Changes in out-of-pocket payments for healthcare in Vietnam and its impact on equity in payments,

Transcription

1 * Title Page (showing Author Details) Changes in out-of-pocket payments for healthcare in Vietnam and its impact on equity in payments, July 2007 Corresponding Author: Anoshua Chaudhuri, PhD Assistant Professor Department of Economics San Francisco State University 1600 Holloway Avenue San Francisco, CA Kakoli Roy, PhD Economist Centers for Disease Control and Prevention Office of Workforce and Career Development 1600 Clifton Road NE Mail stop E94 Atlanta, GA JEL Codes: I1, I3, O5 Keywords: out-of-pocket payments, healthcare, Vietnam, vertical equity, horizontal equity Word Count: 3,647 Disclaimer: The findings and conclusions in this report are those of the authors and do not necessarily represent the views of the respective agency or institution.

2 Abstract EXTENDED ABSTRACT Background: Post independence, Vietnam experienced a gradual breakdown in its wellorganized public health infrastructure. Economic reforms initiated in the late 1980s and early 1990s introduced rapid privatization, an increase in user fees for both public and private health facilities, and subsequently, health insurance. Objective: To examine the relationship between out-of-pocket (OOP) health expenditures and ability to pay (ATP) among Vietnamese households during Data: The data are drawn from and Vietnam Living Standard Survey (VLSS) and 2002 Vietnam Household and Living Standards Survey (VHLSS). Methods: We use a probit model to estimate the probability that an individual will seek treatment. Then, using ordinary least-squares and fixed effects models, we estimate the relationship between household consumption (proxy for ATP) and individual OOP payments for healthcare, as well as the relationship between consumption and OOP share in consumption. Based on the analysis, we estimate the predicted share of individual OOP health payments according to consumption quintiles and selected socioeconomic characteristics. Results: Our results indicate that utilization (payments) increased with increasing ATP, but the consequent financial burden (payment share) decreased with increasing ATP, indicating a regressive system during the first two periods. However, share of payments increased with ATP, indicating a decrease in regressivity in payments during When comparing across years, we find horizontal inequities in all the years with some improvements by 2002.

3 Conclusion: That OOP payment share declines with increasing ATP might be because the rich could avail of health insurance more than those at lower incomes. As a consequence, the wealthy are able to use the healthcare system more effectively without paying a high OOP payment. In contrast, the poor either incur higher OOP payments or are discouraged from seeking treatments until their ailment becomes serious. This inequality becomes exacerbated in 1998 when insurance take-up rates were not high, but the impact of privatization and deregulation was already occurring. By 2002, insurance take-up rates were much higher, and poverty alleviation policies (e.g., free health insurance and health fund membership targeted for the poor) were instituted, which may have resulted in a less regressive system. Word Count: 350 Keywords: out-of-pocket payments, healthcare, Vietnam, vertical equity, horizontal equity

4 SHORT ABSTRACT Economic reforms were initiated in Vietnam during the late 1980s and early 1990s that led to rapid privatization and an increase in user fees for both public and private health facilities. This paper examines the determinants and changes in out-of-pocket health expenditures and share of health expenditures in household consumption among the Vietnamese during , using data from and Vietnam Living Standards Survey and 2002 Vietnam Household Living Standards Survey. Our results indicate that utilization (payments) increased with increasing ability to pay across all three periods. However, the consequent financial burden (payment share) decreased with increasing ability to pay, indicating a regressive system during the first two periods in our study. By 2002, we find that the system became less regressive. Comparing across population characteristics, we find evidence of horizontal inequities, which seems to have improved during Word count: 144 Keywords: out-of-pocket payments, healthcare, Vietnam, vertical equity, horizontal equity

5 * Manuscript (without Author Details) 1. Introduction Vietnam was one of the five poorest countries in the world until the mid-1980s, with an estimated gross domestic product (GDP) per capita of US$130 in 1985 [1]. In 1987, extensive market-oriented Doi moi (renovation) policy reforms were introduced to replace the centrally planned economy with a regulated market economy. Aggressive implementation of these policies resulted in a real GDP growth rate of 7%, a trade surplus, and a reduction in poverty rates [2]. Health-sector reforms were an integral part of Doi Moi that led to extensive changes in healthcare delivery, access, financing, and utilization [3]. The most significant change in the health sector was the introduction of user fees for healthcare encounters. This type of out-of-pocket (OOP) financing for healthcare was in stark contrast with the former socialized system of free medical care accessed through the communes. The increase in share of OOP expenditure related to healthcare imposed financial burdens on Vietnamese households and individuals, creating barriers to seeking adequate quantity and quality of care. Financial burden can render a household into debt, causing an endless cycle of poverty and ill-health. Therefore, a critical task for policymakers is to assess how OOP payments might have impacted individual s and households health status and health utilization. In this paper, we examine the determinants and changes in OOP health expenditures and its share in household consumption for Vietnamese households during a 10-year period by using data from and Vietnam Living Standard Surveys (VLSS) and 2002 Vietnam Household Living Standard Survey (VHLSS).

