Research setting and study design
During 1982, 1993 and 2004, birth cohorts representing all birth to mothers residing in the urban area of Pelotas, Southern Brazil, were enrolled. The three studies relied on primary data collection with the same methodology,[16–18] including recruitment of all hospital deliveries in the city (over 98% of all deliveries). Eligible mothers were interviewed soon after delivery using structured questionnaires covering socioeconomic variables, characteristics of pregnancy, labor, delivery and health care utilization. Similar variable definitions were used in all three studies. Non response rates at recruitment were below 1% in the three perinatal studies.
Data from the perinatal study and from follow-up home visits at 12 and 48 months were used in the present study. In the 1982 and 2004 cohorts, we attempted to locate all babies recruited in the perinatal phase. In the 1993 cohort we sought all low birth weight babies and a 20% sample of the remained; weighted data analyses were used to reproduce the total cohort .
Follow-up losses at 12 and 48 months visits were equal to 20.7% and 15.9% for the 1982 cohort, 6.6% and 12.8% for the 1993 cohort, and 5.9% and 8.2% for the 2004 cohort, respectively.
During each home visit mothers were interviewed by trained personnel who also conducted anthropometric measurements of the children. Tanita electronic scales (Tanita, Tokyo, Japan) with 100g precision were used for weight, the scales being calibrated on a weekly basis. Lengths and heights were measured with portable infantometers with 1mm precision. Birth length was not available for the 1982 birth cohort.
Principal exposure (gestational age)
Gestational age estimation was based on the date of the last menstrual period (LMP). Cases of unknown or implausible LMP were treated as missing information. They corresponded to 15.6% of the births – 21% in 1982, 11% in 1933, and 7.3% in 2004. Gestational age was categorized as <34, 34–36, 37, 38, 39–41 and ≥42 completed weeks.
Mortality surveillance was carried out actively through regular visits to all hospital, cemeteries, state vital registration services and the city's health department. Infant and neonatal mortality were defined as the deaths of live-born infants in the first 364 or28 days of live, respectively. Deaths in these periods were expressed per thousand live births: the infant (IMR) and neonatal mortality rates (NMR).
Information regarding breastfeeding was reported by the mother. Total breastfeeding duration (in months and days) was collected at each follow-up. The earliest available information on breastfeeding cessation was used to reduce recall bias.
Information on hospital admissions in the first 12 months of life was obtained from the mothers and was restricted to re-admissions after the newborn had been discharged from the maternity hospital.
The following anthropometric indices were calculated: weight-for-age (W/A), height-for-age (H/A) and weight-for-height (W/H) z-scores at 12 and 48 months, using the World Health Organization (WHO) growth standards . Underweight was defined as W/A z-score below −2; stunting, as H/A z-score below −2; wasting, as W/H z-score below −2; and overweight, as W/A z-score above +2 standard deviations.
Data on potential confounding factors
The following factors, measured in the perinatal period, were considered to be potential confounders of the association between gestational age and each outcome. Family income in the month prior to delivery was analyzed in quintiles; maternal schooling (completed years of formal education), maternal age (completed years) and height (in centimetres) were analyzed as continuous variables. Women who were single, widowed, divorced or lived without a partner were classified as single mothers. Parity was defined as the number of previous pregnancies resulting in a live birth or a late fetal death. Smoking during pregnancy, regardless of the number of cigarettes, was categorized as yes or no. Breastfeeding was not considered as a confounder, because it could not influence gestational age; in fact, it constitutes a potential mediator in the association between gestational age and the other outcomes.
We tested heterogeneity between the three cohorts in terms of the association between gestational age and each outcome. Because no significant interactions were found (p<0.10), we pooled the data from the three cohorts.
We used t-test or x
2 statistics to study crude associations between gestational age and each outcome. Confounder-adjusted analyses included logistic regression analysis for categorical outcomes and multiple linear regression for continuous outcomes. Analyses were carried out using Stata software, v. 11.0.
The study protocols of the three cohort studies were approved by the Medical Ethics Committee of the Federal University of Pelotas, which is affiliated with the Brazilian Federal Medical Council. In the 1982 and 1993 cohort studies, verbal consent to participate in the study was obtained from mothers. In the 2004 cohort study, mothers were provided written informed consent.