The burden of catastrophic and impoverishing health expenditure in Armenia: An analysis of Integrated Living Conditions Surveys, 2014–2018

Armenia’s health spending is characterized by low public spending and high out-of-pocket expenditure (OOP), which not only poses a financial barrier to accessing healthcare for Armenians but can also impoverish them. We analyzed Armenia’s Integrated Living Conditions Surveys 2014–2018 data to assess the incidence and correlates of catastrophic health expenditure (CHE) and impoverishment. Households were considered to have incurred CHE if their annual OOP exceeded 40 percent of the per capita annual household non-food expenditure. We assessed impoverishment using the US$1.90 per person per-day international poverty line and the US$5.50 per person per-day upper-middle-income country poverty line. Logistic regression models were fitted to assess the correlates of CHE and impoverishment. We found that the incidence of CHE peaked in 2017 before declining in 2018. Impoverishment decreased until 2017 before rising in 2018. After adjusting for sociodemographic factors, households were more likely to incur CHE if the household head was older than 34 years, located in urban areas, had at least one disabled member, and had at least one member with hypertension. Households with at least one hypertensive member or who resided in urban areas were more likely to be impoverished due to OOP. Paid employment and high socioeconomic status were protective against both CHE and impoverishment from OOP. This detailed analysis offers a nuanced insight into the trends in Armenia’s financial risk protection against catastrophic and impoverishing health expenditures, and the groups predominantly affected. The incidence of CHE and impoverishment in Armenia remains high with a higher incidence among vulnerable groups, including those living with chronic disease, disability, and the unemployed. Armenia should consider different mechanisms such as subsidizing medication and hospitalization costs for the poorest to alleviate the burden of OOP.

Introduction there is no comprehensive analysis of the incidence of OOP payments for health care in Armenia and limited data on CHE and impoverishing health spending after 2013. While the global literature indicates that: household economic status, the incidence of hospitalization, the presence of an elderly or disabled household member in the family, and the presence of a family member with a chronic illness are common significant factors associated with household CHE [5,13], there are additional drivers of catastrophic and impoverishing health spending that are context-specific [14][15][16][17][18][19][20][21]. Among some European and Central Asian countries, the high cost of medication, outpatient coverage gaps, and physician-induced demand are cited as contributors to high OOP [22,23].
This analysis aims to examine trends in the current state of Armenia's OOP expenditures and determine who is most at risk for CHEs or impoverishing health spending. It investigates sociodemographic factors, health status and service usage, and current benefit levels to understand their potential impact on catastrophic and impoverishing health expenditure in the Armenian context. By exploring the trends and inequalities in catastrophic and impoverishing health spending, this study both updates and adds nuance to the discussion around the current state of Armenia's financial risk protection and those most impacted by the cost of healthcare.

Data sources
We obtained data from Armenia's Integrated Living Conditions Survey (ILCS) for five years from 2014 to 2018, within which the full complement of variables for the analysis was available and collected consistently [10]. However, the ILCS in Armenia has been conducted annually since 2001 by the National Statistical Service of the Republic of Armenia with support from the World Bank, United States Agency for International Development, and other donor organizations. The survey involves a two-stage stratified sampling approach where households are the ultimate sampling units. In this approach, the county is first stratified into sampling areas based on the location's geographical region, urbanization, and population size. Once stratified, eight households are randomly selected from each sampling area. Since its introduction, Armenia's ILCS has included a sample size of 5,184 households. However, the sample size was expanded to 7,872 between 2007 and 2011 following increased funding by the Millennium Challenge Account-Armenia (MCA-Armenia). The sample size reverted to 5,184 from 2012 except in 2017, where 7,776 households were included, because of budget availability. Ethical clearance for this study is deemed not to be required because it uses publicly available, de-identified data.

