Hardship financing of out-of-pocket payments in the context of free healthcare in Zambia

Despite the removal of user fees on public primary healthcare in Zambia, prior studies suggest that out-of-pocket payments are still significant. However, we have little understanding of the extent to which out-of-pocket payments lead patients to hardship methods of financing out-of-pocket costs. This study analyses the prevalence and determinants of hardship financing arising from out-of-pocket payments in healthcare, using data from a nationally-representative household health expenditure survey conducted in 2014. We employ a sequential logistic regression model to examine the factors associated with the risk of hardship financing conditional on reporting an illness and an out-of-pocket expenditure. The results show that up to 11% of households who reported an illness had borrowed money, or sold items or asked a friend for help, or displaced other household consumption in order to pay for health care. The risk of hardship financing was higher among the poorest households, female headed-households and households who reside further from health facilities. Improvements in physical access and quality of public health services have the potential to reduce the incidence of hardship financing especially among the poorest.


Introduction
The general failure of cost sharing-based policies in healthcare that led to worsening conditions in terms of access and financial protection in health systems around Sub-Saharan African led to new health financing policy regimes. Some countries have experimented with a number of policies such as community-based health financing schemes, user fee abolition, prepayment schemes, social health insurance, or a mix of these [1,2]. These policies were intended to guarantee greater healthcare access and financial protection, especially to the poorest who were worst hit by user charges. Arguably, the boldest of such policies has been the complete removal of user fees in the public sector, starting with primary healthcare which is seen to benefit the poorest more [3,4]. Despite these policy efforts, out-of-pocket payments (OOPs) associated with a visit to a health provider remain high, and in some cases prohibitive, in Sub-Saharan Africa [5,6]. African health systems rely heavily on OOPs as a source of financing for healthcare. In Zambia, healthcare seeking is associated with significant OOPs at the point of use in both public and private facilities.
In this paper, we focus not merely on the amount spent but also on identifying the difficulties or hardships represented by particular means by which money is mobilized to pay for out-a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 Zambia's public health system is structured in a four-tier pyramidal referral set-up, with primary healthcare at the bottom. Primary healthcare consists of a network of health posts, health centres and district hospitals managed under the District Health Office (DHO). Above the primary healthcare level are second level hospitals. Second level hospitals were designed to serve as the highest referral hospitals at provincial level. The next level of care are the third level hospitals which provide tertiary level healthcare and in some cases medical training and research. Finally, at the apex of the health system are specialised hospitals.

The sample and data
The 2014 Zambian Household Health Expenditure and Utilisation Survey (ZHHEUS) is a cross-sectional survey conducted by the Central Statistical Office and the University of Zambia with support from the Ministry of Health. The data were collected using a two-stage stratified cluster sample design. In the first stage, standard enumeration areas were selected within each stratum using the probability-proportional-to-estimated-size procedure. In the second stage, a fixed proportion of 20 households were selected from each enumeration area using a systematic random sampling procedure. The data was collected from 12,000 households sampled from across all the ten provinces in the country. For all selected households, data was collected on each household member giving a total of about 60,000 individual observations. The survey response rate was 99.4%.
Individuals were asked if they had experienced an illness or injury in the 4 weeks preceding the survey, or if they had been admitted to a health facility in the 6 months preceding the survey. The survey included questions on health status (self-rated health status and self-reported illness experience); healthcare utilisation (visits, admission, and type of providers sought), and health expenditure (amount spent during visit or admission, item on which amount was spent, and source of funds). Out-of-Pocket health expenditure included expenditure at the facility (for drugs, consultation, medical examinations, etc.), medical expenditure outside the facility (e.g. for drugs purchased at a private Drug Store rather than the facility visited), non-medical costs associated with seeking healthcare (e.g. travel costs, food, lodging, etc.). This study focused on analyzing hardship financing related to outpatient visits rather than admissions. The data on out-of-pocket payments for admissions was very noisy. For example, of those who were admitted, 92% reported zero expenditure. We strongly believe that this data suffered from recall bias.
The survey included several socioeconomic and demographic variables. Household income was measured indirectly using consumption expenditure in the one month prior to the survey, as is common in similar surveys in low income countries. For education of the head of the household, the survey asked the highest level of schooling completed, while employment type was categorized as either employed or unemployed. Demographic variables on individuals in each household, including age, sex, marital status, residence, employment status, and so on, were also captured. A number of these individual and household characteristics were included as covariates as shown in Table 1.

