Sample Characteristics
Four-hundred households participated and completed the surveys, totaling 1057 people. We included the 925 adult participants in the analyses. Table 1 shows the demographic, socio-economic, health status and the descriptive analysis of the six outcome variables for the overall sample as well as in Gusu (urban) and Jinhu (rural) respectively. The sample included 463 adult participants in the urban area and 462 in the rural area. The gender distribution was almost balanced, and 44.1% were over 60 years old. Participants in the urban area had higher socio-economic status in terms of education, employment and income (P<0.001). In the rural sample 27.9% of participants had never completed primary school education, while the percentage is only 2.6% in the urban sample. The average income per capita of the households in the urban sample was almost twice that in the rural sample. Over 95% of the sample were covered by public health insurance, and therefore we could not explore the effects of having no public health insurance on health care utilization. Over 70% of urban sample were covered by UEBMI while in the rural sample 87.5% were enrolled in the NRCMS. As for NCD status, 45.7% of the sampled population had at least one type of NCD, and this rate was slightly higher in the rural area. Descriptive analysis on the six outcome variables showed that the rural sample had more self-reported emergent illness episodes, used both more informal and formal health care services, had higher OOP health expenditure and higher likelihood of incurring CHE (P<0.01).
Table 2: Basic characteristics of study participants (%)
|
Total sample
|
Gusu
|
Jinhu
|
P value
|
|
|
(urban)
|
(rural)
|
|
|
n=925
|
n=463
|
n=462
|
|
Gender
|
|
|
|
|
male
|
51.2
|
47.08
|
49.6
|
0.308
|
Age
|
|
|
|
|
average (sd)
|
54.5 (17.0)
|
53.5(18.8)
|
55.5(15.0)
|
0.077
|
18-29
|
10.7
|
12.3
|
9.1
|
|
30-59
|
45.2
|
41.3
|
49.1
|
|
>=60
|
44.1
|
46.4
|
41.8
|
|
Marriage
|
|
|
|
|
married
|
85.4
|
85.3
|
85.5
|
0.937
|
Education
|
|
|
|
|
Below primary
|
15.2
|
2.6
|
27.9
|
<0.001
|
primary and junior high
|
48.7
|
41
|
56.3
|
|
senior high school and above
|
36.1
|
56.4
|
15.8
|
|
Employment
|
|
|
|
|
employed
|
55.0
|
37.8
|
72.3
|
<0.001
|
retired
|
27.1
|
50.8
|
3.3
|
|
unemployed
|
21.4
|
17
|
26.2
|
|
Health insurance
|
|
|
|
|
UEBMI
|
40.5
|
73.4
|
7.6
|
<0.001
|
URBMI
|
6.2
|
7.6
|
4.8
|
|
NRCMS
|
49
|
10.6
|
87.5
|
|
other and no insurance
|
4.3
|
8.4
|
0.2
|
|
Mean household income per capita (RMB)
|
|
|
|
average (sd)
|
2135(1395)
|
2807(1462)
|
1462(920)
|
<0.001
|
With NCD
|
|
|
|
|
yes
|
45.7
|
43.8
|
47.6
|
0.249
|
Self-reported emergent illness episodes
|
mean (sd)
|
0.98(1.66)
|
0.40(1.12)
|
1.56(1.90)
|
<0.001
|
Total times of self-treatment
|
|
|
|
|
mean(sd)
|
0.39(0.94)
|
0.06(0.37)
|
0.71(1.19)
|
<0.001
|
Total times of outpatient service
|
mean (sd)
|
0.66(1.34)
|
0.35(0.96)
|
0.97(1.57)
|
<0.001
|
% inpatient service use
|
6.0
|
3.9
|
8.0
|
0.008
|
Total OOP health expenditure
|
|
|
|
|
mean(sd)
|
594(4266)
|
201(1940)
|
991(5697)
|
<0.001
|
% CHE
|
16.8
|
6.5
|
27.1
|
<0.001
|
Factors associated with health care needs
Table 3 shows the association between perceived health care needs and a series of demographic, health status and socio-economic factors, using the ZINB model. Process 1 of the model showed that people with rural residence were much more likely to be at risk of reporting emergent illness as compared to the urban counterparts (OR = 0.02, P = 0.003). Having NCD also increased the probability of such risk (OR = 0.326, P = 0.032). Process 2 of the model shows that NCD is also associated with reporting emergent illness more often. People enrolled in NRCMS/URBMI also tend to report more times of emergent illness compared to those enrolled in UEBM, after adjusting for other factors (IRR = 1.67, p = 0.039). Education level seemed to be negatively associated with the times of self-reported emergent illness, and the association is almost significant for those with highest education level (senior high school and above).
