Assessing the burden of medical impoverishment by cause: a systematic breakdown by disease in Ethiopia

High out-of-pocket (OOP) medical expenses can lead to impoverishment in many low- and middle-income countries [1, 2]. Often, individuals and households resort to a variety of strategies to cope with illness-related bills such as borrowing money from peers or selling assets [3]. Without pre-existing financing arrangements, like social insurance, household medical expenditure can often be categorized as “catastrophic” – defined as exceeding a certain fraction (e.g., 10 %) of total household expenditure [2, 4].

Well-designed health policies are critical for the minimization of financial risks tied to OOP direct medical expenses. In this respect, the World Health Organization’s 2010 World Health Report [5] argued for pathways to universal health coverage (UHC) accounting for the reduction of OOP direct payments and financial risk protection as one essential dimension to measure progress on the way to UHC. Likewise, the Sustainable Development Goal for health (SDG3) included a sub-target on achieving “universal health coverage, including financial risk protection and access to quality essential health services” [6]. The prevention of medical impoverishment, defined here as OOP direct medical costs pushing households below the poverty line or leading to catastrophic medical expenditure, is a major objective of health systems and policy instruments such as insurance and public finance [7]. Therefore, analyzing the composition of OOP direct medical payments per cause can help design health systems better at preventing such catastrophic expenditure, which have been estimated to be substantial [8].

In the majority of low- and middle-income countries, there is not necessarily a clear evidence base for the interventions to be covered by public finance [7, 9]. In particular, there has been no systematic examination of which specific diseases and conditions (e.g., tuberculosis, heart disease) drive OOP direct medical costs and subsequent impoverishment. Such a systematic examination could inform the design of an essential health benefits package.

Understanding the underlying composition of medical impoverishment and its disaggregation per condition can prove useful for several reasons. First, it can point to the links between disease and poverty and be used to examine the poverty consequences of epidemiological changes and guide the implementation of tailored poverty-focused policies. In this case, assessing illness-related poverty would account for all the financial implications of disease, including OOP direct medical costs (e.g., medications and consultation fees), transportation costs, and indirect costs (e.g., time and income losses); this is not the intent of this paper. Second, the disaggregation of medical impoverishment, defined as OOP direct medical costs pushing households under the poverty line, can help within the health sector in the design of appropriate policy solutions which provide high financial protection [10]. This approach can highlight the interventions to be financed by the health sector that can prevent catastrophic medical expenditure. Thus, this paper deliberately focuses on OOP direct medical payments only, as it is one essential dimension of UHC [5].

We propose a preliminary analytical approach to systematically assess the burden of impoverishment due to OOP direct medical costs per condition in a given country. We intend to expose, with simplicity, a framework which we subsequently apply to estimate medical impoverishment by cause in one particular country, as an example, Ethiopia.

The 20 causes of mortality were ranked according to the level of medical impoverishment due to OOP direct costs they incurred, allowing us to infer what inputs might have the largest impact based on cause of death, OOP costs, and utilization (Table 2; Fig. 1).

The three top drivers of impoverishment, defined as OOP direct medical costs pushing households below the poverty line, were diarrhea, with approximately 164,000 poverty cases (95 % UR: 95,000–233,900; approximately 47 % (95 % UR: 32–58 %) of all 350,000 cases estimated); lower respiratory infections, with approximately 59,000 cases (35,800–98,400; 17 % (10–27 %) of all cases); and road injuries, with approximately 45,000 cases (24,200–74,800; 13 % (7–20 %) of all cases). The three bottom conditions were breast cancer, with approximately 50 poverty cases (30–100, < 0.1 % of all cases); cervical cancer, with approximately 70 cases (40–120, < 0.1 % of all cases); and cirrhosis with approximately 390 cases (200–600, 0.1 % of all cases) (Table 2). All three bottom conditions had very low utilization rates (2 %).

We observed substantial impoverishment due to OOP direct medical costs for diarrhea and lower respiratory infections (Fig. 1a). The distribution of medical impoverishment by cause became much tighter (road injuries, TB, and ischemic heart disease, then representing 35 %, 11 %, and 10 % of all remaining poverty cases, respectively), when both diarrhea and lower respiratory infections were removed from the analysis. Moreover, we observed that a large number of non-communicable diseases (e.g., ischemic heart disease, asthma, stroke) led to substantial medical impoverishment (Fig. 1b). Some of the important drivers of the medical impoverishment findings seemed to be health services utilization and the number of illness cases, much less so though when, again, diarrhea and lower respiratory infections were removed. On the other hand, OOP direct medical costs seemed less of a driver.

When we unrealistically set health services utilization to 75 % across all conditions, in other words, when access barriers were fully removed, expectedly, the magnitude of impoverishment tied to OOP direct medical costs was substantially augmented (Table 2; Fig. 2). This confirmed that low utilization would impact on the medical impoverishment findings. We observed a reordering of rankings beyond the first three leading causes of medical impoverishment (Table 2; Fig. 2a), and could point to the contribution of reduced access to care in the estimated small numbers of poverty cases for several conditions in the base case (Fig. 2b).

When we varied key inputs (e.g., disease cases, utilization, OOP direct medical costs) across the income distribution, we found an overall reduction in the estimated burden of medical impoverishment and almost all conditions had such a decrease (Fig. 3a); yet, there were only slight changes in rankings (Table 2). Further, we examined the distribution of poverty cases incurred per condition per income quintile (Fig. 3b). We found that no poverty cases would occur among the bottom quintile due to the use of the poverty case metric (i.e., those already under the poverty line were not counted). We observed a distinct distribution of medical impoverishment across income quintiles depending on the condition: the conditions with large OOP direct medical costs (e.g., stroke, ischemic heart disease) had all four quintiles impacted, contrary to the conditions with small OOP direct medical costs (e.g., asthma, COPD).

