Trends in Out of Pocket Payments and Catastrophic Health Expenditure in the Kyrgyz Republic Post the “Den Sooluk” Health Reform, 2012 - 2018

Background: To examine the trends in out- of- pocket health payments (OOPPs) and the incidence of catastrophic health expenditure post the “Den Sooluk” health reform, we used data from the Kyrgyzstan Integrated Household Survey (2012 – 2018). Methods: Population-weighted descriptive statistics were used to examine the trends in OOPPs and catastrophic health expenditure at three thresholds; 10 percent of total household consumption expenditure (Cata10), 25 percent of total household consumption expenditure (Cata25) and 40 percent of total household non-food consumption expenditure (Cata40). Panel and cross-sectional logistic regression with marginal effects were used to examine the predictors of Cata10 and Cata40. Findings: Between 2012 and 2018, OOPPs increased by about US $6 and inpatient costs placed the highest cost burden on users (US $13.6), followed by self-treatment (US $10.7), and outpatient costs (US $9). Medication continues to predominantly drive inpatient, outpatient, and self-treatment OOPPs. About 0.378 to 2.084 million people (6 – 33 percent) of the population incurred catastrophic health expenditure at the three thresholds between 2012 and 2018. Residing in households headed by a widowed or single head, or residing in rural regions, increases the likelihood of incurring catastrophic health expenditure. Conclusions: The initial progress in the reduction of OOPPs and catastrophic health expenditure gained appear to gradually erode since costs continue to increase after an initial decline and catastrophic health expenditure continues to rise unabated. Efforts needs to be made to check the rising out-of-pocket costs to enable a reduction in catastrophic health payments.


