Equity of access to maternal health interventions in Brazil and Colombia

Maternal mortality has also received significant attention in Latin American and Caribbean regions, which encompass countries at various levels of development and with different health system structures. Brazil and Colombia are two of the biggest countries in the region that follow divergent healthcare financing and provision models. Structural reforms in Brazil in the late 1980s, enshrined in the Brazilian constitution, aimed at achieving universal healthcare coverage. Nevertheless, chronic underfunding coupled with management weakness, has constrained improvements in the quality of services and access to healthcare services [9]. A set of structural reforms in the Colombian healthcare system initiated in the early 1990s had similar aims, though existing evidence indicates that universal healthcare coverage was not achieved until 2011 [10]. Despite these achievements, a significant part of both the Brazilian and Colombian population has supplementary private health insurance. A report by the World Health Organization (WHO) suggests that supplementary private health insurance represents 20.9% and 11.9% of total healthcare expenditures in Brazil and Colombia, respectively [11]. Bearing in mind these two different contexts, both Brazil and Colombia implemented several policies with a specific objective of increasing coverage and enhancing the quality of maternal health services. The aim of these policies was to improve maternal and neonatal health outcomes (Additional file 1: Table S1).

Maternal health has been a component of the Brazilian health agenda before the initiation of reforms aimed at achieving universal health coverage. However, it was not until the watershed moment of the ‘National Programme for the Humanization of Antenatal - Delivery and Post-Partum Care’ (2000) that maternal health became a political priority. The aim of the programme was to achieve significant access, coverage and quality of key maternal health interventions by anchoring maternal health services to the principles of sexual and reproductive rights and, in turn, position such services as a human right. Despite these ambitious goals, the programme has had limited success [12]. Several other policy initiatives such as the ‘Pact for the Reduction of Maternal and Newborn Mortality’ (2004) and the ‘Stork Network Strategy’ (2011) are evidence of the unfinished agenda in maternal health in Brazil. However, all those programmes and initiatives are struggling with the issues of quality in antenatal, childbirth, and postnatal care; system capacity limitations; and lack of integration among maternal health services [13–15].

In Colombia, several directives issued by the Ministry of Health and Social Protection after the 1993 Health System Reform sought to raise the standard of care for maternal health interventions. However, it was only in 2003 that a comprehensive policy for maternal health was enacted. The 2003 Sexual and Reproductive Health Policy approached maternal health issues in the broader context of women’s lifecycle, recognizing that sexual and reproductive rights were human rights that needed to be upheld in order to protect the life of women, mothers and children. This represented an important change in the paradigm; however, implementation difficulties led the government to propose an emergency plan, ‘Crash Plan’ in 2004; this plan was expected to expedite progress in the reduction of maternal deaths. Despite significant efforts to improve policies focused on maternal health outcomes, those responsible for diagnosis, policy analysis, and policy design often came to the same conclusion: that the stagnation in reductions of maternal deaths, despite high levels of coverage achieved with key maternal health interventions, pointed to issues of quality of care [16, 17].

Despite recent efforts, MMRs in Brazil and Colombia are considered high and lag behind Latin American countries at a similar level of economic development (Mexico and Argentina) [18]. Moreover, the improvements in maternal health fell short of the MDG target 5 (improve maternal health), therefore MMR of the two countries remains unacceptably high [18]. Against this background, our study had a three-fold objective: (i) to analyse the equity of access to four key maternal healthcare interventions: (a) ANC of at least four visits; (b) SBA, which for both Brazil and Colombia encompasses doctors and nurses given there is no professional midwifery; (c) C-section; and (d) PNC; (ii) to compare the equity of access Brazil and Colombia; and (iii) to further study some of the drivers of inequitable access, while paying particular attention to private health insurance in Brazil.

Data sources and definition of healthcare intervention variables

Measures of inequality (concentration index) and decomposition analysis

The CI approach to measure inequalities in health sector variables has several advantages [21, 23, 24]. First, it provides a measure of the variations of inequality across the entire income distribution. Second, the CI is a summary measure that provides a numerical measure of inequality; when this is required, it facilitates inter-temporal and cross-country comparisons of levels of socioeconomic-related inequality. Third, the CI allows the calculation of standard errors to check statistical significance of results derived from survey data with different sample sizes and designs. Finally, the CI lends itself easily to decomposition into the determinants of inequality.

Despite these advantages, the CI approach is associated with certain shortcomings, such as the ‘bounds issue’ for bivariate variables. The CI inter-temporal and cross-country comparability is limited when the health care variable of interest is a bivariate variable. The existing literature argues that concentration indices for bivariate variables are bounded by the means of the variables; that is, equal concentration indices of two countries that have different mean rates of utilization of a given service reflect different levels of inequity in access, because “the mean of the distribution places bounds on the possible values of the concentration index” [25]. Thus, minimum and maximum values of the CI are no longer (− 1, + 1), but (μ − 1 − μ) respectively. One way to approach this issue is to normalize the CI dividing it by 1 − μ [21]. We followed this approach in this study.

