The inexorable rise in health expenditure as a share of GDP over the years has been a source of concern among policy makers. This study provides empirical evidence to the drivers of health expenditure growth in the Middle-East region for the time period of 2004 to 2020. Consequently, the objective of the study is to investigate the major drivers of health expenditure growth among the socio-demographic, economic and health system capacity constructs and how it changes with the subsequent introduction of the behavioural construct as a moderator, inter-alia. Data obtained from the World Development Indicators is analysed using Partial Least Square Structural Equation Model (PLS-SEM). The empirical evidence suggests that the three constructs drive up health expenditure growth in the Middle-East region. Further, the introduction of moderation improves the fitness of the model, though it has no direct statistically significant effect on health expenditure growth. Conversely, the moderation effect is significant through the socio-demographic, economic and health system capacity indicators. Overall, the socio-demographic transition has been the major driving force for the exponential growth in health care expenditure. The findings suggest that combining set of related variables may help in better understanding of the determinants of health care expenditure growth and ultimately offer informed policy advice that will eventually improve health outcomes.
Research Article
Impact of Socio-Economic and Behavioural Factors on the Health Care Expenditure Growth of Middle-East Region
https://doi.org/10.21203/rs.3.rs-3874752/v1
This work is licensed under a CC BY 4.0 License
Version 1
posted
You are reading this latest preprint version
Health expenditure growth
Health system capacity
Socio-demographic
Economic
Middle-East region
Behavioural Factor
Globally, the size of health expenditure as a percentage of GDP has been on the rise continuously over the years. The rate of the increase largely depends on the nature of the economy and the existing health system. In the Middle-East and African region, however, data from the World Bank Development Indicators shows that during the period of this study, the level of health expenditure relative to GDP grew by almost 46 percent (World Development Indicators, 2021). As a result, many countries are hitherto grappling with fiscal imbalances since available public revenue could not meet up with the levels of the increase in health expenditure to the provision of efficient health care services. Particularly, the lack of published evidence on the pace of health expenditure growth in the Middle-East context has made it imperative to investigate the principal factors leading to unstoppable growth in health expenditure growth.
However, empirical studies into the principal factors driving health expenditure growth yield mostly inconsistent results in different contexts. Apart from the inability of many studies at providing robust estimates leading to a surge in health spending more than the level of GDP per capita most of these health expenditure determinants were analysed without enough application of the theoretical framework of the demand for health care (Roberts, 1999). However, Grossman theoretical model has been commonly used as a theoretical premise of the demand for care in some empirical studies (Grossman, 1972). Although micro-founded by nature, the Grossman model has been criticized by some analysts that it’s rejected by the data (Zweifel, 2012). In their review of recent achievements in the field of health economics Gerdtham and Jönsson (2000) have maintained that health care theorists need a more macroeconomic framework of health expenditure analysis.
Several studies have provided empirical evidence to the literature using the theoretical framework of the Grossman model. For instance, Wagstaff (1986) used household-level data and validated Grossman’s model. His parameter estimate of the demand for health provides results that are not consistent with what previous studies have reported. In a different study, Wagstaff (1993) shows that the Grossman model has failed to capture the dynamics of the model itself and provided alternative specifications that will give consistent estimates of the model.
Similarly, Nocera and Zweifel (1998) have used a rich dataset and estimated the demand for care from pure consumption and pure investment good view and concluded that most of the differences in the estimates of the previous studies were due to methods applied other than the data. Gerdtham and Jönsson (2000) have investigated the determinants of health expenditure in OECD countries using international health expenditure data. Their empirical results show that assessing primary level care was associated with a lower level of health expenditure. Grossman (2000) reviewed his earlier model and stressed on the theoretical predictions and extensions of the model thereby strongly emphasizing on the previous findings that the model is partly rejected by the data. Recently, Hartwig and Sturm (2018) have tested the Grossman model using OECD panel data and provides robust evidence to the determinants of health expenditure established by the Grossman theoretical model.
However, available literature survey shows that most of the existing studies have provided evidence at testing Grossman theoretical model in spite of the fact that other key factors also drive-up health expenditure growth. This study advances the literature by providing evidence to the major drivers of health expenditure growth by integrating different indicators that determine health expenditure growth in line with the literature.
2.1 Research model and hypothesis
This study developed a conceptual model as depicted in Figure 1 and seven hypotheses for analysing the determinants of health expenditure otherwise known as the demand for care. The conceptual model is set up in line with the literature on the variables which are known to have a significant effect on the growth of health expenditure per capita. The hypotheses are therefore postulated as:
H1: There are significant relationships between the socio-demographic indicators and the level of growth in health expenditure.
