Health insurance and public health care utilization in vietnam

To answer this question, this study will assess the impact of health insurance on health care utilization, particularly public health services through two purposes: medical examination a

VIETNAM THE NETHERLANDS

VIETNAM - NETHERLANDS PROGRAMME FOR M.A IN DEVELOPMENT ECONOMICS

HEALTH INSURANCE AND PUBLIC HEALTH

CARE UTILIZATION IN VIETNAM

BY

TRAN THE HUNG

MASTER OF ARTS IN DEVELOPMENT ECONOMICS

VIETNAM THE NETHERLANDS

VIETNAM - NETHERLANDS PROGRAMME FOR M.A IN DEVELOPMENT ECONOMICS

HEALTH INSURANCE AND PUBLIC HEALTH

CARE UTILIZATION IN VIETNAM

A thesis submitted in partial fulfilment of the requirements for the degree of

MASTER OF ARTS IN DEVELOPMENT ECONOMICS

By

TRAN THE HUNG

Academic Supervisor:

Dr TRUONG DANG THUY

HO CHI MINH CITY, DECEMBER 2014

Vietnam is in the process of improving health system To achieve this goal, the Vietnam Government attempts to expend the coverage of public health insurance which is an effective tool in low and middle income countries to finance health care provision (WHO, 2000) Although the insurance coverage increases significantly over the last ten years, the private expenditure on health is still high It only reduces 6%, particularly from 69.1% of total expenditure on health in 2000 to 62.9% in 2010 (WHO, 2013) This comes up with a question that whether health insurance improves access to care? To answer this question, this study will assess the impact of health insurance on health care utilization, particularly public health services through two purposes: medical examination and treatment A binary probit model is used to estimate the impact of health insurance on public health care utilization Then we investigate determinants of insurance enrollment to increase the number of insurance participators if insurance affects positively significant on health care use Data are obtained from Vietnam Household Living Standard Surveys (VHLSS) in 2010 The empirical results indicate that insurance has a positively significant effect on public health care utilization In other words, we can conclude that health insurance actually improve access to care Moreover, the results of insurance participation show that insurance enrollment is affected strongly by income and interaction terms of frequency

of illness It is also remarked that demand for insurance is different between five income quintiles Finally, household’s characteristics including household’s size, income and illness ratio affect significantly to insurance enrollment

This thesis is not only the result of my own effort, it also consists direct and indirect supports of other individuals and organizations I would like to express my deep gratitude to them

My academic supervisor, Dr Truong Dang Thuy, is the person that I would like

to thank firstly Without his comments and supports, I would not finish my thesis in time and as good as this

Furthermore, I would also like to acknowledge the Scientific Committee, the lecturers and staffs of Vietnam-Netherlands Programme for the knowledge and guidance during the period of studying and writing thesis

Last but not least, I am grateful to my family for create favorable conditions to help me learn better Finally, I would like to thank my friends, especially “HLNTTV Group” for their supports in the whole time of studying

HCMC, December 2014 Trần Thế Hùng

LIST OF FIGURES iii

LIST OF TABLES iii

CHAPTER 1: INTRODUCTION 1

1.1 Problem statement 1

1.2 Research objectives 4

1.3 Research question 4

1.4 Research scope and data 4

1.5 The structure of this study 4

CHAPTER 2: LITERATURE REVIEW 6

2.1 Relationship between health utilization and insurance 6

2.1.1 Health care usage theory 6

2.1.2 Theory of relationship between health insurance and health utilization

11

2.3 Empirical reviews of relationship between health insurance and Health utilization: 14

2.4 Theory of insurance participation: 23

2.5 Empirical reviews of insurance participation: 24

CHAPTER 3: RESEARCH METHODOLOGY 30

3.1 An overview of Vietnam health system and health care use 30

3.1.1 Provider network 30

3.1.2 Access and utilization of medical examination and treatment services

32

3.2 Overview of health insurance 34

3.3 Methodology and data 36

3.3.1 Methodology 36

CHAPTER 4: RESULTS 45

4.1 Descriptive statistic 45

4.2 Empirical results 50

4.2.1 Impact of health insurance on public health care use 50

4.2.1.1 Medical examination 50

4.2.1.2 Treatment 52

4.2.2 Determinants of insurance participation 54

CHAPTER 5: CONCLUSIONS AND POLICY IMPLICATIONS 63

5.1 Conclusion remarks and policy implication 63

5.2 Limitation and further research 67

REFERENCES 69

APPENDIX 75

Figure 2 1: Initial behavioral model of health services utilization 8

Figure 2 2: Modeling the effect of insurance programme on the use of health services 21 Figure 3 1: Proportion of seeking care in 2010 33

Figure 3 2: Timeline and roadmap of universal health insurance coverage 34

Figure 3 3: Trend in health insurance coverage from 1993-2010 36

LIST OF TABLES Table 3 1: Measurement of variables 40

Table 4 1: Descriptive statistics of using public health care services by purpose 45

Table 4 2: Descriptive statistics of insurance participation 45

Table 4 3: Descriptive statistics of continuous independent variables 46

Table 4 4: Public health care use and insurance enrollment by gender 47

Table 4 5: Public health care use and insurance enrollment by employment status 47

Table 4 6: Public health care use and insurance enrollment by area (rural) 48

Table 4 7: Public health care use and insurance enrollment by minor ethnic people 49

Table 4 8: Results of impact of health insurance on medical examination 50

Table 4 9: Results of impact of health insurance on medical treatment 52

Table 4 10: Results of insurance participation (household level) 54

Table 4 11: Results of insurance participation (individual level) 58

Table 4 12: Results of insurance participation by different income quintile 61

Over the period of 2002-2010, healthcare utilization in Vietnam increases dramatically It suggests that people pay more attention to their health As for 2010, the percentage of people having health treatment is about 40.9% Of which, the rates of inpatient and outpatient are 8.1% and 37.1% respectively There are two main kinds of health care services that people use in Vietnam, including public and private health care services The percentage of people using public health care services is nearly seventy percent; particularly, the ratio of inpatient hospitalized in public health services

is around 90.1% of total inpatient and 57.2% is the percentage of outpatient using public health care services in 2010