Measures of hardship financing for OOPs
In line with the literature, we define a measure of 'hardship' financing as methods of raising money for out-of-pocket payments that either signal or induce financial difficulty at the household level. A household is considered to have experienced hardship financing if they reported financing healthcare OOPs using any of the following four mutually-exclusive options: (i) sold assets or borrowed money, (ii) asked a relative or friend to pay, (iii) experienced catastrophic level of OOPs, and (iv) did not seek care in avoidance of unaffordable OOPs [7,8,23,24].
The empirical literature provides justification for the inclusion of the indicators of hardship financing. For example, selling assets or borrowing money are considered hardship means of coping with out-of-pocket healthcare costs because of the economic hardships that both strategies impose on households [9]. While reliance on the generosity of friends or relatives might reflect presence of some degree of social capital, it is most likely that the individuals rather faced financial hardships. The phenomenon of households asking relatives or friends to defray health costs is commonly associated with financial distress due to illness [25][26][27]. Further, as identified by Xu et al [28], CHE presents a form of hardship financing because CHE often implies that households have to forgo or compromise other basic household consumption such as food in order to pay for healthcare. Another study has used CHE as an indicator of hardship financing [29]. And, a household forced into CHE by healthcare OOPs is likely to get into debt or poverty. CHE is calculated as OOPs per visit that exceeds 10% of total household expenditure [2,30]. Finally, forgoing treatment on account of cost is a strategy that indicates that a household faces a real prospect of getting into financial distress due to OOPs.

Statistical model of determinants of hardship financing
Our statistical analysis is divided into two parts. We first run a two-stage sequential response model to demonstrate the association between the likelihood of experiencing hardship financing and a host of individual demographic and socio economic characteristics conditional on an individual reporting an illness and incurring some positive out-of-pocket payment. A sequential response model is a statistical method used to analyse sequential decisions or events [31]. The model assumes that individuals make choices, but often these choices are not made simultaneously but rather sequentially. The sequential response model is essentially a sequence of binary-outcome models that could be either logit or probit models. Exceptions do exist in which a stage in the sequence is another probability model such as a multinomial-response model [32]. The sequential model estimated in this paper combines binary and multinomial logistic models.
In the first stage of the sequential response model, an individual who has visited a health provider is faced with two alternatives, either they pay or do not pay. This means the group of individuals who decide to seek care belong to one of the two categories; paying or not paying. Given only two options we estimate the binary logit model specified as where; h = 1 if the individual did not pay for healthcare and h = 2 if the individual paid to access health services.
In the second stage, we estimate a multinomial logistic regression. Individuals who visited a health provider and paid to access health services are faced with one or more of the scenarios specified in the model below. For estimation purposes, we use the category of individuals who paid for health services using own cash without incurring Catastrophic Health expenditure as the reference category.
j = 0 if the individual paid for healthcare using own cash without incurring Catastrophic Health expenditure j = 1 if the individual borrowed money or sold assets to pay for healthcare j = 2 if the individual requested someone else to pay for them j = 3 if the individual paid above the threshold considered above Catastrophic healthcare expenditure (CHE) Finally, the second part of our statistical analysis estimates a multinomial logit model in order to analyse the factors that are associated with the decision to seek care or not to seek care. All analyses were conducted using STATA/SE 13.0.