6 From an historical perspective, before reunification, North Vietnam had an extensive and well-planned network of public health providers who administered basic curative and preventive services, free of charge, at the commune level by using commune resources. After reunification in 1975, this system was extended throughout South Vietnam, but limited resources ultimately made this network financially unsustainable. During the late 1980s, rapid privatization resulted in the collapse of the cooperative system, leaving the communes resource-starved and unable to provide free medical services. Further disintegration of the Soviet Republic in 1991 stopped the supply of subsidized drugs, leading to serious shortages. To deal with the financial crisis in the healthcare system, user fees were introduced in 1989, followed by a complete deregulation of the healthcare system in This allowed medical personnel to establish private practices, and user fees were introduced for services at both private and public facilities. These changes resulted in raising OOP health expenditures for Vietnamese households. By 1993, private OOP expenditures amounted to 84% of total expenditures in the health sector. Public funds that accounted only for 16% were disproportionately spent on delivery of care through hospitals. Households paid 100% OOP for drug-related expenditures, which comprised 80.1% of total healthcare expenditure [4]. The public sector stopped providing any outpatient services or drugs, but continued to provide almost all of the inpatient services. As a result, to access publicly provided care, individuals had to seek inpatient care in hospitals, whereas most of the rural communes that previously provided outpatient care were severely understaffed and experienced equipment failures and drug shortages. Consequently, the rural residents 2

7 were forced to either avoid seeking professional help or to use more expensive privately provided healthcare paid by OOP expenses. Because public funds were used to provide higher quality hospital care and these hospitals were mostly located in the urban areas, public expenditure disproportionately benefited urban residents [5]. The poor in Vietnam live mainly in the rural areas, further leading to disparities in access to care and OOP payments by households not only between rural and urban areas but also between the poor and the non-poor. To improve equity in health expenditures, Vietnam Health Insurance was introduced in August of The program was initially compulsory for formal sector workers but later made available on a voluntary basis [6]. Because of rising income disparities, healthcare funds and free insurance for the poor were introduced in subsequent phases in 1999 and Given this context, this paper explores the determinants of healthcare utilization and OOP payments and addresses the relationship between households ability to pay (ATP) and OOP payments, and payment shares. Prior studies have examined Vietnamese health utilization by using data from 1993 and 1998 [7, 8, 9]. The poor in Vietnam seem to restrain their health-seeking behavior, but the expenditures they incur are high in proportion to their income [3, 5]. This leads to a regressive system of health payments where the poor pay a higher proportion of their income for healthcare than the non-poor. An important aspect of the paper is that it allows examining changes in the regressive healthcare expenditures across time and to assess whether regressivity has improved, worsened, or stayed the same during We extend the literature by using a new data set (VHLSS 2002) and examining changes in health payments and payment shares 3

8 for individuals during the decade that experienced immense changes in the Vietnamese healthcare system. According to the World Health Organization's fairness in financing index, households should be required to pay for healthcare in line with their ATP [10, 11]. Households of unequal ability should make appropriate dissimilar payments (vertical equity), whereas households of the same ability should make the same contribution (horizontal equity). An existing body of literature that explores equity in healthcare financing makes cross-country and within-country comparisons by using bivariate [3, 12 14] and multivariate analyses [15]. We use a multivariate regression-based approach that incorporates the impact of other socio-demographic variables when assessing the relationship between ATP and OOP payments for healthcare. Household is considered the sharing unit for income and payments, but the individual is the unit of analysis. We use household consumption expenditure to proxy for household s ATP. We examine vertical equity in health payments by examining healthcare utilization and burden of healthcare expenditures across five consumption quintiles. We also explore horizontal equity by examining health payments by ethnicity, age, gender and region of residence. Most significantly, we look at changes in vertical and horizontal equity in health payments across time. The paper is organized as follows: Section 2 describes the methods used in the study; Section 3 contains results; and Section 4 is the discussion and conclusion. 2. Methods 2.1. Analytical model 4