In the last ten years, Vietnam households still have to concern with a burden of health care expenditure The amount of money that people have to spend in health care

is much more than Government spending; private expenditure on health accounts for around 62.9% of total expenditure on health while general Government expenditure on health is around 37.1 in 2010 compared to Thailand with 25% of private expenditure

and 75% of Government expenditure on health (World Health Organization 2013) The major element that makes the large proportion of private expenditure is households’ out-of-pocket payment Out-of-pocket expenditure is about 93% of private expenditure

on health in Vietnam 2010 (WHO, 2013) An increase in out-of-pocket payment on health may lead households to sell their assets to be able to pay the treatment fees Most of households, especially poor households, have to pay such a substantial share

of their income for health service As the result, they are pushed into poverty (World Health Organization, 2004)

Health risk is probably the greatest threat to people’ lives because it impacts on their direct expenditure and it also reduces their health affecting to labor supply and productivity leading to income poverty (Asfaw, 2003) This author suggests that health insurance is an effective tool to deal with health risk for the poor In addition, health insurance is as a part of income protection because it reduces financial burden of treatment at low income levels (Jutting, 2003) Health insurance is also a tool in order

to create an equitable access to health services throughout the population at income countries (WHO, 2000) Ensor (1995) discusses that voluntary health insurance plays an important role in reforming overall health care system by making health service provision more efficient

low-Recognizing the important role of health insurance, many authors study the relationship between health insurance and financial risk protection or health, especially, impact of health insurance on health utilization Saksena et al (2010) state that health insurance has statistically significant positive impact on health care utilization of health services when people are needed For the poor, health insurance is

an effective tool which increases health care usage when they are sick (Jutting, 2003) Health insurance does not only rise health care utilization, but it also increases the

usage of physician services and preventive services and so it improves health (Freeman

et al, 2008)

Health utilization is affected by many determinants including demographic factors; social structures, characteristics of family and community (Anderson, 1995) The author argues that demographic variables such as age, gender, education have low mutability, so they cannot be altered to change utilization; and cultural backgrounds (ie, ethnicity, region) are not changeable to promote health care usage (Anderson & Newman, 2005) while personal/family and community’s characteristics which include

an important factor: health insurance are quite mutable and strongly associated with health utilization For example, the impact of health insurance on health care use has been demonstrated dramatically by The Rand Health Insurance Study such as the studies of Manning et al (1987) and Jutting (2003) As a result, we can conclude that increasing insurance participation is a good choice to accelerate health utilization; and

it is necessary for policy makers to adopt how the impact of insurance on health care utilization is and then assess what are determinants of insurance participation so as to create favorable conditions for people to join health insurance scheme, specially, for the poor who do not have enough resources to use health services

In this situation, the study will examine the effect of health insurance to health care utilization at public health care services with different purposes including health test and treatment In other word, we will hypothesize whether health insurance improves access to health care since many studies use health care utilization as a proxy for access such as Fox (1972); Aday & Anderson (1974; 1995) After measuring the impact of health insurance on health care usage, if the effect is positively significant meaning that health insurance actually improves access to health care, we then investigate determinants affecting to insurance enrollment Then, the results are used to recommend policy implications to improve insurance participation including:

administrating stringently the insurance participation of employees and financial intervention such as subsidies for different income quintiles, especially for low income households with high illness ratio

1.2 Research objectives

This study aims to identify relationship between insurance and public health utilization of people in Vietnam After that, determinants affecting health insurance enrolment are measured in order to improve insurance enrollment As such, there are two main objectives in this study:

- Impact of health insurance on health care utilization at public health services using data from Vietnam Household Living Standard Survey in 2010

- Investigating determinants which impact to join the insurance scheme of people Then, policy implications are recommended to increase the number of insurance participators

1.3 Research question

This research aims to handle the question: Does health insurance actually improve access to care at public services? If yes, how to improve insurance participation?

1.4 Research scope and data

The study examines the impact of insurance on health care usage of individuals and determinants affecting insurance participation of households and individuals using cross section data of Vietnam Household Living Standard Surveys (VHLSS) in 2010

1.5 The structure of this study

There are five chapters in this study which are organized as follow:

Chapter 2: literature review includes theory as well as empirical literature about the relationship between insurance and utilization, also the determinants of insurance

Chapter 3: research methodology which presents regression technique used and data collection

Chapter 4: empirical results The statistic description is presented first, and then explaining the empirical results The coefficients of all factors will be interpreted and discussed

Chapter 5: summarizes the main results and some policy implications

CHAPTER 2: LITERATURE REVIEW

2.1 Relationship between health utilization and insurance

2.1.1 Health care usage theory

The behavior of health utilization has traditionally explained in five different approaches including the sociocultural approach, the socio-demographic approach, the social-psychological approach, the organizational approach, and the social systems approach (Anderson, 1973)

For the sociocultural approach, health care usage is a part of a cultural complex and, as such, related to other social institutions in a society or subculture One example

of Shuval (1970) shows that the utilization of health services depends on the basic latent functions of catharsis, cooperation with social system through contacts with social institution, status achievement through such contacts, and the resolution of conflicts between magic and science Zborowski (1952) founds that responses to pain among ethnic groups are different when he attempted individual utilization behavior It means that cultural condition affects to personal recognition of symptoms and the responses to them

For the socio-demographic approach, variations of utilization behavior can be related to age, sex, education, occupation, ethnicity, socioeconomic status, and income

As the theory of Moore (1969), the utilization of health care can be view as a type of individual behavior which is a function of individual characteristics, characteristics of environment where they live and maybe the interaction of these individual and societal forces The author emphasized the individual characteristics and less paid attention to the societal impacts This means that health utilization affected mostly by characteristic

of individual themselves such as age, education, gender, health status and income, and

so on Moreover, utilization among various groups within a population is also different even when cost barriers are eliminated (Nolan et al, 1969)