Introduction
One of the principal goals of health reforms especially in developing countries is to ensure that households and individuals do not experience nancial hardship from accessing and utilising healthcare (1). This is based on the principle that healthcare is a human right and nancial barriers are a signi cant impediment to healthcare access especially for the most vulnerable (the poor and sick) (2). A key element of health reforms that aim to protect users from nancial hardship is the move from direct payment for healthcare to universal health coverage where risk and nancial resources are pooled for e cient healthcare access and utilisation (3). The emphasis on nancial protection to reduce or remove out-of-pocket payments and catastrophic healthcare expenditure is established in its inclusion in the Sustainable Development Goals (4).
The Kyrgyz Republic like other Central Asian countries inherited a health system from the former Soviet Union characterised by universal health coverage and user-free health access although health systems were centralised which resulted in high bureaucratic costs and ine ciencies (5). With the collapse of the Soviet Union and the withdrawal of economic links established during the Soviet era, most of the Central Asian countries experienced a signi cant reversal in economic growth (6). This resulted in di culties for public funds to manage social and health services and user fees were introduced (7). Thereafter, the burden of access to services, including healthcare, was falling more on the population and more people were being pushed into poverty from accessing services including healthcare (7). Public spending on healthcare as a percentage of GDP dropped from over 6 percent in 1994 to about 2 percent in 2004, and it was estimated that per capita government expenditure on health in 2003 was just US$66 purchasing power parity (6,8). Although the economy experienced important positive growths and increases in GDP in the late 1990s, recovery has been slow and funding for healthcare has suffered (9).
The Kyrgyz Republic has made signi cant progress to date in improving the living conditions of its populace. Poverty rates have declined to about 23 percent (10) and important reforms have been enacted to improve healthcare delivery and to also protect users from nancial barriers to healthcare access and utilisation (11). A chronology of health reforms in the Kyrgyz Republic adapted from Meimanaliev et al. (2005) and Falkingham et al. (2010) can be found in Table 1. In the period covering 1996-2006, the National Health Care Reform Program "Manas" was developed with the support of the World Health Organization with the primary aim of unbundling the Soviet-era health system in the Kyrgyz Republic (5). The main features of this health reform were the creation of an infrastructure that corresponds to population needs in medical care and nancial resources, the decentralization of management and enhancement of administrative and nancial autonomy of health organisations, pooling of health funds, outcome-based provider payment mechanism, and the split of the health sector into providers and purchaser of healthcare services. The "Manas" health reform faced important health nancing challenges. Households which utilized health services at outpatient and/or inpatient levels experienced heavy nancial burden caused by persistence of informal payments and high level of co-payment, and low public health spending as a share of GDP lingered. total government expenditures; (2) the establishment of the Department of Public Health in the Ministry of Health with the goal to provide comprehensive coordination of the healthcare services and their s integration with other health programs; and (3) a wage increase for healthcare workers to redress the continued increase in informal payments which was associated with low wages (14). However, some health system issues persisted. There was a need for a stronger focus on investments that could change population health behaviour and clinical practice in order to improve the e ciency of key health interventions, the increased out ow of human resources has adversely affected both the access to healthcare and its quality, especially for vulnerable populations in remote rural areas, incomplete de nition of roles and responsibilities and limited management autonomy of healthcare providers have generated a governance challenge, and adequate funding remains critical for maximizing population coverage with cost-effective healthcare services due to year of reduced funding. The "Den Sooluk" health reform was proposed to consolidate the "Manas Taalimi" health reform successes and to address identi ed reform gaps.
[ Table 1 here] The "Den Sooluk" health reform The "Den Sooluk" (2012-2016) health reform is considered a logical continuation of the "Manas Taalimi" health reform with a foundation laid by deep analysis of results, problems and experience obtained during previous "Manas" and "Manas Taalimi" years and aimed at maintenance and protection of population health which would contribute signi cantly to poverty reduction (WHO, 2011). Key strengths of the health reform include: (1) the reform proposals to match the identi ed health and health systems problems in a balanced way, (2) the program aims at social health protection (universal coverage, fairness in nancing, equal access to services and prevention of impoverishment using international best practices), (3) the coordination, implementation and management arrangements are based on sound principles and make organisational sense, (4) the need for solving health human resources in general and for strengthening management and supervision capacity in particular are acknowledged, and (5) nancial management and procurement policy and standards are adequate. An assessment of the performance of the health form is underway. It is important to note that a review of the health reforms in the Central Asian region suggests that among the ve countries, the Kyrgyz Republic has undergone the broadest, most sustained, and most successful health sector reform in the region (16).
This paper examines the trends in out-of-pocket and catastrophic health expenditures post the "Den Sooluk" health reform using data from 2012-2018. Although other health reforms have continued since the end of the "Dan Sooluk" health reforms and that there might have taken place other non-health reform interventions which could in uence household health expenditures, this paper focuses on a key objective of "Dan Sooluk" which is protection from economic hardship from seeking healthcare.  (17). On average, 5000 households with 22,000 individuals who were randomly selected by strata are surveyed per data round. The survey contains a module on household food and non-food consumption expenditure and on out-ofpocket payments (OOPPs) for healthcare utilisation, and OOPPs data are collected for inpatient, outpatient, and selftreatment healthcare expenditures. Our analyses focused on the 2012-2018 data rounds in line with our research objective and also since the sample methodology and questionnaire design changed post 2010 hence and this limits data comparability (9).

Out-of-pocket payments
For each category of healthcare expenditure (inpatients, outpatients, and self-treatment), we disaggregated our analysis to examine the components of health spending hence for inpatient care, costs were aggregated under medications, hospitalisation, monetary value of gifts and kinds, and "others". For outpatient care, costs were aggregated under medications, monetary value of gifts and kinds, and "others", and for self-treatment, costs were aggregated under medications and "others". Gifts and kinds were considered as an important cost component of OOPPs since the Kyrgyz Republic like other old Soviet Union countries inherited the tradition of presenting monetary and/or in-kind gifts to health providers and care givers (6,13,18). The "other" category of health spending included costs incurred for medical supplies, diagnostic and lab services, and for inpatient care, costs also included payments to physicians, surgeons, and other hospitalisation supplies. Falkingham et al. (2010) argued that they could not separate formal and informal (gifts and kinds) payments based on the likelihood that some enumerators could have been unclear whether 'charges' demanded by medical personnel prior to consultation were 'o cial' or not. We retained the distinction between formal and gifts and kinds (informal) payments in the study since these payments have persisted over the years and experts with local understanding of healthcare payments in the Kyrgyz Republic observes that these payment categories are clear and usually subtly expected especially in the public health system (authors' personal correspondence).

