This study was approved by the Research Ethics Committee of the University Hospital, Faculty of Medicine of Ribeirao Preto, University of Sao Paulo; and by the Research Ethics Committee of the University Hospital, Federal University of Maranhao, Brazil. All participants gave signed informed consent.
This is a study including two birth cohorts from two different Brazilian cities. These cohorts were set up to investigate perinatal health and to study the effect of early life factors on chronic non-communicable diseases in later life.
The first birth cohort started in 1994 in Ribeirao Preto (RP) in Sao Paulo state. RP is a city with a total population of 461,427 inhabitants, located in the most developed state in Southeastern Brazil and ranked 22nd in the Human Development Index, considering Brazilian cities .
The second birth cohort started in 1997 in Sao Luis (SL) in Maranhao state. SL is a city with a total population of 871,068 inhabitants and is one of the poorest state capitals in the Northeastern region, the least developed area of Brazil, ranked 1112nd in the Human Development Index .
In RP, 2,923 newborns were included corresponding to all live-births during a 4-month period in the 10 maternity hospitals in the city (from April to August 1994). After exclusion of multiple births (n = 65), 2,858 singletons remained. Missing information amounted to 5.8% and was due to early hospital discharge or refusal to participate .
In SL, a systematic stratified sample according to the number of births in each of the 10 maternity hospitals in the city was collected from March 1997 to February 1998. One out of 7 births was selected from each maternity hospital. A total of 2,541 births were initially selected and, after excluding multiple gestations (n = 50) and stillbirths (n = 48), a final sample of 2,443 live-births was included .
Anthropometric measurements were obtained from newborns in the two live-birth cohorts just after birth (birth weight and birth length) by trained staff. Information related to the gestational, delivery and post-delivery periods was obtained by means of a standardized questionnaire.
In 2004/2005, a randomized sub-sample from each of the original live-birth cohort samples was calculated for reassessment. The following categories were selected according to birth weight: <1,500 g [very low birth weight (VLBW)]; 1,500-2,499 g [low birth weight (LBW)]; 2,500-2,999 g [insufficient birth weight (IBW)]; 3,000-4,249 g [normal birth weight (NBW)]; and ≥4,250 g [high birth weight (HBW)]; the latter birth weight category corresponded to +2 SD from the mean. The lowest (< 2500 g) and highest (≥4,250 g) categories of birth weight were oversampled in order to increase the power of the study .
Children were searched at schools or at home in both cities. Parents or persons responsible for all VLBW, LBW and HBW and for a fraction of one out of three children in the remaining groups were invited to participate by telephone or mail. In RP, after exclusion of 48 deaths in the first year of life, 2,810 children were alive at one year of age; from these, 1,150 children were eligible for follow-up. The follow-up rate was 68.7%, 24 of them being VLBW, 145 LBW, 593 IBW and NBW, and 28 HBW, for a total of 790 children aged 10–11 years . In SL, after exclusion of 65 deaths in the first year of life, 2,378 children were alive at one year of age; 926 of these children were eligible for follow-up. With a follow-up rate of 72.7%, 673 children from 7 to 9 years old were evaluated, five of them being VLBW, 76 LBW, 573 IBW and NBW, and 19 HBW. Details of the methodology and flow charts of the studies were published previously .
In both cities losses were due to the impossibility of locating the children, to migration, to the fact that the children were not enrolled in school, to refusal on the part of the parents or because two schools in SL did not allow the research team to contact the children.
The variables obtained at birth and late infancy in both live-birth cohorts considered in this study were: type of delivery (vaginal or cesarean section), maternal smoking during gestation (yes or no), maternal schooling in completed years (≤ 8, 9–11 and ≥ 12), newborn gender (male or female), preterm birth (<37 gestational weeks based on the last menstrual period), maternal weight before pregnancy informed by mothers (continuous variable collected only in SL). Due to small numbers in the VLBW and HBW groups, birth weight in grams was reclassified in three groups for analysis (≥ 3,000 g, 2,500 |- 3,000 g and < 2,500 g).
In 2004/2005, anthropometric measurements (weight and height) were obtained from the two subsamples. Body mass index (BMI) was calculated by dividing weight in Kg by height in m2. Obesity was defined as BMI ≥ 95th percentile according to gender and age in months . Type of schooling was classified as public or private and exclusive breastfeeding duration was categorized into ≤ 30 days, > 30 days, and no breastfeeding.
Samples of 771 individuals have an 80% power to detect a minimal difference of 4% in obesity rates (estimating the prevalence rate of obesity at 10%) between the exposed and the non-exposed group, with a 5% likelihood of type I error .
Due to the complex sampling design, children with low and high birth weights were over-represented in the samples. Therefore, the prevalence estimates and their standard errors were calculated taking into account the different probabilities of selection of each birth weight and preterm birth group. Stratification of the sample by birth weight was also taken into account .
In order to investigate the association between CS and obesity in school-aged children a multivariable logistic regression model was performed, using the covariables described above in the adjustment. Furthermore, a multiple linear regression model was performed in order to investigate the association between CS and BMI z-score, as a continuous variable, also using the same covariables for controlling of confounding. Because of differences in age and sex, BMI was standardized by them as recommended by the World Health Organization [28, 29]. Missing data on covariates were excluded from the final models when missing values were less than 5%. In the case of maternal schooling a dummy variable for missing cases was included in the analysis.