For the Social-Psychological Approach, Stoeckle et al (1963) review much of the analytic literature on the seeking of medical care and outline three major factors in the patient’s decision of seeking care including individuals’ knowledge and attitudes concerning symptoms; attitudes and expectations regarding to health services; and individuals’ definition of illness Similarly, in studying illness behavior, Mechanic (1978) identified the theory of health seeking and found out various circumstances affecting to the decision of seeking care The first one is the salience of deviant signs and symptoms Individuals’ perception and tolerance of symptoms is the second and third Forth, disruption caused by illness affects to individual’s life Fifth is the frequency of illness and its persistence And the final circumstance is the individual’s knowledge and cultural assumptions of the illness

For the organizational approach, the structure of health care system is examined

to account for differences of health care behavior Regarding to Anderson’s study of comparing health services in the United State, Sweden and England (1972), the differences in the supply of physicians and hospitals’ beds leads to the changes of variation in the use of hospital Typically, if the supply of physicians and hospitals’ beds is deficient markedly, the use of health care services will be diminished Moreover, when the admissions increase, the average length of stays will drop The author also pointed out that each country has evolved a pattern of financing and organization that is consistent with the unique characteristics of its social and political systems Hence, intervention strategies are necessary

For the social systems approach, it has emerged as a way of understanding health utilization On the basis of social systems, in 1960’s, Anderson developed the initial

behavior model looking at three categories of determinants such as predisposing characteristics, enabling resources including factors which enable or impede use, and people’s need for care that affects to people’s use of health services (Anderson, 1995)

Figure 2.1: Initial behavioral model of health services utilization

In 1972, Anderson expended and refined the initial behavioral model in order to predict the effect of changes in social structure of population and of supply of health services including the supply of hospital beds, aggregate level of education, employment, income and socio-demographic characteristics such as age, ethnicity and ecological features on health utilization

In addition, the updated utilization model can be characterized by purpose, type and unit of analysis In the case of purpose, health care utilization is as primary care with stopping illness before it begins or secondary care with referring to the process of treatment or tertiary care with providing stabilization for long-term irreversible illnesses such as heart disease or diabetes For type characteristic, health care utilization is as a choice of health services such as Hospital, Physician, Drugs and Medications, Dentist, Nursing Home, and Other A final character describing the utilization is the unit of analysis which includes the contact with a physician during the

Predisposing

characteristi

cs

Enabling resources

Need Use of health

period of time or the using volume of services Although health care utilization has different characteristics, determinants affecting to use of health services are based on characteristics of population and health services (Anderson, 1995; Andersen and Newman, 2005)

In general, the extent of health care is to improve health which should be primitive in the description of consumers’ preferences Health care services would then

be demanded only as an input into the production of health, and the level of demand for services would be determined by the extent to which they satisfied the individual’s underlying preference for health Individuals use their available resources to achieve health, so their preferences for health are represented within a standard utility-maximizing framework All of alternative uses that individuals must have for their

resources to admit a choice are bundled into a generic good denoted c The utility

function of health care use is:

𝑢 = 𝑢(𝑐, ℎ) Where h is level of health that individuals enjoy rather than quantity of health care services consumed

The demand for medical care is not constrained to a choice of how much, but also

of what kind meaning that individual can decide how often to visit, as well as choose visiting various providers such as hospital, clinic, healer After having made these choices, consumers may also face the choice of what kinds of treatments they wish to adopt including the use of drugs and other remedies While many of these input decisions will be based on recommendations made by the provider, such

recommendations may be altered with variations in prices and incomes For an

individual with income m, the price vector defines a consumption vector as

(𝑐0, 𝑐1, … , 𝑐𝑛) = (𝑚 − 𝑝0, 𝑚 − 𝑝1, … , 𝑚 − 𝑝𝑛) The function of health care utility can be rewritten as

𝑢 = 𝑢(ℎ, 𝑚, 𝑝) Where m presents income and p is the price of medical services

The existence of such discrete choices requires more elaborate econometric techniques to estimate the demand curves The discrete choice can be modeled in an integrated fashion using a multilevel approach

𝑒̂(𝑥𝑖) = 𝑝̂ [𝜋𝑖 ̂ 𝑒1𝑖̂ + 𝜋1𝑖 ̂ 𝑒2𝑖̂ + ⋯ + 𝜋2𝑖 ̂ 𝑒𝑛𝑖̂ ] 𝑛𝑖Where: 𝑒̂ = 𝑒𝑗𝑖 ̂(𝑥𝑗 𝑖) is the estimated use of medical care by individual i who consumes service j

xi is a vector of regressors used to explain medical care use such as price, income and demographic variables

Category j = {1…n} assumed as a various types of medical care services including clinics, public hospitals, traditional healers, and so forth

And 𝑝̂ = 𝑝̂(𝑥𝑖 𝑖) is the estimated probability that individual i will consume some quantity of medical care; 𝜋̂ = 𝜋𝑗𝑖 ̂ (𝑥𝑗 𝑖) is the estimated conditional probability that individual i will use medical service j Formally, probability can be estimated as 𝑞̂ =𝑖

𝑝̂ 𝑖𝜋𝑗𝑖

In the case of dichotomous choice, there are only two alternatives (j=0) and (j=1), for example self-care and clinic The equation predicting medical care use above collapses to:

𝑒̂(𝑥𝑖) = 𝑝̂ 𝜋𝑖̂ 𝑒1𝑖̂ = 𝑞1𝑖 ̂ 𝑒𝑖̂ 𝑖This equation is composed of the probability that a clinic visit will be chosen (j=1), times the expected quantity of services purchased, conditional on use If there is an assumption that the quantity conditional on use is fixed, then one interesting thing is estimating probability of health care use, 𝑞̂ 𝑖

From the utility function above, it is clear that utility gained from choosing visit

of a clinic depends on health status, income and price; and utility can also gain from xi According to behavior theory of health care utilization, xi should be a vector of characteristics of individuals and also includes characteristics of households and communities where they live

