Factors associated with low birth weight at term: a population-based linkage study of the 100 million Brazilian cohort

This study employed data from the 100 Million Brazilian Cohort, a database constructed by the Center for Data and Knowledge Integration for Health (Centro de Integração de Dados e Conhecimentos para Saúde-CIDACS), affiliated with the Oswaldo Cruz Foundation (FIOCRUZ) [24]. This is a retrospective and dynamic cohort. The cohort database contains records of 114,001,661 low-income individuals (40,542,929 families) eligible for social assistance programs via the Unified Registry for Social Programs (CadÚnico), who were registered between between 2001 and 2015. Socioeconomic data from the 100 Million Brazilian Cohort were linked to data contained in the National System of Information on Live Births (SINASC), considering the period of Jan 1, 2001 to Dec 31, 2015.

Cohort data were linked to the live birth registry from SINASC according to similarity using the CIDACS Record Linkage algorithm [25]. This novel record linkage tool considers the following attributes in its matching process: mother’s name or newborn’s name, mother’s municipality of residence at time of registry/delivery, newborn date of birth and/or mother’s age. In the current linkage process, the number of linked records was 24,695,618 (55.51%) and the estimated accuracy was over 90%/year.

The study population included live births of women aged 14–49 years who were registered in the 100 Million Brazilian Cohort between 2001 and 2015 (Fig. 1). We included only the most recent live birth reported for each woman, and excluded preterm (< 37 gestational weeks) and post-term births (42 gestational weeks or more). Multiple births and newborns with congenital anomalies were excluded in an effort to avoid bias, as these conditions are known to be strongly associated with low birth weight [13, 16, 26, 27].

Descriptions for the variables of interest are detailed in Table 1. The dependent variable was low birth weight, defined as less than 2500 g, in term births (37–41 completed weeks) [28]. The following covariates were considered: i. socioeconomic characteristics (marital status, self-reported race/ethnicity, maternal schooling, household conditions, urban/rural living environment and geographic region of residence); ii. prenatal assistance (number of visits); iii. maternal- and newborn-related variables (maternal age at birth, newborn sex and birth order).

Maternal and live birth characteristics were summarized using frequency distributions. We also calculated the percentage of TLBW among all those born with LBW (6.7%), prior to excluding preterm and post-term births. Multivariate logistic regression was conducted to investigate the factors associated with TLBW. A conceptual hierarchy-based approach (Fig. 2) was employed to introduce, in subsequent adjusted models, the variables contained in the datasets, considering covariates deemed relevant and plausible in the literature [7, 9‐18]. The TLBW variables were grouped into three blocks representing distal, intermediate and proximal determinants, i.e. socioeconomic characteristics, use of the health services (prenatal care, defined by number of visits) and maternal and newborn characteristics, respectively [7, 29‐31].

The initial model was adjusted for the distal factors. In the second model, all variables contained in the previous model were maintained, with the inclusion of the number of prenatal visits (intermediate factor). The final model included, in addition to the variables contained in the two previous models, the mother’s age at the time of delivery, birth order/parity and sex of the newborn. Odds ratio (OR) values and respective confidence intervals (95% CI) were calculated for each variable maintained in the final model. Data analysis was performed using Stata version 15.1 (Stata Corporation, 153 College Station, USA).

Large sample sizes provide ample data to conduct analyses on subgroups of interest while maintaining sufficient power to gain insights into the direction and size of the effects. Due to changes in the SINASC registry with respect to gestational age, it was necessary to conduct additional analyses to verify the results of our multivariate regression analysis. From 2001 to 2010, gestational age was recorded as a categorical variable (gestational age in completed weeks: <22w; 22-27w; 28-31w; 32-36w; 37-41w; ≥42w). Beginning in 2011, gestational weeks at birth was recorded as a discrete variable (gestational age in number of completed weeks), estimated from the date of the mother’s last menstrual period, physical examination or image results obtained from SINASC records. Due to dynamic aspects of this cohort, such as the differences in the entry time of each woman in the study, as well as changes in 2011 in terms of how information on gestational age was collected/recorded by SINASC, additional analyses were carried out (shown in Suplemmentary material) for verification purposes. These additional multivariate logistic regression models were employed in accordance with the same variable selection method used in the main analysis, yet were stratified according to year of birth before and after 2011 and incorporated “time of exposure” quartiles prior to birth, considering the time difference (in years) between the mother’s inclusion in the cohort a newborn’s date of birth.

The study has some important strengths. First, the large-scale dataset allowed us to comprehensively investigate known factors associated with TLBW. Second, the SINASC registry has high national coverage (over 90%), which provided a truly representative study population [58]. Moreover, the high reliability of the information contained in the SINASC database reinforces the suitability of using this system for epidemiological investigations [58]. With regard to limitations, the use of secondary data subjects the collected information to bias. This further implies that since data collection procedures were not performed as a function of the present study’s objectives, some important variables relevant to the determination of TLBW were not collected, such as maternal smoking habits, weight gain during pregnancy, the mother’s nutritional status and other maternal comorbidities. It is important to highlight that larger samples provide great opportunities for empirical research, but also may lead to equivocal interpretations due to the detection of statistical significance [59, 60].

This population-based study was focused on investigating live births to poor and extremely poor women. With due caution, some of the findings reported herein may be generalized to other populations with similar characteristics. Importantly, these limitations do not greatly detract from the promising potential enabled through research conducted on the “100 Million Brazilian Cohort”.