Long-run measurement of income-related inequalities in health care under universal coverage: evidence from longitudinal analysis in Korea

Abstract

Background

Many countries have sought to promote well-being for their entire populations through the implementation of universal health coverage (UHC). To identify the extent to which UHC has been attained, it is necessary to evaluate equity of access to use of needed care and the cost burden of health services for the country’s entire population. This study considers income-related inequalities in health care utilization and spending in a long-term perspective for the case of the Republic of Korea.

Methods

Exploiting longitudinal data from a nationally representative health survey from 2008 to 2018, this study investigates how income-related inequalities in health care in Korea have varied over time and examines the extent to which need and non-need factors contribute those inequalities, using an in‐depth decomposition analysis, allowing for heterogeneous responses across income groups.

Results

The empirical results show that overall health care utilization is disproportionately concentrated among the poor over both the short and long run. Income-group differences and non-need determinants, such as marital status and private health insurance, make larger pro-poor contributions to inequality in inpatient care use, while chronic disease prevalence greatly pushes outpatient care utilization in a pro-poor direction. The results regarding inpatient care expenses indicate a similar pattern of pro-poor bias. Long-run inequality favors the better-off in terms of outpatient care expenses, where the contribution of income-group differences has the largest impact.

Conclusion

My findings suggest that it is important for health care policy in Korea to focus on improvements in the health status and well-being of low-income groups, as poor people are likely to be in poorer health. Non-need contributors could worsen pro-poor inequalities if the economic disparity across households were to increase due to the demographic transition. Higher spending on inpatient care may be a heavier financial burden for low-income people. Additional supportive measures should be provided to prevent them from suffering economic hardship. By contrast, people in high-income groups may spend most on costly services in outpatient care, including uninsured services, with the help of private health insurance. Nevertheless, the expansion of income disparity should be alleviated even from a health care policy perspective.

Introduction

Many countries seek to promote well-being for their entire populations by achieving universal health coverage (UHC), which is one of the health-related targets proposed among the Sustainable Development Goals. According to the World Health Organization (WHO), UHC has the goal of ensuring that every individual, regardless of their circumstances, including standard of living, should be able to receive safe, effective, and high-quality essential health care services as needed at an affordable cost without the need for financial hardship [31]. Strengthening the health care systems plays an important role in making progress toward UHC: health financing that influences the level of people’s direct payments for the use of health services may be a key policy instrument for providing a population with equal access to needed services, along with other components of health systems, such as the health care workforce and organizations, service delivery, and health information [30].¹ To measure the extent to which UHC is attained, it is necessary to evaluate equity of access to use of needed care and the cost burden of health services for a country’s entire population, including the most vulnerable and disadvantaged in that society [31].

The Republic of Korea (Korea) first introduced mandatory health insurance based on a social insurance system in 1977, and this has been a major financing scheme for health care nationwide since then. Gradually expanding health insurance coverage,² Korea ultimately achieved UHC in 1989, with more than 90% of the population covered by national health insurance and the remaining falling under the tax-financed Medical Aid Program. However, the achievement of UHC does not immediately ensure that all the people have an equal access to essential health services for free of charge or at little out-of-pocket expenses so as not to cause financial hardship [31]. Specifically, there have remained some limitations in the Korean healthcare system such as insufficient benefits coverage and higher rate of catastrophic health expenditures among low-income households due to high out-of-pocket payments and uninsured services [16]. These problems emphasize the necessity of quantifying the degree of income-related inequalities in health care utilization and spending under the universal coverage in Korea.

Within the bounds of this universal health insurance scheme in Korea, managed by a single insurer (National Health Insurance Service), the government has taken a laissez-faire position in providing health services for citizens; health care delivery relies heavily on the private sector to directly respond to the increased demand for health care [13]. Health care providers are generally reimbursed on a fee-for-service basis, where the fee schedule set by the insurer is enforced only for insured services, with higher prices allowable for uninsured services at their discretion to increase their profit margins. In the absence of a gatekeeping system, patients have a high degree of freedom to choose health care providers at any facility level they wish so long as they can afford to pay for the services they need [13].

Despite the rapid accomplishment of universal health insurance coverage within a period of only 12 years, health financing in Korea has been characterized by the shrinking role of government and a limited range of covered services,³ as well as a greater dependence on private spending,⁴ which could result in weak financial protections from the benefits package. According to OECD health statistics for Korea, health care spending from public sources accounted for 57% of total health expenditures (OECD average of 71%), and the proportion of out-of-pocket payments and voluntary health insurance were 34% and 7%, respectively, of total spending (OECD averages of 21% and 4%) in 2017 [20]. The large share of out-of-pocket spending on health care is partly attributable to relatively high cost-sharing for insured services,⁵ and it is also driven by additional payments for increased uninsured services, most of which involve the adoption of new technology and medicines with uncertain levels of cost effectiveness [13]. To cover copayments for insured services and full payments for services not included in the benefits package, many Koreans purchase complementary private health insurance in recent years [22].⁶ The wide coverage provided by voluntary private health insurance, however, is likely to encourage beneficiaries to overuse health services. On the other hand, high out-of-pocket payments may lead to limited access to needed care for low-income groups due to the financial burden,⁷ which has caused inequity in health care utilization by different income groups.