9 Healthcare expenditure of the i th individual in j th household is (1) M ij = P j (q ij ) X Q ij where P j is the price faced by the household; q ij is the quality of care; and Q ij is the quantity of care that the individual seeks. Log transforming the equation, we can express it as (2) Log (M ij )= log[p j (q ij )] + log (Q ij ) where, (3) q ij = f( Y j, X ij, Z j ); that is, quality of care depends on household income (Y j ), individual characteristics (X ij ), and household characteristics (Z j ), and (4) Q ij = f( D ij, X ij, Y j ); that is, quantity or amount of care received depends on the individual s health condition (D ij ), individual characteristics, and household income. Hence, our relevant empirical model is a reduced form expenditure function, (5) M ij = F( Y j, D ij, X ij, Z j ) + error term. We use household consumption expenditure to approximate a household s income or ATP. Expenditures are commonly used to proxy for effective income in developing countries, because expenditures are less likely to vary than income and are less prone to reporting biases. Using household consumption aggregates provides little choice but to assume that all members of each household enjoy the same level of economic well-being. Although using a unitary model ignores the important question of the intra-household distribution of consumption in a bargaining framework, we incorporated household size and age structures in our analysis to obtain insight into intra-household allocation. 5

10 We used the expenditure model to examine the determinants of incurring OOP health expenditures. One drawback is that this excludes individuals who do not report payments because they do not seek care or are otherwise deprived because they cannot afford care. Our primary focus was on examining the burden of health payments conditional on incurring these expenditures, which is an interesting question in itself. However, to reconcile the selection bias from truncating the sample with our objective of finding the level of burden from health OOP payments, we used a two-part model [16]. The first part estimated employing a probit model the determinants of the likelihood of a healthcare encounter (equation 4). In the second part, we estimated the determinants of utilization, particularly the impact of income (proxied by per capita household expenditure [1]) on OOP health payments, conditional on incurring a nonzero health expenditure (equation 5) Data The data are drawn from (henceforth 1992) and (henceforth 1998) VLSS and 2002 VHLSS. The 1992 VLSS comprises 4,800 households; 1998 VLSS comprises approximately 5,999 households; and the 2002 VHLSS comprises 31,854 households. Although the 2002 survey covers substantially more households, it has fewer details on health-related information, compared with the previous data sets. These differences will be elaborated later in this section. All of these data sets are socioeconomic surveys that provide nationally representative 1 information on selfreported illness, and associated healthcare utilization and expenditure patterns VLSS is not nationally representative, but we used sampling weights to produce nationally representative estimates. 6

11 The study samples included all individuals to examine the probability of having an encounter with a healthcare provider. The reference period of encounter is not uniform across the three periods. The first two periods provided information on encounter with a provider during the previous 4 weeks, whereas 2002 data have information regarding provider contact during the previous 12 months. Not all individuals report having incurred health expenditures; hence, in the second part of our empirical model, we restricted our samples to individuals who have spent non-zero amounts on healthcare during the previous 12 months. Not all individuals who are ill have provider encounters, and not all individuals who have health expenditures have an encounter with a provider. By considering all individuals who have non-zero health payments, we tried to capture the extent of financial burden faced by such individuals. Our unit of analysis is the individual (although household would also be useful) because we wanted to focus on capturing the individual-level variation. Our study samples are 24,068 individuals in 1992, 28,623 individuals in 1998, and 138,432 individuals in Statistical analysis In the first part of the study, we estimated, by using a probit model, the determinants of using a healthcare provider. As discussed earlier, the recall period used to collect data during the different years limited our ability to use exactly comparable dependent variables across years. Hence, for 1992 and 1998, we used a categorical variable of whether the individual sought treatment or not during the previous 4 weeks. For 2002, we used a categorical variable indicating whether the individual had met with a provider during the previous 12 months. 7

12 We estimated the second part by using ordinary least-squares (OLS) and fixed effects methods. Our dependent variables are log of individual OOP health-related expenses (annual expenditures measured in thousand Vietnamese Dongs) and log of OOP as share of ATP (consumption expenditure). The share is calculated as the ratio of individual OOP healthcare expenses to aggregate household consumption expenditures. We use a fixed effects model to eliminate all community-level unobservables in the error term that might be correlated with consumption expenditures. Because coefficients from a log-linear model hold little intuitive content, except to give us the direction of correlation, we estimated retransformed spending levels and shares by using Duan s [17] smearing estimator. Independent variables in both models included age (in years); square of age; dummy for sex (male = 1; 0 otherwise); marital status (married = 1; 0 otherwise); education (years of schooling completed); ethnicity (dummy indicating ethnic minority; i.e., not Kinh or Chinese); log of per capita household consumption expenditure; household head s education (years of schooling); dummy for region (urban = 1; 0 otherwise); and provider type (private = 1; 0 otherwise). To control for healthcare utilization, we used a dummy whether the individual was afflicted with an illness or injury in the reference period. 2 This information was not available for 2002; hence, for estimations using 2002 data, we did not use this explanatory variable. All estimates are corrected for clustering at the commune level and robust standard errors are reported. Data for 1998 are not representative for which sampling weights have been used. 2 In 1992 and 1998, we have information on whether an individual was ill or injured during the previous 4 weeks. 8