Considering the utility index associated with the choice of a clinic visit over care, the utility form can be obtained as

self-𝑢1𝑖 = 𝑢1(𝑥𝑖, 𝑚𝑖, ℎ𝑖, 𝑝1) = 𝛼𝑥𝑖+ 𝛽𝑚𝑖+ 𝛾ℎ𝑖+ 𝜑𝑝1+ 𝜀𝑖1 (a)

Where: 𝑥𝑖 is a vector of characteristics of individuals, households and communities 𝑚𝑖 is income and ℎ𝑖 is health status of individual i p1 is the price of medical services that individual i consume

2.1.2 Theory of relationship between health insurance and health utilization

Nyman (2001) states that people purchase insurance in order to obtain the income transfer which is the difference between any given payoff if ill and the premium With this income transfer, people tend to consume more health care than they would without insurance and the income transfer can be described by utility theory

In the absence of insurance, a consumer with initial income, Y0 would like to maximize his utility when he is sick:

max 𝑈𝑠(𝑀, 𝑌) The budget constraint is: Y0 = 𝑀 + 𝑌

Where M is medical care and Y is residual income available for purchases of other goods With the price of medical care M and assume it is normalized by 1, demand for medical care is

𝑀𝑢 = 𝑀(𝑝, 𝑌0) = 𝑀(1, 𝑌0) When people purchase insurance, they have to pay premium R which cover expected expenses, and the price of medical care reduce from p=1 to c The budget constraint now is

Y0− 𝑅 = 𝑐𝑀 + 𝑌 Where: c is the coinsurance rate The demand for care with insurance becomes

𝑀𝑖 = 𝑀(𝑐, 𝑅, 𝑌0) Assume that premium is not fixed (because it covers expected expenses), so R should be a function of

𝑅 = 𝜋(1 − 𝑐)𝑀𝑖Where 𝞹 is the probability of illness and (1-c)Mi is health care expenses paid by the insurer It is also known as a payoff As a result, the ill consumer’s budget constraint after insurance is:

Y0− 𝜋(1 − 𝑐)𝑀𝑖 = 𝑐𝑀𝑖 + 𝑌𝑖

or Y0+ (1 − 𝜋)(1 − 𝑐)𝑀𝑖 = 𝑀𝑖 + 𝑌𝑖 Compared to budget constraint without insurance:

Y0 = 𝑀𝑢+ 𝑌𝑢

The spending with insurance (𝑀𝑖+ 𝑌𝑖) is larger than spending without insurance (𝑀𝑢+ 𝑌𝑢) by (1 − 𝜋)(1 − 𝑐)𝑀𝑖 which is known as income transfer The income transfer provides the additional income for people to consume additional medical care, goods and services

In conclusion, the conventional theory of health insurance holds that becoming insured acts like a reduction in the price of health care Newhouse (1978) argues that as

an important point in studying the relationship between health insurance and demand for care, insurance is like a subsidy for individuals to purchase medical care The reason is insurance lowers the per-unit price of care Health care utilization is now affected strongly by insurance and the price of medical services is distorted Therefore, the utility equation of using medical service at clinics (a) now should be

𝑢1𝑖 = 𝑢1(𝑥𝑖, 𝑚𝑖, ℎ𝑖, 𝑖𝑛𝑠𝑢𝑖) = 𝛼𝑥𝑖+ 𝛽𝑚𝑖+ 𝛾ℎ𝑖+ 𝜑𝑖𝑛𝑠𝑢𝑖 + 𝜀𝑖1

The probability of individual i using medical services at clinic is calculated as:

𝑞̂ = 𝐹( 𝑢𝑖 ̂ ) 1𝑖

Where 𝑢̂ is the fitted value of 𝑢1𝑖 1𝑖; and F is monotonically increasing with R

→[0, 1] It means that if the utility index is higher, the probability of visiting a clinic is higher

2.3 Empirical reviews of relationship between health insurance and Health

utilization:

To assess the impact of health insurance on health care usage, many studies pay attention to behavior of seeking care when people need which bases on analyzing determinants of individual, household and community characteristics

To begin, the study of Manning et al (1987) is one of empirical review done early The study reports on the demand for health services and the role of health insurance in The United State Particularly, the study focuses on the use of medical care measured

by different schemes: probability of any medical use, probability of any inpatient use, the number of outpatient visits rather than dental services or psychotherapy Although the study examines the impact of health insurance on the demand for medical care, the authors also employ other controlled covariates such as site, health status, socio-demographic, and economic variables The unit of analysis is individual level as the authors argued that most major factors belong to individual characteristics rather than family In order to control for other covariates, study applies analysis of variance (ANOVA) and multi-regression method Typically, the authors used two-part model to

be more robust and against selection models although the data are truly generated by a selection model

In the two-part model, one observed random variable is divided into two observed random variable, for example medical expenditure MED is decomposed into

“MED>0” and “MED|MED>0” The two random variables are used in two equations The first regression is the probit model which presents probability of using any medical care during a year of an individual The second equation is log-linear regression which measures total medical expenditure of users during a year

More formally, the probit and log-linear models for the dichotomy are below:

The probability probit model: 𝐼1𝑖 = 𝛽1𝑋𝑖 + 𝜇1𝑖

Where: medical expense is positive if I1i > 0, and 0 is otherwise

Xi is a vector of individual characteristics such as insurance status, age, gender, health status, etc

The log linear model for positive expense:

ln( 𝑀𝐸𝐷𝑖|𝐼1𝑖 > 0, 𝑋𝑖) = 𝛽2𝑋𝑖+ 𝜇2𝑖Where

E( 𝜇2𝑖|𝑋𝑖, 𝐼1𝑖 > 0) = 0

Xi is is a vector of individual characteristics such

𝜇2𝑖 is i.i.d and it is not assumed to be normal distributed

The results show that the use of outpatient services decreases significantly when persons do not have insurance or low percent of insurance plan While insurance plans have no significant effect on inpatient use by children, outpatient use by children and adults responds strongly on health insurance In general, the authors suggest that demand elasticity for medical care responds to cost sharing In other words, insured individuals consume more medical services than they would have if they paid full price

