Maternal education and age: inequalities in neonatal death

ABSTRACT OBJECTIVE Evaluate the interaction between maternal age and education level in neonatal mortality, as well as investigate the temporal evolution of neonatal mortality in each stratum formed by the combination of these two risk factors. METHODS A nonconcurrent cohort study, resulting from a probabilistic relationship between the Mortality Information System and the Live Birth Information System. To investigate the risk of neonatal death we performed a logistic regression, with an odds ratio estimate for the combined variable of maternal education and age, as well as the evaluation of additive and multiplicative interaction. The neonatal mortality rate time series, according to maternal education and age, was estimated by the Joinpoint Regression program. RESULTS The neonatal mortality rate in the period was 8.09‰ and it was higher in newborns of mothers with low education levels: 12.7‰ (adolescent mothers) and 12.4‰ (mother 35 years old or older). Low level of education, without the age effect, increased the chance of neonatal death by 25% (OR = 1.25, 95%CI 1.14–1.36). The isolated effect of age on neonatal death was higher for adolescent mothers (OR = 1.39, 95%CI 1.33–1.46) than for mothers aged ≥ 35 years (OR = 1.16, 95%CI 1.09–1.23). In the time-trend analysis, no age group of women with low education levels presented a reduction in the neonatal mortality rate for the period, as opposed to women with intermediate or high levels of education, where the reduction was significant, around 4% annually. CONCLUSIONS Two more vulnerable groups – adolescents with low levels of education and older women with low levels of education – were identified in relation to the risk of neonatal death and inequality in reducing the mortality rate.


INTRODUCTION
Social disparities in relation to infant and neonatal mortality have been investigated in public health 7,10,19,23 . Among the social determinants studied, maternal education level is one of the most corroborated. The ability to acquire knowledge on health issues and the optimal use of health services is attributed to high educational level 7 .
In several countries, including Brazil, there has been a progression in educational level in the last few decades for women, accompanied by an improvement in infant indicators 7 . In Brazil, progress in child health indicators has led the country to achieve, before the deadline, the Millennium Development Goal of reducing mortality of children under five years old by 2/3 21 .
Besides the improvement in maternal education, the reduction of inequalities in the country 15 and interventions such as complementary sources of income and expansion of primary care 11 contributed to the achievement of this goal. In the context of the public policies directed to children's health, several programs were implemented in the last 20 years. The priority was the reduction of infant mortality through the organization of perinatal and newborn care and the humanization of prenatal care and birth 21 .
Despite the overall reduction in infant mortality, one of the challenges for Brazil is the reduction of neonatal mortality, which accounts for 2/3 of infant mortality and which had been decreasing at a much slower rate than that of post-neonatal mortality 21,23 .
One of the proposed explanations for the persistence of inequalities is that, as the educational level of the population increases, the group of women with low levels of education is selected, with a concentration of risk factors 10 . Another sociodemographic factor related to social disparities in perinatal health is maternal age, both for the group of women over 35 years old 14 and for the adolescent girls 8 .
Our hypothesis is that an interaction between low education levels and extreme maternal ages would increase the risk of neonatal mortality above what would be expected by the isolated effect of each of these factors. Additionally, our interest is the time trend of the disparities, since few studies have evaluated this trend in low and middle-income countries 16,22 .
This study aimed to evaluate the interaction between maternal age and education in neonatal mortality, as well as to investigate the time trend of neonatal mortality in each stratum formed by the combination of these two risk factors, in a retrospective cohort of births.