13 We used a multivariate regression framework adopted in previous studies of this nature [15] to explore the relationship between OOP payments and ATP by controlling for relevant factors that are correlated with ATP and might influence payments. To assess vertical equity, we examined how predicted OOP payments for healthcare, both absolute and as a share of household consumption, varies with increases in ATP (measured by consumption expenditure). Log-transformed mean predicted shares 3 are presented across five consumption quintiles, measuring the first (poorest), second, third, fourth, and fifth (richest), setting all other covariates at their sample means. To assess horizontal equity, we compared whether predicted OOP expenditures, both absolute and as a share of consumption, vary among individuals across two groups distinguished by a non income-related characteristic, but are otherwise similar in terms of ATP and other non income-related characteristics. The non income-related characteristics that are varied to assess horizontal equity across the two groups are their ethnicity (whether Kinh and Chinese or other ethnic minority), gender (male or female), age groups (whether aged 5 years or 60 years), and region (residing in rural versus urban area), controlling for ATP and all other non income-related characteristic that might affect ATP. The relevant categorical variable was set to zero or one (instead of the sample average) to estimate for each comparison group, mean predicted shares, across consumption quintiles, while setting all other covariates at their sample mean. Data management and statistical analyses were performed by using Stata Version 8.0 (StataCorp LP, College Station, Texas) [18]. 3 Mean predicted payments were calculated but have not been reported. 9

14 3. Results Table 1 presents the distribution of key socioeconomic characteristics and health utilization patterns across household consumption expenditure quintiles for all three study periods. A notable difference evident across the years is the increased use of private healthcare providers and the increased take-up of health insurance in the later year. 4. Healthcare utilization is higher among the upper consumption quintiles. The proportions that have health insurance and also receive more private care are higher in the upper quintiles. Upper quintiles report higher contact with healthcare providers but lower incidences of illness, compared with the lower quintiles. Higher proportions of individuals in the lower quintiles live in the rural areas, and a greater proportion of ethnic minorities are in the lower quintiles. Table 2 presents results from the probit regressions for the three periods. The table displays the determinants of the probability of an encounter with a healthcare provider. Table 3 reports the coefficients where log of individual OOP payments is regressed over log of household consumption, controlling for other individual and household characteristics, by using OLS and commune-level fixed effects estimation. Table 4 reports the coefficients for the models where log of share is regressed over log of household consumption, controlling for individual and household characteristics, by using OLS and fixed effects estimation. While the coefficient estimates of all of the covariates are contained in Tables 2 4, we only highlight the principle findings from our regression analyses in this discussion. Individuals who are males, are more educated, or live in urban areas are less likely to report healthcare encounters and are associated with lower OOP expenditures. 4 Provision of health insurance started in 1993; hence, we only have this information for 1998 and

15 Married individuals and children aged less than five years are more likely to report encounters with providers and are associated with higher OOP payments. The ethnic minorities are associated with lower OOP payments. Private care is correlated with higher spending. In 1998, we observe a negative association between OOP payments and having health insurance. However, this reverses in 2002, with OOP payments positively correlated with having health insurance. For a poor individual, having health insurance is associated with higher OOP payments. Although per capita expenditure is positively associated with probability of seeking care as well as making an OOP payment, an increase in per capita expenditure is associated with lower shares of OOP payments in aggregate consumption in 1992 and The shares of OOP payments are, however, positively correlated with per capita consumption in Based on the regression output, we estimated the predicted mean absolute payments and share of health expenditure according to consumption quintiles. OOP payments in absolute terms (measured as thousand Vietnamese Dongs) are progressive (i.e., increases for higher consumption quintiles). Results captured by share of consumption taken up by OOP payments, however, reveal differences in the redistributive effect. The predicted shares are reported in Figure 1. Individuals in the upper quintiles pay a lower share, indicating a regressive system for 1992 and This is consistent with earlier findings. What is most interesting is that the health payments as a share of consumption became less regressive in Although those in the poorest and richest quintiles pay higher shares of their income on health-related expenditures, those in the middle quintiles pay according to their ATP. 11

