Equitable realization of the right to health in Haiti: how household data inform health seeking behavior and financial risk protection

The primary data used to estimate morbidity, health service utilization and CHE rates were obtained from two surveys on living conditions in Haiti conducted in 2012 and 2013 (Enquête sur les Conditions de Vie des Ménages Après Séisme (ECVMAS I and II)) [15]. The 2012 survey, ECVMAS I, had a sample size of 4930 households that was representative at the department and national levels [15]. The 2013 survey, ECVMAS II, was a panel survey with a sample size of 2282 households (which were part of ECVMAS I, including 4930 households). The replacement rate was 8.86%. ECVMAS II included a new module on health, detailed health expenditures and health seeking behavior, which consisted of 21 questions at the individual level)¹.

This study focuses on outpatient service utilization by provider type, and does not examine inpatient data for the following two reasons: 1) Eight percent of the 2013 ECVMAS II sample measured outpatient services delivered by different types of providers, which allowed for a ‘pro-poor²’ assessment of outpatient services by provider type; 2) There were significant gaps in inpatient observation data in ECVMAS II, which constituted less than 3% of the survey sample.

This study utilized the Automated DEC Poverty Tables (ADePT) software developed by the WB [16] to analyze inequalities between wealth quintiles in outpatient health services utilization.³ ADePT estimates CI following the procedures described by O’Donnell et al. for micro-data [17]. Inequalities are estimated as the transformation of a variable of interest (e.g., outpatient providers) on fractional rank of wealth within a given population. Outpatient services range from − 1 to 1, representing an accurate distribution from pro-poor to pro-rich health care services.

A detailed decomposition of the CI for outpatient health care service utilization by provider type is presented in Table 5. In this analysis, we differentiate between inequities and inequalities as follows: Inequities refers to the disparity in rates due to differences in social, economic or healthcare resources (i.e., Is the distribution of resources fair?). These are unjustifiable determinants (e.g., wealth, education, health insurance status). In contrast, inequalities refers to how rates vary based on justifiable standardizing determinants such as age and gender (i.e., Can the distribution of outpatient services be influenced by demographic characteristics? [16, 18]).

Defined by the SDG Framework (Indicator 3.8.2) CHE refers to expenditures that exceed 10% of overall household expenditures using a methodology applied to monitor UHC financial protection [4]. This indicator measures the rate of financial hardship incurred by OOP payments. This study defines CHE based on household consumption data (as the 2012 ECVMAS I did not collect income data). In addition, expenditure data is preferable to income data since it is more reliable and considered a better proxy of permanent income [19].

The numerator (total health expenditures) was estimated using survey questions on health spending in the consumption module in both ECVMAS I and II (rather than data collected in the health module in the case of ECVMAS II).This was chosen because respondents tend to report higher expenditures when questions about health expenditures are asked in a separate health module [20]. Health expenditures (e.g., consultations, medicines, hospitalizations, lab work, glasses and prosthesis and other medical supplies) were captured if they were incurred during the ‘last episode of illness’. Households were asked to estimate their expenditures over the previous three and twelve-month period. Evidence showed that longer recall periods yielded lower reported average spending on an annualized basis [21, 22]. Taking into account this limitation, we utilized data collected over the previous 3-month period to capture a more accurate measure of health expenditures.

The denominator was determined by the consumption aggregate created to measure poverty in Haiti (comprised of consumption and non-food expenditures, including health expenditures). We estimated the consumption aggregate to include all types of health expenditures, as the initial consumption aggregate only included recurrent health expenditures (i.e., consultations and medicines). Estimated CHE rates of both truncated and non-truncated data identified minimal differences (i.e., less than 0.5 percentage points) during both survey years.

A health seeking behavior dummy variable⁴ was generated using a question that asked individuals whether they consulted a provider when they were sick over the last 3 months. Affirmative answers were coded as ‘1’ and negative answers were coded as ‘0’. This variable reflected health service utilization.

Two regression models were utilized for this study. The first regression model examined the determinants of health service utilization in 2013, using the dummy variable for health services utilization as a dependent variable. The second regression model identified the determinants of CHE in 2012 and 2013. The dependent variable was coded as 1 for CHE-affected households, and 0 for households not affected by CHE.

Based on a literature review of the determinants of CHE and health seeking behavior, the independent variables included geography and several household characteristics (e.g., expenditure quintile, household size, education and gender, and having at least one member older than age 65 or younger than age four). The CHE model also included data on provider type (i.e., public, private, CHWs and traditional healers) and health insurance.