Finally, when we used an alternative metric of cases of catastrophic health costs (Additional file 1: Section 2.2.3), there was a change in the medical impoverishment rankings. Notably, all income quintiles would now be impacted by medical impoverishment (including those previously under the poverty line).

The purpose of this paper was to present a preliminary attempt to systematically estimate medical impoverishment by cause, and the difficulties inherent in making such assessments.

In our case study of Ethiopia, we found that medical impoverishment was tied to illness cases and health services utilization. Cancers presented high OOP costs, but the corresponding population-level medical impoverishment was, in fact, small because countrywide cases of cancer are low, and utilization difficult (Ethiopia has only one cancer treatment center for the whole country, located in a specialized hospital in the capital city of Addis Ababa). However, when we omitted the two leading causes of mortality (lower respiratory infections and diarrhea), we observed little overlap between disease-specific mortality and impoverishment due to OOP direct medical costs.

This approach to medical impoverishment estimation could, with more local data, enable policymakers to take health-related payment criteria into account when designing the architecture of health systems. Financial protection is central to achieving equity and fairness, and a key component of UHC. While economic evaluations have tended to focus on estimating health improvements per dollar spent [19], newer methods, such as extended cost-effectiveness analysis (ECEA) [11, 20], allow the quantification of financial protection that can be purchased by health policies. A systematic assessment like ours aims at identifying both the proportion of individuals impoverished by OOP direct medical costs and the causes of such impoverishment. This can provide an evidence base for efficient policy and planning. For instance, in 2011 in Ethiopia, the main reason households faced medical impoverishment was expenditure on drugs [21], which suggested the need for appropriate prepayment mechanisms for accessing healthcare. The government began adopting community-based health insurance and developing social health insurance [22]. A systematic approach quantifying the factors leading to medical impoverishment can point to the causes which social programs should address in order to improve financial protection and equity for all.

Nonetheless, our preliminary attempt has several important limitations. First, we only describe the unidirectional impact of illness on OOP medical impoverishment, and do not include the impact of poverty on health. Second, we deliberately focused on OOP direct medical costs. Yet, there are a number of other factors in impoverishment. Transport costs and indirect costs, including loss of earnings, may contribute significantly to poverty. Illness-related poverty may be correlated with chronic conditions (e.g., epilepsy) among productive adults, leading to more wages lost than acute conditions (e.g., diarrhea). The illness and death of wage earners may impoverish a family over a long time; as this suggests, which individual is impacted in the family (child vs. bread-winner) would matter. However, we were prevented from including these considerations due to the lack of reliable data across conditions. Third, our analysis critically points to the importance of health services utilization data and the inadequacy of current data at the national and global levels. Further utilization data for a large number of conditions is urgently needed. Before new data becomes available, an interim solution would be to estimate utilization for selected groups of conditions (e.g., childhood conditions, major infectious diseases, cancers), depending on incidence, service availability, and recorded patient loads for services in countries. The scarcity of data, including the limited data on utilization but also on OOP costs or private providers (e.g., traditional healers), influenced the choice of our parsimonious modeling approach, relying on secondary inputs and focusing on a limited number of conditions. We restricted our study to leading causes of mortality (and could not include causes of morbidity) as a systematic assessment by cause was solely available for mortality [12]. Yet, by examining high mortality conditions only, our results may be biased against chronic conditions for which payments may be repeated across long time periods. Similarly, for some conditions, we relied on macro-level indicators (e.g., sources of funding for health) and on provider costs to derive OOP costs even though OOP costs are not necessarily based on providers’ costs. We did not include heterogeneity (e.g., geographical variation) in the model due to lack of data. For all these reasons, we preferred a simple framework. We acknowledge that many inputs (e.g., mortality, utilization, OOP costs) present substantial uncertainty; our findings should be interpreted with caution and future work should carefully contextualize our analysis. Fourth, we chose to represent disease-specific impoverishment due to OOP direct medical costs in estimating a headcount of poverty cases in spite of the shortcomings of such threshold-based metrics [2]. Alternative metrics include the estimation of cases of catastrophic health costs [4], insurance risk premiums [20], or measures of downside risk arising from uninsured expenditure [23]. We chose the cases of poverty metric for its simplicity, and pursued a sensitivity analysis estimating cases of catastrophic health costs. Further, OOP payments were simply compared to income, not accounting for household borrowing capacity, which could lead to underestimation of coping mechanisms and overestimation of impoverishment [24].

Finally, our approach may mask the importance of the size of OOP costs and the lack of access to care. Some conditions may lead to large numbers of poverty cases because care is available and hence incurs OOP costs (e.g., road injuries), whereas others may lead to few poverty cases as care is rarely available and thus little in OOP costs can be expected (e.g., cancers). A focus on current medical impoverishment and utilization patterns is not able to distinguish between these two situations. One of our sensitivity analyses is set at 75 % utilization for all conditions in order to acknowledge this limitation. Studying medical impoverishment must be undertaken jointly with the contextual examination of available health services. As greater medical impoverishment may be driven by greater service availability, it would be inadvisable to prioritize services on the basis of greater medical impoverishment only. We emphasize herein the potential financial catastrophes of private financing for health services.

Our intentionally simple approach presents a preliminary attempt to systematically consider the catastrophic consequences of OOP direct medical payments in low- and middle-income countries and the difficulties in such an undertaking. We hope it can be a departure point for substantial empirical and analytical improvements. Our framework represents a stepping stone toward highlighting the importance of the financial protection benefits that could be procured by the scale-up of health interventions. Medical impoverishment findings complemented by the effectiveness of benefits packages and social safety nets can help estimate how much medical impoverishment could be averted by the scale-up of health policies.