Changes in the Prevalence and Correlates of Weight-Control Behaviors and Weight Perception in Adolescents in the UK, 1986-2015

This cohort study uses data from 3 cohort studies in the UK to investigate the changes in prevalence of adolescent weight-loss and weight-perception behaviors and their association with depressive symptoms.

and were eligible for receiving child benefits. The cohort includes children living in non-household situations and children who were not born in the UK, but lived in the UK at recruitment. The study used a stratified clustered framework to ensure disadvantaged and ethnic minority groups were adequately represented. 14 The Multi-Centre Research Ethics Committee (MREC) gave ethics approval for MCS.
Participants gave written consent to take part in these studies.

eMethods 2. Measures of Depressive Symptoms
The Short Moods and Feeling Questionnaire 2 contains 13 statements which the participant could respond in one of three ways (0 = "not true", 1 ="sometimes", 2="true") whose total score ranges from 0 to 26, with greater scores indicating greater depressive symptoms. The 9 questions from the Malaise inventory are scored on a three-point Likert scale (0 = rarely/never, 1 = some of the time, 2 = most of the time), resulting in a total score ranging from 0 to 27, with higher scores indicating greater symptoms. 3

eMethods 3. Body Mass Index
To minimize data missingness, in BCS and ALSPAC we supplemented objective measures of BMI (obtained by fieldworkers in BCS and in clinic assessments in ALSPAC) with self-reported measurements when objective measures were not available (28% of participants in BCS, 11% in ALSPAC). The correlation between the two was good (r=0.69 BCS and r=0.89 ALSPAC). In MCS, fieldworkers objectively measured BMI at study visit with no self-reported measures available. In all samples, we created age-and sex-standardized BMI categories indicating underweight, normal weight, and overweight BMI using the International Obesity Task Force cut-offs. 4-6 eMethods 4. Multiple Imputation We imputed missing outcome and covariate data for participants with at least one outcome variable available using multiple imputation by chained equations and imputing 50 datasets.
In our imputation models, we included all variables used in the analyses, plus an indicator of maternal marital status, as these variables could be harmonized across datasets. We imputed data separately for each cohort and merged them after imputation for analyses. To account for participant attrition from baseline assessments in each dataset, we created attrition weights as the inverse of the probability of having taken part in the sweep of interest. We used indicators of child's sex and ethnicity, paternal social class, and maternal marital status, age, and highest education level to create attrition weights. To do this, we first imputed missing baseline data using single imputation as the proportion of participants in our sample with missing data was minimal (2% BCS, 9% ALSPAC, 22% MCS*), so that all children in our sample would have an attrition weight. We ran all of our analyses in imputed datasets using attrition weights. We additionally ran a number of sensitivity analyses using complete cases and imputed data without weights to check the consistency of our results. *all MCS missing data were on the social class variable. There was no missingness on maternal age, marital status, child's ethnicity and sex.

eMethods 5. Changes From Published Protocol
We made minimal changes to the analysis plan detailed in our protocol these were: 1. We additionally included attrition weights in the analyses to account for attrition within cohorts 2. We ran a model further adjusting for BMI for the the analyses looking at changes in prevalence of current weight intentions.