Factors Associated with Infant Mortality in a Northeastern Brazilian Capital

Abstract Purpose Identify factors associated with infant mortality by a hierarchical model based on socioeconomic, health care, obstetric and biological determinants in a northeastern Brazilian capital. Methods Observational, retrospective cohort study based on secondary data of births and deaths of infants of mothers living in the city of Teresina. Results Based on the distal level of determination of infant mortality, the characteristics that remained statistically significant were maternal age, maternal education and maternal occupation (p < 0.001). In the intermediate level, all variables were statistically significant, particularly the type of pregnancy and delivery (p < 0.001). The gender of the baby was the proximal level feature that had no significant association with the outcome, while the other variables of this level had association (p < 0.001). Conclusions This study evidenced that, in addition to biological factors, socioeconomic status and maternal and child health care are important to determine infant mortality.


INTRODUCTION
The worldwide concern about infant mortality gained visibility with the publication in 2000, of the Millennium Development Goals, in which reduction in infant mortality was highlighted as a goal number 4, which proposes to reduce by two thirds the mortality rate in children below five years old, in the period between 1990 and 2015, in addition to represent a critical point to achieve other goals related to welfare, equity and poverty reduction. Brazil is a signatory of these goals 1,2 .
It is known that, in Brazil, there is a high level of heterogeneity among the regions caused by sharp social and economic and cultural differences, in addition to iniquity in access to health services, and the investigation of infant mortality and its risk factors at the local level is relevant 3 .
The study of risk factors involved in the causal chain of infant mortality is crucial to understand the living conditions of the maternal and child population and to define the priority actions for public policies appropriate for each situation and to develop strategies in order to control and organize the healthcare network, and aiming to reduce infant mortality.
Because of its sensitivity to living and health conditions of the population, infant mortality is determined by factors of different dimensions. However, it is not uncommon to find situations in which social and economic and health care factors are superficially analyzed due to the presence of biological factors 4 .
In this sense, strategies with hierarchical data analysis, which takes into account the temporal precedence of these factors, the potential relationship between them and their relevance to determine the outcome, have been used as alternatives to the analyses based only on statistical associations 5,6 .
This strategy is based on the construction of a conceptual model aiming to establish a relationship of hierarchy between the effects of exposure variables on the outcome studied, assuming that its effect on the occurrence of this event can be direct or through other variables 7,8 .

Data from the official Health Information Systems in Brazil, particularly the Mortality
Information System (Sistema de Informações sobre Mortalidade, SIM) and Live Birth Information System (Sistema de Informação sobre Nascidos Vivos, SINASC), have been widely used to monitor infant mortality in Brazil. However, their use to carry out epidemiological studies depends on the coverage degree of the event in question, in addition to the completeness, reliability and validity of their data 6 . Studies have shown progressive improvement of these systems with increasing level of coverage and improved quality of their data 7,9,10. The possibility to use linkage or another database relationship methodology is highlighted as an important strategy to improve the quality of national healthcare information systems, identifying and recovering information that are present in a system and missing or incomplete in another 11 .
Given the above, this study aims to identify, through linkage, factors associated with infant mortality using a hierarchic model based on social and economic, health care, obstetric and biological determinants, in a northeastern Brazilian capital.