16 The predicted mean shares according to select socioeconomic characteristics are presented in Table 5. Clearly, horizontal inequities exist in the OOP payments where predicted payment shares are higher for females, compared with males; urban residents pay a higher share than rural residents (with the exception of 1992 where the rural rich paid more than the urban rich, possibly to pay for quality care); and ethnic Vietnamese and Chinese pay more than ethnic minorities, indicating inequality in access and utilization. The vertical and horizontal inequities became more pronounced in 1998 but substantial improvements are noted by Discussion Our main conclusion is that OOP payments in Vietnam, which were regressive during the 1990s, demonstrate having become less regressive by Vertical and horizontal inequities in health payments also have improved. Our findings indicate that the poor tend to seek less care despite reporting greater incidence of illness. There is evidence in literature that corroborates this finding [5]. Previous evidence also suggests that the poor in Vietnam have been known to circumvent seeing a provider by obtaining cheap and easily available over-the-counter medication to treat illnesses [8]. The increase in predicted OOP payments at higher income levels, demonstrating the progressive nature of payments, is similar to many developing countries in Asia (e.g., India) [15]. However, the shares of OOP payments are lower at higher income levels, indicating a regressive system of healthcare payments in Vietnam. Studies speculate that Vietnam s regressive nature of health payments might have decreased because the increase in user fees might have deterred the poor from using healthcare services. 12

17 Further, introduction of insurance also might have helped reduce the regressive burden of OOP payments [19]. Although the burden of OOP payments was less regressive, we found that the disparity in payment share between the richest and the poorest increased sharply by Our findings indicate that by 1998, only 16% of the population was insured. Insurance coverage was higher among individuals with higher ATP. The insured were more likely to use publicly provided health services, especially inpatient services in government hospitals. Use of private health facility and drug vendors was higher among the uninsured [8]. As a result, the uninsured spent larger OOP payments for accessing healthcare. Hence, health insurance coverage protected the rich from high OOP payments more than the poor. Other reasons for the disparity possibly were the increase in user fees, shifting payment responsibility from the state to the patients, and increasing use of more expensive private facilities. The disparity in payment share, however, showed improvements by 2002, as indicated by the decline in predicted OOP shares across all income levels, coupled with a more equitable distribution of payment shares across the consumption quintiles. This might have resulted from different factors. By 2002, use of health insurance as a risk protection was far more widespread (approximately 28%), which might have resulted in lower shares of OOP payments for all income levels. The introduction of health funds and free insurance for the poor might have been successful in providing protection against the risk for higher payment shares. By 2002, the poorest quintiles were paying an average of a fourth less (in absolute terms) than the highest quintiles, compared with 33% in In terms of payment share, the poorest quintile was paying similar proportion of expenditure as the richest quintile rather than a higher proportion, as witnessed by the 13

18 regressive system in the earlier years. For the middle three income quintiles, share of payments are indicative of being progressive (i.e., increasing with increases in ATP). There might be two possible explanations for this. Because the poor are known to have a higher price elasticity of demand for healthcare, a reduction in effective prices as a result of insurance coverage or healthcare cards might have contributed to increased utilization of primary and preventive care, making care less expensive for those with payment assistance [6]. Alternatively, the poor might have responded to the market oriented system by not seeking care unless absolutely necessary, thereby incurring lower medical expenditures on average. This has important policy implications. Although health financing in Vietnam might have appeared to become fairer by 2002, it would not in fact be an improved system if the poor were seeking less care despite reporting greater incidence of illness than the rich, as evident in this study. This might have negative consequences on the overall health status of the population in the long term. Future research should examine if disparities in health status exist across quintiles and whether this is getting better or worse over time. Progressive nature of health payments can be interpreted as a positive indicator only if it is accompanied by the poor not being compromised or impeded in seeking needed care. Future research should also explore the effect of OOP payments on debt burden imposed on households and individuals across different income levels and the role insurance or public assistance plays in mitigating the intensity of the burden. 14