Considering OOP payments for health services in the consumption aggregate implied that poor households with substantial health expenditures could shift to a “rich” consumption quintile, even though such expenditures would actually decrease their overall welfare and not make them richer, consumption/expenditure quintile were estimated net of OOP for health [23].

We used ECVMAS I and II to estimate the rate of CHE. However, only 2013 data was available to examine the determinants of health seeking behavior. The 2012 ECVMAS I did not include data on morbidity and health seeking behavior, while the 2013 ECVMAS II examined these issues.

National health expenditures in this study were estimated in Haitian Gourdes (HTG) and geographically geo-deflated⁵.

While health seeking behavior were estimated at the individual level, the rate of CHE was estimated at the household level. The two regression models used a descriptive analysis to identify health utilization and CHE by consumption quintile. Logistic regression was used to predict determinants of health seeking behavior in 2013.

The following variables were used in model 1:

Seemingly Unrelated Regressions (SUR) models were used to estimate the determinants of CHE based on 2012 and 2013 data from each survey year and by applying different explanatory variables. Compared to Ordinary Least Square (OLS), SUR allows dependent variables to have different sets of independent variables [26, 27]. The SUR method simultaneously estimates the parameters of all equations so that the parameters of each single equation also take into account information provided by the other equation. The relationship between these two equations with different independent variables is that the error terms in the two equations can correlate. As a result, SUR may produce more accurate estimates by combining information on different equations rather than running the models separately while allowing joint testing [28].

Two CHE equations (using 2012 and 2013 data⁶) were utilized and run through the SUREG command in STATA 14 [29]. Both the 2012 and 2013 equations were predicted by socio-economic and demographic variables using similar variables to model 1. The following health system variables only available for 2013 were included in the 2013 equation: health insurance, utilization of public and private health facilities, and utilization of CHWs, traditional healers and other ancillary services. Affirmative answers were coded as 1, negative answers were coded with 0. Joint tests utilizing 2012 and 2013 data were also conducted to assess how changes in socio-economic and demographic variables (e.g., wealth quintile, age and household size) effect CHE rates over time.

In Haiti, the high rate of CHE among the poor could stem from the absence of a pro-poor health system. This finding is clearly illustrated in the CI analysis. Despite a low positive coefficient, public health facilities remain pro-rich (CI of 0.05) and are associated with an increased risk of CHE (OR:3.83; P < 0.001). Inequities in access to health services at public facilities may be driven by public hospitals which have the highest positive CI among public facilities (CI of 0.08). In contrast, CI among public dispensaries is close to 0 (CI of 0.02). Overall, the positive association between public facilities and CHE (in the CHE SUR regression model, Table 7) may be related to OOP payments at points of service (e.g., outpatient user fees and drug-related costs) that all wealth quintiles, including the poor, incur on a continual basis. As shown earlier, outpatient user fees and the cost of medicines has been identified as a main driver of CHE in Haiti (Fig. 2), reflecting similar trends throughout the LAC region and other LICs and LMICs. For example, according to a 2018 study on financial protection looking at the drivers of CHE in LICs and LMICs, medicine costs are driving CHE in Guatemala, Sierra-Leone, Burkina-Faso, and Uganda; while outpatient user fees are a key driver of CHE in the context of outpatient care in Guinea, Bangladesh, and Liberia [23].

Unsurprisingly, private facilities are even more pro-rich than public facilities in Haiti and present an even greater risk to vulnerable populations of incurring CHE. This said, poorest wealth quintile households continue to seek health care services at private facilities. Additional research is needed to better understand why poorest wealth quintile households may be willing to risk accrual of significant personal debt in exchange for accessing privately provided health care services.

Results from the CI analysis in Haiti also mirror findings from other studies that have used the same methodology (e.g., estimating CI among public and private health facilities). A study on equity in health service utilization in Ghana, South Africa and Tanzania showed that both public and private health services were pro-rich [31]. As in Haiti, public health facilities were found to be less pro-rich in the three countries than private facilities. Tanzania had similar results to Haiti in that the CI of public health facilities was close to the equality line, yet remained pro-rich. In contrast, the CI of public health facilities was much higher in Ghana and South-Africa than in Haiti. In a separate study, poor population groups in Afghanistan used public facilities more frequently than wealthy populations, who tended to use private facilities instead [32]. In this same study, the CI of public facilities was negative [− 0.14] and truly pro-poor. In some LICs and LMICs, primary health care (PHC) facilities were found to be more pro-poor than public hospitals [33].