Using a logistic regression, Saksena et al (2010) measured the impact of insurance

on health utilization and expenditure in Rwanda Using survey data from Rwanda, the authors contribute the evidence that mutual health insurance (MHI) in Rwanda actually improve access to care by examining MHI effect on health care usage and financial

protection The unit of analysis is individual level who reposted demand A logistic regression is employed to run the utilization model with a binary utilization The form

of utilization model is:

𝐿𝑛Pr (𝑢𝑠𝑒 = 1)

Pr (𝑢𝑠𝑒 = 0) = 𝛽𝑋 Where: use=0 presents the base group of individuals who did not use both

inpatient and outpatient services

Use=1 presents people who use health services

X is a vector of age, sex, whether the household head had completed primary

education, household size, household expenditure quintile, region, household

insurance status

One problem that the authors had is “endogeneity” To deal with this problem, they use the Durbin-Wu-Hausman test to checking the endogeneity between health insurance and utilization and the result was insignificant The authors conclude that health insurance increases significantly health care utilization when people have demand Furthermore, the results also indicate that insured individuals purchased health services as double as uninsured

Third, empirical research of Sekyi and Domanban (2012) studies the relationship between the National Health Insurance Scheme (NHIS) and outpatient utilization of medical care and expenditure in Ghana based on analyses

of a household survey carried out within the Mfantseman Municipality to solicit cross-sectional information on households To assess the effect of NHIS membership on outpatient utilization and expenditure, the authors employed the two-part model developed by Manning et al (1987) The first part is the binary

logit model which presents the impact of insurance on probability of a person visiting a modern health services such as health centres/health post, district hospitals, and private hospitals The model takes the form:

𝑃𝑟𝑜𝑏(𝑣𝑖𝑠𝑖𝑡 > 0) = 𝑋𝑖𝛽 + 𝜇𝑖Where: dependent variable equals 1 if a person visits any modern provider; 0 is not

X is a set of covariates including insurance status, individual and household‘s

characteristics

The second equation is linear model estimating the level of out of pocket expenditure

on health at the point of visit

(𝑜𝑢𝑡 𝑜𝑓 𝑝𝑜𝑐𝑘𝑒𝑡 𝑒𝑥𝑝𝑒𝑛𝑑𝑖𝑡𝑢𝑟𝑒/𝑣𝑖𝑠𝑖𝑡 > 0) = 𝑋𝑖𝜙 + 𝜀𝑖Dependent variable is total out of pocket expenditure including cost of treatment, transports, medicaments (drugs), consultation and any other expenditure related to the use of modern healthcare services and also payment made to private providers not covered by health insurance

According to the authors, endogeneity and/or self-selection are the two challenges

of the study and they are received a lot of attention in many areas In the study, the Durbin-Wu-Hausman (DWH) class of test is employed to control endogeneity occurring when covariance of the insurance variable and the error term differs from zero which leads the coefficient biased and inconsistent The process of DWH test includes two steps Firstly, insurance variable is regressed with all other exogenous variables including dummies of employment status and formal sector workers as

proxies for instrumental variables; as a result, the residual terms- ê, is obtained

Secondly, the residual terms in utilization and expenditure- â is tested with ê If the coefficient ê is statistically significant from zero, one can assume that failure to reject

the null hypothesis: insurance is exogenous The study also includes health status to control for self-selection

The authors conclude that while the uninsured individuals report significantly worse health utilization, health insurance, however, is lower barrier for people to access to care, meaning that the insured would like to use more medical services at modern providers, particularly, outpatient care

Another research is the impact of school health insurance program (SHIP) on access to care in Egypt done by Yip & Berman (2001) According to them, improve access means increasing visits rate and reducing financial burden In other words, they assess the impact of health insurance on health utilization and out of pocket expenditure based on Egypt Household Health Care Utilization and Expenditure Survey in 1994 The authors did not separate medical providers: public and private providers because they want test the effect of SHIP on overall access If visits to public services are only counted, the results will be misinterpretation on overall access For methodology, the two-part model developed as part of the Rand Health Insurance Experiment was employed Specially, a logit model estimating the impact of SHIP on individual child’s probability of visiting a formal provider: public and private providers

is part one The model can be written as follow:

𝑃𝑟𝑜𝑏(𝑣𝑖𝑠𝑖𝑡 > 0) = 𝑋𝑖𝛽 + 𝜇𝑖

A log linear model estimating the level of out of pocket expenditure with positive use of health services is part two The equation can be written as:

𝑙𝑜𝑔(𝑜𝑢𝑡 𝑜𝑓 𝑝𝑜𝑐𝑘𝑒𝑡 𝑒𝑥𝑝𝑒𝑛𝑑𝑖𝑡𝑢𝑟𝑒/𝑣𝑖𝑠𝑖𝑡 > 0) = 𝑋𝑖𝛾 + 𝜀𝑖

Where: X is a set of independent variable including health status, socio-demographic information, health status, employment

β and γ are vectors of coefficient And μ and 𝞮 are error terms

The results of the study confirmed that SHIP has significantly improved access to care for the insured children through health insurance increases probability of visit and reduces expenditure The authors also imply that among the percentage of children with health problem, the proportion of uninsured children seeking care is much lower than children covered by the SHIP In addition, the lowest income children having insurance benefit the most in term of utilization (probability of visits)

Fifth is the study of Jutting (2003) estimating whether health insurance improves access to care using data from a household survey in rural Senegal The author measures the effect of community based health insurance on access by assessing the impact of health insurance on health utilization and out of pocket expenditure The author uses the two-part model developed as part of the Rand Health Insurance Experiment in the United State (Manning et al., 1987) Specially, a logit model assesses the probability of visiting a hospital:

𝑃𝑟𝑜𝑏(𝑣𝑖𝑠𝑖𝑡 > 0) = 𝑀𝑖𝛼 + 𝑋𝑖𝛽 + 𝜇𝑖 Where: Prob(visit>0) is probability of using health services

