yes	61	78		13
Mother’s smoking exposure history during pregnancy			0.0083		<0.001
None	53	30		168
Level 1	44	41		36
Level 2	27	37		14
Kitchen ventilation			0.00628		<0.001
Range hood	59	28		177
Ventilation fan	7	2		13
Chimney	36	66		22
No	22	12		6
Heating history during Childhood			0.2376		<0.001
Centralized heating	9	1		116
Electricity	0	1		14
Coal	50	50		69
Firewood	65	56		19
Current heating			<0.001		<0.001
Centralized heating	68	31		194
Electricity	1	0		1
Coal	29	39		21
Firewood	26	38		2

Note: Data were given as n or mean±SD.

Abbreviation: BMI, body mass index.

Table 4 The Summary of Primary Screening Models for COPD (Training Set with Cross-Validation)

	Accuracy	Precision	Recall	F1	AUC
Multiple linear regression	0.9278	0.9444	0.9140	0.9290	0.928
Stepwise multiple linear regression	0.9361	0.9657	0.9086	0.9363	0.937
Logistic regression	0.9472	0.9563	0.9409	0.9485	0.947
Stepwise logistic regression	0.9528	0.9519	0.9570	0.9544	0.953
Linear discriminant analysis	0.9417	0.9188	0.9731	0.9452	0.941
KNN	0.9361	0.9657	0.9086	0.9363	0.937
Decision tree	0.9278	0.9444	0.9140	0.9290	0.928
SVM	0.9472	0.9418	0.9570	0.9493	0.947

Notes: F₁ value, the principal criterion; accuracy, recall rate, AUC value and precision, the secondary criterion; the bold font in the table indicates that they have the highest score in this category. In the comparison of the primary screening models, we should finally select the model with the most points and the highest score based on the F1 value priority. Second, consider the highest accuracy, precision, recall, AUC value.

Abbreviations: AUC, area under the curve; KNN, k-nearest neighbor; SVM, support vector machines.

Table 5 The Summary of Primary Screening Model for COPD (Test Set)

	Accuracy	Precision	Recall	F1	AUC
Multiple linear regression	0.9333	0.9000	0.9783	0.9375	0.932
Stepwise multiple linear regression	0.9333	0.9545	0.9130	0.9333	0.934
Logistic regression	0.9556	0.9565	0.9565	0.9565	0.956
Stepwise logistic regression	0.9667	0.9778	0.9565	0.9670	0.967
Linear discriminant analysis	0.9333	0.9000	0.9783	0.9375	0.932
KNN	0.9333	0.9545	0.9130	0.9333	0.934
Decision tree	0.9444	0.9556	0.9348	0.9451	0.945
SVM	0.9444	0.9184	0.9783	0.9474	0.944

Notes: F₁ value, the principal criterion; accuracy, recall rate, AUC value and precision, the secondary criterion; the bold font in the table indicates that they have the highest score in this category. In the comparison of the primary screening models, we should finally select the model with the most points and the highest score based on the F1 value priority. Second, consider the highest accuracy, precision, recall, AUC value.

Abbreviations: AUC, area under the curve; KNN, k-nearest neighbor; SVM, support vector machines.

Table 6 Corresponding Mean and Standard Deviation of Primary Screening Model Variables After Standardization

Variables	Mean (μ)	Standard Deviation (σ)
x4	2.02	0.99
x5	2.01	1.56
x6	2.27	0.97
x7	12.27	22.49
x9	–	–
x11	23.54	3.55
x13	1.66	0.95
x14	–	–
x16	–	–
x17	0.62	0.76
x18	0.87	1.10
x19	1.72	1.18

Table 7 Multicollinearity Analysis of Independent Variables

Variable	Vif Value	Variable	Vif Value	Variable	Vif Value	Variable	Vif Value
x4	1.57	x7	1.40	x13	1.18	x17	1.47
x5	2.01	x9	1.39	x14	1.85	x18	1.84
x6	1.69	x11	1.23	x16	1.51	x19	1.30

Abbreviation: Vif value, variance inflation factor value.

Table 8 The Variable Parameters and Significance of Primary Screening Model for COPD

	Estimate	Std.	Error	z-value	Pr(>|z|)
(Intercept)	−1.2562	2.7190	−1.679	0.0931	.
x4	−0.3891	0.3162	−2.236	0.0254	*
x5	1.7996	0.3372	3.961	0.0001	***
x6	0.5102	0.4164	2.026	0.0428	*
x7	1.4980	0.0406	3.088	0.0020	**
x9	0.8077	0.6883	2.300	0.0214	*
x11	−0.5552	0.0912	−2.173	0.0298	*
x13	0.5380	0.3372	1.931	0.0535	.
x14	2.0328	0.9837	3.246	0.0012	**
x16	1.3378	0.7777	2.839	0.0045	**
x17	0.8187	0.5023	3.125	0.0018	**
x18	−0.3890	0.3177	−2.293	0.0218	*
x19	0.6888	0.3094	1.934	0.0532	.

Note: The p is the most significant between 0 and 0.001, indicated by ‘***’; The p is extremely significant between 0.001 and 0.01, indicated by ‘**’; The p is relatively significant between 0.01 and 0.05, indicated by ‘*’; The p is significant between 0.05 and 0.1, indicated by ‘.’.

Table 9 The Confusion Matrix of Primary Screening Mode l (Training Set with Cross-Validation)

PredictionTruth	y2=0	y2=1
y2=0	165	9
y2=1	8	178

Table 10 The Confusion Matrix of Primary Screening Model (Test Set)

Prediction Truth	y2=0	y2=1
y2=0	43	1
y2=1	2	44

Figure 1 The ROC curve of the primary screening model (training set with cross-validation).

Figure 2 The ROC curve of the primary screening model (test set).

Table 11 The Effectiveness of Different Discriminant Models

	Accuracy (Training Set with Cross-Validation)	Accuracy (Test Set)
Multiple linear regression	0.7778	0.7667
Stepwise multiple linear regression	0.7944	0.7889
Linear discriminant analysis	0.8028	0.7889
KNN	0.8139	0.8000
Decision tree	0.8361	0.8333
SVM	0.8278	0.8222

Notes: The bold font in the table indicates that they have the highest score in this category. In the comparison of discriminant models, we have to select the model with the highest accuracy value.

Abbreviations: KNN, k-nearest neighbor; SVM, support vector machines.

Table 12 The Confusion Matrix of Decision Tree (Training Set with Cross-Validation)

Prediction Truth	y3=1	y3=2	y3=3
y3=1	170	3	1
y3=2	11	73	15
y3=3	4	25	58

Table 13 The Confusion Matrix of Decision Tree (Test Set)

Prediction Truth	y3=1	y3=2	y3=3
y3=1	42	2	0
y3=2	3	20	2
y3=3	0	8	13

Figure 3 The discriminant model for COPD.