Risk factors for neonatal mortality among children with low birth weight

OBJECTIVE: To analyze the risk factors associated with neonatal deaths among children with low birth weight. METHODS: A cohort study was carried out on live births weighing between 500 g and 2,499 g from single pregnancies without anencephaly in Recife (Northeastern Brazil) between 2001 and 2003. Data on 5,687 live births and 499 neonatal deaths obtained from the Live Birth Information System and the Mortality Information System were integrated through the linkage technique. Using a hierarchical model, variables from the distal level (socioeconomic factors), intermediate level (healthcare factors) and proximal level (biological factors) were subjected to univariate analysis and multivariate logistic regression. RESULTS: After adjusting the variables through multivariate logistic regression, the factors from the distal level that remained signifi cantly associated with neonatal death were: cohabitation by the parents, number of live births and type of maternity hospital. At the intermediate level, the factors were: number of prenatal consultations, complexity of the maternity hospital and type of delivery. At the proximal level, the factors were: sex, gestational age, birth weight, Apgar score and presence of congenital malformation. CONCLUSIONS: The main factors associated with neonatal mortality among low weight live births are related to prenatal and postnatal care. Such factors are reducible through health sector actions. DESCRIPTORS: Infant, Low Birth Weight. Neonatal Mortality (Public Health). Risk Factors. Socioeconomic Factors. Cohort Studies.


INTRODUCTION
Birth weight of less than 2,500 g is recognized as the most infl uential factor in determining neonatal morbidity-mortality. 16 Low birth weight stems from prematurity and/or delayed intrauterine growth and is associated with around four million neonatal deaths around the world every year, mostly in developing countries. 12onatal mortality results from a complex chain of biological, socioeconomic and healthcare-related determinants.Over recent years, several authors have analyzed the role of these factors according to hierarchical determinant models in which each grouping of factors presents interference with the others, thus allowing separate and collective understanding of their importance. 10,14,15 Brazil, the Mortality Information System (Sistema de Informação sobre Mortalidade, SIM) and the Live Birth Information System (Sistema de Informações sobre Nascidos Vivos, Sinasc) have been greatly used to study neonatal mortality a Sistema de Monitoramento dos Indicadores de Mortalidade Infantil [home page on the internet].Brasília: Fundação Oswaldo Cruz / Ministério da Saúde; 2006 [cited 2006 Jan 12].Available from: http://www.monitorimi.cict.fiocruz.br/fontes.htm#and, especially, its determinants.6,17 In municipalities in which SIM and Sinasc have adequate coverage and good data quality, important characteristics of neonatal mortality have been observed.This has been seen particularly when these databases are integrated, starting with a variable that is common to the two systems and unambiguous for each case ("key variable"): the Live Birth Declaration (LBD) number.This has been widely used in population-based mortality studies, at low cost.1,13 The aim of the present study was to analyze associations between variables coming from Sinasc and neonatal mortality, among low-weight live births.

