Income-related inequalities in health care utilization in Mongolia, 2007/2008–2012

Data

The data used in this study were adopted from the nationwide cross-sectional data
sets, the HSES, collected in 2007/2008 and 2012 by the National Statistical Office
of Mongolia. The aim of the survey is to evaluate and monitor the income and expenditure
of households, update the basket and weights for consumer price index, and it offers
inputs to the national accounts. The survey is conducted every year with three levels
of strata as Ulaanbaatar (the capital city), province centers and rural area by covering
all 21 provinces and the capital city of Mongolia. The HSESs are based on the standardized
questionnaires that reveal information on elements such as demographics, socio-economic
indicators, social transfers, health, housing and education, among others. In the
HSES 11,172 and 12,811 households were included in 2007/2008 and 2012, respectively.
These households consist of 44,510 and 47,908 individuals in total in 2007/2008 and
2012. Our main inclusion criteria was individuals, who were aged 18 and older. Additionally,
we excluded individuals who were: i) a household head or any household student members
away from home for 11 months or more; ii) anyone else away from home for 6 months
or more.

After we applied the inclusion criteria to the data, we removed cases with missing
data. We found that there were only 49 and 17 missing data on income in each year
and we eliminated them. Accordingly, 27,681 and 30,567 individuals retained in the
studies from 2007/2008 to 2012.

Dependent variables

Measurements of outpatient care utilization were based on whether individuals received
outpatient care by visiting any central hospital/clinic, district/aimag hospital/clinic, FGP/soum hospital as well as private hospital during the past 1 month or not (yes/no). In
Mongolia, primary health care services are delivered by FGPs/village health centers
in urban areas and by soum and intersoum hospitals in rural areas. We used the terms FGP and soum hospitals; however, FGP and soum hospitals were renamed family health centers and soum health centers respectively, according to a revision of the health act in 2011. Inpatient
service utilization was measured, if any hospitalization occurred in the past 12 months
(yes/no).

Independent variables

The HSES questionnaires in both years elicited wide range of information about household
income. In the analysis, only household net monetary income earned by the household
members during the reference years was used. We calculated household income on the
basis of sources of income, including wage from work, income from self-employment,
agricultural income, private income and pension, among others for both years. In the
next step, household income per equivalent adult was estimated in accordance with
the OECD modified equivalence scale, adopted by the Statistical Office of the European
Union, which is “1 to the household head, of 0.5 to each additional adult member and
of 0.3 to each child”.

Need variables used in the paper are age, gender and self-reported health. We generated
14 dummy variables based on age and sex (females aged 18–24, 25–34, 35–44, 45–54,
55–64, 65–74, and 75 or older; males aged 18–24, 25–34, 35–44, 45–54, 55–64, 65–74,
and 75 or older). Measurement of health variables is based on four questions which
were directly asked from individuals: (a) ‘Have you got any disabilities? (yes/no)’;
(b) ‘Did you have any health complaints in the past month? (yes/no)’; (c) ‘Did you
miss your work, school or daily activities due to the illness in last month? (number
of days)’; and (d) ‘Have you got any chronic illnesses? (yes/no)’ which is available
only in the 2007/2008 year’s data. Non-need variables are activity status, marital
status, education, location, household size and health insurance coverage. Marital
status is categorized into married/living together, divorced/separated, widowed and
single/never married. Activity status contains employed, herder, self-employed, inactive
and unemployed. Household size is a continuous variable. Location included urban and
rural areas. Health insurance is based on whether an individual is covered by the
social health insurance.

Measuring inequality

A wide range of measuring techniques are used to measure inequality in health and
health care utilization, including simple, regression based, and more advanced techniques
28]. Among them, the concentration index is the most commonly used method to calculate
the degree of income-related inequality in health care utilization 29] due to its direct link to the concentration curve, which shows a complete picture
of a share of health service by cumulative proportions of population ranked by income
1].

The concentration index indicates the covariance of the health care utilization and
the fractional rank of income distribution as:

(1)

where i is an individual, y _i
is the health care utilization, ? is the mean of the health care utilization (y), R _i
is the individual’s fractional rank in the income distribution and t is the year. The concentration index represents the concentration curve as a single
number by summarizing the inequality weights at different points in the income distribution.
The concentration index falls within a range of ?1 and +1. If a value of the concentration
index is a negative, it indicates that health care utilization is concentrated among
the pro-poor. When a positive value index appears, it shows that health care utilization
is concentrated among the pro-rich group 1].