Catastrophic health expenditure
To estimate catastrophic health expenditure, we used two approaches; based on OOPPs as a share of total household consumption expenditure (food and non-food), and OOPPs as a share of total household non-food expenditure. These two approaches utilised household expenditure net of healthcare payments.
For catastrophic health expenditure based on total consumption, we used two thresholds at 10 percent (Cata10) and 25 percent (Cata25) of total household consumption expenditure. These thresholds, called the budget share threshold, were used because they are the o cial Sustainable Development Goals (SDGs) thresholds for estimating catastrophic health expenditure (SDG, 2018). A review of studies that have estimated catastrophic health expenditure by Wagstaff et al. (2018) found that in general, the budget share approach (i.e. total consumption expenditure) was the second most popular method for estimating households' ability-to-pay for healthcare and 29 percent of studies have applied this methodology. However, in economics journals, total expenditure was the most frequently used methodology (49 percent). The thresholds of 10 percent and 25 percent were chosen since they are the most used threshold in studies that have used this approach to estimate catastrophic health expenditure (19).
When estimating catastrophic health expenditure as a share of household non-food expenditure, we used the threshold of 40 percent (Cata40). This approach is in line with the methodology of adjusting analysis to re ect household capacity to pay by subtracting household food expenditure or an allowance from total consumption as suggested by Xu et al. (2003). This is based on the argument that food expenditure is non-discretional and hence does not re ect household capacity to pay for expenditures net of food consumption expenditure (21). Although in general, this methodology is the most popular approach to estimating catastrophic health expenditure (31 percent of studies), in economics journals, it is the least popular approach (13 percent of studies) (19). The threshold of 40 percent was chosen since it is the most used threshold in studies that have used this approach to estimate catastrophic health expenditure (19).
To collect household consumption (food and non-food) and health expenditures, households kept a daily expenditure diary where costs were collected every quarter by enumerators. The frequency of data collection and diary keeping helped to limit the effect of recall bias on expenditures (17). Diaries were examined by enumerators every quarter and follow-up questions were administered where missing or incomplete information were observed. The costs of non-food and health expenditures were aggregated per month and food expenditures were aggregated per 14 days. We extrapolated monthly food expenditure by multiplying biweekly expenditures by two. Costs were converted to 2020 US dollars (22).

Predictors of catastrophic health expenditure
To estimate sociodemographic and economic characteristics that predict catastrophic health expenditure at household and individual levels, we developed two regression models using Cata10 and Cata40 as outcome variables. We estimated only two models since the two approaches to estimating catastrophic health expenditure use different denominators (total vs non-food consumption expenditures) and hence, different sociodemographic and economics factors might predict catastrophic health expenditures differently at these thresholds.

Outcome variables
The two outcome variables (Cata10 and Cata40) were estimated as binary outcomes with "1" indicating catastrophic health expenditure and "0" indicating no catastrophic health expenditure.

Predictor variables
Age of household members -The ages of sick household members were grouped into ve categories and were used as predictor variables based on the hypothesis that the ages of different household members might predict catastrophic health expenditures differently.
Marital status -The marital status of the household head was estimated as a categorical variable and households were grouped under the following categories: legally married and civil unions, divorced and separated, widowed, and single. This predictor variable was used based on the hypothesis that marital status of household heads might in uence household socioeconomic status and catastrophic health expenditure.
Sex of household members -The sex of sick household members was estimated as a predictor variable based on the assumption that sex might predict catastrophic health expenditure differently.
Location -The location of households was used as a predictor variable based on the hypothesis that urban and rural households might incur catastrophic health expenditure differently at the established thresholds.
Consumption expenditure quantile -This variable was developed to categorise households into different socioeconomic quantiles based on household spending adjusted for household size. Consumption spending has been used in numerous surveys and studies to estimate household poverty levels among other indicators (23). This predictor variable was chosen based on the hypothesis that households that belong to different quantiles might incur catastrophic health expenditure differently. Consumption expenditure included all household routine spending including utilities, rent, food, entertainment, clothing etc. but excluded once-off-payments including education.
Regions (oblasts) -This was included as a predictor variable based on the hypothesis that catastrophic health expenditure might be incurred differently based on the region in which a household resides. There are seven regions and two administrative centres in the Kyrgyz Republic. Since a new administrative centre was added in the surveys post 2012 and due to changes in coding system in 2018, regions were excluded in the analyses for these two years.