Table 3: Regression analysis of factors associated with self-reported emergent illness episodes using ZINB model
|
Process 1
|
Process 2
|
|
OR
|
P>z
|
IRR
|
P>z
|
Age
|
|
|
|
|
<30
|
ref.
|
|
ref.
|
|
30-59
|
0.24
|
0.165
|
0.76
|
0.551
|
>=60
|
0.41
|
0.367
|
0.73
|
0.493
|
Male
|
0.88
|
0.805
|
0.91
|
0.561
|
Rural residence
|
0.02
|
0.003
|
0.73
|
0.280
|
Married
|
0.76
|
0.584
|
0.93
|
0.668
|
Education Level
|
|
|
|
|
no education
|
ref.
|
|
ref.
|
|
primary and junior high
|
1.36
|
0.761
|
0.86
|
0.317
|
senior high school and above
|
0.76
|
0.808
|
0.60
|
0.052
|
Employed
|
1.17
|
0.795
|
1.07
|
0.671
|
Insurance
|
|
|
|
|
UEBMI
|
ref.
|
|
ref.
|
|
NRCMS/URBMI
|
1.75
|
0.260
|
1.67
|
0.039
|
Income level
|
|
|
|
|
poorest 33.3%
|
ref.
|
|
ref.
|
|
middle 33.3%
|
0.92
|
0.913
|
0.94
|
0.739
|
richest 33.3%
|
0.75
|
0.669
|
0.85
|
0.390
|
With NCD
|
0.33
|
0.032
|
1.39
|
0.035
|
(OR: odds ratio. IRR: incident rate ratio. Process 1 modeled the likelihood of not being at risk of reporting self-reported illness, process 2 modeled the total number of self-reported emergent illness episodes given that one is at risk. The sample size is the same as described in table 2).
Factors associated with use of self-treatment, outpatient and inpatient service
Table 4 shows the analyses of the total times of conducting self-treatments, outpatient service use and inpatient service use for a series of demographic, health status and socio-economic factors, using different regression models. NB regression of self-treatment on these factors showed that older age, rural residence and having NCD were significantly associated with increased use of self-treatment (P<0.05), and the effect was particularly strong for rural residence (IRR = 6.072), after adjusting for other factors. As for outpatient service use, regression analysis using the ZINB model showed that rural residence was associated with much higher probability of being at risk of using outpatient services (i.e. using any of these services) compared to urban residence (OR = 0.015, P<0.001). On the contrary, enrollment in NRCMS/RBMI significantly decreased the probability of using any outpatient service compared to UEBMI (OR = 13.3, P<0.05), which means NRCMS/RBMI may discourage outpatient service use. Nevertheless, for those who were at risk of using outpatient service, NRCMS/RBMI was significantly associated with more use (IRR = 2.754, P<0.05). Logit regression of inpatient service use showed that rural residence and having NCD were associated with higher probability of using inpatient service (P<0.056), while men were less likely to use inpatient service than female (P<0.05), after adjusting for other covariates.
Table 4: Regression analysis of factors associated with self-treatment, outpatient service and inpatient service use
|
Self-treatment
|
Outpatient service use
|
Inpatient service use
|
|
NB
|
|
ZINB-proc1
|
ZINB-proc2
|
logit
|
|
|
IRR
|
P>z
|
OR
|
P>z
|
IRR
|
P>z
|
OR
|
P>z
|
Age
|
|
|
|
|
|
|
|
|
<30
|
ref.
|
|
ref.
|
|
|
|
ref.
|
|
30-59
|
2.41
|
0.010
|
0.64
|
0.574
|
0.75
|
0.499
|
0.34
|
0.083
|
>=60
|
2.27
|
0.026
|
0.76
|
0.790
|
0.59
|
0.232
|
1.11
|
0.879
|
Male
|
0.94
|
0.692
|
1.66
|
0.423
|
0.98
|
0.919
|
0.37
|
0.008
|
Rural residence
|
6.07
|
0.000
|
0.02
|
0.000
|
0.56
|
0.341
|
3.56
|
0.002
|
Married
|
0.87
|
0.498
|
0.59
|
0.595
|
0.86
|
0.520
|
1.85
|
0.180
|
Education Level
|
|
|
|
|
|
|
|
|
no education
|
ref.
|
|
ref.
|
|
|
|
ref.
|
|
primary and junior
high
|
1.09
|
0.627
|
0.50
|
0.557
|
0.71
|
0.112
|
1.27
|
0.577
|
senior high school
and above
|
0.87
|
0.630
|
1.30
|
0.879
|
0.71
|
0.381
|
1.70
|
0.378
|
Employed
|
0.91
|
0.623
|
1.45
|
0.659
|
1.07
|
0.745
|
0.61
|
0.197
|
Insurance
|
|
|
|
|
|
|
|
|
UEBMI
|
ref.
