Subjects
Pelotas is a city located in Southern Brazil, with a population of approximately 330,000 inhabitants according to the 2010 national census. Three birth cohorts are currently being followed-up in the city, starting in 1982, 1993 and 2004. They share a very similar methodology – in their reference year, all newborns whose mothers lived in the urban area of the city were recruited to the studies. Pelotas had four maternity hospitals in 1982 and five in 1993 and 2004 that were monitored for births during the whole reference year of each cohort. Limiting recruitment to hospitals is not a problem in Pelotas since more than 98% of births are institutional, and those few mothers giving birth elsewhere usually go to a hospital for postnatal checks and care.
Specially trained field workers approached mothers in the first 24 h after giving birth and invited them to participate in the study. After consenting, mothers completed a questionnaire containing information about the family, the current pregnancy and delivery, and their babies were examined and had their length measured. Furthermore, field workers collected information about birth conditions, including Apgar score and weight, from hospital records. A full account of the methods used in the perinatal studies of these three cohorts was published previously [19].
The three birth cohorts recruited decreasing numbers of babies given the rapid decline in fertility observed in Brazil. In 1982, 5914 babies were included in the study, 5249 in 1993 and 4231 in 2004. In all cases, refusals to participate were below 1%. We followed up these three cohorts several times since recruitment, each with somewhat different assessment ages. Details for each cohort are available in specific methods papers [20–22].
C-section and perinatal information
Directly relevant to this study, interviewers obtained the date and time of delivery, as well as type of delivery during the perinatal interview. We also obtained other variables used as potential confounders at this moment: quintiles of SEP, according to Brazilian National Economic Index (IEN) [23], an asset index based on household goods and the household head’s education; maternal age at birth (years); maternal education (years); maternal reported skin colour (white, brown or black, according to the classification used by the Brazilian Bureau of Census, IBGE); parity (1, 2, 3 or ≥4 siblings); smoking during pregnancy (0, <20 or ≥20 cigarettes per day); pre-gestational BMI (based on reported weight before the pregnancy and measured height at the interview); financing of delivery (private insurance vs free public health system); and the child’s birth weight and length (except for 1982 birth cohort, where birth length was not measured).
For the 2004 cohort, we used maternal BMI obtained at the 3-month follow-up (based on height and weight measured during the assessment) instead of pre-gestational BMI since height was missing for about one third of the sample. The correlation between pre-gestational and 3-month follow-up BMI was 0.86 and Lin’s concordance coefficient [24] was 0.82, showing good agreement for those women with complete data.
Anthropometric and body composition information
The latest follow-up of each cohort included a detailed assessment of body composition. We measured participants’ body fat, lean and bone-mineral masses at the research clinic using dual-energy X-ray absorptiometry (DXA; GE Lunar Prodigy densitometer) in a full-body scan. Specially trained technicians carried out the exams, with participants in supine position using light and tight-fitting shorts and sleeveless tops. We asked participants to remove all metal accessories, such as bracelets, earrings or piercings. The examiners assessed the quality of DXA exams with participants still on the machine and repeated it, if necessary.
Height was measured twice by trained technicians using a Harpenden metal stadiometer, with 1 mm precision (Holtain, Crymych, UK). We used arithmetic mean of two measurements. We assessed weight using a high precision scale (0.01 kg), part of the BODPOD machine (Cosmed, Italy, http://goo.gl/7jzfLc) used for further body composition assessments. For each participant, we collected all measures included in this analysis on the same day.
Due to observed discrepancies between measured weight from BODPOD scale and total body mass from DXA, we adjusted some body composition indicators from DXA. Total body mass was obtained by adding up total fat, lean and bone-mineral masses. We calculated the percentage of fat mass by dividing fat mass by total body mass. We obtained adjusted fat mass by applying the DXA percentage of fat mass to the weight, as measured by the BODPOD’s scale.
We calculated fat mass index by dividing the adjusted fat mass (kg) by height (m) squared. For the 1993 and 2004 cohorts, we calculated the z-scores of BMI-for-age using the WHO 2007 growth reference [25]. For the 1982 cohort, we calculated BMI by dividing weight from BODPOD scale (kg) by the square of height (m), and afterwards we standardized the result to provide a similar scale used in the 1993 and 2004 cohorts.
Statistical analyses
We used multiple linear regression analysis to examine the crude and adjusted associations between the three body composition outcomes and type of delivery. We adjusted for potential confounders in three steps: 1. we included SEP at birth, maternal schooling, mother’s skin colour and financing of delivery (public health system or private health system); 2. we included variables of level 1 plus pre-gestational maternal BMI, parity, maternal age at birth and smoking during pregnancy; 3. we included variables in level 1 and 2 plus birth weight and length.
Because SEP information is difficult to measure accurately and SEP is strongly associated with both C-section and adiposity in Brazil, residual confounding is an important concern. We thus did a parallel analysis using height as outcome to check for residual confounding. Since we do not expect any relationship between C-section and height, an association after adjustment would suggest residual confounding. For the three cohort studies, we used current height (cm), and the same multiple linear regression analysis as described above.