Data source

This study was a secondary data analysis of the National Health and Nutrition Examination Survey (NHANES) conducted by the U.S. Centers for Disease Control and Prevention (CDC). NHANES is a continuous cross-sectional survey of the U.S. noninstitutionalized civilian population and provides a nationally representative sample through its stratified, multistage, probability cluster design.^{11} NHANES 2003–2004 and 2005–2006 were used because these data releases uniquely measured physical activity objectively by accelerometry in participants 6 years of age and older, as described in greater detail below.

Participants

Nonpregnant participants aged 6–19 years were included. Participants with missing data or insufficient accelerometry data to estimate physical activity (see below) were excluded from analyses. NHANES oversampled Mexican-American and non-Hispanic black participants. We restricted analyses to participants who identified as non-Hispanic white, non-Hispanic black, and Mexican-American to provide sufficient sample size for the analyses; participants identified as “Other race–Including Multiracial” and “Other Hispanic” were dropped due to low numbers, similar to a previous NHANES analysis on pediatric metabolic syndrome.^{12} The Institutional Review Board of Baylor College of Medicine exempted this secondary analysis on deidentified data from review.

Outcome variables

Blood pressure (systolic and diastolic; SBP and DBP) and anthropometric measurements (height, weight, and WC) were assessed on participants using standardized techniques and equipment in the NHANES Mobile Examination Centers.^{11} BMI was calculated as weight in kilograms divided by the square of height in meters, and BMI *z*-scores were determined from the CDC U.S. growth charts.^{13}

Clinical tests were performed per standard protocol^{11} and selected on the basis of their relationship to metabolic risk (cardiovascular disease or insulin resistance).^{14,15} The following were obtained with detailed methods reported elsewhere^{11}: C-reactive protein (CRP), total cholesterol, and high-density lipoprotein cholesterol (HDL-C).

Glycohemoglobin was obtained on participants ≥12 years old. Serum triglycerides (TG), plasma glucose, and plasma insulin levels were obtained on a subsample of fasting participants ≥12 years old. Fasting low-density lipoprotein cholesterol (LDL-C) was calculated with the Friedewald equation from the fasting subsample of participants.

A metabolic risk score was calculated, based on the sum of the age- and gender-adjusted *z-*scores for SBP, DBP, HDL, TG, and fasting glucose.^{16} The *z*-score for HDL-C was multiplied by −1. The metabolic risk score was composed of the clinical and laboratory components that comprise the metabolic syndrome^{14,17,18} but excluded central adiposity (excess WC), because a mediator should not be a component of the outcome variable.^{10}

Main exposure

Physical activity was assessed by accelerometry as previously described.

^{2} Participants wore an accelerometer (Model 7164, Actigraph, LLC; Ft. Walton Beach, FL) for 7 days. The accelerometers measured the volume and intensity of movement in 1-min intervals and provided a valid and reliable objective measure of physical activity in children.

^{19} Accelerometer data were processed following the criteria in a previous NHANES study on physical activity.

^{2} Habitual physical activity was estimated by using participants' data that had at least ≥4 valid days (10

h or more of monitor wear daily). To ensure consistency with previous NHANES accelerometer studies,

^{2,20,21} we used the age-specific thresholds developed by Freedson for MVPA,

^{22} which was set at four metabolic equivalents. Minutes that met or exceeded the threshold were summed for the daily estimate of MVPA minutes.

^{2} This sum was divided by the number of valid days to calculate mean daily minutes of MVPA. While Trost

^{23} has generally recommended using the physical activity intensity thresholds of Evenson

^{24} due to their classification accuracy for all four levels of physical activity, both the Freedson and Evenson thresholds for combined MVPA exhibited “excellent” classification accuracy and were superior to other cut points.

^{23}Because a large number of participants would be excluded on the basis of having <4 valid days of accelerometer data and because only a limited number of participants were assessed for all components of the metabolic risk score, we sought to maximize sample size for analyses involving the metabolic risk score by using a Bayesian approach

^{2,25} to estimate adherence to the physical activity recommendation for children (≥60

min MVPA on 5 of 7 days). The Bayesian approach required ≥1-valid day (compared to ≥4-valid days for the traditional estimate), which increased eligible participants by ≈300 participants. The estimate is based on the probability of a participant obtaining the recommended amount of MVPA on a given day, taking into account the number of days on which the accelerometer was worn and the number of days in which the recommendation was met.