Across a long period of time, many studies have been conducted to examine socioeconomic inequalities in the use of health care services in European countries. However, there has been little empirical study of inequity in health care utilization in Asian regions including Korea.⁸ Lu et al. [17], in a pioneering work on this issue in Asian economies, compared the equity performance of health systems with the egalitarian goals of Hong Kong, South Korea, and Taiwan around 2000. They showed that Korea appeared to feature almost equal distribution in outpatient visits overall but a strong pro-poor bias for outpatient care in health centers and inpatient admissions, accounted for by non-need factors, such as lower levels of education and unemployment, combined with significant pro-rich inequality in outpatient use of tertiary medical institutions. Kim et al. [8] demonstrated horizontal inequity favoring the better-off in both outpatient and inpatient care for the elderly in the late 1990s and early 2000s, and they also revealed that the prevalence of chronic disease, educational attainment and income level may have significantly contributed to that disproportionate distribution. Kim et al. [11] found that pro-poor patterns appeared in terms of the probability of using secondary care and inpatient care relative to a pro-rich tendency that emerged in the number of visits and inpatient stays in the late 2000s. They also showed a modest pro-rich inequity in the amount of medical expenditures due to the substantial contributions of income, education, and private insurance. Furthermore, Kim et al. [10] separately estimated two age groups, below and above 60 years old, in 2010 and 2011, finding that health care utilization was concentrated on the worse-off in general and equally distributed, especially in emergency care and inpatient care, for the non-elderly.⁹ On the other hand, larger amounts of medical expenses were seen for outpatient and inpatient care services among high-income groups, and pro-rich inequalities appeared to be greater among the elderly, who showed a higher need of health care utilization. Major limitations of these prior studies lie in the fact that they focused on the short-term socioeconomic inequity for several survey years based on the standard method of factor decomposition, while they have shown the mixed results for inequalities in health care depending on a period of time and age group.

Exploiting longitudinal data from a nationally representative health survey from 2008 to 2018, this study investigates how income-related inequalities in health care utilization and spending in Korea have varied over time and examines the extent to which different factors have contributed to them by using an in‐depth decomposition analysis, allowing for heterogeneity. This clearly differs from the previous studies mentioned above that capture a sequence of independent snapshots of inequalities for each year in several ways: I use short-run and long-run concentration indices as measures of the degree of inequality and employ an extended decomposition method that allows for variation in individual responses to need and non-need determinants across income groups. In short, this study adds to the literature by expanding the standard methods of the concentration index and decomposition analysis with the use of the Korean panel data to take into account medium- to long-term inequalities and heterogeneous responses to factor contributions. Longitudinal analysis also enables me to derive policy implications for the long-run mechanism behind the equity performance of the Korean health care system under the universal coverage, which would otherwise be missing from a series of short-term cross-sectional analysis.

The remainder of the paper is structured as follows: Section “Methodology” presents the empirical methods I use to quantify the degree of income-related inequalities and factor decomposition. Secton “Data” describes the data used in this study and presents the summary statistics. Section “Results” outlines the results for the concentration indices and reports the results of estimation in the regression and decomposition analysis. Section “Discussion” discusses the implications and limitations of this study.

Methodology

Concentration indices in the short and long run

The concentration index method developed by Wagstaff et al. [26] and Kakwani et al. [7] is a standard tool used in health economics to quantify the extent of socioeconomic inequalities in a health-related variable. The concentration index ($CI$) can be simply calculated as follows:

$$CI=\frac{2}{\overline{y} }cov\left({y}_{i},{r}_{i}\right)$$

(1)

where ${y}_{i}$ is the health-related measure for individual $i$ ($i=1,\dots ,N$),¹⁰$\overline{y }$ is the mean of ${y}_{i}$ for all individuals ($={\sum }_{i}{y}_{i}/N$), and ${r}_{i}$ is the individual’s fractional rank in the distribution of their socioeconomic status, that is, household income per equivalent household member¹¹; this value ranges from − 1 to 1 and becomes zero when the health outcome is equally distributed among individuals irrespective of their standard of living (the values of − 1 and 1 represent perfect inequality). When the concentration index takes a negative value ($CI<0$), the outcome measure (e.g., the use of health services) is concentrated on the poor, while a positive value ($CI>0$) indicates that it is biased toward the rich.