19 References [1] Glewwe P. An overview of economic growth and household welfare in Vietnam in the 1990s [Chapter 1]. In: Glewwe P, Agrawal N, Dollar, D, eds. Economic Growth, Poverty, and Household Welfare in Vietnam. Washington, DC: The World Bank; [2] Adams SJ. Vietnam s health care system: a macroeconomic perspective, Paper presented at the International Symposium on Health care systems in Asia, Tokyo, Japan, [3] Wagstaff A, van Doorslaer E. Catastrophe and impoverishment in paying for health care: with applications to Vietnam, Health Economics 2003; 12: [4] Gertler P, Litvak J. Access to health care during transition: the role of the private sector in Vietnam [Chapter 8]. In: Dollar D, Glewwe P, Litvak J, eds. Household Welfare and Vietnam s Transition. Washington, DC: The World Bank; [5] Segall P et al. Economic transition should come with a health warning: the case of Vietnam. Journal of Epidemiology and Community Health 2002; 56: [6] Jowett M, Deolalikar AB and Martinsson P. Health insurance and treatment seeking behaviour: evidence from a low-income country. Health Economics 2004;12: [7] Deolalikar AB. Vietnam growing health: a review of Vietnam s health sector. Human Development Sector Unit Report. Washington, DC: The World Bank; [8] Trivedi, P. Patterns of health care use in Vietnam: analysis of 1998 Vietnam Living Standards Survey Data [Chapter 11]. In: Glewwe P, Agrawal N, Dollar D, eds. Economic Growth, Poverty, and Household Welfare in Vietnam. Washington, DC: The World Bank;

20 [9] Deolalikar AB. Access to health services by the poor and non-poor: the case of Vietnam. Journal of Asian and African Studies 2002;37: [10] Wagstaff, A and van Doorslaer, E. Equity in health care finance and delivery, [Chapter 34] in: A. J. Culyer & J. P. Newhouse (ed.), Handbook of Health Economics, edition 1, volume 1, chapter 34, pages , Elsevier. [11] Wagstaff A. Reflections on and alternatives to WHO s fairness of financial contribution index. Health Economics 2002; 11: [12] Van Doorslaer E, O Donnell O, Rannan-Eliya RP, et al. Catastrophic payments for health care in Asia. Health Economics 2007; In press. [13] Kakwani NC. Measurement of tax progressivity: an international comparison. Economic Journal 1977; 87(345): [14] Aronson JR, Johnson P, Lambert PJ. Redistributive effect and unequal tax treatment. Economic Journal 1977; 104: [15] Roy K, Howard DH. Equity in out-of-pocket payments for hospital care: evidence from India. Health Policy 2007; 80: [16] Yip W, Behrman P. Targeted health insurance in a low income country and its impact on access and equity in access: Egypt s school health insurance. Health Economics 2001;10: [17] Duan N. Smearing estimate: a nonparametric retransformation method. Journal of the American Statistical Association 1983;78: [18] StataCorp LP. Stata statistical software: Release 8.0. College Station, TX: StataCorp LP;

21 [19] Wagstaff A. Measuring equity in health care financing: reflections on and alternatives to the World Health Organizations Fairness of Financing Index, World Bank Policy Research Working Paper No Washington, DC: The World Bank;

22 Lowest 20% Fig. 1 Predicted payment share for healthcare costs across income quintiles 18

23 Figure Predicted individual OOP payments as a share of household expenditure 2.5% 2.0% 1.5% 1.0% 0.5% 0.0% 1 st 2 nd 3 rd 4 th 5 th Consumption quintiles

24 Table Table 1 Selected socioeconomic characteristics, by consumption quintiles, conditional on positive out-of-pocket (OOP) health payments st 2 nd 3 rd 4 th 5 th Male 44.9% 47.9% 49.2% 46.6% 47.2% Urban 6.1% 8.0% 12.5% 23.2% 49.4% Ethnic minority 14.6% 17.4% 12.2% 8.4% 4.9% Ill in the past 4 weeks 37.5% 34.6% 34.6% 32.2% 32.2% Treated in past 4 weeks 10.6% 11.8% 11.6% 11.3% 14.4% Seen private provider 0.3% 0.3% 0.2% 0.4% 0.8% Annual OOP health payment (in thousand VND*) Annual household expenditure per capita (in thousand VND) Male 45.3% 48.3% 49.0% 47.1% 48.4% Urban 8.0% 11.0% 18.4% 29.3% 64.3% Ethnic minority 26.0% 16.1% 10.8% 7.6% 1.8% Ill in the past 4 weeks 14.6% 13.9% 13.2% 12.9% 14.1% Treated in past 4 weeks 56.1% 49.2% 46.6% 43.1% 40.2% Seen private provider 7.8% 7.5% 6.4% 7.4% 8.2% Has health insurance 9.8% 12.2% 15.0% 16.9% 24.0% Annual OOP health payment (in thousand VND) Annual household expenditure per capita (in thousand VND) Male 47.4% 49.5% 49.6% 49.8% 49.4% Urban 5.7% 9.0% 15.0% 25.9% 56.5% Ethnic minority 25.0% 19.9% 17.1% 10.7% 4.3% Ill in past 12 months Seen a provider in past 12 months 46.1% 50.5% 54.2% 57.0% 63.6% Seen private provider 4.5% 4.7% 5.7% 6.8% 9.1% Has health insurance 13.8% 22.9% 26.4% 34.1% 42.5% Annual OOP health payment (in thousand VND) Annual household expenditure per capita (in thousand VND) *VND: Vietnamese Dong