The level of socio-demographic composition specific to the country’s health system plays an important role in its demand for health. Thus, the log level of the share of population above the age of 65 years, the log of years of schooling, and the log of compensation of employees were first introduced in the literature by Grossman (1972). Similarly, the log level of female labour force participation rate was used by Gerdtham and Jönsson (2000) and the log of population density was used by Crivelli, Filippini and Mosca (2006). Therefore, the lower the numerical values of the variables constituting the socio-demographic indicator, the lower would be the level of growth in health care expenditure.
H2: There are significant relationships between the performance of economic indicators and the pace of growth in health expenditure.
Several studies have also established the empirical connections regarding the growth in economic indicators and the level of public health spending growth. The log level of per capita GDP was used by Newhouse (1977) and the log level of domestic public expenditure as a percentage of GDP was used by Hitiris (1997). Moreover, the log level of the unemployment rate as a percentage of labour force was used by Christiansen et al. (2006) and the log level of gross capital formation is included following Feng, Xia and Kong (2018) Therefore, the greater the performance of the economic indicators, the higher would be the level of growth in health expenditure, ceteris paribus.
H3: There are significant relationships between the health system capacity indicators and the growth in health expenditure.
The strength and capacity of the health systems indicators are also considered crucial factors determining the level of health expenditure growth. The log level of physician density was used among others (Gerdtham and Jönsson, 2000), the log level of hospital beds is included following Christiansen and Bech (2006) and the log level of nursing and midwives density per 1,000 people by Winkelmann, Muench and Maier (2020). The log level of research and development expenditure as a percentage of GDP was included following the Okunade and Murthy (2002). Hence, the greater the strength of the health systems capacity indicators, the larger would be the growth of health care expenditure.
H4: There is a significant direct effect of behavioural indicators on health care expenditure growth.
H5: The behavioural indicators further moderate the growth in health care expenditure through the socio-demographic indicators.
The rising prevalence of alcohol and tobacco consumption increases the NCDs risk factors leading to a higher premature mortality rate from NCDs among the active workforce further exerting downward pressure on labour supply.
Therefore, the behavioural indicators explain the level of indirect variation in health expenditure through the socio-demographic, economic, and health system capacity indicators.
H6: The behavioural indicator moderates the growth of health care expenditure through the economic indicators.
Clearly, with an increase in the global burden of non-communicable diseases which are linked directly to behavioural risk factors, the stronger the pace of growth in the economy the higher would be the share of resources channelled to the health system for the procurement, maintenance and improvement of the capacity of the health system for in services delivery.
H7: The behavioural indicators also moderate the rate of growth in health care expenditure through the health system capacity indicators. In a given health care system where the prevalence of tobacco and alcohol consumption is high, there would be a high burden of non- communicable diseases that require a functional health system for the provision of preventive, diagnostic and curative care efficiently.
Following Grossman’s theoretical model, the depreciation rate of human capital is attributed to lifestyle variables like tobacco and alcohol consumption as used in the Gerdtham and Jönsson (2000) studies. Therefore, the behavioural indicators moderate both the cost of health care and the three key determinants (socio-demographic, economic, and health system capacity indicators).
2.2 Selection of indicators
The magnitude of socio-demographic indicators of a given health system serves as an important factor in the analysis of health care expenditure growth. In line with the extant literature, this study includes the percentage of the population above the age of 65 years threshold since it’s the age cohort in greater demand for health care (Grossman, 1972). The Grossman theoretical model also considered the wage rate (compensation per employee) as a crucial variable in the demand for care (Grossman, 1972). All else constant, the higher the wage rates the higher the resources for the demand for care. Education level is also a relevant variable in determining access to health care. Differences in education levels among individuals affect utilization of health care and adherence to medical prescriptions. Similarly, population density shows the level of people’s concentration in a given, and thus in a country with higher population density, the demand for care would be high since diseases could be spread more easily and therefore difficult to be controlled (Gerdtham and Jönsson, 2000). Female labour force participation is increasingly considered as an important demographic factor in the analysis of development. In a context where female labour force participation is high, they are economically empowered to have better access for maternal and child care health which are essential to curb the rate of maternal and neonatal maternity (Gerdtham and Jönsson, 2000)
The level of the economic capacity of a country determines the overall share of its health care expenditure. The per capita GDP level measured in (PPP) gives the total income earned by the individuals living in a country determined by the demand for care in a given context. Thus, the higher the income the higher demand for care. The level of government health expenditure as a percentage of GDP is commonly used as an indicator for financing health care (Hitiris, 1997). Gross capital formation refers to the additional investment to existing capital stock or inventory level which results in augmenting productivity level (Feng, Xia and Kong, 2018). An increase in productivity would raise income and employment thereby leading to more economic growth. The rate of unemployment of a country shows the national estimates of the abled workforce that could not get paid jobs at the wage rate prevailing in the market (Christiansen et al., 2006). Prolong unemployment leads to poor health status and in some cases leads to premature death due to in-access or poor access to required health services at the needed time.