Because the concentration index above depicts the degree of inequality at a point in time, it corresponds to the short-run concentration index (${CI}^{t}$) as presented in Jones and López Nicolás [5] and Allanson et al. [1]. Alternatively, following those works, Eq. (1) can be rewritten as

$${CI}^{t}=\frac{2}{{\overline{y} }^{t}}cov\left({y}_{it},{r}_{i}^{t}\right)=\frac{2}{N{\overline{y} }^{t}}\sum_{i}\left({y}_{it}-{\overline{y} }^{t}\right)\left({r}_{i}^{t}-\frac{1}{2}\right)$$

(2)

where ${y}_{it}$, ${\overline{y} }^{t}$, and ${r}_{i}^{t}$ are defined in the same way as above for time period $t$ ($t=1,\dots ,T$). Similarly, they proposed that when longitudinal data are available, the long-run concentration index (${CI}^{T}$) over $T$ periods can be derived as

$${CI}^{T}=\frac{2}{{\overline{\overline{y}}}^{T}}cov\left({y}_{i}^{T},{r}_{i}^{T}\right)=\frac{2}{N{\overline{\overline{y}}}^{T}}\sum_{i}\left({y}_{i}^{T}-{\overline{\overline{y}}}^{T}\right)\left({r}_{i}^{T}-\frac{1}{2}\right)$$

(3)

where ${y}_{i}^{T}$ is the average health measure of individual $i$ after $T$ periods ($={\sum }_{t}{y}_{it}/T$), ${\overline{\overline{y}}}^{T}$ is the mean of ${y}_{i}^{T}$ for all individuals in $T$ periods ($={\sum }_{t}{\overline{y} }^{t}/T$), and ${r}_{i}^{T}$ is the individual’s fractional rank in the distribution of their average equivalized incomes over all $T$ periods. Note that both concentration indices over the short and long run have the same properties as the standard concentration index, in terms of an interpretation of the inequity.

This study utilizes both the short and long run concentration indices because the longitudinal data across 11 survey years are available, allowing for a deeper investigation into medium- to long-term income-related inequalities in health care use and spending in Korea.¹²

Decomposition method with heterogeneity

Inequalities in health-related variables across the income distribution can be decomposed into the contributions of their potential determinants [27]. First, the individual’s health measure ${y}_{i}$ is assumed to be explained by a linear combination of $J$ need variables ${x}_{ji}$ that are likely to directly influence the health outcome (e.g., age, sex, health status, physical condition, etc.) and $K$ non-need variables ${z}_{ki}$, which are generally defined as socioeconomic characteristics, including income level, such that

$${y}_{i}=\alpha +\sum_{j=1}^{J}{\beta }_{j}{x}_{ji}+\sum_{k=1}^{K}{\gamma }_{k}{z}_{ki}+{\varepsilon }_{i}$$

(4)

where ${\beta }_{j}$ and ${\gamma }_{k}$ are their corresponding coefficients, $\alpha$ is the intercept, and ${\varepsilon }_{i}$ is the error term. Wagstaff et al. [27] demonstrated that, based on the linear regression model in Eq. (4), the concentration index ($CI$) can be rewritten as follows:

$$CI=\sum_{j}\left({\beta }_{j}\frac{{\overline{x} }_{j}}{\overline{y} }\right){CI}_{{x}_{j}}+\sum_{k}\left({\gamma }_{k}\frac{{\overline{z} }_{k}}{\overline{y} }\right){CI}_{{z}_{k}}+\frac{2}{\overline{y} }cov\left({\varepsilon }_{i},{r}_{i}\right)$$

(5)

where ${\overline{x} }_{j}$ and ${\overline{z} }_{k}$ are the means of the covariates ${x}_{ji}$ and ${z}_{ki}$, ${CI}_{{x}_{j}}$ and ${CI}_{{z}_{k}}$ are their concentration indices with respect to the fractional rank in the income distribution,¹³ and the final term is the generalized concentration index for the error term reflecting income-related inequality in health that is not explained by any systematic variation in the regressors. In other words, the concentration index in the decomposition method can be defined as the weighted sum of the concentration indices of the explanatory variables ${x}_{j}$ and ${z}_{k}$, where the weights provide the elasticity of the health measure with respect to each factor, evaluated at the sample mean (i.e., ${\beta }_{j}{\overline{x} }_{j}/\overline{y }$ and ${\gamma }_{k}{\overline{z} }_{k}/\overline{y }$), plus the residual component [19]. Therefore, each term in Eq. (5) comprises factor contributions to the overall concentration index.

