Motivation and Research Question

Data

Empirical Results on Insurance Utilization . . . . . . . . . . . . .

Conclusion

Determinants and Impacts of Health Insurance Nonutilization

Hisaki Kono

September 20, 2012

Motivation and Research Question

Data

Empirical Results on Insurance Utilization . . . . . . . . . . . . .

Conclusion

Motivation and Research Question

Many developing countries have attempted to expand the coverage of health insurance to the poor and disadvantaged group in recent years. Impact of health insurance on service utilization and out-of-pocket (OOP) payment varies across countries:
Service utilization:
positive: Waters (1999, Ecuador), Trujillo et al. (2005, Colombia), Yip and Berman (2005, Egypt), Wagstaff et al. (2009, China) no impact: Thornton et al. (2010, Nicaragua), Waters (1999, Ecuador), Wagstaff (2010, Vietnam)

OOP payment:
reduction: Thornton et al. (2010, Nicaragua), Yip and Berman (2005, Egypt), Wagstaff (2010, Vietnam) no impact: Wagstaff et al. (2009, China)

The effect of the health insurance will depend on the demand side factors and supply side factors, so it can be quite context-dependent.
demand side: price elasticity of the healthcare utilization supply side: the accessibilities to healthcare facilities.

Need to identify what makes healthcare utilization and OOP not so responsive to the health insurance provision in that country or region.

Motivation and Research Question

Data

Empirical Results on Insurance Utilization . . . . . . . . . . . . .

Conclusion

Motivation and Research Question An issue missing in the literature: discriminated service against the insurance holders and nonutilization of insurance. Insurance is helpful only when it is utilized. Insurance provision will have a limited impact if insurance utilization rate is low. A story of a patient visiting a central cancer hospital in Hanoi (Ha, 2011):
if he would use it, it would take two days for him to be fully checked. it took him a morning to receive the same services when he paid with his own money.

Vietnam Household Living Standard Survey 2006

insurance utilization (free health insurance card): 0.80 for inpatient treatment, 0.55 for outpatient treatment Three most cited reasons for the nonutilization
cumbersome procedures (21.9%) lower quality of health care service when using the insurance (11.9%) little prospect of receiving reimbursement (9.0%)

But we do not know when and where they chose not to utilize insurance.

We have detailed information on insurance utilization at each hospital visit. Correcting sample selection and endogeneity of the facility choice.

Motivation and Research Question

Data

Empirical Results on Insurance Utilization . . . . . . . . . . . . .

Conclusion

Overview of the Result

The likelihood of insurance utilization is signicantly and substantially lower at the public central hospital.
Reecting low incentives of the public, especially crowded, healthcare facility to increase the number of patients?

The time cost, proxied by the respondents average working hour and the household income per capita, affects the insurance utilization choice.
Only we can provide is the indirect evidence on the discrimination as we do not have direct observations on the waiting time or the objective measure of service quality.

Less prospect of utilizing health insurance at the central hospital discourages people from visiting there. The average effect of insurance nonutilization on the healthcare expenses per visit is around 1 million VND, which is 1.5 times larger than the average healthcare expenses per visit.

Motivation and Research Question

Data

Empirical Results on Insurance Utilization . . . . . . . . . . . . .

Conclusion

Data 2,673 adults aged 18 or older living in 60 communes, selected by the multi-stage stratied random sampling with oversampling poor communes. Health insurance is quite familiar. 68.7% have some health insurance, 16.2% do not have but did have some health insurance.
Table : Summary statistics (A) Demographic variables (1) Mean 0.576 (0.016) 43.201 (0.553) 0.236 (0.030) 8.548 (0.322) 13.729 (0.065) 3.020 (0.082) 0.963 (0.006) 0.320 (0.015) 2.329 (0.021) 0.687 (0.027) 0.162 (0.014) 2426

gender: female Age of the respondent Respondent is an ethnic minority Years in school log(1+household income per capita) ln(work hour) can read Kinh chronic disease last illness: severity (1-4) have HI does not have but did have HI Observations

Motivation and Research Question

Data

Empirical Results on Insurance Utilization . . . . . . . . . . . . .