The capacity of the health system of a country would be the first to consider when assessing its efficiency in health services delivery. This study, therefore, considers four indicators as of paramount importance in this regards in line with the literature i.e. physician density following Gerdtham and Jönsson (2000) studies, hospital beds following Christiansen study, nurses and midwives following Winkelmann, Muench and Maier (2020) studies, and expenditure on research and development as a share of GDP following Okunade and Murthy (2002) studies. The densities of physicians, hospital beds, and nurses as well as midwives all are measured per 1,000 people and shows the strength of the health systems in providing both in-patient and out-patient care as well as during emergency periods. Also, in a context where expenditure on R&D is reasonably budgeted and used for the intended purposes, scientific advances through innovations and discoveries results which eventually increase the capacity of the health systems for efficient care delivery.
In the last one and a half-decade, the menace of raising premature mortality from non-communicable diseases becomes a common threat to longevity across the globe. The prevalence of alcohol and tobacco consumption are the main behavioural factors attributed to the growing risk factors of the burden of non-communicable diseases and their associated mortality. The behavioural indicators are assumed to moderate the depreciating state of health capital further escalating the pace of health expenditure growth in two ways. First, is the direct effects on driving health expenditure growth and secondly through the socio-demographic, the economic, and the health system capacity indicators. As a consequence, this study includes alcohol and tobacco consumption prevalence following Gerdtham and Jönsson (2000) studies.
The outcome variable of this study includes both public and private health expenditure measured in PPP international US$. Though in many studies that examine the growth of health expenditure and its determinants like Hartwig and Sturm, (2018) among others used only per capita public health expenditure due to the nature of the study, this study includes per capita private health expenditure to capture its dynamics that is usually unaccounted for in the previous studies and also to balance the endogenous construct.
Table 1 Descriptive Assessment of Index Descriptors
|
Construct |
Code |
Indicator |
Observation |
Mean |
SD |
Min. |
Max. |
|
|
Exogenous (Multiple Indicator Construct) |
Socio-demographic (SD) |
X11 |
Population ages 65 and above (% of total population) |
255 |
1.240 |
0.723 |
-0.377 |
2.574 |
|
X12 |
Population density (people per sq. km of land area) |
255 |
1.96 |
0.545 |
0.86 |
3.26 |
||
|
X13 |
Labor force, female (% of total labor force) |
255 |
1.30 |
0.179 |
0.90 |
1.67 |
||
|
X14 |
Years of Schooling |
255 |
2.015 |
0.410 |
0.182 |
2.564 |
||
|
X15 |
Compensation of employees (% of expense) |
255 |
3.489 |
0.414 |
1.238 |
4.248 |
||
|
Exogenous (Multiple Indicator Construct) |
Economic Factor (EF) |
X21 |
GDP per capita, PPP (constant 2011 international $) |
255 |
4.389 |
0.405 |
3.42 |
5.09 |
|
X22 |
Domestic general government health expenditure (% of GDP) |
255 |
2.762 |
0.442 |
1.16 |
3.52 |
||
|
X23 |
Gross capital formation (% OF GDP) |
255 |
1.382 |
0.119 |
0.97 |
1.69 |
||
|
X24 |
Unemployment, total (% of total labor force) (modelled ILO estimate) |
255 |
0.693 |
0.446 |
-0.82 |
1.21 |
||
|
Exogenous (Multiple Indicator Construct) |
Health System capacity (HSC) |
X31 |
Physicians (per 1,000 people) |
255 |
0.436 |
0.596 |
-1.518 |
1.348 |
|
X32 |
Hospital beds (per 1,000 people) |
255 |
0.655 |
0.569 |
-4.605 |
1.509 |
||
|
X33 |
Nurses and midwives (per 1,000 people) |
255 |
1.032 |
0.551 |
-0.352 |
1.967 |
||
|
X34 |
Research and development expenditure (% of GDP) |
255 |
-1.013 |
1.183 |
-4.055 |
1.505 |
||
|
Exogenous (Multiple Indicator Construct) |
Behavioural Factor (BH) |
X41 |
Prevalence of current tobacco use (% of adults) |
255 |
1.357 |
0.147 |
0.98 |
1.64 |
|
X42 |
Total alcohol consumption per capita (litres of pure alcohol, projected estimates, 15+ years of age) |
255 |
-0.168 |
0.654 |
-2.30 |
1.09 |
||
|
Endogenous (Single Indicator Construct) |
Health Care Expenditure (HCE) |
X51 |
Domestic general government health expenditure per capita, PPP (current international $) |
255 |
0.389 |
0.193 |
-0.68 |
0.80 |
|
X52 |
Domestic private health expenditure per capita, PPP (current international $) |
255 |
2.585 |
0.287 |
1.83 |
3.14 |
||
Source: Authors’ Calculation.
