The status of CHE under different definition criteria
The incidence of CHE, AGCHE, and RGCHE showed a decreasing trend as the defining criteria (subsequently denoted by Z) increased (See Table 1). The decline between Z=15% and Z=25% (16.30 percentage points) was lower than the decline between Z=25% and Z=40% (23.70 percentage points).
Table 1 CHE occurrence among households with diabetic patients
Z
|
Number of occurrences (households)
|
Incidence of CHE(%)
|
AGCHE(%)
|
RGCHE(%)
|
15%
|
203
|
75.19
|
18.62
|
24.76
|
25%
|
158
|
58.89
|
11.96
|
20.43
|
40%
|
95
|
35.19
|
5.09
|
14.47
|
Distribution of CHE by income quintile
We divided all households into five groups based on annual per capita household income, from the poorest (1st quintile) to the wealthiest (5th quintile) households. The overall trend of the incidence of CHE declined as the economic level rose under the same defined criteria. The AGCHE did not show a clear changing pattern across economic levels. RGCHE was higher in the low-income group than in the other groups (except for Z=40%). The defined criteria showed a negative correlation with each group's incidence of CHE, AGCHE, and RGCHE at a given economic level (See Table2).
Analysis of equity
By analyzing the CI with different definition criteria(See Table3), we found that when the definition criteria were 15%, 25%, and 40%, Wagstaff. CIs were all negative (-0.1234, -0.1044, and -0.1885), indicating that low-income households easily experienced CHE. Comparing the Wagstaff. CI of households with different living environments, the above phenomenon was more pronounced in rural areas. In particular, when Z was 40% (Wagstaff.CI=-0.3475), low-income households living in rural areas were more likely to be stuck in CHE (See Table3). The Fig.2 represented the concentration curves under different criteria. We could see that no matter where households lived in rural or urban areas, the cumulative curve was above the equity line, suggesting that CHE occurred more in low-income households. The tendency was more noticeable when Z=40% (See Fig.2c).
Table 2 Percentages of occurrence of catastrophic in households with diabetes at different economic levels
Z
|
Household economic statusa
|
Incidence of CHE(%)
|
AGCHE(%)
|
RGCHE(%)
|
15%
|
Quintile 1
|
84.21
|
25.65
|
30.46
|
|
Quintile 2
|
82.61
|
16.75
|
20.27
|
|
Quintile 3
|
62.50
|
13.38
|
21.41
|
|
Quintile 4
|
80.39
|
23.04
|
28.66
|
|
Quintile 5
|
69.23
|
14.66
|
21.17
|
30%
|
Quintile 1
|
84.21
|
25.65
|
30.46
|
|
Quintile 2
|
70.18
|
18.14
|
25.85
|
|
Quintile 3
|
56.52
|
9.78
|
17.31
|
|
Quintile 4
|
48.44
|
7.89
|
16.29
|
|
Quintile 5
|
66.67
|
15.55
|
23.32
|
40%
|
Quintile 1
|
56.14
|
8.71
|
15.51
|
|
Quintile 2
|
28.26
|
3.64
|
12.88
|
|
Quintile 3
|
23.44
|
2.94
|
12.56
|
|
Quintile 4
|
43.14
|
7.44
|
17.25
|
|
Quintile 5
|
25.00
|
2.75
|
11.01
|
aQuintile 1 is the poorest and quintile 5 is the wealthiest.
Table 3 Equity of CHE in households with people with diabetes
Z
|
Urban
|
Rural
|
Total
|
Wagstaff.CI
|
Std.
|
Wagstaff.CI
|
Std.
|
Wagstaff.CI
|
Std.
|
15%
|
-0.7344**
|
0.2329
|
-0.1754
|
0.1126
|
-0.1234
|
0.1022
|
25%
|
-0.4507
|
0.1424
|
-0.1674*
|
0.8222
|
-0.1044
|
0.0713
|
40%
|
-0.0774
|
0.1422
|
-0.3475***
|
0.0833
|
-0.1885*
|
0.0729
|
*means P<0.05; **means P<0.01; ***means P<0.001.