METHODS
A nonconcurrent cohort study which used a linked database of birth and neonatal death records of a population of live-born children of women residing in the state of Rio de Janeiro between 2004 and 2010. We chose this period due to the quality of the data, mainly the possibility of relating the bases and the comparative perspective with the policies implemented by the Stork Network in 2011 a .
The integration of the databases of the identified live births records, the Live Birth Information System -SINASC (n = 1,519,095) and the neonatal mortality records from the Mortality Information System -SIM (n = 15,540 deaths with less than 28 days of life) from the period between 2004 and 2010 was performed using the method of probabilistic linkage of records 3 , using the program RecLink 3 . With this technique, we linked the information of 14,827 of the 15,540 neonatal deaths recorded in the SIM, with a sensitivity of the linkage procedure above 95%.
For inclusion in the study' s population, the registry should belong to a single-pregnancy live birth, the mother had to reside in the state of Rio de Janeiro at the time of delivery, with birth weight equal to or greater than 500 g and estimated gestational age equal to or greater than 22 weeks. Multiple deliveries were excluded because of the probability of false positive pairs occurrence in the probabilistic linkage. Records with ignored information were about 2% of the total.

Statistical Analyses
Neonatal mortality rates were estimated along with their respective confidence intervals calculated by the exact method. Following the recommendations by Knol and Vanderweele 13 , a composite exposure variable was created, with six levels combining maternal age and education. The odds ratios (OR) of each level of the composite variable were presented using the group of 20-34-year-olds with education ≥ 4 years as reference 13 . We estimated the OR for neonatal mortality through the logistic regression model, using the Stata software (version 9). The additive interaction measures were calculated in an electronic spreadsheet developed by Andersson et al. 1 , which was powered with estimates of coefficients and covariance matrix. Three measures were calculated: the relative excess risk due to interaction (RERI), the proportion attributable to the interaction (AP), and the synergy index (S), as well as the respective confidence intervals estimated by the delta method 1 . These measures are calculated according to the following formulas: RERI = OR 11 -OR 10 -OR 01 + 1; AP = RERI/OR 11 ; S = [OR 11 -1] / [(OR 10 -1) + (OR 01 -1)]; where 0 in the subscript index indicates the reference category and 1 indicates the exposure category. When RERI = 0, AP = 0 and S = 1, no interaction is indicated.
Next, the effect of educational level in each age group was estimated, as well as the effect of age at each educational level, using logistic models. Finally, the multiplicative interaction was evaluated by means of a logistic model in which each variable was included separately (age and education) and a term of interaction between them was also included.
The time trend of the neonatal mortality rate (NMR) according to the same combined maternal age groups and education was described. This time trend analysis was performed through a joinpoint regression analysis, using the software Joinpoint Regression Program b , testing annual trends. The annual percentage change (APC) is estimated and considered significant when the curve differs from zero using the Monte Carlo Permutation method.
In the analysis, the category of low education levels was valued, with a cut-off point in four years compared to the others. It is understood that an education period lower than four years characterizes functional illiteracy and that this level impacts human development c .
Regarding the ethical aspects, the confidentiality of the data was guaranteed and the project was approved by the Ethics Committee of the Instituto de Estudos de Saúde Coletiva of the Universidade Federal do Rio de Janeiro (Declaration 69965, dated 8 August, 2012 -CAE 05373812.1.0000.5286).
In the analysis of the joint distribution of maternal education and age (Table 1), the most vulnerable groups are evident. For neonatal mortality, both adolescents and older women with low educational levels had an NMR greater than 12 deaths per thousand live births, close to twice the NMR of the 20-34 age group.
In all age strata, children of mothers with less than four years of education presented a greater chance of neonatal death, when compared to the children of mothers with at least four years of education (Table 2, in bold). In the last two columns of Table 2, the OR related to the association between maternal age and the chance of neonatal mortality in each stratum of maternal education are presented. Children of adolescents and women aged 35 years old and older presented a higher chance of mortality when compared to mothers aged between 20 and 34 years, in both strata of maternal education. increased the chance of neonatal death by 25% (OR = 1.25, 95%CI 1.14-1.36). The isolated effect of age on neonatal death was higher for adolescent mothers (OR = 1.39, 95%CI 1.33-1.46) than for mothers aged ≥ 35 years (OR = 1.16, 95%CI 1.09-1.23). Children of mothers in the extremes of age and with low education levels presented approximately 1.7 times greater chance of evolving to neonatal death, when compared to the children of mothers between 20-34 years old and with education ≥ 4 years. However, only for the group of mothers aged ≥ 35 years old, there was a significant interaction between age and education in the occurrence of neonatal death, both in the additive and on the multiplicative scale, although in the latter the statistical significance was borderline (p = 0.06). Table 3 shows the time trend of neonatal mortality in both periods. In adolescents with low education levels, an increase in NMR was observed, although it was not statistically significant. For women of low education levels in other age groups, the downward trend did not reach significance. On the other hand, all women with intermediate or high education levels had a significant reduction in NMR, regardless of age.