Ethical considerations
Since the survey did not involve collection of human samples, ethical exemption for this study was granted under the provisions of the Census and Statistics Act Number 127 of the laws of Zambia. No identifying information of individuals or health institutions were collected in the survey. In observance of the ethical requirements, only participants aged at least 15 years were interviewed after giving written (signature or thumb print) informed consent. Table 1 summarises the characteristics of sampled individuals. The majority of households are headed by the males who are the primary decision makers of the household. Consistent with other national surveys, about 60% of the population reside in rural areas [33,34]. Education levels are generally low with more than half having either no formal schooling or only primary education. About 79% of heads of households were in employment (salaried or self-employed) while the rest were classified as unemployed (e.g. student, home-marker, aged, etc.). The average nominal per capita monthly household expenditure is Kwacha 214.7 (equivalent of US $21.47).

Health service utilization and hardship financing of OOPs
In Table 2 we present results on illness experience, healthcare seeking options, out-of-pocket expenditure and measures of hardship financing. About one fifth of the sample reported an illness or injury in the four weeks preceding the survey. Of the persons who reported illness or injury, 62% visited a health facility while 29% self-medicated. Furthermore, 9% could not seek care due to cost, illness not being serious, religious reasons, etc. More than half of the respondents who reported an illness four weeks prior to the survey suffered from either Malaria or Fever. Other prominent illnesses reported include Diarrhea (10%), Respiratory infections (4%) and headache (5%).
The majority of the patients visited health centers and health posts. Only 3% visited private facilities. In Table 3, the Pearson's correlation test shows that household income (proxied by household consumption expenditure) is positively associated with choosing a private health facility, which suggests that the poor are more likely to choose public facilities. Similarly, visiting a private facility is positively correlated with the level of education. An association is also found between choosing a private facility and urban residence, whereby the urban residents are more likely to visit private facilities. Private health facilities are almost exclusively located in urban areas. One patient in every five incurred health expenditure during visits to a healthcare provider. The average expenditure per visit is K14.90 (US$1.49), with a standard deviation Overall, the results show that 11.4% of households who incurred health expenditure faced one form of hardship financing or another. When transportation costs are excluded, the incidence of hardship financing reduces to about 5%, showing that transportation costs are a major part of healthcare seeking costs. The major source of hardship financing was CHE (8.5%). About 3% either asked someone else to pay or borrowed or sold assets to meet their out-of-pocket expenditure. Further, it is also noteworthy that 7.5% of those who reported an illness did not seek care due to cost, which likely indicates avoidance of financial hardships. Table 4 presents the results of the sequential logit model estimation of the association between individual characteristics and the likelihood of facing any of the forms of hardship financing. The first part of the model shows the characteristics that affect an individual's likelihood of making a positive out-of-pocket expenditure. Household consumption expenditure is associated with significantly increased odds (or likelihood) of incurring a positive out-ofpocket health expenditure (OR = 1.37). Residing in the urban region is associated with increased odds of paying OOPs compared to rural regions. Paying for health services is more prevalent in urban areas because urban areas tend to have better quality health services and more facilities such as pharmacies. The result could reflect the fact that monetised transactions are limited in rural areas. Additionally, distance to a health facility is found to be positively associated with higher odds of incurring OOPs. The results further show that paying for health services is more likely among those who visited a public health centre and public health posts, compared to those who visited a public hospital, whereas the odds increase with visiting a private health facility. The odds of making payments for outpatient services at primary facilities are lower because of the policy of free healthcare.