2.3. Data Source
In this study, 15 countries of the Middle-East region are included based on their geographic location as visualised in Figure 2 and data that is publicly available for 2004 to 2020 periods. Apparently, a substantial difference exists among the member countries under study as documented by World Development Indicators (2020). In spite of these noticeable differences, these countries are always reported as one homogeneous unit. The exceedingly higher rate of health expenditure among these countries has warrants investigation into the principal causes behind the exponential surge for appropriate policy intervention. Specifically, the data of this study is obtained from the World Bank Development Indicators and the detailed descriptions of the indicators included in this study are illustrated in Table 1.
2.4. Model selection - PLS-SEM Path Model
Owing to the predictive nature of this study, variance-based Structure Equation Modelling (PLS-SEM) is employed to establish a linkage between the set of latent variables involved in the study. SEM-based on a two-step approach has two well-known estimating techniques: CB-SEM (Covariance Based – Structural Equation Modelling) and PLS-SEM (Partial Last Square – Structural Equation Model). CB-SEM works on the fundamental to reduce the deviation between estimated data and sample matrices through model estimates but PLS-SEM outweighs by using an iterative sequence of OLS regression to maximize the impact of exogenous constructs on endogenous construct (Astrachan, Patel and Wanzenried, 2014). Furthermore, both SEM techniques stemmed from the same basis but PLS-SEM has become more popular in recent years due to its capability of operating multiple latent constructs in a model (Zhu et al., 2019). Moreover, the Partial Least Square-Structural Equation Modelling (PLS-SEM) is a general estimation technique that involves alternate least square algorithms which extend principal component analysis (PCA) and canonical correlation analysis (CCA). The outer model or measurement model and inner model or structural model are two formal sets of linear equations that are used to define the PLS path model. Figure 3 represents the PLS path model adopted for evaluation in the study.
The structural model is evaluating the construct-to-construct plausible relationships. The development of the path model is from left to right. The latent variables on the left side are exogenous variables and variables on the right-hand side are endogenous variables. The inner model relationship between the latent variables is represented as:
Where β represents the standardised path coefficients and ζ represents the inner model residuals assumed to satisfy the expected relationship and non-correlation.
A comprehensive description of workflow methodology for hypothesised model validation is represented in Figure 4.
The data analysis section details the estimates of the SEM-PLS approach obtained using Smart-PLS 2.0 (Kura, Shamsudin and Chauhan, 2013). The structural equation modelling based on the two-step approach is employed for the analysis (Anderson and Gerbing, 1988). Firstly, the reliability and validity of the data is ascertained in the outer model. Secondly, path coefficients among constructs are evaluated in the inner model to test the proposed relationships.
3.1 Measurement Model Assessment
A measurement model is assumed to satisfy the linear relationship between the explicit variable and the latent construct in the PLS algorithm (Eq. 2).
Xij = λij ηi + 𐐁ij {i = 1, 2, 3...........n; j = 1, 2, 3 ...m} (2)
Where η represents a vector of endogenous latent variable; X represents the indicators; λ represents outer loading, 𐐁 is the residuals. The measurement relationship is subject to predictor specification stating no correlation among the error term and the latent variable which further reduces Eq. (2) to Eq. (3).