25 Table 2 Probit estimates of the probability of an encounter with a health provider Age (yrs) ** *** (2.3229) (1.1600) ( ) (Age in yrs)^ *** *** (0.5522) (3.2838) (8.5217) Male *** *** ** (3.6150) (5.3973) (2.2359) Married *** *** *** (2.6018) (7.5577) ( ) Years of schooling *** *** ** (2.8352) (5.0137) (2.1880) Urban *** *** (0.1213) (6.9090) (2.5879) Household head s * *** schooling in years (1.1737) (1.6542) (6.8058) Log of per capita *** *** *** household expenditure (4.9897) (8.2889) ( ) Ethnic minority ** ** (2.0177) (2.4506) (1.1204) Proportion of *** individuals 60 yrs (0.4168) (0.4810) (4.3273) Proportion of children *** *** *** aged 5 yrs ( ) (5.1624) ( ) Whether has health *** *** insurance (6.8782) (3.2416) Constant *** *** *** ( ) ( ) ( ) Observations Wald Chi-squared *** *** *** Degrees of freedom Robust z statistics are in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%.

26 Table 3 Regression estimates of the determinants of out-of-pocket health expenditure OLS FE OLS FE OLS FE Age (yrs) 0.03*** 0.03*** 0.01*** 0.02*** 0.01*** 0.01*** (9.29) (9.00) (4.86) (6.01) (5.06) (7.31) (Age in yrs)^2-0.00*** -0.00** (3.22) (2.05) (1.12) (0.97) (0.06) (0.37) Male *** -0.09*** -0.03*** -0.03*** (0.60) (1.30) (4.58) (5.83) (4.59) (5.26) Married 0.16*** 0.17*** 0.24*** 0.22*** 0.08*** 0.08*** (5.10) (5.81) (8.16) (9.04) (5.13) (8.36) Yrs of schooling -0.02*** -0.01** -0.03*** -0.02*** -0.01*** -0.01*** (4.64) (1.99) (7.27) (7.42) (6.53) (9.22) Urban *** -0.29*** (1.09) (0.15) (4.19) (27.00) Yrs of head s -0.04*** -0.01* -0.04*** -0.01*** -0.01*** -0.01*** schooling (5.82) (1.68) (7.72) (5.98) (3.97) (6.62) Log of per capita 0.74*** 0.55*** 0.62*** 0.55*** 0.83*** 0.84*** expenditure (16.19) (24.49) (14.94) (26.48) (39.00) (118.54) Private 0.62*** 0.71*** 1.12*** 1.16*** 0.99*** 0.99*** (5.11) (4.91) (25.70) (37.97) (24.10) (73.20) Ethnic minority -0.21* *** *** -0.38*** (1.76) (1.37) (3.45) (0.94) (9.18) (39.50) Whether ill in past 1.12*** 1.22*** months (16.81) (46.76) Proportion of *** 0.21*** individuals 60 (1.08) (0.69) (1.15) (0.21) (5.96) (9.57) yrs Proportion of 0.47*** 0.51*** 0.11** 0.19*** 0.23*** 0.23*** children aged 5 (10.78) (13.18) (2.38) (5.34) (20.04) (14.41) yrs Whether has ** -0.06** 0.10*** 0.11*** health insurance (2.08) (2.36) (5.93) (14.73) Constant -3.07*** -2.06*** -1.03*** -0.77*** -2.81*** -2.87*** (9.95) (12.89) (3.37) (4.74) (17.99) (52.28) Observations Adjusted R squared Number of communes OLS: Ordinary least squares; FE: fixed effects; Robust t statistics in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%.