However, the standard decomposition method often involves the drawback that it only captures homogeneous responses to need and non-need determinants over the entire sample, due to the fixed parameters that are on average adjusted by the sample means. In addition, the contribution of the residuals is likely to be sufficiently large unless the regression model is well specified. Following Jones and López Nicolás [6] and Van de Poel et al. [24], I thus employ an extended decomposition method that allows for heterogeneity across certain socioeconomic groups. I hypothesize a heterogeneous responsiveness of health care to need and non-need factors according to individual income levels. Suppose that each individual belongs to one of $G$ groups differentiated by the level of equivalized income. Then, Eq. (4) can be transformed into the similar linear function of a set of the same need and non-need variables, excluding the indicators of the income group $g$ ($g=1,\dots ,G$), such that

$${y}_{i}={\alpha }_{g}+\sum_{j=1}^{J}{\beta }_{jg}{x}_{ji}+\sum_{k=1}^{K}{\gamma }_{kg}{z}_{ki}+{u}_{i}, \forall i\in g$$

(6)

where ${\beta }_{jg}$ and ${\gamma }_{kg}$ are the differential parameters by income groups, ${\alpha }_{g}$ is the group-specific intercepts, and ${u}_{i}$ is the error term. Based on the estimation of separate regressions for each group in Eq. (6), the concentration index in Eq. (5) can also be further decomposed into detailed factor contributions as follows:

$$\begin{array}{c}CI=\sum\limits_{j}\left({\beta }_{j}\frac{{\overline{x} }_{j}}{\overline{y} }\right){CI}_{{x}_{j}}+\frac{2}{\overline{y}N }\sum\limits_{j}\sum\limits_{i}{x}_{ji}\left({\beta }_{jg}-{\beta }_{j}\right)\left({r}_{i}-\frac{1}{2}\right)\\ +\sum\limits_{k}\left({\gamma }_{k}\frac{{\overline{z} }_{k}}{\overline{y} }\right){CI}_{{z}_{k}}+\frac{2}{\overline{y}N }\sum\limits_{k}\sum\limits_{i}{z}_{ki}\left({\gamma }_{kg}-{\gamma }_{k}\right)\left({r}_{i}-\frac{1}{2}\right)\\ +\frac{2}{\overline{y} }cov\left({\alpha }_{g},{r}_{i}\right)+\frac{2}{\overline{y} }cov\left({u}_{i},{r}_{i}\right).\end{array}$$

(7)

The first and third terms in Eq. (7) are the same as the first two terms in Eq. (5), obtained from the pooled regression, which indicates the homogeneous contributions of need and non-need factors, respectively, as their effects are constant over the entire sample. The second and fourth terms represent the heterogeneous contributions of the need and non-need determinants, respectively, defined as covariance between the differential parameters across income groups and fractional rank in income distribution, weighted by the values of the corresponding covariates. The fifth term refers to the direct contribution of income-group differences to income-related inequalities in the health outcome. We understand that it is transformed from the contribution of income level in the second term of Eq. (5), which is no longer captured in Eq. (7). The sixth term is the unexplained residual component of the concentration index, which is expected to be smaller than the last term in Eq. (5) due to the better specification, allowing for heterogeneity [24].

Data

Korea health panel survey

This study uses individual-level longitudinal data from the Korea Health Panel Survey (KHPS) for 2008 to 2018 (Version 1.7.2).¹⁴ The KHPS is a nationwide comprehensive survey carried out by the Korea Institute for Health and Social Affairs and the National Health Insurance Service on a household or individual basis, using a dually stratified cluster sampling frame of the National Population and Housing Census. It provides a variety of information on individuals’ health status and behaviors, health care utilization, and expenditure by type of care service (e.g., emergency care, inpatient and outpatient care, childbirth, long-term care, and medication utilization), covering the demographic and socioeconomic characteristics of individuals as well. The survey data also include sampling weights to enable adjustment for unequal selection probabilities and non-responses based on the distribution of population totals, which enable nationally representative estimates to be obtained.

Participants in the KHPS are required to collect receipts for each instance of health care expenses to alleviate the problems of recall bias and increase the credibility of the survey data. The complete dataset contains a full sample of 195,032 person-years in 68,347 household-years across the entire survey that are all available in this study as a 11-year unbalanced panel data set.¹⁵ New samples were selected and added to the panel in 2012 to ensure the reliability of the survey in response to the decreasing number of households and household members originally included in the sample who persisted in supplying data. These new participants’ data became available from the 2014 survey data as an aggregated panel with the original sample.