(Xij|𐐁i) = λij𐐁i {i = 1, 2, 3...........n; j = 1, 2, 3 ...m} (3)
In this study the measurement model is evaluated using the procedure outlined by Gefen and Straub (2005) to test the convergent and discriminant validity. The convergent validity measures the extent to which the scale items are related in reality as compared to the existing literature. The indicator loading, composite reliability, and average variance extracted are used to ensure convergent validity. As presented in Table 2, internal consistency among scale items is established as the composite reliability (CR) value is found to be greater than 0.7 (Nunnally, 1978). Furthermore, the indicator loading ranges in between 0.778–0.925 and are quite above the defined threshold level of 0.7. All the AVE values (from 0.639 to 0.805) were above 0.5, thus convergent validity is established (Fornell and Larcker, 1981).
The discriminant validity confirms whether the indicator measures the construct or not. The correlation matrix represented in Table 3 is used to assess the discriminant validity. Table 3 presents the correlation matrix with the square root of AVE placed diagonally. All off-diagonal correlation values are found to be below the square root of AVE and hence the discriminant validity among the construct is established (Fornell and Larcker, 1981). The reliability and validity tests established provide empirical evidence for advancing the analysis of the structural model.
|
Construct |
Indicator |
Scale |
Loadings |
AVE |
CR |
|---|---|---|---|---|---|
|
Socio-demographic (SD) |
X12 |
Reflective |
0.890 |
0.731 |
0.891 |
|
X11 |
0.860 |
||||
|
X15 |
0.813 |
||||
|
Economic Indicators (EI) |
X22 |
Reflective |
0.982 |
0.805 |
0.925 |
|
X21 |
0.905 |
||||
|
X24 |
0.795 |
||||
|
Health system capacity (HSC) |
X32 |
Reflective |
0.786 |
0.662 |
0.854 |
|
X33 |
0.738 |
||||
|
X31 |
0.908 |
||||
|
Behavioural Indicators (BI) |
X41 |
Reflective |
0.701 |
0.677 |
0.805 |
|
X42 |
0.929 |
||||
|
Health care expenditure (HCE) |
X51 |
Reflective |
0.720 |
0.639 |
0.778 |
|
X52 |
0.872 |
||||
| Note: Indicators X13, X14, X23, X34 with factor loading < 0.7 were dropped from the measurement model | |||||
Source: Authors’ Calculation.
|
Latent Variable Correlation (LVC) |
Criterion |
|||||
|---|---|---|---|---|---|---|
|
BH |
CMC |
EI |
HSC |
SD |
√AVE > LVC |
|
|
BH |
0.823 |
Yes |
||||
|
HCE |
0.544 |
0.799 |
Yes |
|||
|
EI |
-0.024 |
0.347 |
0.897 |
Yes |
||
|
HSC |
0.426 |
0.670 |
0.479 |
0.813 |
Yes |
|
|
SD |
0.687 |
0.792 |
0.110 |
0.749 |
0.854 |
Yes |
| Note: √AVE are represented diagonally in bold | ||||||
Source: Authors’ Calculation.
3.2 Structural Model Assessment
Assessment of the inner model involves its evaluation in terms of its predictive relevance and path coefficient. The structural model examines the direction and strength of the relationships between exogenous and endogenous constructs using path coefficients (β). Table 4 provides the path coefficient and t statistics for the hypothesised relationships guiding this study. The statistical significance of the path coefficient is ascertained by employing standardised bootstrapping with 500 re-sampling. Numerical estimates with t ≥ 1.96 are considered statistically significant in either accepting or rejecting the hypothesised relationships. The estimates provide support for three effect hypothesis; SD (H1: β = 0.796, t = 22.294), EI (H2: β = 0.302, t = 13.608) and HSC (H3: β=-0.088, t = 2.307). By contrast, the effect of BI (H4: β = 0.042, t = 1.438) on HCE was found to be insignificant as the t value is below the defined threshold of 1.96.
|
Hypothesis |
Predictor (X) |
Moderator (M) |
Outcome (Y) |
Path X→ Y |
Path M→ Y |
Moderation X*M→ Y |
Result |
|||
|---|---|---|---|---|---|---|---|---|---|---|
|
β |
t |
β |
t |
β |
t |
|||||
|
H1 |
SD |
-- |
HCE |
0.796 |
22.224 |
-- |
-- |
-- |
-- |
Significant |
|
H2 |
EI |
-- |
HCE |
0.302 |
13.608 |
-- |
-- |
-- |
-- |
Significant |
|
H3 |
HSC |
-- |
HCE |
-0.088 |
2.307 |
-- |
-- |
-- |
-- |
Significant |
|
H4 |
BI |
-- |
HCE |
0.042 |
1.438 |
-- |
-- |
-- |
-- |
Insignificant |
|
H5 |
SD |
BI |
HCE |
0.949 |
15.735 |
0.310 |
3.214 |
-0.394 |
3.082 |
Significant |
|
H6 |
EI |
BI |
HCE |
0.202 |
4.474 |
-0.077 |
1.413 |
0.176 |
2.323 |
Significant |
|
H7 |
HSC |
BI |
HCE |
0.070 |
0.912 |
0.320 |
2.865 |
-0.418 |
2.614 |
Significant |
Source: Authors’ Calculation.