27 Table 4 Regression estimates of the determinants of out-of-pockets health payments as a share of total household expenditure OLS FE OLS FE OLS FE (1) FE (2) Age (yrs) 0.04*** 0.04*** 0.03*** 0.03*** 0.01*** 0.01*** 0.01*** (10.50) (11.58) (8.23) (9.82) (8.34) (13.07) (13.03) (Age in yrs)^2-0.00*** -0.00*** ** -0.00** -0.00*** -0.00*** (4.26) (4.38) (1.41) (2.07) (2.24) (3.33) (3.28) Male * -0.10*** -0.10*** -0.04*** -0.04*** -0.04*** (1.63) (1.71) (5.49) (5.78) (5.33) (5.74) (5.76) Married 0.13*** 0.12*** 0.16*** 0.13*** 0.04* 0.03*** 0.04*** (3.34) (4.03) (5.05) (5.12) (1.76) (3.24) (3.26) Yrs of schooling -0.02*** -0.01*** -0.03*** -0.03*** -0.01*** -0.01*** -0.01*** (3.29) (3.13) (6.41) (8.67) (7.89) (11.48) (11.42) Urban -0.18** *** -0.41*** -0.41*** (2.15) (1.00) (4.37) (34.87) (34.86) Yrs of head s -0.03*** -0.01* -0.04*** -0.02*** ** 0.00** schooling (3.93) (1.95) (7.22) (9.23) (1.10) (2.06) (2.13) Log of per capita -0.17*** -0.23*** -0.22*** -0.19*** *** 0.03*** expenditure (3.54) (9.78) (5.02) (8.79) (0.53) (4.10) (4.45) Private 0.57*** 0.71*** 1.18*** 1.22*** 1.01*** 1.01*** 1.01*** (5.17) (4.67) (25.33) (38.07) (24.13) (69.78) (69.75) Ethnic minority -0.30*** -0.11** -0.39*** *** -0.49*** -0.48*** (2.77) (2.26) (3.75) (1.61) (10.93) (47.04) (46.90) Whether ill in past 1.18*** 1.28*** 12 months (16.09) (46.74) Proportion of 0.13* 0.14** 0.13* 0.14** 0.36*** 0.36*** 0.36*** individuals 60 yrs (1.91) (2.24) (1.78) (2.55) (8.47) (14.94) (14.93) Proportion of 0.62*** 0.64*** 0.22*** 0.26*** 0.31*** 0.31*** 0.31*** children 5 yrs (13.65) (15.76) (4.58) (6.98) (24.82) (17.69) (17.68) Whether has ** ** 0.01 health insurance (0.85) (2.09) (0.65) (2.04) (1.21) Has health 0.30*** insurance*poor (6.34) Constant -5.63*** -5.47*** -4.12*** -4.54*** -5.97*** -6.09*** -6.11*** (17.45) (32.63) (12.68) (26.58) (34.24) (103.71) (103.92) Observations Adjusted R squared Number of communes Robust t statistics in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%.

28 Table 5 Predicted shares of out-of-pocket health payments, by expenditure quintiles Male Female Rural Urban Minority Kinh Infants Seniors st quintile 1.87% 2.23% 1.99% 2.09% 1.65% 2.13% 1.40% 4.43% 2 nd quintile 1.63% 1.84% 1.71% 1.89% 1.47% 1.80% 1.25% 3.74% 3 rd quintile 1.56% 1.85% 1.74% 1.62% 1.41% 1.75% 1.20% 3.99% 4 th quintile 1.48% 1.69% 1.71% 1.39% 1.54% 1.60% 1.07% 3.98% 5 th quintile 1.26% 1.45% 1.67% 1.39% 1.67% 1.33% 0.90% 3.51% st quintile 1.79% 2.19% 1.91% 2.36% 1.09% 2.32% 1.14% 4.53% 2 nd quintile 1.46% 1.77% 1.57% 1.92% 1.02% 1.77% 1.17% 3.70% 3 rd quintile 1.34% 1.61% 1.49% 1.47% 1.02% 1.56% 1.16% 3.39% 4 th quintile 1.28% 1.54% 1.49% 1.41% 1.13% 1.47% 1.07% 3.68% 5 th quintile 1.06% 1.31% 1.38% 1.13% 1.47% 1.19% 0.98% 3.37% st quintile 1.18% 1.30% 1.19% 1.37% 0.65% 1.45% 0.92% 2.82% 2 nd quintile 1.12% 1.18% 1.13% 1.16% 0.69% 1.27% 0.97% 2.61% 3 rd quintile 1.12% 1.22% 1.19% 1.09% 0.78% 1.27% 0.94% 2.77% 4 th quintile 1.14% 1.26% 1.28% 1.07% 0.88% 1.25% 0.94% 3.04% 5 th quintile 1.11% 1.26% 1.41% 1.02% 1.13% 1.16% 0.95% 3.36%