Outcome variables and need/non-need determinants

The health-related outcome measures of primary interest in this paper are health care utilization and spending in a year. I use six types of outcome variable: (1) length of hospital stay, (2) number of outpatient visits, (3) number of instances of emergency care use for health care utilization, (4) amount of inpatient care expenses, (5) amount of outpatient care expenses, and (6) total amount of medical expenses for health care spending. All of these outcomes are assumed to be continuous non-negative variables starting from 0.¹⁶

The need determinants of health care utilization and spending are proxied by individual’s age, sex,¹⁷ number of chronic diseases, and whether he/she is physically handicapped. The needs for health care services could also include variables such as self-reported health status, mental health problems, or various risk factors (e.g., smoking, drinking, eating habits, exercise, etc.), which are partly available in the KHPS. However, it would be better not to use these variables to prevent selection bias due to attrition. On the other hand, following previous studies on socioeconomic inequalities in health care, non-need determinants are defined as follows: individual’s income level,¹⁸ educational attainment,¹⁹ labor force participation,²⁰ marital status,²¹ number of household members, residential area,²² whether he/she receives public assistance, whether he/she has a private health insurance policy, and total amount of monthly premium for private health insurance. A binary variable of private health insurance represents the fact that individuals cannot be reimbursed for an identical health care use from multiple private health insurance schemes in Korea. To reflect the growing popularity of the purchase of voluntary private health insurance, even under the UHC, I also use the variable that captures variation in capacity to pay for insurance.

In addition to need and non-need determinants, survey year fixed-effects are also taken into account in the regression and decomposition analysis. Note that monetary variables, expressed in ten thousand Korean won (i.e., equivalized income, medical expenses, and monthly premium for private health insurance) are transformed into real values adjusted by the consumer price index for each year to compare them across survey years.

Descriptive statistics

Table 1 reports the descriptive statistics, including concentration indices for outcomes and need/non-need variables across the entire sample. The concentration indices for health care utilization show negative values, indicating that it is disproportionately concentrated on poorer people as a whole. Nevertheless, the utilization for inpatient care (about 2 days on average per year) is more biased toward the poor than outpatient and emergency care use (on average, 15.4 and merely 0.1 times per year, respectively). However, the concentration indices for health care spending demonstrate a different tendency: inpatient care expenses show a pro-poor concentration, while the inequality favors the better-off in outpatient care spending, which is higher than the former on average. Total medical expenses are almost equally distributed among all of the samples available, even if the concentration index has a small positive value with no statistical significance. The concentration indices for outcome variables in descriptive statistics are slightly different from but nearly equal to those calculated in the regression and decomposition analysis, where some of the observations are dropped due to missing values of other covariates than equivalized income.

Table 1 Summary statistics for outcomes and need/non-need variables

A graphical representation of the concentration indices for outcome variables is shown in the form of the concentration curves in Appendices A and B. The concentration curve plots the cumulative percentage of a health-related variable against that of the population according to socioeconomic status, from poorest to richest. The concentration index is equal to twice the area between the concentration curve and the 45-degree line of perfect equality [7]. If the health variable is concentrated among the poor (rich), the concentration curve lies above (below) the line of equality [19]. The concentration curves for health care utilization and inpatient care spending are plotted above the 45-degree line, due to the negative values of the concentration indices, while the opposite is true for the case of outpatient care spending. However, it is worth noting that the concentration curve for the total amount of medical spending apparently crosses the line of equality.

Table 1 also indicates that older people and females are more likely to belong to the poorer population, and having more chronic diseases and disabilities is more common among the poor. Individuals who have completed education beyond high school are concentrated in the richer groups, and those with lower education are biased toward the poorer groups. Approximately half of those sampled are married and have worked during the survey year, and these respondents are more prevalent among the wealthier people. Those who live in the capital regions were more than 40% of the samples and also showed a pro-rich prevalence, while public assistance recipients accounted for only 4% and were strongly concentrated in the poor group. Moreover, the richer population is likely to purchase private health insurance that has higher monthly premiums.

Results

Short-run/long-run concentration indices

Figure 1 shows changes in concentration indices for six outcome measures over the short and long run (Eqs. (2) and (3)), using a weighted average for the short-run concentration indices. The confidence intervals for concentration indices are also obtained from the linear regression. As with the descriptive statistics across the entire sample, both the concentration indices of health care utilization (inpatient, outpatient, and emergency care) show negative values with sufficient statistical significance, implying a disproportionate concentration of overall health care utilization among the poor over the short and long run. The concentration indices for inpatient care spending also demonstrate a pro-poor concentration, while outpatient care spending is consistently biased toward the rich over the long run (although this relationship shows no statistical significance over the short run in some later years). The total amount of medical expenses, however, is more or less equally distributed across the population, as the concentration indices are not statistically different from zero in most survey years.