METHODS
This is an observational, retrospective cohort study based on secondary data of births  This step was included in order to find potential deaths that, perhaps, were not included in the digital system SIM. Stillbirths in 2011 were searched in order to identify potential death misclassification.
Ten original DO were not found in this file, and a copy of these originals was requested by the responsible department to the health facilities where the deaths occurred in order to be restored and filed. For the Statements of Live Births (DNV) not found in the system, active searches were carried out at the health facilities where the deaths occurred in the DNV file departments by date of birth.
Determinist linkage was carried out through the variable "DNV number" filled in the DO. Deaths not matching with their DNV, deterministically, were paired in a probabilistic manner using the variables "mother's name", "date of birth of the newborn" and "gender of the newborn". All pairs were reviewed to minimize the presence of bias. Information contained in SINASC were considered gold standard in the case of disagreement or missing information on SIM.
Categorization of independent variables ( year of study completed, a 10-year old at least 2 years of study completed and so on 15 .
Based on this premise the category "low" was considered to all ages as those with no year of study; for mothers aged 17 -19 years, those with 1 -3 years of study; and for mothers aged greater than or equal 20 years, those with 1 -7 years of study. For category "intermediate", the following was included: mothers aged 13 -16 years who had 1 to 3 years of study; mothers aged 17 -19 years who had 4 -7 years of study and mothers aged greater than or equal 20 years who had 8 -11 years of study. The category "high" comprised mothers aged 9 -12 years who had at least 1 year of study; aged 13 -16 years, at least 4 years of study; aged 17 -19 years, at least 8 years of study and mothers aged 20 years and higher who had at least 12 years of study.
The variable "prenatal care" was obtained from the correspondence between the number of prenatal visits and gestational age, and categorized, according to the recommendations of the Ministry of Health, as: "not performed", "inappropriate" and "appropriate". The first prenatal visit should start early in the first trimester and should be regular, with visits performed monthly until the 28th week, every two weeks between the 28 th and 36 th week and, weekly between the 36 th and 41 st week 16 .
The category "not performed" was considered to mothers who did not have any prenatal visit. For category "inappropriate" the following newborns were included: gestational age 28-36 weeks and 1 to 3 prenatal visits; and those with gestational age of 37 weeks or higher and 1 to 6 visits. The category "appropriate" was considered for newborns with gestational age of 27 weeks or older and at least 1 visit performed; between 28 and 36 weeks of gestational age with at least 4 visits; gestational age of 37 weeks or older with at least 7 visits performed.
The variable "congenital malformation" was defined by the presence or absence of malformation recorded on SINASC, along with the analysis of the underlying cause of death recorded on SIM. Some birth defects are not diagnosed at birth and, therefore, they are absent on SINASC, and sometimes can be identified as one of the underlying causes of death The multicollinearity test required for MLR was performed by VIF (Variance Inflation Factor), taking as cut-off point for the diagnosis of multicollinearity one VIF above four; however, the test did not detect multicollinearity between the variables studied. Hosmer and Lemeshow test was used as a quality measure of MLR adjustment 19 .
In the final model, a 5% level of significance was used for analysis, and the variables were considered significant with p-value ≤ 0.05 (p = Wald test), and 95% confidence interval (CI95%). The association measurement used was Odds Ratio adjusted (ORadj), and a reference category was set (ORadj equal to 1) to the lower risk for occurrence of the outcome 20 .

RESULTS
For the distal level for determination of infant mortality, the characteristics that remained statistically significant after applying hierarchical MLR model were maternal age, maternal education and maternal occupation, p<0.001 (Table 3). Mothers aged 20-34 years (ORadj=5.11; CI95% 3.80 -6.46) had the strongest association with infant mortality, followed by teenage mothers, who had higher chance of death (2.82 times compared to mothers aged 35 years or more). The results for maternal education were unexpected, since mothers with low education were the group with stronger association with the outcome (ORadj=1.85; CI95% The baby's gender was a determinant factor at proximal level that did not have significant association with infant death. The other variables of this level had an association with a p-value <0.001. Babies born preterm had a 13.88-time increase in risk of death compared to those born at term. Also, a stronger association with the outcome studied was observed with those born post-term (ORadj=1.08; CI95% 1.01 -1.16) compared to those born at term.
The low birth weight represented greater chance of infant mortality [12.3 times (CI95% 8.76 -16.14) than those who were born weighing 2,500g or more. For Apgar score at 1st minute and 5th minute, group scoring 7 to 10 had the lowest chance to death if compared to the others. Babies identified with congenital malformation were more associated (ORadj=42.33; CI95% 30.11 -54.48) with death before completing the first year of life compared to those who have not been identified with congenital malformation.

DISCUSSION
Among the distal level factors related to social and economic aspects of the mothers, the characteristics that remained statistically significant were maternal age, education and occupation. found women over 34 years were more likely to lead to infant death compared to mothers aged between 20 and 34 years for both periods analyzed, reflecting the increased proportion of Brazilian mothers in this age group and possibly representing the biological disadvantages of pregnancies in mothers with more advanced age 6 .
On the other hand, women who decide to get pregnant later can usually be also those with higher education and social and economic level, stable marital status, as well as have a planned pregnancy with greater attention to prenatal visits, factors that could explain the finding of this study in relation to this variable.
Teenage mothers were almost three times more likely to lead to infant death compared The association between teenage mothers and infant mortality can be discussed, both from a social and biological standpoint. One of the outcomes of poor social and economic status for the adolescent and her family is the teenage pregnancy, a situation that precedes and creates difficulties to have access to health services, as well as adverse perinatal outcomes, which demonstrates the need to strengthen the family planning activities recommended by primary healthcare 7 .
The immaturity of the female reproductive system and the continued growth and de- For the two cohorts studied in Londrina (PR), education was associated with the outcome, but with different behavior between the two periods. In the older cohort, low education had higher chance of death; however, in the most recent cohort, low education was a protective factor for infant mortality 6 . Importantly, in the latter study, education was categorized similarly to the present study, adjusting the years of study to the maternal age, unlike the others in which education was provided only by the amount of years of study completed, regardless the mother's age.
Analyses associating fewer years of study from the mother with more likelihood to infant death assume that mothers with higher education are those belonging to higher social classes, fewer children and greater access to knowledge about child care and appropriate prenatal care, factors that would increase the protection to the outcome 1,7 .