Conclusion

Insurance holdings

The majority of health insurance is health insurance for the poor, followed by voluntary government health insurance, obliged government health insurance (typically held by employees in public sector and formally registered enterprises), and health insurance for the priority group (free).
(B) Insurance holdings across insurance types (1) Mean student HI 0.042 (0.007) Poor household HI 0.326 (0.043) priority group HI 0.131 (0.013) obliged gov. HI 0.176 (0.030) voluntary gov. HI 0.287 (0.031) private HI 0.023 (0.006) other HI 0.022 (0.005) Observations 1929
Standard errors in parentheses

Motivation and Research Question Pattern of Health Insurance Utilization

Data

Empirical Results on Insurance Utilization . . . . . . . . . . . . .

Conclusion

Pattern of Health Insurance Utilization Use the healthcare facility visit level data for the last disease in the last 12 months. Information on the type of the healthcare facility visited, insurance utilization, and the amount of healthcare expenses for each visit is available. Use observations which have health insurance, experienced illness in the last 12 months, and visited any healthcare facilities. Insurance utilization rate is slightly lower for the ethnic minority and female.

Table : Insurance Utilization (1) Full sample last illness: use HI Kinh Ethnic_Minority Observations 0.782 0.744 1177 (0.028) (0.034) last illness: use HI male female Observations (1) Full sample 0.803 0.753 1177 (0.029) (0.028)

Standard errors in parentheses

Standard errors in parentheses

Motivation and Research Question Pattern of Health Insurance Utilization

Data

Empirical Results on Insurance Utilization . . . . . . . . . . . . .

Conclusion

Pattern of Health Insurance Utilization across Facilities Private facilities: often no contract with the insurance agency, especially small clinics.
In the analysis, we exclude the observation visiting the private facilities. The number of observation is also small.

Among the public facilities, the province/city hospitals and district hospitals/policlinics have high insurance utilization rates. The public facility with the lowest insurance utilization is the central hospital.
in spite of relatively high administration and management ability. consistent with the discrimination story in the crowded public facilities. Table : Insurance Utilization across healthcare facilities (1) Treatment or examinations only last illness: use HI Central Province District Commune Private_hospital Private_clinic Observations 0.789 0.949 0.960 0.886 0.712 0.034 1025 (0.047) (0.030) (0.016) (0.022) (0.140) (0.028)

Standard errors in parentheses

Motivation and Research Question Pattern of Health Insurance Utilization

Data

Empirical Results on Insurance Utilization . . . . . . . . . . . . .

Conclusion

Health Insurance Utilization across Facilities and Policies

Standard errors are not reported because the sample size in each sampling units are too small to calculate the clustered standard errors. The priority group always utilize insurance at the central hospital, while the insurance utilization rate for people with the obliged health insurance is quite low at the central hospital.

Table : Insurance Utilization across Health Facilities and Types of Insurance (1) HI poor last illness: use HI Central Province District Commune Observations 0.837 0.880 0.902 0.866 491 (.) (.) (.) (.) (2) HI priority 1.000 1.000 0.954 0.993 135 (.) (.) (.) (.) (3) HI obliged 0.676 1.000 0.993 0.976 108 (.) (.) (.) (.) (4) HI voluntary 0.804 0.903 1.000 0.855 153 (.) (.) (.) (.)

t statistics in parentheses

Motivation and Research Question Pattern of Health Insurance Utilization

Data

Empirical Results on Insurance Utilization . . . . . . . . . . . . .

Conclusion

Econometric Issues

Sample selection bias: Insurance utilization is observed only when the individual had an insurance and chose to visit any health facilities.
Individual may not go to clinics/hospitals because they expect that they will not utilize the insurance. Government may have provided the insurance to those who tend to be unfamiliar with insurance and are less likely to utilize insurance. We correct sample selection bias by the inverse probability weighting.

Endogeneity of facility choice: Which health facility to visit will depend on the expectation on how likely they will use insurance there.
People who visited the central hospital may be those who have relatively higher prospect of utilizing insurance. The insurance utilization rate at the central hospital itself does not tell us how difcult to utilize insurance there. We use 2SLS with commune dummies as instrumental variables for hospital choice. We show in the paper that we can obtain conservative estimates of the effect of health facility on insurance utilization under the assumption that individuals in a commune where the commune members are more likely to visit the central hospital are not less likely to utilize insurance at the central hospital.