Fig. 1

Concentration indices for health care outcomes

Note: CI stands for a confidence interval. SRCI stands for the short-run concentration index. LRCI stands for the long-run concentration index

There exist some differences between the long-run concentration indices and the weighted average of the short-run concentration indices for every outcome measure, stemming from a systematic association between changes in individual income ranking and differences in measures of his/her health care outcomes over the given time period [1, 5]. Although digging into the mechanism behind these deviations is not the mainstream of this study, a series of the short-run inequalities over the longer period of time are likely to be underestimations or overestimations of the long-run inequality [29].

Regression analysis

The estimation results of the pooled regressions over the entire sample (Eq. (4)) and separate regressions across income groups (Eq. (6)) for six outcome measures are fully reported in Tables 2 and 3. They indicate a linear association between health care outcomes and need/non-need determinants while allowing for heterogeneous responses according to income group. Note that the fixed parameters over the entire sample in Eq. (5) and the differential parameters by income groups in Eq. (7) constitute a certain portion of the homogeneous and heterogeneous factor contributions, respectively. Thus, it is important to present the sign and magnitude of each coefficient with respect to need and non-need factors.

Table 2 Estimation results for health care utilization

Table 3 Estimation results for health care spending

Among the need factors, age is positively associated with inpatient care utilization/spending and total medical expenses, but it is negatively correlated with outpatient and emergency care use ($p<0.01$). Females are less likely to use inpatient and emergency care than males, and they tend to use outpatient care more and to spend more on it, with higher spending on total medical care ($p<0.05$), and this effect tends to grow as income level grows. The number of chronic diseases shows a positive relationship with health care utilization and spending, as expected ($p<0.01$), and their impacts become smaller for health care use but greater for expenditures as income level goes up. Being physically handicapped is also significantly associated with increasing frequency of overall health care utilization and higher amount of medical spending ($p<0.05$), with the exception of spending for outpatient care.

Among the non-need factors, lower educational attainment than graduation from high school is significantly correlated with greater use of health care as a whole and greater expenses for outpatient and total medical care ($p<0.05$). On the other hand, those who have achieved higher education than high school graduates are less likely to utilize and spend on outpatient care (inpatient/emergency and total medical care only in the poorest group) ($p<0.05$). The working population reveals a negative association with health care utilization and spending, as expected, likely due to the healthy worker effect ($p<0.05$). People who are married tend to use more outpatient and emergency care and spend more on overall medical care ($p<0.05$), but they also show shorter hospital stays for inpatient care ($p<0.01$). The number of household members is negatively associated with health care utilization (except for inpatient care) and spending ($p<0.05$). Living in the capital regions is significantly associated with higher spending on outpatient and total medical care ($p<0.05$), although it is reverse-correlated with a decreasing frequency of overall health care use and lower expenses for inpatient care ($p<0.05$). We also find a clear contrast such that public assistance recipients are likely to utilize more health care services but spend less on them, owing to the tax-funded Medical Aid Program ($p<0.05$). Purchasing private health insurance raises the probability of spending more on outpatient and total medical care but leads to reduced utilization for inpatient and emergency care ($p<0.05$). Those who pay higher monthly premiums tend to increase their overall health care use and expenses ($p<0.05$). Nevertheless, positive gradients were not found across income levels in the effects of private health insurances on health care utilization and spending, as had been expected.

Decomposition analysis

The decomposition results of the concentration indices allowing for heterogeneity for six outcomes (Eq. (7)) are graphically displayed in Figs. 2, 3, 4, 5, 6 and 7. The corresponding results, expressed in numerical values and percentage shares, are also presented in Appendices C and D. It can be recalled that the homogeneous contributions of need and non-need determinants are evaluated as the product of the elasticity of health care measures with respect to each explanatory variable and the concentration index for each variable, whereas the heterogeneous contributions depend on the covariance of the differential parameters across income groups, with a fractional rank in the income distribution weighted by the values of the corresponding covariates. Note likewise that the direct contribution of income-group differences can be obtained from the covariance between the group-specific intercepts and the fractional income rank.