In Table 4, we also present results which show the factors that are associated with the likelihood of households to resort to one form of hardship financing or another in coping with outof-pocket expenses, conditional on having incurred an out-of-pocket payment. Higher household income (as measured by consumption expenditure) is associated with a lower likelihood of incurring hardship financing. For each unit increase in log-transformed per capita expenditure (which is equivalent to K2.72 in natural kwacha terms), the odds of borrowing/selling assets or asking someone else to pay or making catastrophic OOPs decrease by 33%, 46% and 59%, respectively. This means that households with less resources are more prone to borrowing or selling assets to meet healthcare payments. Also, in such households, facing unaffordable OOPs increases the odds that healthcare OOPs will displace spending on other important basic households needs. Further, the odds of borrowing or selling assets or relying on someone else to help meet OOPs or making catastrophic OOPs are predicted to significantly increase with the distance to the health facility. For each additional kilometer travelled to the facility, the odds of experiencing hardship financing increases by 2-3%, across the three types of hardship financing.
In addition, the results show that after household expenditure and other variables are controlled for, female-headed households are significantly more likely to depend on asking someone else to meet unaffordable healthcare payments (OR = 1.44) and making payments which are catastrophically high (OR = 1.27). Visiting a private health facility was associated with a higher likelihood of facing financing hardships in the form of a patient requesting someone else to pay for them. The log odds of requesting someone else to pay are significantly greater (OR = 2.3) for those who visited private health facilities, compared with those who visited public hospitals. Similarly, visiting primary healthcare facilities (health centres and health posts) was associated with a reduced likelihood of requesting someone else to pay. Households who visited public health centres, health posts or mission hospitals had a reduced likelihood of facing CHE while those who visited private facilities had a significantly higher likelihood of experiencing CHE, compared to those who visited public hospitals. Other factors such as the region of residence, level of education and the type illness are found to be non-significant predictors of experiencing hardship financing.  Table 5 shows the results of the multinomial logistic regression of factors associated with an individual's inability to seek care on account of cost. The results show that living in an urban area increases the likelihood of failing to seek care on account of cost. Higher household expenditure is associated with reduced likelihood of failing to seek care on account of cost. Higher level of education is associated with lower likelihood of failing to seek care on account of cost.
In terms of factors associated with the likelihood of not seeking care due to non-cost factors, it is shown that income is not significant. Being from a female-headed household is significantly associated with a lower probability of failing to seek care due to non-cost reasons such as religious or cultural beliefs, illness not being serious, self-medication, etc. Patients from households headed by individuals without any formal education was associated with a higher likelihood of failing to seek care due to reasons other than cost. Patients of diarrhea and headache were less likely to seek care due to non-cost reasons compared to malaria patients.