Motivation and Research Question Pattern of Health Insurance Utilization

Data

Empirical Results on Insurance Utilization . . . . . . . . . . . . .

Conclusion

A test for the hypothesis that low prospect of insurance utilization affects the healthcare facility choice

The endogeneity problem will arise mainly because the healthcare facility choice will depend on the expectation how likely they will use the insurance there. If the magnitude of the estimates obtained by the 2SLS is signicantly larger than that obtained by the OLS, which can be tested by Durbin-Wu-Hausman test, it is implied that the low prospect of utilizing insurance at a given healthcare facility actually discourages people from visiting that healthcare facility.
Another cause of the endogeneity is the measurement errors. But in that case, the magnitude of the coefcient when correcting the endogeneity becomes smaller due to the attenuation bias.

Motivation and Research Question Pattern of Health Insurance Utilization

Data

Empirical Results on Insurance Utilization . . . . . . . . . . . . .

Conclusion

Central hospital Province/City hospital commune health center Poor household HI Priority group HI gender: female ethnic minority Years in school ln(HH income per cap) any savings ln(work hour) can read Kinh chronic disease severe illness very severe illness HI_poorcentral hosp HI_prioritycentral hosp ln(work hour)cent hosp Observations

(1) Probit -0.153 (0.039) 0.039 (0.053) -0.033 (0.029) -0.013 (0.039) 0.107 (0.026) -0.037 (0.022) -0.057 (0.033)

(2) Probit -0.133 (0.031) 0.021 (0.054) -0.045 (0.029) -0.062 (0.046) 0.103 (0.023) -0.038 (0.022) -0.043 (0.031) -0.002 (0.003) -0.047 (0.021) -0.052 (0.036) -0.045 (0.026) 0.102 (0.048)

(3) OLS -0.140 (0.050) 0.005 (0.036) -0.048 (0.033) -0.058 (0.055) 0.085 (0.036) -0.027 (0.028) -0.042 (0.043) -0.004 (0.005) -0.040 (0.029) -0.065 (0.048) -0.018 (0.015) 0.102 (0.073)

(4) 2SLS -0.284 (0.104) 0.027 (0.085) -0.154 (0.073) -0.040 (0.053) 0.092 (0.037) -0.021 (0.029) -0.032 (0.043) -0.003 (0.005) -0.043 (0.029) -0.059 (0.049) -0.010 (0.020) 0.116 (0.071)

(5) IPW 2SLS -0.254 (0.072) -0.071 (0.082) -0.170 (0.061) -0.081 (0.048) 0.073 (0.039) -0.035 (0.030) -0.069 (0.053) 0.005 (0.007) -0.083 (0.035) -0.104 (0.050) -0.020 (0.021) 0.061 (0.080)

(6) IPW 2SLS -0.298 (0.089) -0.154 (0.095) -0.179 (0.064) -0.100 (0.048) 0.062 (0.041) -0.026 (0.028) -0.072 (0.052) 0.003 (0.007) -0.076 (0.034) -0.114 (0.050) -0.025 (0.023) 0.043 (0.085) 0.082 (0.039) -0.077 (0.030) 0.102 (0.073)

(7) IPW 2SLS -0.565 (0.130) -0.272 (0.093) -0.182 (0.065) -0.145 (0.066) -0.169 (0.080) -0.087 (0.041) -0.057 (0.059) 0.001 (0.006) -0.066 (0.032) -0.060 (0.052) -0.055 (0.027) 0.009 (0.092) 0.106 (0.043) -0.055 (0.028) 0.133 (0.067) 0.160 (0.235) 0.716 (0.236)

(8) IPW 2SL -0.390 (0.199) -0.243

(0.092) -0.179 (0.062) -0.139

(0.065) -0.143 (0.081) -0.086

(0.041) -0.062 (0.058) 0.003 (0.007) -0.070

(0.033) -0.071 (0.055) -0.045

(0.023) -0.009 (0.102) 0.118

(0.047) -0.055 (0.028) 0.133

704

704

704

704

697

697

697

(0.072) 0.047 (0.261) 0.674 (0.224) -0.050 (0.053) 697

Motivation and Research Question Pattern of Health Insurance Utilization

Data

Empirical Results on Insurance Utilization . . . . . . . . . . . . .