Pareto char of CHE
According to the different annual per capita income groups corresponding to the cumulative percentage of incidence of CHE, there were three intervals: the cumulative percentage of occurrence within 80% was the A interval, which was the main economic interval in which CHE occurred and was also the main factor interval we need to pay attention to; 80% to 90% was the B range, which is the secondary economic range; the cumulative percentage of occurrence of 90% or more was the C range, which is the least important factor. Figure 3 indicated that households with annual per capita income groups of 0-740USD (0-5,000 CNY) were most likely to have CHE. The main groups where CHE occurred under the three defined criteria were under 2,962 USD (19,994CNY), under 4,444 USD (30,000 CNY), and under 3,704 USD (25,000 CNY) or less (See Fig.3).
Single determinants of catastrophic health expenditure
Regarding demographic and sociological factors, we could see that as age increased, the incidence of CHE became higher. In terms of household elements, the number of household members, whether the household has older people over 65 years old, annual per capita income, and whether it was a low-income household were the main factors affecting the occurrence of CHE. In terms of health and morbidity factors, the results of the univariate analysis of CHE were significant for physical pain, the number of types of chronic diseases, whether health checks were conducted in the last 12 months, and self-rated health status. Concerning health service utilization, CHER was higher for diabetic patients with inpatient service utilization (46.38%) than for diabetic patients without hospitalization within one year (31.34%). (See Table4).
The importance of prediction variables in the cart decision tree model
Fig.4 showed that of all the variables, decision tree analysis identified 4 important variables in the evaluation that influenced the occurrence of CHE in households with people with diabetes. No health services variables in the decision tree were founded in the decision tree. We could see that the most important predictor, which formed the root node, was household economic status. The other predictors were the number of chronic diseases, household size, and health checks in the past 12 months. The results also showed that suffering from only 1 chronic disease and physical examination in the last 12 months were protective factors for the occurrence of CHE, regardless of economic status. The incidence of CHE for households with economic status quintile 1 or 4 was 63%. The CART model also showed an interaction between household economic status and household size, with a household economic status of quintile 1 or 4 and a decrease in incidence of CHE from 55% to 44% for households with a household size of 3. There was an interaction between the presence or absence of health checks in the last 12 months and the number of chronic diseases in households with economic status of 1 or 4 and a household size of 1 or 2. Households with a household economic status of quintile 1 or 4, a household size of 1 or 2 persons, and no health check-up in the last 12 months had the highest relative incidence of CHE (88%), while those with a household economic status of quintile 1or 4, a household size of 3 or more persons, and a health check-up in the last 12 months had the lowest incidence of CHE (58%).
Binary logistic regression model for the probability of incurring CHE
Whether CHE occurred was used as the response variable, and nine statistically significant variables for univariate analysis were included as independent variables. The variable screening was performed using the forward stepwise LR method, and the criteria for including and excluding were 0.05 and 0.10, accordingly. Households with larger household sizes (3 or more than 3)were less likely to experience CHE(OR=0.207). The risk to occur CHE in households with 2, 3 and more than 3 members was 0.790 and 0.207 times higher than in those living alone. There was a significant negative correlation between the incidence of CHE and household economic status. It is obvious that CHE was less likely to occur in wealthier households. Besides, we can also see that patients with 3 or more 3 types of chronic diseases were associated with a high risk of CHE(OR=3.783). The risk of CHE in patients with 2, 3, or more chronic diseases was 1.807 times and 1.724 times higher than in people with diabetes only (See Table 5).
ROC analysis
The Area Under Curve (AUC) for the decision tree model was 0.7403, and the AUC for the logistic model was 0.7551. The ROC curves were plotted with the predicted probabilities from the decision tree model and the logistic regression model. The ROC curves of both models were relatively close and above the reference line. The two models could be combined in related studies to explore the influential elements collectively (See Figure 5).