Fig. 2

Decomposition results for inpatient care utilization

Fig. 3

Decomposition results for inpatient care spending

Fig. 4

Decomposition results for outpatient care utilization

Fig. 5

Decomposition results for outpatient care spending

Fig. 6

Decomposition results for emergency care utilization

Fig. 7

Decomposition results for total amount of medical spending

Within the result for inpatient care utilization, age makes the largest positive contribution in total to the income-related inequality (− 0.254), where the positive heterogeneous contribution (i.e., the effect on length of hospital stay is stronger for high-income groups) overwhelms the negative homogeneous contribution, which is derived from its positive association with inpatient care use and pro-poor inequality in its distribution. In contrast, the number of chronic diseases forms a negative contribution to income-related inequality, mainly due to the negative heterogeneous contribution, in which the positive correlation is stronger for low-income groups. The total contribution of need factors takes a positive value (0.052) due to the greater positive effect of the heterogeneous contribution. Among the non-need determinants, marital status makes the largest negative (heterogeneous) contribution, where married people, who have their strong association with shorter days of hospital stay, belong to higher income groups. Another large negative (heterogeneous) contribution of private health insurance is shown in that the negative correlation with inpatient care use is stronger for high-income groups in the context of pro-rich inequality in its distribution. The total contribution of non-need factors (− 0.113) accounts for 44% of the income-related inequality, and the largest contributor is the direct impact of income-group differences (− 0.191), which accounts for 75%.

The result for inpatient care spending is similar to that for inpatient care use with respect to the direction of each need factor,²³ but the homogeneous and heterogeneous contributions of need determinants compensate for each other to some extent (− 0.009). The number of household members produces the largest negative (heterogeneous) contributions among the non-need factors, suggesting that individuals living with more family members, who have their significant association with reduced inpatient care expenses, are likely to become high-income groups. Marital status and private health insurance are also larger for negative (heterogeneous) contributors. As a results, the total contribution of non-need determinants (− 0.065) accounts for 72% of income-related inequality in inpatient care spending (− 0.091). It is noteworthy that the direct effect of income-group differences (− 0.015), which accounts for 17% of inequality, is also the important contributor, implying that individuals who belong to lower income groups are likely to spend more on hospitalization, despite the greater financial burden on them.

The decomposition result for outpatient care utilization shows a different picture, although outpatient care is also disproportionately concentrated on the poor. Income-related inequality in outpatient care use (− 0.129) is mostly attributable to the contribution of need factors (− 0.125), which accounts for 98%, among which the prevalence of chronic disease makes the largest negative (homogeneous) contribution, due to its positive association with utilization for outpatient care and pro-poor inequality in its distribution. Among the non-need determinants, educational attainment is the largest negative (homogeneous) contributor, in terms of the combination of the positive/negative effects and pro-poor/rich inequalities in the education dummies, whereas marital status,²⁴ residential area,²⁵ and the effects of survey years contribute positively (and heterogeneously) on a larger scale to income-related inequality. Thus, the total contribution of non-need factors (− 0.02) results in a smaller share of 15%, and the direct impact of income-group differences also makes less of a contribution in the opposite direction (0.016).

The result for outpatient care spending reveals an insightful pattern of homogeneous and heterogeneous contributions. We find that the status of public assistance is the greatest positive (homogeneous) contributor, deriving from its negative correlation to outpatient care expenses and strong pro-poor concentration, while number of family members and working status²⁶ make larger negative contributions among the non-need determinants. Consequently, the homogeneous and heterogeneous contributions of non-need and need factors turn out to nearly cancel out (0.007 and 0.001 in total, respectively). Therefore, a large share, 62%, of income-related inequality in outpatient care spending (0.036) can be accounted for by the direct contribution of income-group differences (0.023), suggesting that the better-off tend to have expenses from costlier outpatient care, probably including uninsured services.

The decomposition result for emergency care utilization shows that income-related inequality (− 0.072) is mostly explained by the contribution of non-need factors (− 0.066), which accounts for 91%, among which the number of household members, public assistance status,²⁷ and private health insurance make relatively larger negative contributions. It also shows a negative contribution according to need determinants (− 0.04), accounting for 55%, where the number of chronic diseases plays the most important role. However, the direct contribution of income-group differences involves a smaller share in the opposite direction (0.032). Finally, the result for overall medical care spending is found to be similar to that for outpatient care expenses, in terms of the contribution of each component. However, the offsetting effect of the contributions of need/non-need factors and income-group differences leads to small income-related inequality (0.003), which is close to perfect equality, indicating that total amount of medical expenses is almost uniformly spent across the population, irrespective of their income level.

These different patterns of the decomposition results reflect the demographic (age, sex, health status, and residential area) and socioeconomic (education, working status, household characteristics, private health insurance, and income-group disparity) differences in responses to health care outcomes. In addition, the differential effects of a welfare policy (public assistance to the poor) on health care utilization and spending could partly explain the different results for factor decomposition as an institutional background.