However, the value of R2 measures the predictive ability of the model against the suggested range of 0.19, 0.33, and 0.67 indicating weak, moderate and substantial effect, respectively (Chin, 1998). As shown in Fig. 9, the value of R2 for the relationship between SD, EI, HSC, BI, and HCE is 0.700. This indicates that the exogenous construct exerts more than 70% variance in the endogenous construct. Also, the Stone-Geisser Q2 criterion is employed to obtain the quality of PLS model through bootstrapping and cross-validated redundancy. The identified Q2 value established the relevance of exogenous construct to predict endogenous construct as the values are found to be 0.318 which is greater than the threshold limit of 0 (Hair Jr et al., 2014). The Goodness of Fit (GoF) index was calculated to ascertain the global validity and power of the model. The GoF is used to measure the overall predictive power of a large complex model as it considers the performance of both the measurement model and the structural model and hence it may be used as a better model fit index as compared to any other measure (Shah et al., 2018). The Goodness of Fit (GoF) index is the geometric mean of average communality and average R2 of all endogenous constructs (Tenenhaus et al., 2005). A good model fit indicates that the model is parsimonious and plausible. The range for global validation of the path model lies in between 0 to 1 with the specified criterion of GoFsmall=0.10; GoFmedium=0.25 and GoFlarge=0.36. Basic Eq. (4) represents the method to calculate GoF for the path model.
$$GoF = \sqrt{(AVE \times {R}^{2}})$$
4
$$GoF = \sqrt{(0.572 \times 0.770}$$
5
$$GoF = 0.632 \ge 0.361$$
6
The identified GoF index 0.632 provides empirical support that the data fits the model satisfactorily and provides substantial support in explaining the growth of health care expenditure in the Middle- East region using the indicators included.
3.3. Moderation
This study applies the product indicator approach to examine the moderation effects. As illustrated in Fig. 5, firstly, the main effect of the exogenous construct is measured and represented in Eq. (7):
Y = a + b · X + c · M + p (7)
Here, a is the intercept, b and c are the slope for X and M, respectively. The idea behind moderating variables is that the slope of the exogenous construct no longer remains the same, but differs with the introduction of the interaction term (X * M) in the structural model. The impact of interaction term can mathematically be formulated as follows:
Y = a + b · X + c · M + d · (X * M) + p (8)
Y = (a + c · M) + (b + d · M) · X + p (9)
The term X * M, represents an additional latent variable in the structural model covering the product of the independent variable and the moderator. Eq. (9) provides a different view to look at the model where (a + c · M) is the intercept and (b + d · M) represents the slope of the construct X and p is error term.
In this model, two behavioural indicators are used as an interaction term for moderating the direct effect relationship between SD, EI, HSC, and HCE. The direct effect of BI on HCE is statistically insignificant while assessing the structural model. As given in Table 4, the BI significantly moderates the relationship between SD (H5: β=-0.394, t = 3.082); EI (H6: β = 0.176, t = 2.323); HSC (H7: β=-0.418, t = 2.614) and HCE. On the interaction plot, the parallel line represents no interaction effect but it can be seen from Fig. 6 (SD*BH), Fig. 7 (EF*BH) and Fig. 8 (HSC*BH) that the difference in slope creates an intersection point and establishes the moderation effect. Furthermore, it is evident from the interaction plot that the BI negatively moderates the relationship from SD to HCE and HSC to HCE. This means, the prevalence of alcohol and tobacco consumption in the Middle-East region are not augmenting the increase in the cost of health care through increasing the demand for preventive care in the health systems targeted at reducing the menace of premature mortality from non-communicable diseases. Similarly, the BI positively moderates the relationship from EI to HCE. Thus, risky health behaviours lead to the emergence of new diseases that may require more expenditure on health for the management of those conditions. As shown in Fig. 9, with interaction terms, the model accounts for 73.26% (R2 = 0.7326) variation in health care expenditure as compared to 70% in the structural model. Notably, by adding the interaction term, it has the effect of significantly increasing the predictive power of the model (∆R2 = 3.26%). The identified Q2 value of 0.341has also established the predictive relevance of the moderated model.