The concentration index depends on a mean value of the health variable (health care
utilization). Thus, Wagstaff stated that the concentration index has a limitation,
which occurs when health care utilization is binary, because as the mean increases,
the concentration index shrinks 30]. Consequently, Erreygers introduced the Erreygers’ concentration index (EI) as a
solution for the drawback of the standard concentration index 31], and it is more compatible with binary variable and formulated as this:

(2)

where C(h) represents the standard concentration index presented in equation 1. The ? is the mean of health care utilization in population. b _n
and a _n
are the upper and lower bound of health care utilization. This study used EI owing
to the variable’s binary nature.

Horizontal inequity

In this study, we estimated horizontal inequity to assess avoidable inequity in health
service utilization in the population. Apparently, health care utilization differs
among and across the populations as regards the income differences because health
care needs differ in the population due to, for example age, gender, health status,
and this difference is unavoidable. Therefore, in order to assess if health care utilization
is equally distributed in the population regard to income distribution, one should
control varying need variables. Thus, horizontal inequity is expressed by a difference
of actual inequality in the population and need-standardized utilization of health
care. In other words, standardization for differences in need explains unavoidable
inequity in health care utilization and a difference between the concentration index
and unavoidable inequity demonstrates avoidable inequity in health care utilization
1].

Need standardization

We used the indirect standardization method to measure horizontal inequity in health
care utilization. Owing to the nature of a binary variable, in general, a non-linear
estimation is applied. However, studies on health equity, which have used both a linear
and non-linear estimation, showed that the results were consistent in both models.
Therefore, we used ordinary least square regression (OLS) 1], 32]. First, coefficients of OLS for actual health care use (y _i
) were obtained by the following formula:

(3)

where y _i
is health care use of individual, In inc _i
represents the logarithm of household income per equivalent adult; ? _k
is a set of need variables including age, sex and health needs; z _p
is a set of non-need variables consisting of location, insurance, activity status,
household size, education and marital status; ?,??,?? _?
, and ? _p
are the parameter vectors, and ? _i
is an error term.

Second, based on equation 3, we generated need-predicted values of health care utilization (? _i^x
) using the parameter vectors (?,??,?? _?
,?? _p
), individual values of the need variables (? _?,i
 ), sample means of the logarithm of household income (In inc _i
 ), and non-need (z _p,i
) variables. The equation of the need-predicted value is written as:

(4)

Finally, the estimate of indirectly standardized health care utilization (? _i^IS
) was simply obtained from the difference between actual (y _i
) and need-predicted health care utilization (? _i^X
), and the sample mean (y ^m
) was added 1].

(5)

Decomposition analysis

It is evident that how much various factors contribute separately to income-related
inequality in health care utilization with the decomposition analysis 1]. There has been argument that decomposition analysis is not developed for a linear
regression model and when it is used in a non-linear model for binary outcome, it
introduces an approximation error. However, the decomposition analysis only requires
using the OLS coefficients, not the predicted values; thus, this is not a problem
3].

Regarding the transformation of health care utilization, the EI is equal to the decomposition
of the concentration index multiplied by 4 and ?_h
. Thus, the EI for health care utilization can be written as:

(6)

where ? represents the mean, j and k are vectors of variables z _j
and x _k
, ? and ? represent the coefficient of the variable z and x, respectively. C represents the concentration index 31].

The main interest of this work was to analyse how horizontal inequity changed between
2007/2008 and 2012; and in order to accomplish that, the Oaxaca decomposition analysis
was used 33], 34].

(7)

An alternative of the Oaxaca decomposition analysis can be written as:

(8)

where ? _kt
represents the elasticity of variable k, t is the year, and ? denotes differences. The Oaxaca decomposition allows to show changes
in income-related inequality in health care use as i) changes in inequality in the
determinants of health care use; and ii) changes in the elasticities of the correspondent
determinants by cross-sectional unit or over time 1].

In addition, we used the bootstrapping method with 1000 replications to obtain confidence
interval for the concentration index and horizontal index. We performed statistical
analysis with the STATA MP 12.1 (StataCorp LP, TEXAS).