Discussion

This study investigates long-term, income-related inequalities in health care utilization and spending in Korea, and it examines the extent to which need and non-need factors contribute in a longitudinal setting using an extended decomposition analysis, allowing for heterogenous responses across income groups. The empirical findings are summarized and discussed as follows: First, we find a disproportionate concentration of overall health care utilization among the poor over the short and long run. Income-group differences and non-need determinants, such as marital status and private health insurance, make larger pro-poor contributions to inequality in inpatient care use, while the prevalence of chronic diseases greatly pushes outpatient care utilization in a pro-poor direction. Income-related inequality in emergency care use is mostly explained with the contribution of non-need determinants, such as the number of household members, as well as health status as a need factor, proxied by the distribution of chronic diseases. The pro-poor concentration of health care utilization and its decomposition results suggest that poor people consume more health care services because they are likely to be in physically worse condition and the chronic disease prevalence always becomes a large negative contributor. This finding is consistent with some of the previous studies such as Lee and Shaw [15] and Kim et al. [9]. It is important for health care policy in Korea to focus more on improvement in the health status and well-being of low-income groups. Additionally, this study demonstrates that household characteristics and private health insurance as non-need determinants also make a heterogeneously substantial contribution to pro-poor inequalities in health care use. These contributors could worsen the pro-poor concentration if the economic disparity between unmarried and married people or small and large families were to increase due to the demographic transition, given their negative association with health care utilization. Likewise, the growing popularity of private health insurance among higher income groups could also exacerbate the disproportionate tendency toward the poor.

By contrast, income-related inequalities in health care spending unveil insightfully different patterns, depending on types of care services, although total amount of medical care expenses is almost equal across the population, regardless of income level. Inpatient care expenses are biased toward the poor, and the decomposition result shows that non-need factors, such as household characteristics and private health insurance, and the direct effect of income-group differences contribute to most of the income-related inequality. This implies that higher spending especially on inpatient care may be a heavy financial burden to low-income people. Although the cost-sharing for insured inpatient care is set at the relatively lower rate of 20% and the cost-sharing ceiling scheme also works for insured care services, extra payments for uninsured services such as special treatments and room charges account for a large amount of high out-of-pocket expenditure on hospitalization [18].²⁸ Lee and Shaw [15] and Kim et al. [9] point out that poor people are likely to be provided with less sufficient or advanced care services, as the quality and intensity of care increase in direct proportion to income level, which could bring about longer periods of hospital stays with higher spending for them. Furthermore, an increase in the out-of-pocket payment for inpatient care is highly correlated with the probability of facing catastrophic health expenditure that could occur more often among vulnerable low-income groups [15, 18]. Thus, additional financially supportive measures should be provided for low-income people to mitigate their heavy burden of inpatient care spending and prevent them from suffering economic hardship. This may also lead to institutional issues in terms of the charging of inpatient care services. On the other hand, we find that long-run inequality favors the better-off in outpatient care expenses, while the direct contribution of income-group differences accounts for the largest share of overall pro-rich inequality. This finding implies that people in high-income groups are more likely to spend costly services for outpatient care, including uninsured services with the help of voluntary private health insurance, which currently brings about a policy debate on how to regulate uninsured health care services and the growing market for private health insurance. Nevertheless, the expansion of income disparity should be alleviated even from a perspective of health care policy, as a large part of pro-poor inequalities in inpatient care services are also driven by income-group differences.

This study has some limitations. First, the need and non-need determinants of health care utilization and spending, as defined above, might omit other potentially influential variables. For example, as noted, potential needs for health care services could include such variables as subjective health status, mental health condition, and lifestyle-related risk factors, which are not fully available for analysis. Other possible non-need factors could include such socioeconomic variables as individual expected rate of copayment or out-of-pocket payment, health insurance premium rate, and distance to nearest health care facilities, which are all difficult to calculate from the available dataset. Nevertheless, the residual components in decomposition analysis that are explained by a set of omitted or unobservable factors show small enough contributions, owing to the detailed specification allowing for heterogeneity. Secondly, individual heterogeneity is adjusted for only by sampling weights, although one of the benefits of using panel data is being able to control for individual fixed-effects as time-invariant unobserved heterogeneity. However, employing the fixed-effects model usually has the side effect of cancelling out other time-invariant variables, such as gender and educational attainment which contributions are preferred to be estimated in decomposition analysis. Again, relatively small contributions of residual components may imply that individual fixed-effects are also sufficiently small. Finally, as outcome measures in this study are defined by general types of health care (i.e., inpatient, outpatient, and emergency care), they do not take into account differences in quality of care. Decomposition results suggest that people in low-income groups are likely to utilize insured basic care services that are necessary for them, while the better-off tend to use and spend more on premium services, especially in outpatient care, that are not usually covered by national health insurance. Room remains for future research on examining socioeconomic inequalities in the use of quality-adjusted care services in the context of universal coverage.

Appendices

Appendix A

Fig. 8

Concentration curves for health care utilization

Appendix B

Fig. 9

Concentration curves for health care spending

Appendix C

Table 4 Decomposition results for health care utilization

Appendix D

Table 5 Decomposition results for health care spending