Analysis
The data were analysed using Stata 15.1 statistical software (24). Univariate and bivariate descriptive analyses were used to summarize the data characteristics and to examine trends in OOPPs. Univariate and bivariate analyses were populationweighted to ensure that the estimated could be extrapolated from the Kyrgyz Republic population. To achieve this, household sample weights were divided by household size. To examine the predictors of catastrophic health expenditure, populationaverage estimator panel logistic regression analysis was speci ed, and marginal effects are reported. Population-average estimator panel regression was used since we believe that this approach provides a more useful approximation of observed association as recommended by Hubbard et al. (2010). Since the panel was rotational and hence unbalanced, we conducted the panel regression on the pooled data while each year was also analysed as cross-sectional surveys where marginal effects of logistic regressions were examined. We used this approach to examine if there were any important changes in the predictors of catastrophic health expenditure due to the annual change in data composition. All regression analyses were population-weighted, and signi cance was established at 95 percent and 99 percent con dence intervals.

Results
The description of the sample statistics is contained in Table 2    [ Table 3 here] Trends in out-of-pocket health expenditure For households which sought inpatient care, the cost of medication was the highest driver of OOPPs (37 percent), hospitalisation costs were 19 percent, and other costs including medical supplies and payment of physicians contributed 35 percent (Fig. 1). Household which sought inpatient care incurred a substantial cost through payments for gifts and kinds to healthcare providers as this represented 9 percent of total inpatient OOPPs. While the share of inpatient OOPPs attributed to [ Figure 1 here] Trends in catastrophic health expenditure Examining the share of households that suffered catastrophic health expenditure at 40 percent of total household non-food consumption expenditure, the incidence of catastrophic health expenditure was 26 percent and incidence increased by about 14 percentage-points; from 19 percent in 2012 to about 33 percent in 2018.
[ Figure 2 here] Although more rural households incurred catastrophic health expenditure at Cata10, Cata25, and Cata40 ( This suggests that while more rural households incurred catastrophic health expenditure at 40 percent of total non-food consumption expenditure, there was a more rapid increase in Cata40 in urban areas relative to rural areas. [ Figure 3 here] While households who belonged to the highest consumption expenditure quantile incurred a higher catastrophic expenditure relative to those who belonged to the lowest quantile across the three thresholds, over the years, the incidence of catastrophic expenditure increased more rapidly for households who belonged to the lowest quantile (Fig. 3). In 2012, equal proportion of households who belonged to the lowest and highest consumption quantile incurred Cata10 (13 percent) and by 2018, the incidence has increased to 23 percent for the lowest quantile households and 22 percent for the highest. In 2012, equal proportion of households who belonged to the lowest and highest total consumption expenditure quantiles incurred Cata25 (6 percent) and by 2018, the incidence increased to 10 percent of households who belonged to the lowest quantile and 8 percent for those who belonged to the highest quantile). Similarly, the incidence of Cata40 was higher among households who belonged to the highest consumption expenditure quantile in 2012 by about 5 percentage-points, by 2018, equal proportion of households who belonged to the lowest and highest consumption expenditure quantile incurred catastrophic health expenditure at Cata40 (34 percent). (Fig. 4). These ndings suggest that catastrophic health expenditure at different thresholds have increased over the years albeit in varying magnitude with households who belonged to the lowest quantiles experiencing more growth.
[ Figure 4 here] The incidence of catastrophic health expenditure was most concentrated among households in Naryn, Chui, Batken, and Osh regions while Bishkek, Jalal-Abad, and Talas regions recorded the least incidence at the three thresholds (Fig. 5). [ Figure 5 here]