METHODS
A birth cohort study composed of live births weighing 500 to 2499 g, born to mothers living in Recife, Pernambuco, between January 1, 2001, and December 31, 2003, was conducted.Live births from multiple pregnancies and cases of anencephaly were excluded.Over this period, there were 73,854 births in the city of Recife, of which 5,670 live births were included within the selection criteria.Information on neonatal survival or death was gathered until January 27, 2004.
Data from Sinasc and SIM were gathered from the Municipal Health Department of Recife.These systems present information that is considered "satisfactory" according to the infant mortality indicator monitoring system (MonitorIMI) that has been put into operation by the Oswaldo Cruz Foundation with support from the Ministry of Health.a The 5,670 live births and 537 neonatal deaths constituted two separate databases.As a fi rst step, the EpiInfo 6.04d software was used to integrate these databases deterministically, using the LBD number as the "key".However, the LBDs corresponding to 55 deaths could not be located in the live birth database, and these were investigated in the state Sinasc database and in the hospitals where they occurred.
Initially, the state Sinasc database was consulted.From this, 35 LBDs relating to these 55 deaths were located.According to the LBDs for these children, fi ve were twin live births, two had birth weights of less than 500g and 28 were births to mothers who did not live in Recife.Since these 35 cases did not fulfi ll the selection criteria adopted for participation in the cohort, they were excluded.
Among the remaining 20 deaths, after investigation in the hospitals where these births and deaths occurred, it was found that three were fetal deaths and the others had not had LBDs issued.These were cases of live births with weights close to 500g in which the infants died within the fi rst hours of life.For these 17 children, LBDs were constructed on the basis of data in the medical records and delivery room books.Thus, these cases were included in the database of live births.The fi nal population forming the study cohort consisted of 5687 live births, of which 499 died within the neonatal period and 5188 survived.These two databases were then correlated to form the fi nal database that was used in this study.
The independent variables relating to the exposure were hierarchically ranked at three determinant levels: distal, intermediate and proximal.The positioning of the variables followed a previously established order based on a model that described the logical or theoretical relationships between the variables with regard to determining the outcome.Thus, at the distal level, the following socioeconomic variables were selected: To construct this variable, all live births with an Apgar of less than 7 at the fi fth minute were considered to present hypoxia and were classifi ed according to the Apgar score at the fi rst minute as severe (between 0 and 3), moderate (between 4 and 6) and mild (7 or more); (f) presence of congenital malfunction.
Some variables available in the Sinasc database, such as the mother's occupation, were not studied because the percentage of unknown information exceeded 5%.The variables that were not obtained directly from the original live birth database (living conditions and poverty density in the district of residence, type and complexity of the hospital where the birth took place, and evolution of the Apgar score) were constructed and added to the database.Aggregation of the levels of the variables was done while taking conceptual notions (epidemiological and clinical) into consideration.
The statistical analyses were performed using the SPSS statistical package, version 12. Firstly, univariate analysis was performed between the independent variables at each level of determination and the neonatal deaths (dependent variable), using logistic regression.
Although this was a cohort study, it was decided to present the results using odds ratios (ORs) as the association measurement, with their respective 95% confi dence intervals, given that logistic regression would be used in the multivariate model.This measurement has the disadvantage, compared with relative risk, of overestimating the association when the outcome studied presents a high frequency in relation to the exposure.The reference category was defi ned as the one that would theoretically present least expected risk of neonatal death.
To adjust the ORs, a hierarchical approach to entering the variables was used, in conformity with the three determinant levels defi ned previously for multivariate logistic regression, using the "enter" method.The variables that presented statistical signifi cance at p-values ≤ 0.20 in the univariate analyses were previously selected for regression analysis, except for the variable of evolution of the Apgar score from the fi rst to the fi fth minute of life, since this was constructed from two variables that were already present (Apgar scores at the fi rst and fi fth minutes of life).Although the mother's age and the poverty density presented p-values > 0.20, they were included in the model because they were considered to be exposure variables of interest.
The strategy used for adjusting the ORs used the following procedures: fi rstly, the variables at the distal level (the mother's schooling level, conjugal situation and age; the number of live births; poverty density; and type of hospital where the birth took place) were included in a regression model in order to make adjustments between them.The variables that presented p-values ≤ 0.20 were kept in the model, even if they happened to lose statistical signifi cance through the inclusion of variables from lower levels.This same strategy was maintained with the introduction of variables from the intermediate level (number of prenatal consultations, complexity of the hospital where the birth took place and type of delivery) and from the proximal level (sex, gestational age, birth weight, Apgar at the fi rst and fi fth minutes of life and occurrences of congenital malformations), which were adjusted between each other and using the variables from the distal and intermediate levels.
This study was approved by the Research Ethics Committee of Hospital Oswaldo Cruz, Universidade de Pernambuco.

RESULTS
With regard to the distal determinants of neonatal mortality among children with low birth weight (Table 1), univariate analysis showed that three of the fi ve variables studied (mother's schooling, mother's conjugal situation and type of hospital where the birth took place) presented statistically signifi cant associations, with ORs between 1.39 and 1.82.
With regard to the intermediate determinants (Table 2), all the variables subjected to univariate analysis presented statistically signifi cant ORs, with values between 0.67 (birth in a hospital without a neonatal intensive care unit) and 5.71 (number of prenatal consultations between 0 and 3).In relation to prenatal consultations, OR was seen to decrease with increasing numbers of consultations.At this analytical stage, the protective effect of birth in a hospital without a neonatal intensive care unit was observed.
Among the proximal determinants (Table 3), all the variables presented significant associations with neonatal death.Gestational age, birth weight and the variables relating to Apgar presented the highest OR values.Among the premature live births, gestational age ≤ 31 weeks (OR = 96.94)was the exposure factor with the greatest chance of neonatal death.The birth weight category < 1,500 g (OR = 38.73)also presented a signifi cant chance of neonatal mortality, compared with the weight category from 1,500 to 1,999 g (OR = 3.94).The children whose Apgar evolution from the fi rst to the fi fth minute of life revealed severe hypoxia presented a chance of death that was 44 times greater than among those without hypoxia.
Table 4 presents the adjusted ORs at the three levels of determination.At the distal level, out of the six variables included in the logistic regression, the mother's schooling level and age and the density of poverty were excluded from the model because they presented p > 0.20.Mothers with six or more live births presented a chance of neonatal death (OR = 1.91) that was greater than the exposure factor among mothers without a partner (OR = 1.38).At the intermediate level, out of the three variables included, all of them presented signifi cant associations with neonatal death after adjustment between each other and for the variables of the distal level.At this level of determinants, number of prenatal consultations between 0 and 3 (OR = 6.63) was the exposure factor with the greatest chance of death.Finally, at the proximal level, all of the six variables included in the model maintained signifi cant associations with the outcome, after adjustment for the variables of the intermediate and distal levels.Children born with a gestational age ≤ 31 weeks presented the exposure factor with the greatest chance of death (OR = 7.89), followed in decreasing order by those born with weight < 1,500 g (OR = 6.87),Apgar between 0 and 3 in the fi rst minute of life (OR = 6.08), presence of congenital malformation (OR = 5.60), Apgar between 0 and 3 in the fi fth minute (OR = 4.16) and male sex (OR = 1.64).

