The results of this study proved that the conceptual framework adopted, the CMHC, is a useful tool for the analysis of the effects of social determinants of health in an upper-middle-income country, but with distinct subregional characteristics, under the effects of an inclusive institutional, social and health policies framework. Those results were only possible with the use of the multilevel panel data model with fixed effects nested within-cluster. The method presented allowed the use of the variables provided by the conceptual framework by applying aggregated data that could hardly be used by other methodologies without leading to incorrect estimations. Our models were able to isolate the effects of the variables under study from factors not observed, which are subject to estimation errors due to different degrees of error homogeneity within and between clusters.
Although other studies on infant mortality in Brazil relied on larger datasets for the analysis of the factors impacting infant mortality in Brazil after the implementation of FHS and BFP, our models relied on a longer observational window that allowed us to infer more about specific factors related to infant mortality rates such as the relation between the employment rate and different indicators of infant mortality and between the BFP and the neonatal mortality rate or the threshold of household income according to minimum wage bracket which acts as a protective factor for infant mortality. Furthermore, the use of the methodology of clustered observations at different levels of fixed effects is a low cost-benefit solution, since it relies on a small volume of data when compared to conventional panel data studies.
In this section, we will address each of the variables and their relationship with mortality rates and conduct the analysis in line with CMHC.
Employment and infant mortality
Regarding the health capabilities approach, the findings may be interpreted as a possible effect of employment on the childbearing decision as part of reproductive choice at the household level, which may be a result of a reasonable period of increasing employment that impacted a substantial number of households and therefore neonatal mortality that represents more than 70% of all infant mortality rates. On the other hand, the association between a one-year time lag of employment rate and IMR that represents almost 90% of the total U5MR [41], may be related to a better socioeconomic condition and the household ability, or freedom, to child-caring, feeding, identifying an emergent health issue and searching for best treatments for death prevention.
Few studies have addressed the relation between employment and unemployment and infant mortality in Brazil. In a study using a panel data over populational health and economic downturn in Latin America, Williams et al. [42] found that besides income and inflation, unemployment is also strongly related to under-5 mortality. Although the authors reported that unemployment data in Brazil were not available for the study period (1981 to 2010). In a mixed study with data based on interviews collected in a small town near São Paulo, Ventura et al. [43] concluded that among adults who lived in the same household, the fact of having or not having a job was an important factor in determining the degree of stability and vulnerability of families, which is not in disagreement with the capabilities approach.
Income and infant mortality
Changes in the income signals according to strata and different effects on infant mortality may be related to the association between income and access to health services. The change of signal above two minimum wages stratum suggests that the higher the proportion of families earning up to 2 minimum wages on average (about US$ 525.00 in 2015 at current prices), the higher infant mortality tends to be, except for infant mortality (IMR) and under-five mortality (U5MR) and stratum “A” household income, that are not significantly associated. Therefore, a household income of less than two minimum wages increases the odds of infant death and a slight improvement in household income over 2 minimum wages may have a considerable impact on infant mortality in all age brackets.
Such results suggest that an income threshold above two minimum wages per household provides more freedom to prevent infant deaths.
In a geospatial study on the inequality of infant mortality in Brazil, conducted between 2006 and 2010, Oliveira et al. [24] concluded that low household income, fewer prenatal visits and fewer neonatal intensive care unit beds are correlated, forming a cluster in the North and Northeast macro-regions of the country.
Bolsa Família Program and infant mortality
Our findings are in accordance with the results of other studies. Many authors have highlighted the importance of the BFP in reducing socioeconomic inequalities that hinder the access to primary healthcare provided by the FHS and improving nutritional status with positive effects on infant health and mortality [5, 23, 26].
Nevertheless, some studies show that the interaction between the BFP and the Family Health Program (FHP) is associated with higher average prenatal visits only in the Northeast states [26] and that the BFP has little or no impact on neonatal mortality [6]. Other studies confirmed the impacts of BFP on IMR and U5MR [5, 23]. Those findings are somewhat controversial, considering that another study pointed that increased neonatal mortality in the Northeast macro-region between 2006 and 2010 was linked to lower numbers of prenatal visits and socioeconomic conditions [24].
An aspect to be considered when analyzing those differences is that our study was conducted over a 12-year period after the implementation of the BFP and that those studies were conducted over shorter periods, from 5 to 7 years after the implementation of the BFP [6, 23] or in ecological analysis with predefined periods or over a single period [5, 23]. Our observational window may have identified different effects of BFP coverage over time, with important impacts on neonatal mortality as well, which is consistent with a higher number of prenatal visits.