A reduction in Infant Mortality Rate (IMR) is observed among mothers with low education,
and increase or stabilization of this rate among mothers with higher education, which usually have more advanced ages, likely to have cesarean delivery, with a higher ratio of premature births and low birth weight. This scenario depicts the changes in the health care, social and reproductive characteristics of Brazilian women in recent years 6 .
Mothers who have paid work were significantly associated with infant mortality, which was two times higher than stay-at-home mothers. Few studies were found reporting this association. However, in a study conducted in Salvador (BA), maternal occupation took part in the multivariate analysis, reaching statistical significance, with lower probability of infant death in children of self-employed professionals and mid-level technicians compared to children of housewives, students and house maids. Additionally, the chance of death for children of house maids was higher than among children of housewives and students 22 .
In this context, the authors attributed the less likelihood found in women with paid work to their best social and economic conditions. In contrast, house maids were analyzed as a separate group due to their poor working conditions that not always follow the labor laws and, despite receiving payment, their working hours contribute for them to not receive appropriate prenatal care, in addition to the limited time available to stay at home, which affects the quality of maternal care to their children.
Nevertheless, the continuous improvement to the house maids' work conditions are evident and guaranteed by current labor laws in Brazil in a time when poor work conditions of Brazilian women can be found in several types of paid work. Thus, this study was performed employing only two strata for this category: with and without remuneration. As a result, women with paid work were two times more likely to have infant death, which could have been attributed to the stress of modern life, where women take several responsibilities related to their work environment, in addition to carry out their functions within the home environment, factors that could contribute for adverse perinatal outcomes.
To be born in hospitals not convened with SUS showed stronger association with infant mortality compared to deliveries occurred in hospitals providing care by SUS in Teresina, but they were not statistically significant. This contrasts with the study performed in Salvador (BA), which showed significant association between birth in public hospitals and death of children under one year, suggesting the existence of deficiencies in the care of newborns at risk in the population assisted by public healthcare facilities 22 .
In 2011, it was observed that the vast majority of live births of women living in the city under study reported by SINASC occurred in facilities integrating SUS network; however, the number of births in the private network is also high for a total of only 8 beds for neonatal intensive care of the private network in Teresina, while in the public sector convened with SUS, there are 27 beds for this purpose, which is also considered low to meet the demand of the public sector 14 .
Although Teresina is considered a reference in health services for the municipalities and even neighboring states, there is insufficient amount of high complexity beds for the pediatric care both in the public and private sector, particularly for the neonatal period, a main component of IMR in this city.
Obstetrical history was significantly associated with the outcome, and it was more likely to lead to infant death for first-time mothers. Other studies have reported the association between multiparous women and infant mortality, linking high parity to low social and economic level of the family as a determinant of mortality in children under one year 23 . On the other hand, mothers coming from higher social classes have a current tendency to have fewer children, which may also be related to mothers who have paid work and high education, consistent with the current study.
Multiple pregnancies had higher chance to lead to infant death compared to single pregnancies, with statistical significance. The same finding was reported in the analysis conducted in Londrina (PR) in two different periods, in which both had a high probability of death in children under one year born of multiple pregnancies 6 . Some studies excluded this variable from their analysis; however, we chose to keep this subgroup of live births, controlling its effect in the multivariate analysis.
Births of twins are related to increased rate of adverse perinatal outcomes, such as premature birth and low birth weight. This emphasizes the importance of special care for women with this risk profile in the city in question.
The type of delivery performed showed statistical significance in this study showing that children born vaginally have stronger association with infant mortality compared to those born by caesarean section. A study conducted in nine municipalities of the west region of Paraná state and another performed in Londrina also belonging to the same state in southern Brazil, found that caesarean section is a protective factor to infant mortality 6,24 .
An analysis performed on newborns at high risk in neonatal ICU in northeastern Brazil found that cesarean section increases maternal and newborn morbidity and mortality, substantially increasing the spend in health care compared to vaginal delivery. However, operative delivery in high-risk pregnancies is considered an important procedure to reduce perinatal risks, increasing the baby survival 25 .
Failure to perform or inappropriate prenatal care proved to be a determinant of infant mortality in Teresina. Importantly, this variable was achieved by adjusting the number of prenatal visits performed according to gestational age at delivery, as model recommended by the Brazilian Ministry of Health 16 . Reclassification was required since this study included from extremely premature infants to births that occurred after completing 42 weeks of pregnancy, which, therefore, had unequal time opportunities to attend the visits. In Brazil, the coverage and average number of prenatal visits seem to be increasing.