Table 5 Binary logistic regression model for probability of incurring CHE
Variables
|
β
|
P
|
OR
|
95%CI
|
Household size(1)
|
|
|
|
|
|
2
|
-0.236
|
0.608
|
0.790
|
0.320
|
1.950
|
3 or more
|
-1.576
|
0.002
|
0.207
|
0.077
|
0.553
|
Household economic statusa (Quintile (1)
|
|
|
|
|
|
Quintile 2
|
-1.397
|
0.003
|
0.247
|
0.100
|
0.613
|
Quintile 3
|
-1.436
|
0.001
|
0.238
|
0.104
|
0.547
|
Quintile 4
|
-0.776
|
0.071
|
0.460
|
0.198
|
1.070
|
Quintile 5
|
-1.915
|
0.000
|
0.147
|
0.058
|
0.373
|
Number of chronic diseasesb (1)
|
|
|
|
|
|
2
|
0.592
|
0.075
|
1.807
|
0.943
|
3.463
|
3 or more
|
1.331
|
0.001
|
3.783
|
1.724
|
8.304
|
a Quintile 1 was the poorest and Quintile 5 was the wealthiest.
b1 meant patients with only diabetes; 2 meant diabetics suffer from another chronic disease; 3 or more meant diabetic patients with two or more chronic diseases
Table 4 Analysis of single determinants influencing the occurrence of CHE
Variables
|
|
Z=15%
|
|
Z=25%
|
|
Z=40%
|
|
|
Incidence of CHE (%)
|
c2
|
P
|
|
Incidence of CHE (%)
|
c2
|
P
|
|
Incidence of CHE (%)
|
c2
|
P
|
|
Demographic characteristics
|
|
|
|
|
|
|
|
|
|
|
|
Gender
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Male
|
|
76.98
|
0.526
|
0.468
|
|
58.99
|
0.027
|
0.871
|
|
34.53
|
0.054
|
0.817
|
|
Female
|
|
73.28
|
|
|
|
58.02
|
|
|
|
35.88
|
|
|
|
Ethnicity
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Han
|
|
75.30
|
7.300
|
0.069
|
|
58.17
|
0.181、
|
0.670
|
|
35.86
|
0.705
|
0.401
|
|
Minority
|
|
73.68
|
|
|
|
63.01
|
|
|
|
26.32
|
|
|
|
Age group
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15-
|
|
41.67
|
0.797
|
0.671
|
|
25.00
|
7.480
|
0.024
|
|
16.67
|
11.156
|
0.004
|
|
45-
|
|
74.81
|
|
|
|
56.30
|
|
|
|
27.41
|
|
|
|
65
|
|
78.86
|
|
|
|
64.23
|
|
|
|
45.53
|
|
|
|
Education level
|
|
|
|
|
|
|
|
|
|
|
|
|
Illiteracy
|
|
83.56
|
1.069
|
0.785
|
|
63.01
|
8.014
|
0.046
|
|
41.10
|
6.601
|
0.086
|
|
Elementary school
|
77.22
|
|
|
|
65.82
|
|
|
|
40.51
|
|
|
|
Middle school
|
70.00
|
|
|
|
57.14
|
|
|
|
32.86
|
|
|
|
High school and above
|
66.67
|
|
|
|
41.67
|
|
|
|
20.83
|
|
|
|
Marital status
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Married
|
|
76.79
|
0.638
|
0.425
|
|
60.27
|
1.658
|
0.198
|
|
35.71
|
0.161
|
0.688
|
|
Other
|
|
67.39
|
|
|
|
50.00
|
|
|
|
32.61
|
|
|
|
Variables Z=15%
|
|
|
Z=25%
|
|
|
|
Z=40%
