Screen-time, Physical Activity And Bmi

Abstract

PURPOSE: Observational studies often adjust for confounding by socioeconomic status (SES) but rarely report the magnitude of confounding bias. Also, different indicators of SES may not meet the definition of a confounder for some associations. This study will determine whether different measures of SES have different confounding effects on the associations of screen time with BMI and of moderate/vigorous physical activity (MVPA) with BMI. METHODS: In a 15 year prospective cohort study, 1830 young adults (age 25 - 36) reported height and weight (used to calculate BMI in kg/m2), hours/day of screen time, hours/week of MVPA and nine measures of SES in 2015. SES at baseline (2000) was based on parent education and employment. BMI, screen time and MVPA were regressed with linear regression on each SES measure and a composite measure that combined young adults’ income and education. Crude associations of screen time with BMI and of MVPA with BMI were assessed with linear regression. The SES measures were added individually to these models to determine the percent change in the associations (percent confounding). RESULTS: Seven measures of SES were associated with BMI; eight with screen time; and four with MVPA. Education, income, financial strain and living with parents all confounded the association of screen time with BMI by more than 10%. No measure confounded the relationship of MVPA with BMI by more than 10%. The composite measure was significantly associated with screen time (β = -0.39, 95% CI [-0.47 to -0.31]) and BMI (β = -1.29, 95% CI [-1.58 to -0.99]), but not with MVPA. This composite substantially confounded the association of screen time with BMI (29%) but not of MVPA with BMI (3%). CONCLUSIONS: The composite SES measure had similar properties to individual SES measures: it was associated with screen time and BMI but not with MVPA and it had substantial confounding bias effect on the association of screen time with BMI, but not on the association of MVPA with BMI. Knowing the magnitude of confounding can help assess results from other studies that do not adjust for SES. A composite measure may be useful for the varied analyses done with data from large cohort studies. However, because of the potential for over fitting models, associations with measures of SES should be carefully considered before using them to adjust for confounding.