The inexorable increase in the share of health expenditure relative to GDP has been a source of concern over the years. By and large, it is of paramount importance from the policy point of view to apprehending the major root causes behind its exponential surge. Particularly, this study focuses on examining the determinants of health expenditure growth over the years in the Middle-East under the newly developed conceptual framework. Three constructs were established as determinants of health spending growth based on a detailed literature search. The technique of PLS-SEM is applied to estimate the model and eventually assess its relevance. As a result, three main hypotheses are formulated for investigating the direct effect of the socio-demographic, economic, and health system capacity indicators on health expenditure growth. Similarly, for moderating the relationships, four more hypotheses are formulated to examine the moderating effects of the behavioural indicators on health care expenditure and the three different constructs included in the model. Results from the structural model show that there is statistically significant evidence to conclude that socio-demographic, economic, and health system capacity indicators drive up health expenditure growth in the Middle-East region.
Similarly, evidence shows no direct effect of behavioural indicators on health care expenditure growth. Also, there is enough statistical evidence regarding the indirect moderation effect from behavioural indicator through the socio-demographic, economic, and the health system capacity on health spending growth. The negative moderation effect from the socio-demographic and health system capacity indicators contradicts the hypotheses that established the relationships i.e. H5 and H6. What may have contributed to this finding warrants further investigation. Moreover, the estimates show a better fit in the model after introducing the moderating effects on the basis of the coefficient of determination. Though the β coefficient associated with socio-demographic and economic indicators are statistically significant, the estimate are positively related to health care expenditure growth and thus strongly supports H1 and H2 hypotheses. The exact opposite is also true with the β coefficient related to health system capacity indicators represented by H3. Notably, the exponential growth in health care expenditure in the Middle-East context has been greatly driven by a socio-demographic transition over the years. Also, apart from that, NCDs risk factors have also amplified the surge, though through the socio-demographic variables but in an indirect manner.
The results of this study found that though all the three constructs significantly drive up health expenditure growth, the socio-demographic indicators weigh the most in the analysis. Therefore, policy initiatives that can result in improving health among adults above the age of 65 years, better female participation rate, guaranteed better educational outcome, improve employee compensation, and good housing condition may ensure health expenditure growth along with better health outcome over time. Therefore, the socio-demographic transition should be the prime target to slow the face of health care expenditure unchecked growth. Moreover, health expenditure growth is also linked to the levels of economic indicators such as income per capita, gross capital formation, etc. as well as the capacity of the health system to handle and manage different conditions. Similarly, the introduction of the moderation effect improves the estimates of the overall model. The use of behavioural indicators as moderating effect has no statistically significant influence on health expenditure growth, whereas it has a significant effect on health expenditure growth through the socio-demographic, economic, and the health system capacity indicators.
However, for the sake of policy analysis, classifying the indicators into three constructs established in this study can help identify the major drivers of health care expenditure growth in the Middle-East in the last one and a half-decade.
Though to the author’s knowledge, this is the first study that combined different indicators under three constructs as major determinants of health expenditure growth, this initiative can help policy makers in a proper understanding of the situation and adopt the right way to ensure improved health outcome in the Middle-East.
Acknowledgements: We would like to acknowledge the Symbiosis International Deemed University, Pune for providing the database access and research software required for the completion of this study.
Funding: The authors did not receive any financial support for this research.
Conflict of Interest: The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Ethical approval: Ethical approval for this type of study is not required by our University.
Author’s Contribution: SS & NK conceptualise and designed the proposal of the study; CJR and MK collected the data; SS defined the methodology, analyzed, and interpreted the results. NK drafted the manuscript; MK and SS critically revised the manuscript finally completed the proof-reading. All authors read and approved the final manuscript.
- Anderson, J. C. and Gerbing, D. W. (1988) ‘Structural equation modeling in practice: A review and recommended two-step approach.’, Psychological bulletin, 103(3), p. 411.
- Astrachan, C. B., Patel, V. K. and Wanzenried, G. (2014) ‘A comparative study of CB-SEM and PLS-SEM for theory development in family firm research’, Journal of Family Business Strategy, 5(1), pp. 116–128.
- Chin, W. W. (1998) ‘The partial least squares approach to structural equation modeling’, Modern methods for business research, 295(2), pp. 295–336.