|
|
|
|
|
Incidence of CHE (%)
|
c2
|
P
|
|
Incidence of CHE (%)
|
c2
|
P
|
|
Incidence of CHE (%)
|
c2
|
P
|
|
Urban
|
|
80.00
|
2.105
|
0.147
|
|
60.00
|
0.085
|
0.77
|
|
40.00
|
0.961
|
0.327
|
|
Rural
|
|
73.50
|
|
|
|
58.00
|
|
|
|
33.50
|
|
|
|
Household size
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1
|
|
78.57
|
1.590
|
0.451
|
|
67.86
|
16.596
|
<0.001
|
|
46.43
|
12.767
|
0.002
|
|
2
|
|
85.00
|
|
|
|
70.00
|
|
|
|
44.17
|
|
|
|
3 or more
|
|
64.75
|
|
|
|
45.08
|
|
|
|
|
|
|
|
One or more elderly1
|
|
|
|
|
|
|
|
|
|
|
|
|
Yes
|
|
73.43
|
0.847
|
0.357
|
|
55.24
|
1.342
|
0.247
|
|
44.09
|
8.346
|
0.004
|
|
No
|
|
77.17
|
|
|
|
62.20
|
|
|
|
27.27
|
|
|
|
Health care insurance
|
|
|
|
|
|
|
|
|
|
|
|
|
UEBMI2
|
|
76.47
|
0.168
|
0.92
|
|
60.29
|
2.139
|
0.343
|
|
37.25
|
3.966
|
0.138
|
|
Other
|
|
73.33
|
|
|
|
55.00
|
|
|
|
31.67
|
|
|
|
None
|
|
50.00
|
|
|
|
|
|
|
|
0.0
|
|
|
|
Household economic characteristics
|
|
|
|
|
|
|
|
|
|
|
Household economic status3
|
|
|
|
|
|
|
|
|
|
|
|
Quintile 1
|
84.21
|
9.080
|
0.059
|
|
70.18
|
8.272
|
0.082
|
|
56.14
|
19.595
|
<0.001
|
|
Quintile 2
|
82.61
|
|
|
|
56.52
|
|
|
|
28.26
|
|
|
|
Quintile 3
|
62.50
|
|
|
|
48.44
|
|
|
|
23.44
|
|
|
|
Quintile 4
|
80.39
|
|
|
|
66.67
|
|
|
|
43.14
|
|
|
|
Quintile 5
|
69.23
|
|
|
|
51.92
|
|
|
|
25.00
|
|
|
|
Poverty-stricken household
|
|
|
|
|
|
|
|
|
|
|
|
Yes
|
|
85.00
|
0.799
|
0.371
|
|
70.00
|
1.173
|
0.279
|
|
70.00
|
0.429
|
0.512
|
|
No
|
|
74.40
|
|
|
|
57.60
|
|
|
|
32.40
|
|
|
|
Variables
|
|
|
Z=15%
|
|
|
|
Z=25%
|
|
|
|
Z=40%
|
|
|
|
Incidence of CHE (%)
|
c2
|
P
|
|
Incidence of CHE (%))
|
|
|
|
Incidence of CHE (%)
|
c2
|
P
|
|
Destitute Household characteristics
|
|
|
|
|
|
|
|
|
|
|
Yes
|
|
87.50
|
0.799
|
0.371
|
|
75.00
|
1.903
|
0.168
|
|
62.50
|
0.878
|
0.349
|
|
No
|
|
74.41
|
|
|
|
57.48
|
|
|
|
33.46
|
|
|
|
Health and disease characteristics
|
|
|
|
|
|
|
|
|
|
|
Physical pain
|
|
|
|
|
|
|
|
|
|
|
|
|
No problems
|
|
68.75
|
3.340
|
0.068
|
|
53.75
|
3.679
|
0.055
|
|
29.38
|
5.813
|
0.016
|
|
Medium or serious problems
|
84.55
|
|
|
|
65.45
|
|
|
|
43.64
|
|
|
|
Anxiety or depression
|
|
|
|
|
|
|
|
|
|
|
|
|
No problems
|
|
73.85
|
0.154
|
0.695
|
|
57.34
|
0.648
|
0.421
|
|
33.49
|
1.433
|
0.231
|
|
Medium or serious problems
|
80.77
|
|
|
|
63.46
|
|
|
|
42.31
|
|
|
|
Number of chronic diseases4
