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

Background Out-of-pocket (OOP) medical expenses often lead to catastrophic expenditure and impoverishment in low- and middle-income countries. Yet, there has been no systematic examination of which specific diseases and conditions (e.g., tuberculosis, cardiovascular disease) drive medical impoverishment, defined as OOP direct medical costs pushing households into poverty. Methods We used a cost and epidemiological model to propose an assessment of the burden of medical impoverishment in Ethiopia, i.e., the number of households crossing a poverty line due to excessive OOP direct medical expenses. We utilized disease-specific mortality estimates from the Global Burden of Disease study, epidemiological and cost inputs from surveys, and secondary data from the literature to produce a count of poverty cases due to OOP direct medical costs per specific condition. Results In Ethiopia, in 2013, and among 20 leading causes of mortality, we estimated the burden of impoverishment due to OOP direct medical costs to be of about 350,000 poverty cases. The top three causes of medical impoverishment were diarrhea, lower respiratory infections, and road injury, accounting for 75 % of all poverty cases. Conclusions We present a preliminary attempt for the estimation of the burden of medical impoverishment by cause for high mortality conditions. In Ethiopia, medical impoverishment was notably associated with illness occurrence and health services utilization. Although currently used estimates are sensitive to health services utilization, a systematic breakdown of impoverishment due to OOP direct medical costs by cause can provide important information for the promotion of financial risk protection and equity, and subsequent design of health policies toward universal health coverage, reduction of direct OOP payments, and poverty alleviation. Electronic supplementary material The online version of this article (doi:10.1186/s12916-016-0697-0) contains supplementary material, which is available to authorized users.


Inputs and methods for the estimation of the burden of impoverishment due to
out-of-pocket direct medical costs. Table S1. Detailed derivation of key epidemiological and cost inputs used in the estimation of Ethiopia's burden of impoverishment due to out-of-pocket direct medical costs.

Cause of mortality Explanation
Lower respiratory infections (LRI) • 27% of sick patients seek care [1] including 18.3% seeking outpatient care and 8.7% seeking inpatient care [2]. • Case fatality ratio of 1.1% derived from the ratio of the numbers of annual LRI cases (3,367,561) and annual LRI deaths (37,269) for 0-4 year-olds for Ethiopia in 2010 [3]. • Out-of-pocket cost estimates extracted from [4].

Tuberculosis (TB)
• 62% of sick patients seek care, derived from the case detection rate of all forms of TB for Ethiopia [14]. • Case fatality ratio of 17.0% derived from the ratio of TB mortality (30,000+5,600 deaths) and TB incidence (210,000 cases) in 2010 for Ethiopia [14]. • Out-of-pocket cost derived from: provider cost of US$2006 116.5 [15]
• Case fatality ratio of 6.2% derived from 760,000 people living with HIV in 2012 and 47,000 HIV-related deaths in 2013 in Ethiopia [25]. • Out-of-pocket cost derived from: (i) [26] with provider cost of US$2004 265 of which 70% was for antiretroviral (ARV) drugs so non-ARV drug component of US$2004 79.5 inflated to US$2013 98 using U.S. CPI; (ii) [27] with provider cost of US$2010 186 of which $103 was for ARV drugs leading to $110 for ARV and $89 for non-ARV (US$ 2013) using U.S. CPI; (iii) [28] with US$2003 75 and 67% of which was for ARV drugs leading to US$2013 295 for non-ARV drugs. We used $110 for ARV drugs and the average of $295, $98 and $89 for non-ARV drugs leading to a provider cost of $271 (US$ 2013). Finally, we multiplied by 0.02 the share of HIV costs borne out of pocket by households [13] and eventually obtained US$2013 5 for out-ofpocket cost.

Road injury
• Utilization of 50% (based on expert opinion).

Cirrhosis
• Overall utilization of 1.7% taken the same as for breast and cervical cancer (based on expert opinion). • Case fatality ratio extracted from [35].
• Out-of-pocket cost derived as the average of out-of-pocket inpatient cost for ischemic heart disease, stroke, cervical cancer, breast cancer, and diabetes.

Measles
• Utilization and out-of-pocket costs similar to the diarrhea inputs (based on expert opinion). • Case fatality ratio of 3.5% extracted from [36] corresponding to the average of 3% and 4%.
Whooping cough • Utilization and out-of-pocket costs similar to the LRI inputs (based on expert opinion). • Case fatality ratio of 2.8% extracted from [37].

Diabetes mellitus
• Utilization (based on expert opinion).

Mathematical methods used in the analysis
This subsection describes the mathematical methods used for assessing impoverishment due to out-of-pocket (OOP) direct medical costs per condition.
Denote the household income , the total number of households , the condition , and the corresponding total number of cases . Let denote the probability of seeking care for the treatment of the condition conditional on having , and the OOP direct medical costs for the treatment of .
Before condition occurred, we assumed the income was . When there is exposure to the risk of condition , the expected value of income becomes: .
We then estimated the number of poverty cases attributed to OOP direct medical costs due to condition . To do so, we first counted the number of households for which OOP direct medical costs would be incurred, which corresponded to households.
Second, among those households (out of the total ), we counted those Concerning , as there was no income distribution readily available for Ethiopia, we derived a distribution of income drawn from a simulated gamma distribution whose shape and scale parameters were based on gross domestic product per capita (2013US$ 505, the mean of the distribution) and Gini coefficient (0.33) (both available from the World direct medical costs associated with a particular condition among households, we sampled an annual income extracted from the income distribution. Subsequently, we could estimate the number of households (among those households) for whom the size of OOP direct medical costs would push them under the poverty line .
A poverty case was counted when first household income was above the poverty line ( ) and second household income minus OOP direct medical costs was below the poverty line ( ).

