Pelotas (population 320,000 inhabitants) is located in southern Brazil in a relatively developed part of the country. In 1993, a birth cohort study was initiated in the city. Mothers of all hospital-born children, representing >99% of all deliveries in Pelotas, (n = 5,265) were invited to join the study. More than 99% (n = 5,249) of mothers agreed to participate in the study. Since 1993, the national mortality system has been regularly monitored, and deaths of cohort members are linked to the study database. Mothers were interviewed soon after delivery on family characteristics, gestation-related variables, and behavior. Socioeconomic status was evaluated using an assets index. It included questions on the possession (and quantity) of household assets, such as television, radio, air conditioning, phone lines, fridge, among others, and a standardized procedure was used to generate a continuous score through principal component analysis that was later divided into quintiles. Parental education was expressed in years of schooling.
In 2004–2005, we conducted a follow-up visit to all participants of the cohort. We were able to locate 4,452 participants, which represent 87.5% of the cohort taking into account the 141 subjects known to have died. In 2008, another wave of data collection was conducted, and we were able to locate 4,325 cohort members. In both visits, participants were reimbursed for any transportation or food expenses, but no other incentives were given for taking part in the follow-up visits. A detailed description of the cohort has been published elsewhere 
Transport and leisure-time physical activity were measured using a validated questionnaire 
. The questionnaire initially asks about the primary mode of transportation to and from school. A list of several culturally relevant leisure-time physical activities, including sports, was later shown to participants, and they were asked whether they have practiced those activities over the past week. For each positive answer, information on weekly frequency and duration was obtained. Frequency was multiplied by time, generating a weekly score for each activity; later the scores were summed, producing a leisure-time physical activity score. For this analysis, we opted to ignore commuting time for three main reasons: (a) a large proportion of total physical activity in this age-group in Brazil is practiced in the transportation to and from school 
; (b) transport-related physical activity at this age is much more a matter of poverty than a choice in Brazil 
; and (c) the intensity of walking or cycling to school tends to be low 
. The continuous score of leisure-time physical activity was transformed into quartiles for the analysis.
The reliability and concurrent validity of the physical activity questionnaire were tested in a previous study 
. Its reliability was good (rho: .62; p
< .001); 73% of the subjects were classified consistently in a 7-day test–retest exercise. The kappa value was .58. The concurrent validity of the questionnaire was tested against pedometers; the Spearman correlation coefficient was .26 (p
= .02), and 57% of the subjects were classified consistently as physically inactive in the questionnaire and with pedometers (using a cutoff point of 10,000 steps/d).
School failure was defined as having ever repeated a grade at school. We calculated the incidence of school failure from age 11 to 15 years by first excluding from the analyses all subjects who experienced a school failure before 11 years of age, and then categorizing as “positive” all those who reported failing at school from age 11 to 15 years. This strategy was used owing to the basic cohort study principle that subjects should be disease-free at baseline.
School failure was analyzed as a dichotomous variable (yes vs. no). Confounding variables were sex, parental schooling, family assets index, type of school, age at school entry, and child labor (yes vs. no). These variables were selected because the literature search indicated a possible association with both the outcome and the exposure variable.
Analyses were conducted using Stata 11.0 (StataCorp LP, College Station, TX). First, we calculated the proportion of school failure according to quartiles of maternal and paternal schooling. Second, we explored the association between physical activity and socioeconomic position. These analyses were carried out to understand the confounding structure of our data. The main analyses included estimating odds ratios for having experienced a school failure between 11 and 15 years of age according to leisure-time physical activity indicators at 11 years of age. Finally, we stratified the association between sports practice and school failure by gender due to a significant interaction.
All phases of the 1993 Pelotas (Brazil) Birth Cohort Study obtained institutional review board approval. Written informed consent was signed by mothers or guardians in all visits, and verbal consent was given by adolescents at the 11- and 15-year-old follow-up visits.