Conclusion

Correcting the endogenous healthcare facility choice (Column (4)) substantially increases the magnitude of the coefcient of central hospital and commune health center.
If a patient visits the central hospital, the probability of utilizing insurance will be 25-30 percentage points lower than visiting the district hospitals.

Durbin-Wu-Hausman tests rejects the null hypothesis that these facility choice variables are exogenous (p-value of 0.078).
The low prospect of health insurance utilization at the central hospital actually discourages people from visiting there.

Patient with chronic disease are more likely to utilize insurance; patient with severe illness are less likely to utilize the insurance, suggesting that in the case of the severe illness, the patients prioritize the treatment, and insurance utilization is the second matter. However, if the illness is very severe, in which case the medical expenses could be huge, patients seem to try to nd a way to utilize the insurance. The low insurance utilization at the central hospital is not due to the health characteristics of the patients. Rather, once we control the health and illness status, the insurance utilization at the central hospital is even lower.

Motivation and Research Question Pattern of Health Insurance Utilization

Data

Empirical Results on Insurance Utilization . . . . . . . . . . . . .

Conclusion

Our preferred specication is Column (7) as it allows the differential patterns of insurance utilization across insurance policy at the central hospital, which captures important variation in insurance utilization. The negative coefcients on the working hours and the household income would indicate that those whose time costs are higher are less likely to utilize the insurance. Consistent with the anecdote. But the income and working hours little explains the high insurance utilization rate of the priority group. When we include the interaction terms of these variables with central hospital variable, the coefcient changes little. So some other factors than working hours or income are working for this population. Need further studies.

Motivation and Research Question

Data

Empirical Results on Insurance Utilization . . . . . . . . . . . . .

Conclusion

The magnitude of the effect of insurance nonuntilization

The magnitude of the effect of insurance nonuntilization

We have shown that the likelihood of insurance utilization is quite lower at the central hospital, especially for the group other than the priority group. But how important is the insurance nonutilization in terms of the health expenditure? For examining the effect of health insurance nonutilization on the health expenditure, we use the self-reported health care expenditure which is elicited by the question In total, how much did you pay for this visit?. The drawback of this question is that it is not clear if the respondent reports the health care expenditure net of health insurance reimbursement. It is likely that some respondents report the expenditure net of reimbursement, and others report the gross health care expenditure. We use a second outcome variable which is obtained by subtracting the amount of health insurance reimbursement elicited by the question How much were you paid by health insurance in total at that time?. The true value of the expenses should be somewhere between the rst outcome variable and the second one.

Motivation and Research Question

Data

Empirical Results on Insurance Utilization . . . . . . . . . . . . .

Conclusion

The magnitude of the effect of insurance nonuntilization

The amount of the self-reported health care expenditure for the visit at the last illness

count not use HI last illness: last illness: last illness: last illness: last illness: last illness: use HI last illness: last illness: last illness: last illness: last illness: last illness: Total last illness: last illness: last illness: last illness: last illness: last illness: pay for the visit pay for examination pay for medication pay for incentive money for health staff pay for transport pay for other category pay for the visit pay for examination pay for medication pay for incentive money for health staff pay for transport pay for other category pay for the visit pay for examination pay for medication pay for incentive money for health staff pay for transport pay for other category 78 81 83 81 82 83 473 372 387 385 383 386 551 453 470 466 465 469

mean 1037730.8 84252.1 639999.9 28888.8 52073.1 156024.0 544564.5 64089.8 217278.7 13506.4 58018.1 237745.9 614377.5 67694.9 291929.5 16180.1 56969.7 223283.4

sd 3735050.8 406606.1 2731860.4 176167.5 233920.9 690499.9 3147995.1 578916.0 1317363.0 77600.2 185152.1 1114456.1 3238389.6 551730.2 1661084.3 101716.5 194371.7 1051787.0

Motivation and Research Question

Data

Empirical Results on Insurance Utilization . . . . . . . . . . . . .