Incidence of CHE and impoverishment
For each year between 2014 and 2018, we computed the incidence of CHE and impoverishment following widely applied methods [24,25]. First, drawing on reported household spending, we calculated the total OOP expenditure incurred by patients while accessing care, summing individual expenditure at the household level. The OOP expenditure included spending on medicines, laboratory and radiology investigations, and other direct payments to the cashier or medical staff during inpatient or outpatient visits, including informal payments. However, the calculated OOP expenditure does not include the transportation costs to and from the health facilities, as this data is not available in the ILCS. We excluded any payments covered by health insurance. Cost variables for outpatient visits were based on a 30-day recall period. We annuitized these by multiplying the total OOP for outpatient visits by 12. However, inpatient costs were based on their last visit and a 12-month recall period. Therefore, we obtained total annual OOP costs for inpatient care by multiplying their respective inpatient costs in the last visit with their total number of visits in the last 12 months. We imputed inpatient costs for those with inpatient visits but missing inpatient costs by multiplying the median day costs per admission by the average length of stay for admission in their last visit and the total number of visits in the last 12 months. We imputed zero for the length of stay for those who did not have inpatient visits to include the length of stay variable in the model. Then, we adjusted the outpatient, inpatient and total healthcare costs to 2018 constant Armenian Drams (AMD) using Consumer Price Index data from the World Bank Database. CHE was measured by estimating the catastrophic headcount. A catastrophic headcount was defined as the percentage of households whose OOP expenditure for health care was greater than an established threshold. Two thresholds were used to assess the incidence of CHE. A household was considered to have incurred CHE if their per capita annual expenditure on health care exceeded: 1) 10 percent of the per capita total annual household consumption expenditure, or 2) 40 percent of the per capita annual household non-food consumption expenditure [26]. The two thresholds were used for comparison as a result of their differences in denominators (total household expenditure versus non-food expenditure), the threshold levels as a percentage of the denominator, and given the fact that there is no consensus about which of the two is the better threshold for assessing CHE [26,27].
We assessed the incidence of impoverishment in Armenia through several steps. First, we defined the poverty line. We adopted two poverty lines for comparison: 1) the US$1.90 per person per day international poverty line and 2) the US$5.50 per person per day recommended for upper-middle income countries as suggested by the World Bank. Second, we used the GDP deflator data for Armenia to convert the poverty line to 2014, 2015, 2016, 2017, and 2018 values. Third, we converted this international poverty line for each respective year by multiplying by its respective World Bank purchasing power parity conversion factor (LCU per international dollar) into Armenian Drams (AMD). Finally, we calculated the percentage of households pushed further into poverty due to health expenditure by examining changes in total per capita expenditure before and after the household incurred health spending (standardized by household size).

Inequalities in the distribution of CHE and impoverishing health expenditure
We computed the inequalities in the incidence of CHE and impoverishment using well-established methodologies described by Wagstaff et al. [28]. We calculated the rich-poor difference, defined as the difference in the percentage of households incurring CHE or facing impoverishment between the households in the richest quintile (Q5) to the poorest (Q1) and ratios (Q5/ Q1). Although these two measures of inequality are easier to calculate and interpret, they do not consider poor (Q2), middle (Q3), and rich (Q4). Consequently, we generated the concentration curves and Wagstaff's concentration index for the CHE using both thresholds and impoverishing health expenditure for 2017 and 2018. The concentration curve plots indicated the cumulative percentage of CHE/impoverishment on the y-axis versus the cumulative population percentage ranked by wealth from poorest to richest [26]. When every household experiences the same share of CHE/impoverishment, the concentration curve transforms into a 45-degree line. CHE/impoverishment can be interpreted as concentrated among the rich when the curve lies below the 45-degree line and vice versa. The further the curve is from the 45-degree line, the higher the inequality. The concentration index is a summary measure obtained from the concentration curve and ranges from -1 to 1. A concentration index of 0 indicates equality in experiencing CHE/impoverishment, whereas the concentration index is a positive (negative) value when inequality is concentrated among the rich (poor) [29]. For this analysis, the inequalities in impoverishment were calculated using the US$5.50 per person per day poverty line only as it is the recommended poverty line for upper-middle countries like Armenia.