- Christiansen, T. et al. (2006) ‘Demographic changes and aggregate health care expenditure in Europe, paper for AHEAD project’, WP VI.
- Christiansen, T. and Bech, M. (2006) Demographic changes and aggregate health-care expenditure in Europe. CEPS.
- Crivelli, L., Filippini, M. and Mosca, I. (2006) ‘Federalism and regional health care expenditures: an empirical analysis for the Swiss cantons’, Health economics, 15(5), pp. 535–541.
- Feng, W., Xia, Y. and Kong, Y. (2018) ‘Public Health Expenditure and Labour Productivity: A Tentative Interpretation Based on the Science of Brain Cognition’, NeuroQuantology, 16(5).
- Fornell, C. and Larcker, D. F. (1981) ‘Evaluating Structural Equation Models with Unobservable Variables and Measurement Error’, Journal of Marketing Research, 18(1), pp. 39–50. doi: 10.1177/002224378101800104.
- Gefen, D. and Straub, D. (2005) ‘A practical guide to factorial validity using PLS-Graph: Tutorial and annotated example’, Communications of the Association for Information systems, 16(1), p. 5.
- Gerdtham, U.-G. and Jönsson, B. (2000) ‘International comparisons of health expenditure: theory, data and econometric analysis’, in Handbook of health economics. Elsevier, pp. 11–53.
- Grossman, M. (1972) ‘On the concept of health capital and the demand for health’, Journal of Political economy, 80(2), pp. 223–255.
- Grossman, M. (2000) ‘The human capital model’, in Handbook of health economics. Elsevier, pp. 347–408.
- Hair Jr, J. F. et al. (2014) ‘Partial least squares structural equation modeling (PLS-SEM)’, European business review.
- Hartwig, J. and Sturm, J.-E. (2018) ‘Testing the Grossman model of medical spending determinants with macroeconomic panel data’, The European Journal of Health Economics, 19(8), pp. 1067–1086.
- Hitiris, T. (1997) ‘Health care expenditure and integration in the countries of the European Union’, Applied Economics, 29(1), pp. 1–6.
- Kura, K. M., Shamsudin, F. M. and Chauhan, A. (2013) ‘Examining the dimensions of workgroup norms construct: A validation study in the Nigerian education context’,
- Newhouse, J. P. (1977) ‘Medical-care expenditure: a cross-national survey’, The Journal of human resources, 12(1), pp. 115–125.
- Nocera, S. and Zweifel, P. (1998) ‘The demand for health: an empirical test of the Grossman model using panel data’, in Health, the medical profession, and regulation. Springer, pp. 35–49.
- Nunnally, J. C. (1978) ‘Psychometric theory (2nd edit.)’, New York.
- Okunade, A. A. and Murthy, V. N. (2002) ‘Technology as a ‘major driver’of health care costs: a cointegration analysis of the Newhouse conjecture’, Journal of health economics, 21(1), pp. 147–159.
- Roberts, J. (1999) ‘Sensitivity of elasticity estimates for OECD health care spending: analysis of a dynamic heterogeneous data field’, Health Economics, 8(5), pp. 459–472.
- Shah, S. A. R. et al. (2018) ‘Road safety risk assessment: an analysis of transport policy and management for low-, middle-, and high-income Asian countries’, Sustainability, 10(2), p. 389.
- Tenenhaus, M. et al. (2005) ‘PLS path modeling’, Computational statistics & data analysis, 48(1), pp. 159–205.
- Wagstaff, A. (1986) ‘The demand for health: some new empirical evidence’, Journal of Health economics, 5(3), pp. 195–233.
- Wagstaff, A. (1993) ‘The demand for health: an empirical reformulation of the Grossman model’, Health Economics, 2(2), pp. 189–198.
- Winkelmann, J., Muench, U. and Maier, C. B. (2020) ‘Time trends in the regional distribution of physicians, nurses and midwives in Europe’, BMC Health Services Research, 20(1), pp. 1–15.
- World Development Indicators DataBank (2021). Available at: https://databank.worldbank.org/source/world-development-indicators (Accessed: 17 June 2020).
- Zhu, X. et al. (2019) ‘Measuring the efficiency and driving factors of urban land use based on the DEA method and the PLS-SEM model—a case study of 35 large and medium-sized cities in China’, Sustainable Cities and Society, 50, p. 101646.
- Zweifel, P. (2012) ‘The Grossman model after 40 years’, The European Journal of Health Economics, 13(6), pp. 677–682.
No competing interests reported.