Conclusion

The magnitude of the effect of insurance nonuntilization

The amount of the self-reported health care expenditure for the visit at the last illness While almost half of those who used insurance paid nothing for the visit, more than a third of those who did not use insurance paid 100,001 500,000 VND. The ratio of those who paid more than 1 million VND for the visit is also higher for those who did not use the health insurance. More than a forth of those who did not use the health insurance eventually pay no less than 50,000 VND, in which case there is not so much benet for utilizing the health insurance.
not use HI
.5

.3462 .2692 .1795 .1154 .0513 .0385

Density

use HI
.5
.4672

.1839 .1395 .0698 .0465 .074 .019

~50k

~100k

~500k

~1,000k

~5,000k

5,000k~

Motivation and Research Question

Data

Empirical Results on Insurance Utilization . . . . . . . . . . . . .

Conclusion

The magnitude of the effect of insurance nonuntilization

Econometric analysis

The characteristics of those who used the health insurance and those who did not are different. To estimate the average treatment effect on the treated of insurance nonutilization, we use the nearest neighbor matching and the regression adjustment with propensity score weighting.
The latter is doubly robust in the sense that consistency of the estimators only requires either the conditional mean model or the propensity score model to be correctly specied, not both.

We include as the same set of the covariates used in the analysis of insurance nonutilization. Though these variables may not sufce to explain the difference between the two groups, it is likely that the estimates are conservative ones.
If there remains some unobservables which affect the expenses, a major factor will be selection and moral hazard. People do not utilize insurance because their expenses are not so high. People who use insurance will receive more healthcare service as the costs are discounted.

Motivation and Research Question

Data

Empirical Results on Insurance Utilization . . . . . . . . . . . . .

Conclusion

The magnitude of the effect of insurance nonuntilization

Impact of Insurance Nonutilization on Healthcare Expenditure Columns (1) and (2) use the self-reported healthcare expenditure (In total, how much did you pay for this visit?). Columns (3) and (4) use a second outcome variable which is obtained by subtracting the amount of health insurance reimbursement elicited by the question How much were you paid by health insurance in total at that time?. The true value of the effect of the health insurance nonutilization should be somewhere between Columns (1) or (2) and Columns (3) or (4). The estimated effect of insurance nonutilization on the healthcare expenses is around 0.9-1.1 million VND, which is about 1.5 times higher than the average of the healthcare expenses.
Table : Health care expenses for the visit (1) NNM 843084.5*** (198194.2) 549 (2) Reg Adj PSW 1084540* (571180.2) 124 (3) NNM 974463.9*** (187435.4) 549 (4) Reg Adj PSW 1181121* (659005.8) 124

HI nonutilization Observations

Motivation and Research Question

Data

Empirical Results on Insurance Utilization . . . . . . . . . . . . .

Conclusion

Conclusion Though many has been attempting to expand the coverage of the health insurance, it may not be sufcient for protecting people from health-related nancial shocks. If insurance is not utilized or claims are not admitted, having health insurance will be of no use. In Vietnam, insurance nonutilization is actually an issue. The likelihood of insurance utilization is signicantly and substantially lower at the public central hospital, which in turn discourages people from visiting there.
Probably due to the lack in incentives in overcrowded public hospitals? In private facilities, they will concern attracting many patients to raise prot, and ensuring insurance utilization will help them increase the number of patients by providing discount for the medical fees. Staff in public facilities may not care for attracting more patients as it will not increase their income. Further, more patients mean more work given the same salary, and administrative works for insurance utilization put additional works. In crowded public facility, staff are already faced with a heavy workload and thus tend to discriminate the insurees.

The average effect of insurance nonutilization on the healthcare expenses per visit is around 1 million VND, which is 1.5 times larger than the average healthcare expenses per visit of our sample. Barriers to insurance utilization may partly explain the mixed results of health insurance on healthcare utilization and OOP payment across countries. Self-targeting?? Target for those whose time costs are low.