Correlates of CHE and impoverishing health expenditure
To assess the correlates of incurring CHE and impoverishment (using US$5.50 per person per day poverty line) from OOP expenditure on health care, we fit both bivariate and multivariable logistic regression models for each of the two outcomes: 1) whether a household incurred CHE at the 10 percent threshold (Yes/No) and 2) whether a household was pushed into poverty as a result of OOP expenditure on health (Yes/No). In each of these models, we incorporated sociodemographic factors including gender (Male/Female), age group (<25 years/25-34 years/45-54 years/55-64 years/65+ years), and level of education of the household head (None/ Primary/Secondary/Tertiary), whether household has at least one member with hypertension (Yes/No), whether at least one household member has private health insurance (Yes/No), whether at least one household member belongs to a group considered socially vulnerable or eligible for the social package (disabled, pensioner, receives social benefits, military, children), whether at least one household member has access to coverage for vulnerable groups under the BBP (Yes/No), whether at least one household member had some paid work (Yes/No), household location (Urban/Rural), household size (1-2 members/3-4 members/5+ members), and the household's socioeconomic status proxied by quintiles (poorest, poor, middle, rich, or richest quintile) of an asset index generated through principal component analysis. These independent variables were included in this analysis based on their relevance to the Armenian context (e.g., access to the BBP for vulnerable groups) or have previously been associated with the incidence of CHE and/or impoverishment in other studies [17-21, 24, 30]. Sampling weights were also applied to ensure that estimates were nationally generalizable while standard errors were clustered at the sampling area level. Additionally, we employed a complete case analysis where participants with missing data were excluded from the analysis. Overall, 98 percent of households had complete data on all variables included in the multivariable models.

Inclusivity in global research
Additional information regarding the ethical, cultural, and scientific considerations specific to inclusivity in global research is included in S1 Checklist. Table 1 presents the sociodemographic characteristics of the households surveyed in Armenia's 2014-2018 ILCS. Overall, the majority of surveyed households lived in urban areas; had elderly heads of household (65+ years); had at least one member with paid work; and were comprised of 3-4 household members. The proportion of female household heads in the study sample ranged from 31.4 percent in 2016 to 33.6 percent in 2017. Nearly all heads of surveyed households had a secondary education, half of whom received a tertiary education. Fewer than one out of ten surveyed households had one household member with hypertension. The proportion of households with a disabled member decreased from over 17 percent in 2014 to 14 percent in 2018. The poorest households made up a slightly larger share of surveyed households, ranging between 22.2 percent in 2017 and 24.4 percent in 2016.

Results
Sampled households had varied access to governmental benefits or additional financial support. Whereas about half of surveyed households had at least one pensioner, fewer than one in ten sampled households had a member who received social benefits. By 2018, nearly one in five surveyed households had a member covered by the BBP for vulnerable groups, while just over a tenth of the surveyed households had a member with access to private health insurance. Less than two percent of surveyed households had a member in the military, while approximately three percent had a child who qualified as a part of the children's group [8].

PLOS GLOBAL PUBLIC HEALTH
The burden of catastrophic and impoverishing health expenditure in Armenia the mean total OOP payments by households declined by 3.68 percent, with a larger share of this contributed by OOP payments for inpatient care, which declined by 11.57 percent while outpatient care also declined by 3.19 percent.  Table 4 shows the results of the impoverishing effects of OOP payments using the US$1.90 per day international poverty line. Before incurring any health-related expenditure, 1. When applying the US$5.50 per day poverty line, there was a considerable shift in the proportion of households below the poverty line before and after spending on healthcare

PLOS GLOBAL PUBLIC HEALTH
The burden of catastrophic and impoverishing health expenditure in Armenia (Table 5)

PLOS GLOBAL PUBLIC HEALTH
The burden of catastrophic and impoverishing health expenditure in Armenia  . Impoverishing health expenditure was also characterized by a higher incidence among the poor relative to the other quintiles across all years examined but was only significantly higher in 2017 ( Table 6).   (Table 7). Table 8 presents the uOR, aOR, and p-values for the correlates of impoverishing health care expenditures in Armenia using the 2018 ILCS.

Correlates of CHE and impoverishing health expenditure
Although the current marital status was not a significant factor in bivariate analyses, in the adjusted analysis, households with heads that were married had significantly 44 percent (aOR = 1.44, 95 percent CI: 1.05-1.99, p-value = 0.024) increased odds of being impoverished as a result of OOP payments. Additionally, households with at least one hypertensive member (aOR = 2.51, 95 percent CI: 1.56-4.03, p-value<0.001), at least one member belonging to the pensioner group (aOR = 1.94, 95 percent CI: 1.28-2.93, p-value = 0.002), at least one member belonging to the military social group (aOR = 2.17, 95 percent CI: 1.23-3.81, p-value = 0.007), or was located in urban areas (aOR = 1.76, 95 percent CI: 1.10-2.83, p-value = 0.019) were significantly more likely to be impoverished compared to households without hypertension, a member belonging to a pensioner or military social groups, or located in rural areas. Again, households with high socioeconomic status were protected from becoming impoverished because of OOP payments for health compared to households with a low socioeconomic status (Table 8).