|
|
|
|
|
|
|
|
|
|
|
|
1
|
|
65.31
|
7.769
|
0.021
|
|
50.00
|
5.074
|
0.079
|
|
24.49
|
10.202
|
0.006
|
|
2
|
|
79.17
|
|
|
|
61.67
|
|
|
|
37.50
|
|
|
|
3 or more
|
|
84.62
|
|
|
|
67.31
|
|
|
|
50.00
|
|
|
|
Health Record
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Yes
|
|
75.58
|
3.826
|
0.148
|
|
58.53
|
2.462
|
0.292
|
|
34.10
|
1.001
|
0.606
|
|
No, but understand
|
|
85.71
|
|
|
|
85.71
|
|
|
|
28.57
|
|
|
|
No, but do not understand
|
71.74
|
|
|
|
54.35
|
|
|
|
41.30
|
|
|
|
Household doctor
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Yes
|
|
75.25
|
1.116
|
0.572
|
|
58.42
|
0.063
|
0.969
|
|
33.66
|
2.124
|
0.346
|
|
No, but understand
|
|
82.61
|
|
|
|
60.87
|
|
|
|
30.43
|
|
|
|
No, but do not understand
|
71.11
|
|
|
|
57.78
|
|
|
|
44.44
|
|
|
|
Variables
|
|
Z=15%
|
|
|
|
Z=25%
|
|
|
|
Z=40%
|
|
|
Incidence of CHE (%)
|
c2
|
P
|
|
Incidence of CHE (%)
|
c2
|
P
|
|
Incidence of CHE (%)
|
c2
|
P
|
|
Health check in the past 12 months
|
|
|
|
|
|
|
|
|
|
|
|
Yes
|
|
77.30
|
0.394
|
0.53
|
|
60.54
|
0.990
|
0.32
|
|
39.46
|
4.708
|
0.030
|
|
No
|
|
70.59
|
|
|
|
54.12
|
|
|
|
25.88
|
|
|
|
Self-assessment of health status5
|
|
|
|
|
|
|
|
|
|
|
|
Quintile 1
|
|
85.71
|
6.406
|
0.171
|
|
57.14
|
20.028
|
<0.001
|
|
57.14
|
18.244
|
0.001
|
|
Quintile 2
|
|
94.44
|
|
|
|
83.33
|
|
|
|
50.00
|
|
|
|
Quintile 3
|
|
91.15
|
|
|
|
69.91
|
|
|
|
46.02
|
|
|
|
Quintile 4
|
|
81.36
|
|
|
|
44.92
|
|
|
|
22.03
|
|
|
|
Quintile 5
|
|
78.57
|
|
|
|
50.00
|
|
|
|
28.57
|
|
|
|
Health services utilization
|
|
|
|
|
|
|
|
|
|
|
|
|
Biweekly Clinic
|
|
68.97
|
6.449
|
0.011
|
|
54.31
|
1.484
|
0.223
|
|
31.03
|
1.536
|
0.215
|
|
Yes
|
|
79.87
|
|
|
|
61.69
|
|
|
|
38.31
|
|
|
|
No
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Hospitalization episode
|
|
|
|
|
|
|
|
|
|
|
|
|
Yes
|
|
84.06
|
0.993
|
0.319
|
|
68.12
|
3.517
|
0.061
|
|
46.38
|
5.091
|
0.024
|
|
No
|
|
|
72.14
|
|
|
|
55.22
|
|
|
|
31.34
|
|
|
a referred to the household with one or more persons aged 65 or older.
breferred to Urban Employees Basic Medical Insurance.
cQuintile 1 is the poorest and Quintile 5 is the wealthiest.
db1 meant patients with only diabetes; 2 meant diabetics suffer from another chronic disease; 3 or more meant diabetic patients with two or more chronic diseases
eQuintile 1 was the worst and Quintile 5 was the best.