M stands for health insurance status

X is a set of individual, household and community characteristics

The log linear model estimates the level of out of pocket expenditure with

positive use of health services is part two:

𝑙𝑜𝑔(𝑜𝑢𝑡 𝑜𝑓 𝑝𝑜𝑐𝑘𝑒𝑡 𝑒𝑥𝑝𝑒𝑛𝑑𝑖𝑡𝑢𝑟𝑒/𝑣𝑖𝑠𝑖𝑡 > 0) = 𝑋𝑖𝛾 + 𝜀𝑖

When modeling the impact of health insurance on health care use, Jutting experienced the two challenges of both endogeneity and self-selection These are also the two problems that other authors encounter while they estimated the impact of health insurance on various outcomes such as health demand and financial protection (Waters, 1999; Yip & Berman, 2001) For endogeneity problem, the author applied the Hausman test: a reduction of participation form in a health insurance scheme is estimated firstly; then the predicted and actual observed variables are included in the equation of health care usage If the coefficient of predicted variable for membership is not significant, it means that membership is exogenous To control self-selection issue, the author includes a proxy for health status in the study Moreover, the total sample include sick and non-sick, member and non-member is included to control for a sample selection bias

The results show that membership has a strong positive effect on probability of hospitalization and it also has significantly negatively affected to expenditure in the case of hospitalization From the results, Jutting confirmed that community based health insurance actually improve access to care, particularly, hospitalization

Sixth, Water (1999) assessed the impact of public-financed health insurance on health utilization in Ecuador using data from the 1995 Ecuador Living Standards Measurement Survey The author employed bivariate probit estimation techniques to analyses the effect of insurance on the use of health care services In this study, Water experienced the endogeneity problem and gave a solution to solve it; and one implication is that using bivariate probit model can be correction for only one endogenous variable in each equation He firstly built up the model of health care demand (M) which is posited as a function of a set of exogenous variables (X) as well

as three potentially endogenous variables: health insurance affiliation (I), health status (H)

𝑀 = 𝑓(𝑋, 𝐼, 𝐻, 𝜀) Where: X is a combination of individual, household and community characteristics

The potential selection bias leads to potential endogeneity The first reason is the choice of enrollment or not in the insurance shceme The second is from the distinction between those sick and those not sick For example, if environmental and community factors influences health status; hence they also affect health care usage Health status then is potentially endogenous to health utilization (Gertler & Gaag, 1990)

Figure 2.2: Modeling the effect of insurance programme on the use of health

services

Source: Water (1999)

The equation estimates the impact of health insurance on health care usage is below:

𝑀𝑖∗ = 𝛽𝑋𝑖 + 𝛼𝑦𝑖+ 𝜇𝑖

For individual i, M i * presents health care utilization

y refers to any one of the potentially endogenous variables

The probit model implies that dependent variable M i * takes value of 0 and 1

M i * = 1 individual i seeks care and M i * = 0 means not seeking health care

y is expected with a positive value for the suspected endogenous variable and y is a

function of some or all of exogenous variables (X), as well as one or more identifying variables (Z):

𝛾𝑖∗= 𝛼𝑋𝑖+ 𝜑𝑍𝑖+ 𝜀𝑖

y i = 1 individual i selected positively if y i * > 0

= 0 individual i selected negatively otherwise

In this case, yi presents membership of insurance scheme and takes value 1 is member and 0 is not

Endogeneity or selection bias occurs when the y has correlation with error term μ

In other words, if the correlation between 𝞮 and μ differs zero, the y coefficient will be bias Maximum likelihood calculations are applied to estimate the model The model is identified when giving the assumption that both variance μi2 and 𝞮i2 equal to one as long as at least one identifying variable Z is not in X

The results indicate that the General Health Insurance program has a strong positive effect to the use of curative health care and no significant impact on preventive care after correcting for endogeneity resulting from selection bias

2.4 Theory of insurance participation:

According to Arrow (1963), risks involved in health care include two types Firstly, the risk of becoming ill which leads to loss in quality of life, cost of medical care, loss of productivity and more serious is death The second type is the risk of total

or incomplete or delayed recovery In order to against these risks, individuals would like to be insured because people are normally risk averse, and therefore, they pool risk

to insurers Regarding to theory of expected utility, there is an assumption that people strive to maximize the expected value of a utility function (in this case, it is the utility

of health insurance participation) and so we assume that the expected utility associated with each health insurance option is a function of attributes (Xi) and a vector of a household's socioeconomic characteristics (Ri), plus a stochastic error term (𝞮), according to Kirigia et al, (2005)

A household’s decision process can be expressed as:

𝐸𝑈𝑖𝑗 = 𝑔(𝑋𝑖, 𝑅𝑖) + 𝞮

Where: 𝐸𝑈𝑖𝑗 is the utility that ith household expects to derive from choosing jth health insurance option; j = 1 if a household has health insurance; j = 2 if a

household has no health insurance If 𝐸𝑈𝑖1 > 𝐸𝑈𝑖2, the ith household is a

membership; if 𝐸𝑈𝑖1 < 𝐸𝑈𝑖2, it refers no health insurance; if 𝐸𝑈𝑖1 = 𝐸𝑈𝑖2, the ith household is indifferent between the two options

The probability of membership is

𝑝𝑖1 = 𝑝( 𝐸𝑈𝑖1 > 𝐸𝑈𝑖2) The binary probit model assessing the probability of health insurance participation can

be written as:

𝑝𝑖𝑗 = 𝛽𝑋𝑖+ 𝜑𝑅𝑖+ 𝞮 (𝒃) Where: pij = 1 if individual i have insurance meaning that j = 1, and 0 (j=0) is no insurance β and φ are the estimated coefficients 𝞮 is the stochastic error term

2.5 Empirical reviews of insurance participation:

To assess the determinants of insurance participation, most of studies focus on analyzing utility functions, particularly, the choice of health insurance depends on income of household or individual; individuals’ and households’ characteristics; and/or community characteristics where they live