Women were ranked according to socioeconomic status of the household in which they lived, and the sample was then divided into quintiles. For each quintile, coverage indicators were calculated. The chosen measure to capture the socioeconomic status of households was a wealth index, based on the principal-components analysis [26]. While the Colombian DHS contains a wealth index already calculated by the survey providers, to make the results comparable between the two countries, we constructed a similar index (based on the same components) for Brazil (details of this exercise, including limitations, are reported in Additional files 2 and 3).

The CI analysis was coupled with a standard decomposition analysis of the socioeconomic-related inequality affecting access. Socioeconomic-related inequality affecting a health variable of interest (captured by the CI) can be expressed as the result of the socioeconomic related inequalities of its determinants. Given the problems associated with the usage of linear models in conducting the decomposition analysis, we relied on a methodology for the decomposition analysis that used a probit model and its ‘partial effects’ [27, 28].

Consequently, we conducted a decomposition analysis of the socioeconomic-related inequality affecting access to SBA and access to PNC in Brazil using a probit model with its marginal effects. The concentration indices for these two interventions were calculated again for microdata [29]. We have selected these two measures for the decomposition analysis, as the existing research on the epidemiology of maternal mortality suggests that most maternal deaths occur during labour, delivery, or the first 24 h postpartum. Discovering the reasons behind the inequities in access to SBA and PNC could provide further evidence concerning why Brazil was unable to make progress on the reduction of maternal deaths.

For the decomposition analysis, the dependent variable (i.e. access to the two key maternal health interventions, SBA and PNC) was explained as a function of need (pregnancy status and demographic characteristics, such as age); certain predisposing factors (whether the pregnancy was planned or wanted, total number of live births, marital status, and race); and enabling resources (the mother’s educational attainment, the mother’s wealth index, having a complementary private health insurance plan, being—or not—a beneficiary of the cash-transfer program ‘Bolsa Familia’ (a conditional cash transfers programme targeted towards the poor and vulnerable, which, inter alia, serve as a proxy for social status), and—as a proxy for community level factors—the region and the location of the residence in a rural or urban area), following the behavioural model of health service use [30]. Deliveries in a private hospital any by a doctor were considered additional determinants of SBA and PNC, respectively.

Colombian DHS did not allow us to combine child and mother datasets, which could have been used to run linear regression models to decompose SBA and PNC. For this reason, we focused on Brazil. All analyses were conducted using STATA (version 13.0). For all calculations, the sample weights related to the survey design were taken into account.

Coverage rates of maternal health interventions and the extent of inequality across interventions and countries

Table 2 provides a summary of coverage rates for each of the four key maternal health interventions in each country. Figure 1 provides a visual representation of mean coverage rates in each country by quintile for each intervention. ANC and SBA displayed average coverage levels above 90% in both countries. PNC coverage mean rates were below 50% in both countries, while the mean C-section rate for the two countries were well above the 15% threshold recommended by WHO [31]. Overall differences in mean coverage rates between Brazil and Colombia were not significant.

Interventions	Brazil						Colombia
	Quintiles						Quintiles
	1	2	3	4	5	Overall mean coverage rate	1	2	3	4	5	Overall mean coverage rate
ANC (4 visits or more)	87.4%	94.1%	81.1%	96.1%	98.7%	91.3%	85.8%	92.1%	93.9%	95.7%	98.3%	92.8%
SBA	81.9%	92.2%	96.1%	95.5%	98.6%	92.9%	82.7%	96.1%	98.7%	99.3%	99.4%	94.3%
C-section	18.7%	32.2%	36.4%	46.2%	62.1%	39.1%	25.4%	34.1%	36.4%	40.6%	43.9%	35.0%
PNC (within 2 days of birth)a							1.1%	1.1%	1.2%	1.3%	1.8%	1.2%
PNC (within 2 months after birth)	11.6%	21.5%	31.2%	39.9%	58.6%	32.5%	18.7%	24.9%	27.9%	33.5%	35.3%	26.9%

SBA skilled birth assistance, PNC postnatal care

^aOnly the Colombian DHS (2010) asked for attendance of postnatal care within 2 days of delivery

Concentration index

When first considering non-normalized CI, we found that inequity levels in both countries were similar. Further, ANC and SBA were the most equitable interventions (both of which had a mean coverage rate above 90%). The most inequitable intervention in Brazil was PNC. Although PNC in Colombia was more equitable than that in Brazil, PNC was the most inequitable maternal health intervention in Colombia. Once normalization was considered, the overall findings changed the dimension of inequity, not the presence or absence of inequity. Greater changes were observed for those interventions with the highest mean levels of coverage in the two countries after normalization. Before normalization, CI results indicated almost no inequity for ANC and SBA, as CI was close to zero.