|
|
ref.
|
|
|
|
ref.
|
|
NRCMS/URBMI
|
1.48
|
0.255
|
13.29
|
0.027
|
2.75
|
0.026
|
1.10
|
0.832
|
Income level
|
|
|
|
|
|
|
|
|
poorest 33.3%
|
ref.
|
|
ref.
|
|
|
|
ref.
|
|
middle 33.3%
|
0.89
|
0.540
|
0.26
|
0.271
|
0.79
|
0.186
|
1.52
|
0.301
|
richest 33.3%
|
0.61
|
0.092
|
0.21
|
0.460
|
0.71
|
0.200
|
1.07
|
0.869
|
With NCD
|
1.49
|
0.007
|
0.06
|
0.208
|
1.21
|
0.365
|
2.65
|
0.002
|
(OR: odds ratio. IRR: incident rate ratio. Proc: Process. Process 1 of ZINB modeled the likelihood of not being at the risk of using outpatient service, and process 2 modeled the total times of outpatient service use given that one is at that risk. The sample size is the same as described in table 2).
Out-of-pocket (OOP) payment and financial burden across income groups
Table 5 shows the results of regression analyses of factors associated with OOP health expenditure using a two-part model combining logit regression and GLM, as well as factors associated with CHE using a logit model. Similar to the results of the analysis on inpatient service use, NCD and rural residence were significantly associated with higher probability of incurring medical expenditure and CHE (P<0.001). For those who had out-of-pocket health expenditure, men tended to spend less than women, and men were also less likely to incur catastrophic expenditure (P<0.05). People in NRCMS/RBMI were also twice likely to incur CHE as those enrolled in UEBMI (P<0.05), after adjusting for other variables.
Table 5: regression analysis of factors associated out-of-pocket health expenditure and CHE
|
OOP health expenditure
|
Catastrophic health expenditure
|
|
part1-logit
|
part2-GLM
|
logit
|
|
|
OR
|
P>z
|
Coef.
|
P>z
|
OR.
|
P>z
|
Age
|
|
|
|
|
|
|
<30
|
ref.
|
|
|
|
|
|
30-59
|
1.46
|
0.286
|
57.7
|
0.952
|
1.01
|
0.979
|
>=60
|
1.07
|
0.863
|
1898.1
|
0.156
|
1.13
|
0.796
|
Male
|
0.80
|
0.183
|
-2207.4
|
0.042
|
0.58
|
0.013
|
Rural residence
|
6.60
|
0.000
|
1094.2
|
0.258
|
2.92
|
0.000
|
Married
|
1.16
|
0.534
|
595
|
0.337
|
0.94
|
0.830
|
Education Level
|
|
|
|
|
|
|
no education
|
ref.
|
|
|
|
ref.
|
|
primary and junior high
|
0.68
|
0.124
|
1788.4
|
0.148
|
0.97
|
0.904
|
senior high school and
above
|
0.62
|
0.144
|
2199.7
|
0.145
|
0.73
|
0.406
|
Employed
|
0.90
|
0.629
|
712.9
|
0.519
|
0.61
|
0.057
|
Insurance
|
|
|
|
|
|
|
UEBMI
|
ref.
|
|
|
|
ref.
|
|
NRCMS/URBMI
|
1.33
|
0.273
|
238.5
|
0.595
|
2.02
|
0.024
|
Income level
|
|
|
|
|
|
|
poorest 33.3%
|
ref.
|
|
|
|
ref.
|
|
middle 33.3%
|
1.06
|
0.803
|
-237.4
|
0.744
|
0.72
|
0.209
|
richest 33.3%
|
1.09
|
0.703
|
877.2
|
0.572
|
0.57
|
0.069
|
With NCD
|
1.99
|
0.000
|
212.8
|
0.775
|
2.97
|
0.000
|
(Part 1 of the two-part model used logit regression to estimate the likelihood of incurring OOP health expenditure, and part 2 used GLM to model the amount of OOP health expenditure if occurred)
Separate analysis on the rural and urban sample
We further explored the effects of demographic and SES factors on these outcomes of interests for urban and rural population separately. Gender played a role in the rural but not in the urban area. Compared to women, men in the rural area tended to report fewer emergent illnesses, use less inpatient and outpatient services, and thus less often incurred catastrophic expenditure. It is also noticeable that for the rural sample, people enrolled in NRCMS/URBMI were more likely to incur CHE compared to those enrolled in UEBMI, and being in the richest tertile also decreased the likelihood of incurring CHE. Nevertheless, insurance category and income were not significantly associated with the possibility of incurring CHE in the urban sample, and only NCD status seemed to have an effect on CHE (P<0.05) (see supplemental materials for all result tables).