.In comparison to public and private health facilities, health care services provided by CHWs were found to be pro-poor in Haiti, with a negative CI of − 0.22. Households consulting CHWs were 3.5 times less likely to incur CHE. The literature from other LICs shows that services provided through CHWs has helped to expand the availability of health care coverage, while offering financial protections for the poor [34, 35]. Interestingly, seeking care from traditional healers was found to be pro-poor in Haiti (CI of − 0.18), yet was also associated with higher CHE (Table 7). Although this finding was not significant, it points to a concerning trend that the poor may be incurring high OOP payments to access tradition healer services without the benefits and protections of quality control in the delivery of these alternative services. More research is needed to better understand the profile of patients seeking health care services from traditional healers.

The increase of CHE rates between 2012 and 2013 for the poor, and the fact that poor are three time less likely to consult health care services when sick than the rich, suggests a potential explanation behind low UHC tracer coverage across the lowest wealth quintiles (Institut Haïtien de l’Enfance (IHE), ICF International 2018)). Such findings confirm that the national health system poses ever-growing inequities for the poor. The rate of CHE increased nationwide by 22% from 9.43% in 2012 to 11.54% in 2013, compared to an increase of 10% over a 10 year period throughout the LAC region (i.e., from 13.4% in 2000 to 14.8% in 2010 [9]).The rate of CHE increase was most notable among the poorest wealth quintile with an increase of 57% from 11.62% in 2012 to 18.20% in 2013. In contrast, the rate of CHE declined by 53% among the richest wealth quintile from 9.49% to 4.46% during the same period. These results mirror previous research conducted throughout LICs in which poor households were at higher risk of facing CHE than rich households. A 2011 study on the determinants of CHE in 12 Latin American countries found that poor households incurred higher rates of CHE using a 30% threshold of total consumption [36] than did rich households. A 2018 assessment of financial protection conducted by the Global Financing Facility (GFF) in 16 LICs and 8 LMICs found the rate of CHE by income quintile was more concentrated among the poorest groups [23]. A similar finding was observed in a 2017 assessment on CHE in LICs [37]. Continuing this trend, findings from a study on the determinants of CHE in Nigeria in 2015 showed that CHE rates were three times higher among lower income groups than among richer income groups [38]. In Senegal, Séne and Cissé (2015) also used SUR to assess the determinants and magnitude of CHE impact. Predictably, findings showed that the risk of CHE jeopardized household welfare, particularly among the poor [39].

National health accounts may give some insight as to the root causes of deteriorating financial protection for the poor between 2012 and 2013 ( [13]). There was a significant increase in OOP payments for health as a share of current health expenditures, shifting from 31% in 2012 to 40% in 2013 [13]. This increase coincided with a sharp reduction in external assistance, which decreased from 61% to 48% of current health expenditures over the same time-period [13]. In 2010, user fees were exempted across Haiti, but were reinstated in 2013 to compensate for the decline in external donor funding. Indeed, the 2013 Service Provision Assessment (SPA) confirmed that 94% of health facilities collected user fees in 2013 [40]. Additionally, the increase in household expenditures for medicines and medical supplies between 2012 and 2013 may be associated with decreased donor funding for the health sector (which includes disaster relief aid⁸) over these same years, and may have also contributed to increased rates of CHE.

Demographically, Haitian households with older members appear most vulnerable experiencing financial risk. This finding has also been observed in other LICs. For example, in Uganda households with elderly and unemployed family members were more likely to incur CHE [37]. Recognizing the cost of medicines as a key driver of OOP payments, we hypothesize that older populations in this study may have incurred debt due to the costs of medicines needed to treat non-communicable diseases.

While health insurance was positively associated with health service utilization in Haiti, it was also associated with CHE and may therefore not be a viable solution for preventing financial hardship among the nation’s poorest population groups. Similarly, access to health insurance may also push households towards an over-consumption of care without protecting them from financial hardship. Other LICs with health insurance have experienced similar deteriorations in CHE protections [9] [41] [42]. For example, the expansion of health insurance in the Philippines coincided with a worsening of financial protection for the poor because essential drugs were excluded from the national health care benefit package, resulting in a main driver of catastrophic spending [43]. In Haiti, national health insurance policies may also be a key driver of CHE because they do not adequately cover the costs of drugs. Additional research is needed to gain a better understanding of Haiti’s health insurance benefits package and its correlation with CHE among vulnerable populations.