DISCUSSION
Despite the drop in neonatal mortality at the national and global levels 21,23 , our study identified that changes in neonatal mortality in the state of Rio de Janeiro were accompanied by social inequality, showing a reduction only for women with intermediate and higher education. Vulnerable groups were formed 5 , combining extremes of age and low education levels.
Although the combined effect of maternal age and low education levels has increased the odds of neonatal death by about 70% at both extremes of maternal age, the significant interaction between maternal age and low education levels was only observed among mothers aged ≥ 35 years old. One possible explanation for this finding is the greater magnitude of the lowest maternal age category on neonatal mortality.
Our results suggest that the simultaneous presence of these two risk factors characterizes population groups that, over the course of their lives, are exposed to multiple risk factors and to the development of various health problems d . Population interventions generally do not reach such groups, which may increase social disparities in health 6 . According to the authors 19 , for neonatal mortality between 1991 and 2000, the total reduction in the highest income quintile was 49%, while in the poorest quintile it was reduced only by 24%.
McKinnon et al. studied, through absolute and relative indicators of inequality, the reduction of the neonatal mortality rate (NMR) in low and middle-income countries 16 . The average reduction in NMR was associated with changes in indicators of educational inequality -moderate for absolute reduction and weak for relative reduction. Victora and Barros commented on these results, questioning which interventions in the health area were most effective for each type of reduction 22 . In countries where the NMR is not very high, the challenge is to make the most sophisticated interventions -which are technology dependent -accessible to all 20,22 . This may be the case in several more developed regions of Brazil, such as the one in this study.
When analyzing two variables usually investigated as distal factors of neonatal death 24 -maternal age and education -we corroborate its association with neonatal death.
As a limitation of this study, the non-inclusion of the skin color variable, still limited in the information systems, especially in SIM 4 , may have prevented a better understanding of the interrelationships between these sociodemographic factors 2 . However, based on other studies that analyzed the same variables, we believe that, even with adjustment for skin color, age and education would continue to have magnitude and statistical significance 17 .  On the other hand, one of the strengths of this study is the use of population data of universal scope (considering the coverage of SIM and SINASC in the country), which avoid selection bias present in adherence-dependent cohorts at follow-up or restricted to care in the Brazilian Unified Health System (SUS).
Future studies should deepen the understanding of the effects of disparities in relation to the proximal determinants (low birth weight, prematurity, and fetal growth) and intermediary determinants, such as differences in access to health services and maternal habits, such as smoking.
Also important in the future will be the analysis of the period from 2011 to 2015, after implementation of the Stork Network, compared with the period analyzed in this study, to evaluate the changes resulting from this strategy in maternal and child care.
It is worth emphasizing that population strategies without a due focus on the most vulnerable groups can paradoxically increase the disparities; on the other hand, the combination of intersectoral activities (e.g., in the field of education) and the social participation of the population itself are crucial in achieving the objectives 18 .
Although our work lacks the scope to identify the best strategy, we believe that, among different policy approaches 6 , broadening access to effective health services for defined vulnerable social groups is a viable health policy and should be on the agenda of governments, in the perspective of women and their babies.