A study design was a prospective cohort study design. The study sample were participants of the 1993 Pelotas (Brazil) Study . The 1993 Pelotas study was a birth cohort study that recruited 5,249 newborns (99.7% of all births) in the city of Pelotas, Brazil, in 1993 and followed them over time. For an assessment at age 18 years (2011 and 2012), 4,563 cohort members were located, of whom 4,106 (90.0%) attended the clinic visit. For an assessment at age 22 years (2015 and 2016), 4,933 cohort members were located, of whom 3,810 (77.2%) attended the clinic visit. Participants were ineligible for accelerometry assessment if participants did not live in Pelotas at the time of data collection, had physical disabilities, or were unable to wear an accelerometer due to work restriction. Participants were ineligible for dual-energy X-ray absorptiometry (DXA) assessment if participants used a wheelchair or had osteoarticular deformities, implanted metal pins, screws, plates and non-removable metallic objects (body piercings and/or chains), or were extreme obesity, height over 1.92 m, or pregnant . The current report used the data from 1993 Pelotas study participants who completed accelerometer assessments and DXA assessments at 18 and 22 years of age. All study protocols were approved by the Ethics Committee of the Federal University of Pelotas Medical School (register number 05/2011 and 1.250.366). Written informed consent was obtained from individual participants.
The exposure of interest was change in accelerometry-measured MVPA from age 18 to 22 years. Because of device availability, different brands of accelerometers were utilized: GENEActiv Accelerometers (range of ± 8 g; 85.7 Hz; 5-s epoch; ActivInsights, Kimbolton, UK) at age 18 and ActiGraph GT3X + (range of ± 8 g; 60 Hz; 5-s epoch; ActiGraph Inc., Pensacola, FL, USA) at age 22. A prior study  showed a high agreement for an acceleration magnitude metric (intraclass correlation coefficient for Euclidian Norm Minus One [ENMO] = 0.99) between the two devices. During clinic visits at age 18 and 22 years, participants were asked to wear an accelerometer on the non-dominant wrist for 24 h and for 4 to 7 consecutive days, including at least 1 weekend day. Detailed information regarding the protocol can be found elsewhere [19, 24].
Accelerometer data were analyzed using the R-package GGIR . Briefly, the GGIR data processing includes automatic calibration, detection of sustained abnormally high values, detection of non-wear, detection of wake/sleep, and quantification of dynamic acceleration magnitude. Data were analyzed per 24 h from midnight to midnight (MM windows). To be included in data analysis, at least 3 wear days and at least 16 wear hours per day were required . The activity-related acceleration metric, ENMO, was calculated. ENMO (mg) is one omnidirectional measure of body acceleration, calculated by subtracting the value of gravity from vector magnitude [√(x2 + y2 + z2)-1] . MPA minute was defined as a minute with 100 mg ≤ ENMO ≤ 400 mg for ≥ 48 s (80% of 60 s) . VPA minute was defined as a minutes with ENMO ≥ 400 mg for ≥ 48 s . MVPA minutes were calculated by summing MPA minutes and VPA minutes. In addition, because in the adult PAG (i.e., ≥ 150 min/week of MPA, ≥ 75 min/week of VPA, or an equivalent combination of MPA and VPA) [1, 2], one minute of VPA is considered to be equivalent to two minutes of MPA, we calculated MPA-equivalent minutes (minutes/day) at age 22 years by summing MPA minutes and twice VPA minutes (MPA minutes + 2 × VPA minutes) .
MVPA at age 18 years was categorized into two groups: 0–59 or ≥ 60 min/day (no [N] or yes [Y] for meeting the youth PAG). MVPA at age 22 years was categorized into three groups: 0–21, 22–59, or ≥ 60 min/day (N, Y22, or Y60 for not meeting the adult PAG, meeting the adult PAG, or meeting the youth PAG, respectively). The combination of these groups created six MVPA groups (N&N, N&Y22, N&Y60, Y&N, Y&Y22, or Y&Y60). In addition, we also created six weighted-MVPA (wMVPA) groups using MPA-equivalent minutes, instead of MVPA minutes, at age 22 years.
The primary outcome was change in fat mass index (FMI) from age 18 to 22 years (∆FMI). The secondary outcome was change in fat mass from age 18 to 22 years (∆FM). We also explored change in body mass index (BMI) from age 18 to 22 years (∆BMI) as an outcome. During clinic visits at age 18 and 22 years, participants underwent a whole-body DXA scan (GE Lunar Prodigy, USA). Detailed study procedures have been described elsewhere . Scan images were analyzed using the in-built GE Lunar enCore software. Fat mass (kg) was derived from the image analysis. FMI was calculated by dividing fat mass (kg) by height squared (height2).
Several potential confounders were considered: the exact age at the age 22 assessment, wealth index quintiles at age 18 years, education level at age 22 years (0–8, 9–11, or ≥ 12 years of schooling), energy intake quintiles at ages 18 and 22, and FMI at age 18. Wealth index, education level, and energy intake were grouped to allow for non-linear trends in association between the confounders and the outcome. Wealth index was calculated using a principal component analysis based on a set of goods and assets, from which the first component was extracted and then divided into quintiles . Missing data for wealth index quintile (n = 2) was imputed to the middle category. Daily energy intake (kcal/day) at ages 18 and 22 years was estimated using a food frequency questionnaire . Daily energy intake (kcal/day) was categorized into five groups based on sex-specific quintile cut-points. Missing data for the energy intake quintile at age 22 years (n = 1) was imputed with the energy intake quintile at age 18 years. Energy intake quintile at age 18 years and change in energy intake quintile from age 18 to 22 years were used to account for the energy intake effects.
All analyses were conducted separately by sex using SAS 9.4 (Cary, NC). To address potential selection bias, we conducted Chi-square analyses to compare the frequencies of the key characteristics (i.e., sex, birthweight, maternal education level, and family income level) between those who were included in the analysis and those who were excluded. Descriptive analyses, including means and standard deviations of the exposure and outcome variables, were conducted. Analysis of variance (ANOVA) was conducted to compare the exposure and outcome variables among the six MVPA groups. A multivariable linear regression model was used to estimate ∆FMI by the MVPA group variable (reference group = Y&Y60), adjusting for age, family wealth index quintile at age 18, education level at age 22 years, energy intake quintile at age 18 years, change in energy intake quintile from age 18 to 22 years, and FMI at age 18 years. The analysis was repeated for the ∆FM outcome. The statistical significance level was set at 0.05.