Social inequalities in healthcare utilization during Ecuadorian healthcare reform (2007–2017): a before-and-after cross-sectional study

This paper presents a cross-sectional study based on a secondary analysis of the “Living standards measurement survey” conducted in 2006 and 2014 by the National Institute of Statistics and Census (INEC, Instituto Nacional de Estadísticas y Censos in Spanish) in Ecuador. The sample selection was probabilistic, stratified and proportional to the population size. Respondents older than 20 were included in the present study, corresponding to 14,639 men and 15,815 women in 2006, and 30,115 men and 32,369 women in 2014. A series of similar questions regarding socioeconomic characteristics, self-reported health and healthcare utilization were included in both surveys. The surveys were applied for one year and conducted in Spanish. To reduce potential bias, several actions were implemented; for instance, the local interviewer properly trained made as many visits as necessary to obtain direct information from the people face-to-face in the selected households [17]. Overall, the proportion of missing data was low (less than 1% in the wealth index) or non-existent in the rest of the social variables and outcomes. Therefore, no measure to deal with missing data was applied.

Unmet healthcare needs (UHCN) were used as the healthcare utilization outcome. UHCN, referred to as refraining, is commonly defined as failure to seek healthcare despite an individual’s needs [13, 18]. In addition, self-reported UHCN is widely used as an indicator to evaluate barriers to healthcare utilization [19, 20]. To capture UHCN from the survey, three questions were selected and combined. First, respondents were asked, "In the last month, did you have any illness or injury?" Those who answered affirmative were asked, “Due to that illness or injury, did you visit a doctor, nurse, or traditional healer?” Those who answered no to this second question were categorized as participants with UHCN, except if the reason for refraining was that the illness or injury was mild. This was captured by the answer option “mild illness” to the third question “Why didn’t you visit a doctor, nurse or traditional healer?”.

Residence, ethnicity, education, household wealth, sex and age were used as independent variables that could potentially affect UHCN. Place of residence was defined as living in either an urban or rural area, where parishes with less than 5,000 inhabitants were considered rural. Ethnicity was based on self-identification and divided into two groups: non-Indigenous (including white, mestizos, afro-Ecuadorians, and Montubios) and Indigenous people. Educational level was grouped into four categories: incomplete primary (including illiterate, literate but no formal education and initial education categories); primary; secondary (middle secondary and technical); and higher education (undergraduate and postgraduate). Finally, a household wealth index was calculated based on the household infrastructure and assets (household entrance paving; roof, wall, and floor material; type of house; cooking facilities; cooking fuel; type of toilet; water source; lighting source; landline telephone; home internet; satellite TV; and household waste disposal) using principal component analysis (PCA). The index was generated from a total of 15 categorical variables. Then, factor scores were computed, principal components were extracted, and the first was used to measure economic status [21]. The obtained index was divided into quintiles, the first one representing the richest.

Population characteristics were first summarized with descriptive statistics disaggregated by sex and year. The UHCN by each socioeconomic variable was also stratified by sex and year. To examine the absolute and relative socioeconomic inequalities in UHCN, the slope index of inequality (SII) and the relative index of inequality (RII) were calculated. The advantage of these measures resides in their capacity to summarize the health inequality in one number, which makes comparisons over time more interpretable and weight for the population size in the different social categories [22, 23]. This study used the most socially advantaged category (urban, non-Indigenous, highly educated and high wealth index) as the reference group. A log-binomial regression model was applied with the log link function for calculating the RII and the identity link function for SII between the health outcome and the predictors variables. Ridit scores, corresponding to the average cumulative proportion of the population in each socioeconomic indicator category, were first created and used to estimate the SII and RII in men and women separately for each socioeconomic variable adjusted by age in each of the analysed periods (2006 and 2014). Values of 0 in the case of SII and 1 in RII indicate no inequalities. Conversely, values above 0 and 1 in SII and RII indicate a higher UHCN among the most socially disadvantaged compared to the least socially disadvantaged groups. After merging the two national surveys, a multiplicative interaction term between ridit and time was included in the regression model to estimate changes in the inequalities over time. Sample weights were applied to all analyses, and 95% confidence intervals (95% CI) were calculated to assess statistical significance in the regression analysis. All analyses were conducted using the Stata 15.1 statistical software. Finally, the study protocol was performed following the relevant guidelines.

Tables 2 and 3 display the SII and RII for UHCN in 2006 and 2014 for men and women, respectively. Overall, we observed significant reductions in socioeconomic inequalities over time in absolute and relative terms.

In detail, among men, all disadvantaged social groups had a statistically significantly higher prevalence (absolute and relative) of UHCN than the better off in both years. Statistically significant reductions in absolute socioeconomic inequalities (SII) were observed in all socioeconomic variables over time. The magnitude of the reductions was significantly large and up to a decrease of over 10 percentage points in the case of education (95% CI: − 12.98, − 7.45). Ethnicity (SII: 4.62; 95% CI: 1.17–8.07) and residence (SII: 5.38; 95% CI: 3.47–7.28) showed the lowest values of inequalities in 2014. Similarly, the inequalities in relative terms decreased in all socioeconomic groups. For instance, the RII in education moved from 2.46 in 2006 to 1.90 in 2014, the difference being statistically significant (RII = 0.85; 95% CI: 0.74, 0.97).

Among women, and similar to men, all disadvantaged social groups had a statistically significantly higher prevalence (absolute and relative) of UHCN than the better off in both years. Considerable and significant reductions in absolute differences over time were observed in all social groups, with declines over 12 percentage points, especially in the case of ethnicity (SII difference: − 16.76; 95% CI: − 23.20, − 10.33), wealth (SII difference: − 15.39; 95% CI − 18.21, − 12.56) and education (SII difference: − 12.12; 95% CI: − 14.83, − 9.42). A similar pattern was observed for the relative measurements, with reductions in all socioeconomic variables over the two periods. A striking 37% difference decrease was observed for ethnicity (RII = 0.63; 95% CI: 0.51, 0.79) with important statistically significant reductions for wealth (RII = 0.71; 95% CI: 0.63, 0.80), residence (RII = 0.74; 95% CI: 0.64, 0.85) and education (RII = 0.86; 95% CI: 0.76, 0.96).

This study has strengths and limitations that should be considered when interpreting the results. One main strength is the national representation of the different socioeconomic groups before and during the reform implementation. In addition, the use of two complex measures of health inequalities, both in absolute and relative terms, contributes to providing a broader picture of the impact on social inequalities in healthcare utilization and to assessing the evolution over time. However, the questions in the survey regarding UHCN were related to healthcare utilization for curative but not preventive needs. Besides, one month is possibly a short time to evaluate UHCN, which means that the magnitude of the total utilization could only be partially evaluated. Also, the question in the surveys did not allow separate analysis between formal medical care and traditional healers. In addition, response and recall bias could have been present for some questions due to the nature of the study design, despite the thorough training of the interviewers.

Further, we are aware that the before-after cross-sectional study design does not allow to infer a causal attribution to the reform since other factors could have also influenced the UHCN. However, this was the only possible design to be applied given the available data in the country. Finally, although the health reform was implemented during 2007–2017, the study included only two points in 2006 and 2014, which limits the assessment of trends. Because the reform continued until 2017, it would have been interesting to observe if the reform results were different, but no data were yet available.