However, in the current studies on infant mortality using SINASC database, the assessment of quality in prenatal care is not available. In this context, there is evidence that poor quality is a more serious problem than simply performing fewer visits 26 .
Male gender had stronger association with infant mortality in this study, but was not statistically significant. Female gender has shown to be a protective factor in some studies attributing the early lung maturation in females during the neonatal period with a consequent decrease in the incidence of neonatal respiratory problems, resulting in a greater number of hospital days in the first year of life in males 6,24 . The loss of effect of this variable in the current context in Teresina may result from the expansion of perinatal care, with the hierarchical organization and regionalization of health services and greater access to neonatal technologies, such as intensive care and use of corticosteroids and surfactant in preterm infants.
The length of gestation showed a strong association between preterm birth and infant mortality; as well as births occurring after 41 weeks, to a lesser extent, but yet significant.
Several authors found a higher likelihood of infant death with premature births, which was statistically significant in several cities in Brazil, of all regions in diverse periods 6,7,22,23,24,27 .
Preterm birth can be not only triggered by biological factors, but also by several social factors with the change of women's lifestyle in recent decades, such as the use of alcohol and tobacco, social stress, occupational stress and poor diet, among others.
On the other hand, two studies carried out in the state of Rio Grande do Sul, in Porto Alegre (RS) and Passo Fundo (RS), found no significant association between prematurity and infant mortality, which may be related to the level of social and economic development, as well as greater access to human and technological resources for child health in that region 1,7 .
Study performed on trends in infant mortality in Porto Alegre (RS), found that the considerable decline in IMR in that city is a phenomenon that may reflect the process of poverty reduction in Porto Alegre, which in 2009, reached the lowest poverty rate among the six main metropolitan regions in Brazil. This demonstrates that Porto Alegre is ahead in the demographic and epidemiological transition phenomenon, trending to reduce IMR, neonatal and post-natal and the number of live births 28 .
Post-term infants also showed significant association with the outcome under study.
The Brazilian Ministry of Health has established that, between the 36th and 41st week of pregnancy, prenatal visits should occur weekly. This increased frequency of visits in late pregnancy aims to assess the perinatal risk and clinical and obstetric complications, which are more common in this trimester. Placental function reaches its fullness around the 36th week, declining thereafter. Senile placenta has calcifications and other changes that are responsible for decreased nutrition and oxygen supply to the fetus and, therefore, is associated with increased perinatal morbidity and mortality 16 . This shows the importance that should be given to these births. Few studies on risk factors associated with infant mortality refer to those prolonged pregnancies.
Low birth weight had a strong and statistically significant association with infant death, consistent with the vast majority of studies addressing this topic 1,6,7,22,23,24,27 .
The risk factors described in the literature that have a greater association with infant mortality are low birth weight and prematurity, and they should not be studied as isolated risk factors, but rather associated with biological and social and economic characteristics of the mother, which are also considered determinants for death in children under one year 24 .
Apgar score at the 1st and 5th minutes is a simple and useful indicator used to measure the vitality of the live birth and to assess responses to resuscitation maneuvers, and it is an important risk indicator for perinatal morbidity and mortality 6,7 .
The findings of the study show that the lowest scores in this index, both at the 1st and  In places where there is decreased infant mortality from preventable causes and where there is improvement in the quality of prenatal care, congenital anomalies become the leading cause of infant mortality. A few teratogens affecting Brazilian population have been described, highlighting rubella, acute toxoplasmosis among the infectious diseases in pregnancy, drugs such as thalidomide, radiation therapy and chemotherapy, when their use cannot be prevented during pregnancy; and genetic factors. It is also known that pregnant women aged over 40 years are more likely to have malformed children, compared to younger age groups 29 .
Keeping prenatal care under conditions to identify potential teratogenic risks is another potential goal to be met, by becoming available examination system capable to identify agents with potential teratogenic risk; and more complex and specific tests, when early identification of birth defects is required. Another factor that could reduce mortality due to congenital malformations is structuring the service network to detect/provide clinical genetic counseling for SUS, in addition to primary care in the identification of families at risk, performing preventive diagnosis and preventing the birth of malformed children 30 .
In conclusion, it was evidenced that, in addition to biological factors in determining infant mortality, social and economic status and healthcare to the pregnant women and newborn are also important, since they are factors susceptible of change guaranteed by public policies of quality, through increased investment and qualification of health care for prenatal care, childbirth and throughout the neonatal period at local level.