DISCUSSION
Several studies in Brazil have addressed the determinants of neonatal mortality. 1,14,15,17Because of the large social inequalities that exist in this country, it is The factors associated with neonatal morbidity are complex, interrelated and common to those that contribute toward low birth weight.It is therefore important to use hierarchical models to study these factors. 5,22ong the distal (socioeconomic) determinants analyzed in the present study, the living conditions and poverty density of the district where these newborns lived were not seen to be associated with deaths among low-weight live births.Guimarães et al 8 demonstrated a relationship between neonatal mortality and these variables at the ecological level in Recife.At the individual level, Aquino et al 3 did not fi nd that living conditions in the district were a risk factor for perinatal and infant mortality.This variable was probably not revealed as a risk factor for mortality at the individual level because of the internal heterogeneity presented by the different districts of Recife.Thus, the relationship at the collective level should not be transposed directly to the individual level.
Regarding the relationship between the mother's schooling level and neonatal death, some authors have found that a lower number of years of schooling is a risk factor, 5,9 whereas others have not observed any relationship between the variables. 10,17In the present study, only the mother's schooling was not associated with neonatal death.Memory bias may have led to imprecision in declaring this variable, in relation to others such as the type of hospital where the birth took place, and this infl uence may have caused the absence of statistical signifi cance.
In relation to neonatal death, the presence of a partner for the mother is probably refl ected in the care given to the newborn, because of the fi nancial contribution and psychosocial support.This has a positive effect on the child's survival, as indicated by Monteiro et al 16 in a study conducted in the city of Sao Paulo.In the present study, the absence of a partner for the mother was seen to be a risk factor at the distal level and it remained a risk factor after analysis.
In the present study, the age group from 20 to 34 years was taken as the reference category for analyzing exposure factors.No association between the mother's age and neonatal death was observed in this age group, unlike what was suggested by Horon et al. 10 This fi nding probably results from the fact that the population of this cohort was solely composed of children with low birth weight, a condition under which the mother's age is considered to be a risk factor. 7Thus, among the live births studied, the role of the mother's age was already expressed by means of the low birth weight.
The low-weight newborns whose mothers had already had six or more live births presented greater risk of neonatal death in the univariate and multivariate analyses.This fi nding among children with low birth weight may be explained by the occurrence of shorter intervals between the deliveries. 2 Birth in a hospital belonging to the SUS network was seen to be a risk factor for neonatal mortality among children with low birth weight, at the distal level and after adjustment for the exposure factors of the other determination levels.A similar result was found by Morais Neto & Barros in Goiânia (State of Goiás) for all newborns, regardless of birth weight. 17Along with refl ecting the family's socioeconomic situation, this association suggests that high-risk pregnancies and newborns (such as cases of low birth weight) have limited access to obstetric and neonatal interventions of greater complexity.
An apparent controversy was found in relation to the complexity of the hospital where the birth took place (intermediate determinant): univariate and multivariate analyses showed that absence of a neonatal intensive care unit was a statistically signifi cant protective factor for neonatal death.It is likely that, in hospitals with neonatal intensive care units, the greater occurrence of neonatal deaths is mainly related to the characteristics of the population attended, which consists of high-risk pregnant women and newborns. 19ditionally, with regard to care for pregnant women and newborns, many diffi culties have been reported in Brazil, such as inequality of access, disorganization and fragmentation of the healthcare system, and technicalscientifi c inadequacies in the care provided.Concerning prenatal care, hierarchical provision, guaranteed access and quality of care (and not just the quantity of consultations) are undoubtedly key points for improving the care provided. 22As observed by other authors, 12,18 the present study found that there was a direct relationship between greater numbers of consultations during pregnancy and decreased neonatal mortality.The gestational age of the majority of the children studied (57.7%) was less than 37 weeks, which probably had an infl uence on the lower number of prenatal consultations.Nevertheless, fewer than seven prenatal consultations remained a risk factor after adjustment for the other intermediate-level exposure factors and for factors from the other levels, including gestational age.