In the study of Wang et al, (2005), determinants of community based health insurance include: income and health status which play important roles in people’s decisions to enroll in health insurance because a share of low income people is greater than the high income people who pay for the same amount of premium and the less healthy have a greater potential to utilize medical services, and are therefore more willing to join health insurance schemes (Cutler and Zeckhauser, 2000); and other variables belong to socio-demographic characteristics In the study, a logistic regression is applied to measure probability of participation in the community based insurance scheme and data were collected from household surveys (Chinese Rural Health Insurance Study Group)

The results show that income and health status affect strongly on insurance enrollment Farmers with low income are less likely to join the scheme than medium income while high income people do not show their attitudes clearly because the high income coefficient is insignificant In term of health status, farmers with good health status do not refer to buy insurance than the poor health, however, people receive medium health are unclear since its coefficient is insignificant In addition, there is no different enrollment between male and female

Using probit model, Chankova et al (2010) study the national health insurance scheme in Ghana The authors use data conducted from baseline and endline household surveys which collected information about socio-demographic characteristic of

households and individual characteristics of household head, health insurance membership and health care use also The probit equation shows probability of individual participation is:

𝑃𝑟𝑖 = 𝑝(𝑋𝑖, 𝑍𝑖, 𝐻𝑖, 𝐶) + 𝜇𝑖Where: Xi is a set of individual characteristic such as gender, age which is

divided in four groups: 0-4; 5-17; 50-69; >70

Zi and C are Household’s and community characteristics

Hi is a vector of characteristics of a household head who controls activities in a family

The results indicate that female individuals refer to join the national health insurance scheme or people belong to a household which controlled by a female head

or a household participating in a community solidarity group are also likely to join the scheme Age also affect to join insurance scheme, particularly, children and elderly (>70 ages) would like to enroll in the scheme than other groups Maybe the reason is the age-based premium exemption policies Moreover, probability of enrollment increases together with an increase of head education and household income Specially, person who reported a chronic illness are more likely to join the national health insurance scheme

In the case of subsidized insurance program, Ataguba and Goudge (2012) employed probit model to estimate probability of health insurance enrollment in South Africa The model is:

𝑃𝑟𝑜𝑏𝑖 = 𝑝(𝑋)

According to Ataguba and Goudge, probability of children under 16 years old is influenced by characteristics of household head such as education, employment status, the civil status; hence variables X include also characteristics of head; household and individual characteristics The model also includes health status which uses a chronic health condition as a proxy A nationally representative household survey—the SACBIA survey taking place in all nine South Africa provinces is data in the study

In the results, health status is not a predictor of a scheme membership as its coefficient is not statistically significant while income quintiles have strong positively significant effect to insurance enrollment Individuals who belong to households with a formally employed head (government employee) or Wealthier households refer to join the scheme Probability of insurance participation increases if a person lives in a household with higher education, particularly, at least secondary education Children and elderly are more likely to join the scheme compared with adults and middle age people

Another empirical review about determinants of health insurance ownership was done by Kirigia et al, (2005) The study measures the factors affecting to insurance participation of women in South Africa Using data from a cross-sectional national household sample derived from the South African Health Inequalities Survey (SANHIS), the authors applied a logit model to examine the relationship between insurance ownership with demographic, economic and educational characteristics of South Africa women

Based on theory of expected utility, Kirigia et al developed a binary logit model indicating the household’s decision process to join an insurance scheme The model can be expressed as:

𝐸𝑈𝑖𝑗 = 𝑔(𝑋𝑖𝑗, 𝑅𝑖) + 𝞮

Where: 𝐸𝑈𝑖𝑗 is the utility that ith household expects to derive from choosing jth health insurance option; j = 1 if a household has health insurance; j = 2 if a household has no health insurance Xi is a vector of attributes Ri is a set of a household's socioeconomic characteristics And 𝞮 is a stochastic error term

In this study, the authors made an assumption that: if 𝐸𝑈𝑖1 > 𝐸𝑈𝑖2, the ith household is

a membership; if 𝐸𝑈𝑖1 < 𝐸𝑈𝑖2, it refers no health insurance; if 𝐸𝑈𝑖1 = 𝐸𝑈𝑖2, the ith household is indifferent between the two options Therefore, probability of insurance participation of household ith is the case that 𝐸𝑈𝑖1 > 𝐸𝑈𝑖2

𝑝𝑖1 = 𝑝( 𝐸𝑈𝑖1 > 𝐸𝑈𝑖2) The binary logit model assessing the probability of health insurance participation can

be written as:

𝑝𝑖𝑗 = 𝛽𝑋𝑖+ 𝞮 Where: pij = 1 if individual i have insurance meaning that j = 1, and 0 (j=0) is no insurance β and φ are the estimated coefficients 𝞮 is the stochastic error term Variable Xi include: health rating indicates health status; economic factors include income, occupation and employment; demographic factors include age, household size; social factors are marital status and education; Spatial and environmental factors include residence and environment rating; and behavioral factors include contraceptive use, alcohol use and smoking

The results indicate that people who reported excellent or good health have low demand for insurance In term of economic factors, the probability of insurance participation increases considerably as household income moves up while proportion

of people with insurance decreases as people are involuntary unemployment The authors also confirmed that people living in large households are less likely to join the scheme The higher level of education people have, the more demand for insurance Moreover, if people are married, they refer to buy insurance compared the single or

devoted people Behavior of people also affects to participate in an insurance scheme There are some interesting findings that people who are smoking are likely to have insurance while people using alcohol are not Finally, if people live in a good or excellent environment, probability of enrollment goes up considerably

Measuring determinants of participation in community based health insurance, Jutting (2003) employed a probit model using household data from a rural area in Senegal Following an approach of Weinberger and Jutting (2001), the author states that the insurance enrollment depends on the rational choice of individual weighting costs and benefits of membership There is an assumption that insurance participation

of a household (p) is a function of characteristics of individual household’s head (H) who makes a decision of enrollment or not, household income (y), household

characteristics (Z) and community’s characteristics (C) where they live The following equation is below:

𝑝𝑖 = 𝑓 (𝑦𝑖, 𝑍𝑖, 𝐻𝑖, 𝐶) The binary probit model estimating probability of insurance participation is:

The author experienced adverse selection in the model; hence to control it, the health status of a household (illness ratio) and of an individual (frequency of illness) is captured

In the study, Jutting measured determinants of insurance participation at two levels: household level and individual level to confirm largely the results

At the household level, the results indicate that probability of insurance participation of household increases as household income goes up Specially, high income households refer to join community based health insurance program while the low income are less likely to buy insurance The other important variable is religion, if household belongs to a specific religion such as Christian, probability of enrollment increases strongly In addition, ethnic variable also has a statistically significant effect

to have insurance Other variables such as age, gender, education of household head, and especially illness ratio which is a proxy for health status are insignificant

At individual level, gender affects strongly to buy insurance; however, male people would not like to join the insurance scheme The results also confirmed that individuals who can read or write or belong to other organizations refer to buy insurance Moreover, the elderly are likely to take part in the insurance scheme while children and adults are unclear because their coefficients are not statistically significant Other factors such as income and ethnic variables have similar effects on insurance participation as household level

These empirical studies show that individual and household characteristics affect strongly on health insurance participation Typically, income and health status play important roles Although health status variable is sometimes insignificant, it is an effective tool to control for a special problem such as adverse selection

CHAPTER 3: RESEARCH METHODOLOGY

3.1 An overview of Vietnam health system and health care use

3.1.1 Provider network

Vietnam health care system is divided into two main sectors including public and private In term of curative care services, there are 13,500 public facilities; and the number of private facilities is quite large and growing considerably in form of clinics and pharmacies Public sector provides inpatient health services with 1087 hospitals and 188,613 patient beds while there are only about 102 private hospitals with 7,124 private beds Private hospitals, however, is growing in numbers, especially in municipalities and provincial capitals providing an important source of competition It

is the reason that the number of hospital beds increases continuously and reach 21.9 per 10,000 inhabitants

Grassroots health care is identified as a priority because of its access advantages

in term of finance and geography Grassroots health care includes district, commune and village levels For district level, almost all districts have general hospital; in addition, there are also some regional polyclinics or regional maternity clinics in several districts The total amount of commune health station is around 10,926 which cover nearly 98.6% of all communes in the country Moreover, there are around 99,409 village health workers which accounts for 84.4% of all

The national hospital plays a leading role in providing tertiary care although high specialized care can be provided by provincial hospital or even private hospitals Although patients prefer going to public hospital, the share of outpatients visiting private clinics is relatively large People who face difficulty in accessing primary care services, either due to cost and distance refer to use traditional medicine which relies

on locally available herbs and other remedies There are four main used forms of traditional medicine such as herbal medicine, acupuncture, tradition al Chinese medicine and Tam Quat offered by most of commune health station (79.3%)

Vietnam health care system includes two types of curative care services: public and private (WHO, 2012)

The public health care service has five levels of services such as commune, district, provincial, national and sectorial The health care service at commune level is commune health station with 10,926 stations Most primary care services and national targeted health programs is delivered through this level to the population, especially in rural and mountainous areas Commune health stations provide hygiene, vaccinations, antenatal care, safe delivery and health education, screening examination, treatment and referrals for outpatients; and they also collaborate in out-reach activities with village health workers (VHW) There are 686 regional polyclinics and 615 hospitals at district level While regional polyclinics deliver some primary care services and act as satellite facilities, district hospitals provide basic inpatient treatment, emergency care, and treatment of common diseases The provincial level includes 376 hospitals and 53 traditional medicine hospitals which provide general services and traditional medicine Moreover, specialty clinics also belong to provincial level with 47 clinics providing outpatient services The national level has 44 hospitals which deliver curative care, sanatoria and traditional medicine services with intensive specialization and modern technologies National hospitals also provide technical support to services at lower levels and implement research and many specialties such as oncology and endocrinology The final level of public service is sectorial which includes 52 hospitals delivering basic and specialized curative care, including sanatoria and 759 clinics consisting of Sanatorium, polyclinics and health stations of other sectors like military, police, transportation

For the private care service, there are 102 hospitals that provide general and specialized curative care, primarily located in urban areas The private sector also consist a lot of registered private traditional medicine facilities with approximately 10,000 facilities that provide diagnosis and treatment using herbal medicines, acupuncture or other traditional medicine techniques learned in traditional medicine professional schools or family tradition

3.1.2 Access and utilization of medical examination and treatment services

In 2010, there is a high level of the total number of consultations and hospital admissions The number of outpatient visiting for examination and treatment in hospitals increases substantially in 2010 compared 2009, particularly, hospitals provide for 111,128,460 outpatient visits Moreover, the amount of inpatient admissions are 9,908,758, account for 3.6% increase over 2009 in general However, this constituted a 10.9% increase in the private sector; it is much higher than average increase

Access to medical examination and treatment services is affected by different factors including geographic, cultural, economic which is as ability to pay and social factors Typically, the proportion of seeking care between the poor and rich is quite equivalent in the case of inpatient care while outpatient care used by the poor is much lower than the rich

Figure 3 1: Proportion of seeking care in 2010

The share of people received health care services in government hospitals

increases over the year In 2010, outpatient visits in state hospitals accounts for 37.1% and 83.2% for inpatient care This share is different significantly between rural and

urban areas While the proportion of people receiving medical examination and

treatment in state hospitals is 51.5% of outpatient care and 90% of inpatient care,

outpatients in rural areas take 29.9% and 81% of inpatient

The health utilization among gender is lightly different; particularly, proportion of male using medical care in government hospitals is 37.5 of outpatient care and 84.3%

of inpatient admissions; and for female is 36.7 of outpatient care and inpatient admission is 82.4% Moreover, different age groups have different health care use While child group (0-4 ages) takes 22.9% of outpatient care at state hospital, the proportion of elderly group (>60) is much higher and accounts for 44.8%

Source: General Statistics Offices, VHLSS 2010