Discussion
This study has examined the extent and nature of financial hardships experienced by patients due to unaffordable out-of-pocket healthcare payments in Zambia. Overall, we estimate that about 11% of patients who reported a visit experienced financial distress associated with OOPs, which by any account indicates a significant prevalence of unaffordable burden of OOPs. Hardship financing was exemplified by borrowing money, asking someone else to pay for them, or selling personal or household items, and CHE, which entails that OOPs displaces basic household consumption. This level of hardship financing is much lower than what was reported in a study by Tahsina et al [35], mainly because of the removal of user fees, though similar to a study in Bangladesh.
Furthermore, findings from the sequential logit model shades light on socioeconomic factors which are associated with the likelihood of experiencing one form of financing hardship or another: having a low household income, visiting a private facility or hospital, being under a female-headed household, and living further from a health facility. These findings have been corroborated and established in a number of past studies [7,36]. For example, studies have shown that households with severely limited incomes or resources are associated with increased likelihood of facing financial distress in meeting healthcare payments. What we also find interesting is that female-headed households are more likely to experience financial hardships in form of CHE (because they have very little income to start with), or relying on someone else to meet their healthcare costs. They often do not have capacity to borrow money or have assets to sell. Vulnerability of female-headed households to financing hardships is rooted in gender-based structural characteristics of the Zambian society. In our data, two thirds (66%) of heads in female-headed households are either widows or divorcees. In patriarchal societies such as Zambia, a woman loses access to productive household assets, such as land, housing, animals, cars, etc., when her husband dies or when she gets divorced [37]. A study in Nigeria reported that female headed-households reported higher cost burdens and affordability problems [38].
A number of factors explain, at least partially, the level of hardship financing of OOPs in Zambia. First, hardship financing is likely occasioned by gaps in the facility healthcare quality especially in terms of shortage of drugs, inadequate medical examination facilities and overcrowding. Studies have shown that good quality of facilities is of central importance for realizing the potential financial relief of free public health services [11,39,40]. Further, as Nabyonga et al [10] observed, poor patients can also be forced to go to more expensive private facilities or hospitals in fear of poor quality of care in primary facilities.
Second, it is clear from the analysis that costs of travel to health facilities, and related expenses, which constitute a significant portion of total healthcare OOPs do cause financing hardships. These hardships are generally higher in rural areas where distances are longer. Vulnerability to distress associated with unaffordable OOPs has been shown to increase with distance from a facility [36]. Third, patients may face hardships when they choose to go to private facilities which charge much higher user fees. Although in interpreting this association, we were unable to control for the selection of poorer patients into private facilities for outpatient care. It seems reasonable to assume that patients of limited financial capacities would decide to go to public facilities. Thus, the likelihood of hardship financing is generally associated with use of private facilities. However, even poor patients in urban areas do visit private facilities because of the perception that public healthcare is of poor quality [41,42].
Another possible source of high OOPs is that health staff at the primary facilities which are designated to provide care could introduce quasi-formal user charges as a response to insufficient funding by the government for medical and non-medical supplies. Future studies should explore the issue of informal or quasi-formal healthcare payments. A study by Damme et al [43] in Cambodia reported similar findings. Unfortunately, the survey did not ask respondents if they were aware that primary healthcare is free in the public sector.
From a policy perspective, the evidence of vulnerability to hardship forms of coping with unaffordable OOPs highlights the need for policy attention to the ongoing problem of hardship financing in healthcare. Although in this study we have not investigated the health and welfare implications of hardship financing on households, studies have shown how various forms of hardship financing have caused impoverishment [8,44]. Similarly, financial hardships have also been reported to cause patients to take inadequate treatment [45][46][47]. In order to realise the policy goal of universal protection against financial catastrophe of OOPs, there is Hardship financing of out-of-pocket payments in the context of free healthcare a need for not only increased funding for public health services but also more equitable allocation of those resources. A study by Lepine et al [14] demonstrated that healthcare benefits and financial relief from free healthcare in Zambia accrue disproportionately to well-off households. When healthcare resources become scarce, the poorest will bear a disproportionate burden of financial distress of OOPs [48,49]. Finally, this study has a number of limitations. First, household surveys are subject to recall bias in reporting out-pocket expenditure which could affect our calculation of CHE-based hardship financing. Second, self-reported illness is also susceptible to recall bias. Third, we are unable to establish the quality of care received by patients. For example, if patients who reported zero expenditure went without adequate treatment, it means we underestimate the true burden of financial hardship. The fourth and final limitation is that the survey did not ask about the source of borrowing (i.e. whether these were commercial or interest free loans) or the amount borrowed. Local money lenders charge so high interest that the amount to be repaid is much higher than the amount borrowed. Such information could be useful to analyse more fully the level of financing hardship imposed by health payments. These limitations notwithstanding, the findings from this study are consistent with similar studies on financial hardships associated with healthcare visits.

Conclusion
Despite the removal of user fees on primary healthcare in the public sector, this study estimates that up to 11% of Zambian households who experienced an illness still experienced one form of hardship financing or another when coping with out-of-pocket healthcare payments. Furthermore, the disproportionate share of the burden of hardship financing falls on households with very low incomes, female-headed households, and individuals with less education. These findings demonstrate the ongoing policy challenge of reducing financial catastrophe associated with OOPs in Zambia. Finally, the results suggest that improving physical access and quality of care (particularly availability of drugs) in public facilities would likely reduce the incidence of hardship financing especially among the poorest.