The most inequitable intervention was SBA in Brazil, followed by SBA in Colombia, ANC in Brazil, and PNC in Brazil. On the contrary, our study found that C-section mean coverage rates are well above the 15% threshold recommended by WHO in the two countries [31]. And even for the poorer groups of the population, both in Brazil and Colombia, the C-section coverage rate was higher than the 15% threshold. Hence, the uneven distribution of caesarean deliveries across the groups, ranked according to their socioeconomic status, does not seem to indicate that there are access issues to medically prescribed caesarean delivery. While at first these results suggest the existence of inequality, it should not be considered as such, given that C-sections are not recommended in most cases.

Decomposition analysis of SBA and PNC in Brazil

Table 4 provides an overview of the contributions to socioeconomic-related inequality in access to SBA and PNC (within 2 months of birth) for each of the variables considered as determinants of access in the probit regression models. Figure 2 provides a visual presentation of the percentage contributions to inequity of the main determinants of access to SBA and PNC.

	SBA					PNC
	Elasticities	Concentration indices	Contributions to C	Contributions to C (%)	P values	Elasticities	Concentration indices	Contributions to C	Contributions to C (%)	P values
Mother’s age	−0.0082	0.0429	−0.0004	−1.70%	0.618	0.5588	0.0422	0.0236	8.59%	0.006
Wanted pregnancy	−0.0004	0.0702	0.0000	−0.13%	0.837	−0.034	0.068	−0.002	− 0.86%	0.295
Number of total children born alive	0.0088	−0.0774	0.0007	3.33%	0.020	−0.1018	−0.0769	0.0078	2.85%	0.114
Married or in union	0.0039	0.0396	0.0002	0.75%	0.397	−0.0293	0.0387	−0.0011	− 0.41%	0.711
Race (white, non-white)	−0.0004	0.1924	−0.0001	−0.39%	0.769	0.0036	0.1912	0.0007	0.25%	0.876
Mother’s years of schooling	0.0051	0.1333	0.0007	3.30%	0.356	0.2170	0.1327	0.0288	10.49%	0.014
Wealth index	0.0193	0.2479	0.0048	23.44%	0.003	0.4136	0.2473	0.1023	37.29%	0.000
Private health insurance plan	−0.0017	0.4892	−0.0008	−4.13%	0.341	0.0797	0.4817	0.0384	14.00%	0.000
Benficiary of cash-transfer program ‘bolsa familia’	0.0026	−0.3469	−0.0009	−4.44%	0.004	−0.0140	− 0.3432	0.0048	1.76%	0.467
Delivery in private hospital	0.0060	0.5611	0.0034	16.40%	0.000
Delivery by a doctor						0.2118	0.0623	0.0132	4.81%	0.035
Region	0.0096	0.2345	0.0023	11.06%	0.001	0.1730	0.2344	0.0405	14.78%	0.000
Living in urban area (1 = urban, 0 = rural)	0.0069	0.0875	0.0006	2.95%	0.043	0.1423	0.0883	0.0126	4.58%	0.013
Concentration Index w/micro data			0.0103					0.2692
		0.0204	50.4%				0.2743	98.12%
Total Contribution				50.44%					98.12%
Total Contribution (only statistically significant values				52.74%					94.53%

“In both the logit and probit models all the regressors are involved in computing the changes in probability”

“for the logit model the rate of change in the probability of an event happening is given by ßjPi (1-Pi), where ßj is the (partial regression) coefficient of the jth regressor. But in evaluating Pi, all the variables included in the analysis are involved”

“In the probit model, the rate of change in the probability is somewhat complicated and is given by ßj f (Zi), where f (Zi) is the density function of the standard normal variable and Zi = ß1 + ß2X2i + ... + ßkXki, that is, the regression model used in the analysis”

SBA skilled birth assistance, PNC postnatal care

In case of SBA, variables that had a statistically significant impact on inequality to access were as follows: wealth inequality (contributed to 23.4% of total inequality in access), inequality in delivery in a private hospital (16.4%), and regional driven inequalities (11%). The inequality in access to the cash-transfer program Bolsa Familia appears to have a negative contribution of − 4.4% to overall inequality, favouring the poor. The number of children (3.3%) and living in an urban area (2.9%) had lower contributions to overall inequality in access to SBA. The sum of the contributions of the included variables amounted to 50.44%, resulting in a considerable large error term; this suggests that omitted variables explain a significant portion of inequalities in access to SBA.

In the case of PNC (within 2 months of birth), statistically significant variables included the following: wealth inequality (37.2%), regional driven inequalities (14.7%), inequality in coverage of private health insurance (14%), inequality in mother’s years of schooling (10.4%), and inequality in mother’s age (8.5%). Inequality in delivery by a doctor (4.8%) and living in an urban area (4.5%) contributed to overall inequality, albeit to a lesser extent. The sum of the contributions by the included variables amounted to 98.1%.

It is evident that the chosen variables were better suited for decomposing the socioeconomic-related inequalities of PNC than those of SBA, as it appears when comparing the different elasticities of the variables with regard to SBA or PNC. However, it is necessary to consider that the chosen non-linear model (the probit model in this case) gives only an approximation to the decomposition of inequality, so that the residual component of the model reflects an ‘estimation error’ and an ‘approximation error’ [32].