A study pointed out that in 1990 post-neonatal mortality (infant deaths occurred between the 28th to 364th day of life) represented about 44% of the total U5MR, while in 2015, early neonatal mortality (ENMR: infant deaths occurred between birth and the seventh day of life) was the main component of child mortality in Brazil, representing 41% of total deaths [41]. Thus, the findings suggest that there have been changes regarding the age structure of child deaths in recent years.
Prenatal visits, access to health professionals, and infant mortality
Another point that may explain differences in previous studies regarding the controversy of increased or decreased prenatal visits in the Northeast states and neonatal mortality, is the fact that the number of live births per prenatal visits, as a proxy of the quality of prenatal care provided, is statistically significant for neonatal mortality only. In this sense, there may be a confounding factor involving the results of previous studies, considering that not only the number of prenatal visits, but also the quality of care provided emerges as a major factor in determining neonatal mortality. Our results are in line with a study that stresses the importance of prenatal quality for neonatal mortality [44].
Studies have related neonatal mortality with perinatal causes and, although prenatal care represents a protection factor, mortality is strongly associated with the availability of primary care physicians [22]. In this sense, in our methodological proposal, we considered the overall access to health professionals; physicians, and nurses; as a proxy of the access to comprehensive healthcare. Our findings suggest that in addition to prenatal care, access to health professionals is substantially related to all infant mortality indicators and is better adjusted for IMR and U5MR.
Fertility rate and infant mortality
The fertility rate was significant for infant mortality and under-five mortality rates. As mentioned in the Method section, this covariate was used as a control variable but also indicates the ability of the household to make reproductive choices. Our findings are consistent with studies that have reported declines in fertility rate accompanied by declines in illiteracy rate and socioeconomic improvements, all related to declines in under-five mortality rates [21, 23, 26, 44].
In fact, according to the IBGE, the fertility rate in Brazil is undergoing a continuous decreasing trend as in 2000 the country recorded 2.39 live births by women at reproductive age, while in 2015 this coefficient decreased to 1.72 [45].
Educational attainment and infant mortality
The relation between our dependent variables and educational attainment was statistically significant in all estimations. Educational attainment in our models was applied to evaluate the capacity of households to convert capabilities into functionings.
As part of a criticism formulated by Tengland [18] regarding what the author interprets as a political liberalism conception of health capabilities approach proposed by Sen [16] and Nussbaum [14], job seeking, reproductive health and reproductive choices, as well as education capabilities are part of health as a holistic multi-dimensional phenomenon. Thus, measuring health functioning (infant mortality for instance) is the same as measuring education attainment or employment.
This statement seems to take the capabilities approach to an extreme, however, when exploring the individual's capabilities, Tengland puts the development of competences as depending on a basic degree of education and special training. In this line, Nussbaum states that as fundamental capabilities, every individual, at least when he or she comes of age, has to be equipped with a decent degree of health and primary and secondary education. In addition, Tengland also stresses that capabilities, in fact, the actualization of capabilities into functionings, is not an excluding or concurrent process. It is possible, and in fact, desirable, for one to actualize multiple capabilities simultaneously, although in some cases some capabilities are not turned into functionings. Thus, it is logical to expect that among households, the capabilities approach suggests that there might be differences concerning their motivation for educational attainment, family planning, and job-seeking as outcomes of the interaction of the dimensions proposed by Ruger, given individual internal characteristics.
This reasoning may also be supported by an apparent contradiction in Nussbaum's statement regarding her conception of capabilities as plural elements of the quality of life of individuals. Nussbaum stresses that capabilities are qualitatively distinct, such as integrity and bodily health, or education, among other aspects, considered as indivisible and not reducible to a simple metric without distortion. Probably in this statement, Nussbaum was referring to empirical studies aiming to synthesize well-being and happiness in a single scale. In this regard, Anand conducted a study aiming to test the operationalization of variables according to the capabilities approach proposed by Nussbaum based on secondary data from the British Household Panel Survey. The study found evidence suggesting that a wide range of capabilities had a statistically significant association with well-being. The study relied on secondary data sources and subjective well-being concepts according to a scale of life satisfaction from 1 to 7 [46].
Continuing with Nussbaum, the key question to ask when comparing societies is “what is each person capable to do and to be in terms of opportunities?” This reasoning is in line with Tengland's vision of capabilities as a holistic multi-dimensional phenomenon. Thus, capabilities may be assessed by health and educational attainment at an aggregated level if one intends to assess or to propose public policies aiming to promote a fruitful environment that allows a constant actualization of capabilities. The list of basic capabilities proposed by Nussbaum implies the idea of what the State can do in this sense [14, 18].