Probabilistic sensitivity analysis
We conducted a Monte Carlo probabilistic sensitivity analysis to estimate aggregate uncertainty from key inputs. Parameters were given values using probability distributions (details are given in Table S2).

Univariate sensitivity analyses
We pursued three univariate sensivity analyses, which we describe in detail below.

Impact of health services utilization
We set health services utilization to 75% equally across all twenty conditions (e.g. tuberculosis shifted from 62% to 75% utilization). 75% was closest to 68%, the HIV coverage, which was the highest utilization rate in the base case (table 1 in the main text).
For the conditions that implied inpatient and outpatient care (e.g. diarrhea), we set outpatient care utilization to 75% (e.g. from 30.5% to 75% for diarrhea), and then scaled inpatient care utilization proportionally to the base case ratio between inpatient and outpatient care utilization (e.g. 0.5% divided by 30.5% for diarrhea, led to 75*0.5/30.5 = 1.3% for the scaled utilization of diarrhea inpatient care).

Impact of key inputs varying across income
We varied three key inputs across the income distribution (i.e. by income quintile): (i) disease-specific cases; (ii) health services utilizations; and (iii) OOP direct medical costs.
For disease-specific cases, due to lack of data per condition, we tentatively used the mean of the ratios between the bottom and top wealth quintiles of: under-five mortality (137 and 86 per 1,000 live births, respectively), prevalence of acute respiratory infections (7.6% and 4.0%, respectively), prevalence of fever (18.5% and 15.5%, respectively), and prevalence of diarrhea (15.0% and 11.2%, respectively). These four indicators were extracted from the Demographic and Health Survey [1] and were among the sole diseasespecific indicators we could find available per wealth quintile for Ethiopia. We obtained a mean ratio of 1.51 between poorest and richest, which we distributed evenly across all income quintiles (1.51, 1.38, 1.25, 1.13, 1.00; from the poorest to the richest income quintile).
For health services utilization, due to lack of data per condition, we used the mean ratios between the bottom and top wealth quintiles of: coverage of 3 rd dose of Diphtheria-Tetanus-Pertussis (26.0% and 61.5%, respectively), utilization for treatment of diarrhea (22.4% and 53.0%, respectively), utilization for treatment of fever (16.0% and 40.4%, respectively), and utilization for treatment of acute respiratory infections (15.5% and 61.7%, respectively). These four indicators were extracted from the Demographic and Health Survey [1] and were among the sole utilization-specific indicators we could find available per wealth quintile for Ethiopia. We obtained a mean ratio of 2.81 between richest and poorest, which we distributed evenly across all income quintiles (1.00, 1.45, 1.90, 2.36, 2.81; from the poorest to the richest income quintile).
For the OOP direct medical costs, no information was available across socioeconomic groups. Hence, we tentatively used as a proxy the ratios between private and public care of OOP direct medical costs for the treatments of mild diarrhea, severe diarrhea, mild pneumonia, and severe pneumonia, as empirically estimated by Memirie and colleagues [4]. This led to a mean ratio of 2.94, which we distributed evenly across all income quintiles (1.00, 1.48, 1.97, 2.45, 2.94; from the poorest to the richest income quintile).
As described above (section 1.2), we assigned an annual household income drawn from a simulated gamma distribution of income whose shape and scale parameters were based on the country's gross domestic product per capita and Gini coefficient [11,48,50] . This allowed us to define an annual income for each household impacted by condition . The household income was also used to define the income quintile into which each household belonged.
Subsequently, based on the key inputs described above, per income quintile , and per condition we distributed: disease-specific cases , health services utilizations , and OOP direct medical costs . We also sampled incomes from the income quintile of the simulated gamma distribution of income, and then could estimate for each condition and each income quintile , the number of household poverty cases counting the number of households for which first and second .

Impact of the use of an alternative metric for the estimation of impoverishment due to out-of-pocket direct medical costs
We used an alternative metric for quantifying medical impoverishment and estimated the number of incurred cases of catastrophic health costs, per condition. It corresponded to counting the number of households for which, per condition , the OOP direct medical costs would exceed a given threshold (e.g. 20%) of income . In other words, sampling from the simulated gamma distribution of income, and using the same approach as described above, we counted the number of households (among those households) for which: .
(b) Numbers of cases of catastrophic health costs (with a threshold of 20% of income) due to out-of-pocket direct medical costs and deaths incurred by each of twenty leading causes of mortality in Ethiopia, when the two leading causes of mortality, lower respiratory infections and diarrhea, were omitted. Non-communicable diseases and injuries are indicated in blue; communicable diseases, maternal and neonatal causes are indicated in red. Note: Lri = lower respiratory infections, Dia = diarrhea, St = stroke, TB = tuberculosis, Ihd = ischemic heart disease, Pr = preterm birth complications, Mal = malaria, Rti = road traffic injuries, Neo = neonatal encephalopathy, Men = meningitis, Wc = whooping cough, Msl = measles, Cir = cirrhosis, DM = diabetes mellitus, Copd = chronic obstructive pulmonary disease, Cc = cervical cancer, Bc = breast cancer, Ep = epilepsy, Ast = asthma.