Several Brazilian studies have shown that normal delivery is associated with greater neonatal mortality than is observed with cesarean delivery. 6,17,21This association can be attributed to the poor quality of care for vaginal delivery, the high incidence of cesarean delivery in Brazil and the distortions in indicating the delivery method, such that cesarean delivery is performed on low-risk pregnant women and vaginal deliveries on women at high risk of neonatal death. 6,18In the municipality of Goiânia (State of Goiás), the protective effect of cesarean delivery was concentrated on children with low birth weight and on those in private hospitals.The clientele of the private hospitals presented better socioeconomic conditions and had other characteristics that favored survival during the neonatal period. 6The protective effect of surgical delivery may be related to other factors such as type of access to healthcare services and delivery care. 19In the present study, which was restricted to live births with low weight, vaginal delivery was a risk factor for neonatal mortality, even after adjusting for other exposure factors, such as hospital type, gestational age and birth weight range.Other types of study relating to health service evaluation might elucidate these results regarding delivery method more appropriately.
Male sex represented a risk of neonatal mortality that was around 1.6 times greater.This remained significant after adjusting for the other study variables, thus ratifying the results found by other authors. 4,21The protective factor of female sex is attributed to the faster maturation of the lungs and consequent fewer respiratory complications.
Among the proximal determinants studied, it was observed that gestational age and birth weight played roles.In the univariate analysis, the high risk of death among live births weighing less than 1,500 g may have been related to the young gestational age.Even after adjustment by means of logistic regression with variables from all determination levels, weight less than 2000 g and gestational age less than 37 weeks remained important risk factors for neonatal death, in the same way as found in other studies. 14,15 observed by other authors, 13,21,22 the Apgar score was an important risk factor for neonatal mortality.This fi nding stresses the role of organizing obstetric and neonatal care towards minimizing the factors that could lead to perinatal hypoxia and consequent neonatal death.
Children with congenital malformations presented greater incidence of death than did those without this factor.This result was similar to fi ndings from Blumenau, by Santa Helena et al. 19 In the present study, the type of malformation was not characterized, except for anencephaly, since this malformation is in principle incompatible with life, and such infants were excluded from the cohort.It is likely that there would have been greater association between malformations and neonatal death if all the cases of more severe organic dysfunction had been categorized separately.
Although the factors that lead to neonatal death are varied and interact with each other with different intensities, it is possible to use the Sinasc and SIM information systems for studying the infl uence of different variables that determine mortality during the fi rst weeks of life.
The present study provides the public authorities with the elements to address this problem at local level.Among the risk factors identifi ed, the importance of those relating to healthcare for pregnant women and newborns can be seen.These can be reduced through actions within the healthcare sector.It is therefore necessary to look more deeply into prenatal, delivery and newborn care.It is also fundamentally important to evaluate the structure of the perinatal care network and the quality of care offered by the municipality.
30Deaths with unknown variables were excluded: mother's schooling level(19 deaths), conjugal situation (21), mother's age(1), number of live births(5), living conditions (1), poverty density (1), hospital where birth took place (3).b Survivors with unknown variables were excluded: mother's schooling level (95 survivors), conjugal situation (64), mother's age(2), number of live births (30), living conditions (2), poverty density (2), hospital where birth took place(16).importanttoinvestigatelocalriskfactors.In the city of Recife, Sarinho et al20and Aquino et al3analyzed the risk factors for neonatal and perinatal mortality, respectively, using SIM and Sinasc as the data sources.The present study adds to the work produced by those authors by focusing on factors associated with neonatal death among the population of live births with low birth weight, using a hierarchical determination model.

Table 2 .
Intermediate risk factors (unadjusted) for neonatal mortality among children with low birth weight.Recife, Northeast Brazil, 2001-3.Deaths with unknown variables were excluded: number of prenatal consultations (25 deaths), complexity of the hospital where birth took place (3), type of delivery (1).
a b Survivors with unknown variables were excluded: number of prenatal consultations (59 survivors), complexity of the hospital where birth took place(16), type of delivery (3).ICU: intensive care unit

Table 4 .
Hierarchical logistic regression on the determining factors for neonatal mortality among children with low birth weight.Recife, Northeast Brazil, 2001-3.
Distal level: adjusted for mother's schooling level and age and for poverty density Intermediate level: adjusted for the variables of this level and of the distal level Proximal level: adjusted for the variables of this level and of the distal and intermediate levels The variables that did not present statistically signifi cant OR (p < 0.05) are not shown in the Table.These were: at the intermediate level, complexity of the hospital where birth took place (with or without neonatal ICU); and at the proximal level, the mother's age (< 35 years or 35 years and over).