Water supply, sewage services, and infant mortality
Although the lack of a statistically significant association between safe water supply and sewage services and all infant mortality indicators is a controversial result in relation to other studies that found an association between those factors, it is worth noting that most of those studies relied on interpolated data from long periods after the 2000 Census or data covering only part of the municipalities of the country, which may not reflect the evolution of socioeconomic data linked to the sanitary infrastructure.
A study of Guanais based on data of 4853 municipalities of Brazil between 1998 and 2010 found a strong negative association between water supply coverage and infant mortality [26]. That study applied interpolated techniques to obtain data between the National Census from 2001 to 2009, excluding from the analysis a considerable number of municipalities located in rural areas in the North macro-region of the country (n = 449) due to the unavailability of socioeconomic data until 2003 [30]. Rasella, in a longitudinal panel data study on the effects of FHS and BFP, used inadequate sanitation coverage as a control variable that encompasses safe water supply and sewage services together [23]. The study was conducted over a three-year period, 2006–2009, with 2853 municipalities of a total 5565, based on interpolated socioeconomic data from the 2000 National Census.
Another possibility for explaining the lack of association between sanitation and water supply and infant mortality indicators may be related to the fact that structural socioeconomic variables change slowly over time in relation to other socioeconomic variables and the changes regarding water supply and sewage services were probably not captured by our model.
We must emphasize that our data related to safe water supply and sewage services coverages are somewhat redundant, as the sum of coverage rates of urban and rural areas exceeds 100%, which suggests that there must exist redundancies in water supply and sewerage systems and/or overreporting errors that must be considered when interpreting our results.
The health capabilities approach and infant mortality in Brazil
In specific contexts, such as extreme poverty, Sen suggests that one should consider a relatively limited number of central and important functionings and corresponding basic capabilities (such as the ability to be well-nourished and sheltered or escaping from premature death). In other contexts, the number of capabilities and functionings could be much higher and more diversified. One must choose what are the relevant functionings in a specific context and what might be considered negligible [17].
The list of basic capabilities proposed by Nussbaum is far longer than those of our study proposal. Although Tengland reduced this list to a central set of capabilities [18] our methodological proposal restricted those possibilities to the factors interacting with the social and health policies recently implemented in Brazil, to the specificities of the country and data availability.
Tengland’s perspective of health regarding the definition of health capabilities is dynamic in the sense that although some capabilities are impossible for one to convert into functionings, health capabilities must be actualized or turned into functionings. This perspective of health capabilities is more in line with the concept of capability proposed by Nussbaum, for whom capabilities, other than functionings, may be listed as State priorities of actions that may allow individuals to exercise the freedom to choose the life they want to live. On the other hand, for Sen, there is no room for one to define capabilities priorities, and goals. Taking this into account, although Ruger did not make any reference to Nussbaum’s conception of capabilities, the CMHC, and its perspective of health capability as the result of State paternalism and agency seems to be more aligned with a pragmatic perspective of the capabilities approach, having functionings as the "results" to be measured [18].
This study has several strengths. To our knowledge, this is the first study to use the CMHC to study the determinants of infant mortality. The second strength of this study lies in a multi-level data panel with fixed effects nested within-cluster to use aggregate data nested within macro-regions to study the determinants of infant mortality. Third, this is the first time the employment rate is used as an independent variable associated with infant mortality in Brazil. Finally, this study used the longest observational period after the implementation of the BFP in 2003. Despite those forces, this study also has limitations that must be considered when interpreting the results. First, in the application of the CMHC we did not find a variable to control individual characteristics. After conducting a scoping review, we concluded that, in Brazil, regional inequalities can lead to contradictory results such as advanced maternal age as a protective factor for low birth weight. In some regions of the country maternal age conflicts with the level of maternal education. This creates a confounding factor in specific regions of the country. It is a phenomenon known as the “Low Birth Weight Paradox” [47]. On the other hand, we have not found, nor have we been successful in developing a proxy that could be used as a control for the internal dimension in the CMHC. As we have mentioned in the “Independent Variables” section we applied interpolation techniques for specific periods of our independent variables. Although those are minor interpolations and estimations, they must be taken into account when interpreting our findings. Also, in some states, the total coverage of safe water and sewage services exceeds 100%, suggesting the existence of overreporting or more than one contract per household, which should be considered when interpreting the results. Finally, there are some limitations regarding the external validity of our study due to the specificities of Brazil. Although the World Bank classifies Brazil as an upper-middle-income country [48], it ranks 73rd in terms of per capita income. The country has a population of about 212 million living in the world’s fifth-biggest territory and is the only Portuguese-speaking country in the Americas. Although in terms of absolute GDP value, Brazil ranked eighth in the world in 2018, economic inequalities in the country have reached extreme levels and are one of the worst in the world.