Discussion
This study presents the most recent comprehensive update of catastrophic and impoverishing health expenditures in Armenia since 2013. Moreover, it is one of the first studies to examine the correlates of Armenia's catastrophic and impoverishing health expenditures. Based on our analysis, Armenia's OOP payments declined between 2014 and 2018 due to declining outpatient and inpatient costs. Although there was a general decrease in CHE incidence among Armenian households between 2014 and 2018, CHE dramatically increased between 2013 and 2014, and in 2017 -coinciding with a fall in per capita public health spending between 2016 and 2017-before dipping again in 2018 [32]. Similarly, impoverishing health spending fell overall between 2014 and 2018, despite a peak in 2017.
The variation in CHE and impoverishment between 2014-2018 can be attributed to various reasons. First, over time, there have been changes to the benefits package composition, such as services, tariffs, and qualifying groups [33,34]. For example, the maximum age for children eligible for more generous coverage under the BBP has continued to change, from three years old in 2001 to eighteen in 2019 [34]. Moreover, the BBP's service list differs over time as the

PLOS GLOBAL PUBLIC HEALTH
The burden of catastrophic and impoverishing health expenditure in Armenia  [35]. The increases in wealth translated to growth in household consumption. Monthly adult consumption increased overall between 2014 and 2018 from 47,622 AMD to 48,575 AMD-an annual increase of 24 percent [36]. The rise in consumption expenditure and parallel falls in OOP, translated to reductions in catastrophic health spending. In addition, the proportion of the population living below the UMI poverty line of $5.50 a day decreased between 2014 and 2016 by seven percent before trending upwards in 2017 (increase of four percent) [37]. Despite the overall falls in poverty and rise in household consumption in the study period, 2017 was marked by a decline in public health spending that shifted the burden of health care to households and led to a rise in impoverishing and catastrophic health spending. The improvement in poverty rate and increasing consumption over the studied period strengthens the argument that OOP payments are the main driver of increasing CHE. In other words, despite the increasing prosperity enjoyed by Armenians, it was insufficient in protecting households against the impoverishing effects of CHE. Despite efforts by the government to target vulnerable groups with generous coverage, the incidence of catastrophic and impoverishing health expenditures was disproportionally concentrated among Armenia's more vulnerable socioeconomic groups between 2014 and 2018. Examining the impoverishing health expenditure concentration curves and index values demonstrates that the poorest Armenians are much more likely to be pushed into poverty from health spending than any other wealth quintiles. These disparities have also grown over time. While the wealthiest were less likely to experience CHE and impoverishment in 2018 compared to 2014, Armenia's poorest were more likely to experience CHE and impoverishment in 2018 than in 2014 (i.e., more people in the lowest quintile experienced CHE and impoverishment in 2018 compared to 2014). The concentration of financial barriers to health care access among vulnerable groups suggests a need to increase or better target public spending for health benefits. Given ongoing efforts to target health benefits, these inequalities may also highlight the importance of universality in designing health benefits, which may improve equity and reduce inefficiencies arising from the cost of implementing targeting mechanisms. Our findings on the concentration of CHE and impoverishment among the poor are similar to those reported by other studies [3,18,38,39]. Besides a household's level of wealth, a significant predictor of CHE was whether a household member had hypertension, a common risk factor for chronic noncommunicable diseases. After accounting for all other factors, households with a hypertensive member were more than five times as likely to incur a CHE. Households with elderly over the age of 65 were also 47 percent more likely to experience a CHE than households without any elderly members, even if they could receive some form of social benefits. These findings are not just reflective of global trends but were expected as elderly and individuals with chronic conditions often tend to have more facility visits than young and healthy individuals [5]. Our findings point to the importance of strengthening financial protection for people living with NCDs in Armenia, which is essential to prevent the development of complications that are relatively more expensive to manage, and productivity losses due to premature death and preventable disability. Similarly, households with a disabled member had more than twice the odds of incurring CHE compared to households without any disabled members, regardless of their benefits or insurance status. Interestingly, in contrast to other countries' experiences [40][41][42], urban residents had a greater odds of incurring CHE than their rural counterparts controlling for all other factors, including participation in social benefit programs. This may be explained by the fact that urban residents tend to bypass primary health care providers for expensive specialist care in the urban polyclinic model in which the scope of care of specialists and family physicians overlaps [43]. A critical protective factor against CHE was whether a household had at least one member with paid work. Regardless of one's level of wealth or other characteristics, households with at least one member with paid work had 45 percent reduced odds of incurring CHE compared to households without paid work. The presence of paid work-informal or formal-may be a proxy for a higher ability to pay for health services, buttressing the fact that the poorest households continue to incur a disadvantage in terms of access to care.