Predictors of catastrophic health expenditure
The predictors of catastrophic health expenditure are presented in Tables 4 and 5, and only signi cant associations are present in text.   Although the magnitude of association was small, while households who belonged to the lowest consumption expenditure quantile were more likely to incur Cata10, households who belonged to the middle quantile was less likely to incur Cata10 when a member was sick and sought care. Belonging to the lowest quantile was associated with a 0.8 percentage-point increase in the likelihood of incurring Cata10 while belonging to the middle quantile was associated with a 0.9 percentagepoint decrease in the likelihood of incurring Cata10. These associations were also observed in the disaggregated analyses.
Relative to the richest and most developed region in the Kyrgyz Republic (Bishkek), household who resided in Issyk-Kul, Naryn, Osh, Osh city, and Chui were more likely to incur Cata10 when a member was sick and sought care. The magnitude of the association strongest in Naryn, Chui, and Osh city regions suggesting a stronger association in these regions.
[ Table 4 here] Households with a sick member who was between 11-20 (1.5 percentage-point increase in likelihood), 41-50 (1.7 percentage-point increase in likelihood), and 51-60 years old (5.5 percentage-point increase in likelihood) were more likely to incur Cata40 relative to when a sick member was between 31-40 years old. The magnitude of the signi cant association remained relatively the same when analyses were disaggregated ( percentage-point when analyses were disaggregated by year. Although at varying magnitudes, households that belonged to lower and middle consumption quantiles were less likely to incur Cata40 when a household member was sick and sought care (see Table 5 for more details). Relative to richest and most developed region (Bishkek), while household who resided in Issyk-Kul, Jalal-Abad, Osh, and Talas were less likely to incur Cata40 when a member was sick and sought care, households who resided in Naryn, Chui and Osh city were more likely to incur Cata40 when a member was sick and sought care. The magnitude of the association in Naryn, Chui, and Osh city regions were stronger than those in the other regions suggesting a stronger association (see Table 5 for more details).

Discussion
Our data suggest that the initial progress made in reducing the level of out-of-pocket payments for healthcare and the incidence of catastrophic health expenditure by the "Manas Taalimi" and the "Den Sooluk" health reforms is beginning to erode since costs are rising and the incidence of catastrophic health expenditure is increasing. Although a key priority of the "Den Sooluk" health reform was to improve on social protection for healthcare utilisation, our data suggests that out-of-  (26) and Kazakhstan (27).
Medication remains the highest driver of out-of-pocket payments and while medication costs as a share of out-of-pocket inpatient costs declined over the years, as a share of outpatient costs, there has been a noticeable increase. Baschieri and Falkingham, (2006)  Comparing this nding to our study, ndings indicate that catastrophic health expenditures have continued to increase using the SDGs thresholds post "Manas Taalimi" and "Den Sooluk" health reforms. Estimating catastrophic expenditure using 40 percent of total non-food consumption expenditure as the threshold, ndings also suggest that the incidence of catastrophic There is limited comparable data that have examined the predictors of catastrophic health expenditure especially in Central Asia. Although we could only examine associations and not causality, our study ndings suggest that residing in households headed by a widowed or single head, or residing in rural regions (Naryn, Osh, and Chui), when sick, increases the likelihood of households incurring catastrophic health expenditure when care is sought. While households in the lowest consumption expenditure quantile were more likely to incur Cata10 relatively to the highest quantile, they were less likely to incur Cata40. This suggests that perhaps food expenditure for low status households impacts on their capacity to pay for healthcare (i.e. food expenditure takes a signi cant chunk of their spending). These ndings adds to the limited literature and provides comparable data for future studies in the Central Asia region.
Our study has several limitations. The KIHS do not collect information on the nature of ill-health and hence, we could not determine if catastrophic health payment was related to speci c healthcare expenditures. The KIHS also discontinued the collection of detailed information on out-of-pocket payments including on co-payments, timing of informal payments, and type of facility utilised (private, public, etc.). These are important questions in the analysis of out-of-pocket payments and catastrophic health expenditure.
In conclusion, the initial progress in reducing the growth in out of pocket payments and catastrophic health expenditure experienced with the introduction of the "Manas Taalimi" and "Don Sooluk" health reforms appears to be gradually eroded over the years since costs continue to increase after an initial decline and catastrophic health expenditure continues to rise unabated. Although the Kyrgyz Republic has been at the forefront of health reforms in Central Asia, the sustainability of health reforms is crucial in redressing and limiting the decline in the successes achieved through signi cant health reforms.

Abbreviations
OOPPs Out-of-pocket payments Cata10 Catastrophic health payment at 10% of total household consumption expenditure Cata25 Catastrophic health payment at 25% of total household consumption expenditure Cata40 Catastrophic health payment at 40% of total household non-food consumption expenditure