As for the impoverishing impact of health spending, urban households, households with members belonging to the military social group, households with at least one pensioner, and households with a hypertensive patient had higher odds of IHE. Although households with a disabled member were more likely to face impoverishment, these odds were not statistically different from their non-disabled counterparts. Surprisingly, access to social benefits or Armenia's BBP did not appear significantly impact the odds of experiencing impoverishing health expenditures, indicating Armenia's benefits package does not offer a sufficient level of support to meaningfully protect Armenians from impoverishing health spending. Hence, there is a need to re-examine the depth and service scope of coverage in the BBP in Armenia to ensure that it confers adequate financial risk protection to all Armenians.
Armenia has made strides to expand services covered under the BBP, including providing primary health services for the general population as well as inpatient services for the poor and other vulnerable groups. However, with Armenia's growing burden of chronic noncommunicable diseases and the high cost of outpatient and diagnostic treatment for most, the country's current health system and BBP cannot address the growing disparities in catastrophic and impoverishing health spending. Similar to other European and Central Asian countries, outpatient medications are a key driver of OOP payments in Armenia [44,45]. Without pharmaceutical pricing regulation, the VAT tax on medication, and a negative perception of generic drugs among patients and providers, Armenia's reliance on OOP payments to finance expensive outpatient treatment puts many households at financial risk [45].
Increasing the government's financing for health would allow the government to strategically finance more health services with prepaid public resources [46]. For example, in a study of Eastern European countries, Estonia not only has the second highest public spending on health and second lowest levels of OOP payment but also offers a generous package that covers most hospital care with cost-sharing elements for pharmaceuticals, dental care, and therapeutic appliances [47,48]. Several considerations may inform the options for financing an expanded benefits package in Armenia, including the population age structure, formal employment rates, the strength of tax administrative mechanisms, and the broader fiscal policies [5,9,31].
The study has several key limitations. The analysis of Armenia's ILCS demonstrates correlations between factors and outcomes and does not identify any causal relationships between specific factors and the levels of health expenditures. The analysis cannot directly ascertain the individual circumstances that lead to differing household healthcare spending and garner further study. Similarly, while the authors conducted a literature review of common correlates of OOP payments, CHE, and IHE, the study may be at risk of omitted variable bias if there are other context-specific variables that we did not account for in our analysis. Given the survey's focus on Armenia, the results of our analysis may not apply to other contexts outside of Armenia.
This study did not include direct non-medical costs, such as transportation, due to data unavailability, which have proven to increase the incidence of CHE and IHE in other settings [18,39]. Lastly, although examining healthcare expenditures precludes an analysis of those who may not seek care, analyzing the burden of catastrophic and impoverishing health expenditures provides insight into how cost affects healthcare use broadly amongst Armenians, which may also have implications for those who do not seek out healthcare. Future studies should include and explore these factors.

Conclusion
This study offers a detailed examination of the burden of catastrophic and impoverishing health spending in Armenia to date. As the analysis draws from a nationally representative survey of Armenia, the results are generalizable throughout the population, providing a detailed picture of what Armenians are likely to experience on average. Investigating the association between socioeconomic factors and healthcare spending trends in Armenia provides greater insight into the progress Armenia has made on healthcare spending and which groups need more robust support to reduce the incidence and intensity of CHE and IHE. Considering nearly a fifth of Armenian households experienced a CHE in 2018, Armenia's efforts to expand UHC and offer patients greater financial risk protection may require the country to undertake reforms to increase prepaid pooled financing to finance an expanded benefits package for the whole population.