^{25} The Bayesian approach provided an estimate of adherence to the physical activity recommendations, termed MVPA adherence, and was only used for analyses with the metabolic risk score.

Covariates

Several sociodemographic and dietary/lifestyle variables were included as covariates that might confound the relationship between MVPA and adiposity or metabolic risk. The following covariates were included: (1) Age in years, (2) gender, (3) race/ethnicity categorized as non-Hispanic white, non-Hispanic black, and Mexican-American; and (4) income, *i.e*., the poverty-to-income ratio (PIR). PIR values <1 were below the poverty threshold. PIR was provided by NHANES in the following six categories: <1, ≥1 <2, ≥2 <3, ≥3 <4, ≥4 <5, and ≥5 PIR.^{11}

Two dietary variables were included as covariates: Total dietary energy intake (all participants) and alcohol intake (participants aged 12 years and older) calculated from the mean of two 24-h dietary recalls.^{26} Alcohol intake was an important covariate because: (1) Up to one-half of adolescents reported consuming alcohol during the past month in the 2007 U.S. Youth Risk Behavior Survey,^{27} and (2) studies have reported a positive relationship between alcohol intake and the Metabolic Syndrome.^{28,29}

Cigarette smoking was a covariate because it is common among adolescents^{30} and has been independently associated with the metabolic syndrome.^{29,31} It was assessed by two methods: (1) Serum cotinine^{32} (participants ≥3 years old) and (2) the question, “During the past 5 days, did [you/he/she] use any product containing nicotine including cigarettes, pipes, cigars, chewing tobacco, snuff, nicotine patches, nicotine gum, or any other product containing nicotine?” with those who answered “yes” considered as smokers (participants 12–19 years old).^{33}

Analyses

Frequencies, percentages, means, and standard deviations were calculated to describe participant characteristics. Participants with missing data were excluded on an analysis-by-analysis basis. Differences between excluded and included participants were examined using chi-squared tests of independence and analysis of variance.

A series of linear regression analyses were conducted that examined the BMI *z*-score as a potential mediator of the relationship between MVPA adherence and the metabolic risk score (sum of the *z*-scores for SBP, DBP, HDL-C, TG, and fasting glucose), while controlling for WC. For the first linear regression analyses, termed the “C path” (see ), the relationship between MVPA adherence and the metabolic risk score was examined. Covariates included age, gender, race/ethnicity, PIR, total energy intake, total alcohol intake, smoking status, and WC. We next conducted linear regression analyses with MVPA adherence as the independent variable and BMI *z*-score as the dependent variable to determine if MVPA adherence was inversely associated with BMI *z*-score (“A path” of ). Covariates were identical to those listed above for the first linear regression analysis. The relationship between BMI *z*-score and the metabolic risk score, controlling for covariates (“B path” of ), was also examined. A fourth set of regression analyses was conducted with both MVPA adherence and BMI *z*-score as independent variables and the metabolic risk score as the dependent variable to determine if BMI *z*-score mediated any potential relationship between MVPA and the metabolic risk score (“C path” of ). The product-of-coefficients method^{10} was used to assess whether BMI *z*-score was a mediator using the Sobel formula.^{10} This method is a refinement^{10,34} of the original causal steps approach,^{35} but does not require large sample sizes. In parallel analyses, a series of linear regression analyses was conducted with WC as the potential mediator, while controlling for BMI *z*-score in analyses of the A, B, C, and C′ paths.

Similar to the mediation analyses examining BMI *z*-score or WC as mediators for the metabolic risk score, analyses of the A, B, C, and C′ paths were conducted examining BMI *z*-score or WC as mediators for several individual metabolic risk factors (SBP, DBP, CRP, glycohemoglobin, total cholesterol, HDL, TG, LDL-C, fasting glucose, or fasting insulin). In this series of regression analyses, the mean daily minutes of MVPA value was used as the main independent variable because sample size was not as constrained compared to analyses involving the metabolic risk score.

Analyses were conducted using the survey weights and commands in SAS 9.1.3 (SAS Institute Inc., Cary, NC), *e.g*., PROC SURVEYREG, to take into account the complex survey design. For analyses involving nonfasting clinical and laboratory variables (CRP, SBP, DBP, and glycohemoglobin), special sampling weights were used for the subsample of participants with ≥4 valid days of accelerometer data, which reweights the data to be nationally representative.^{2} For analyses involving fasting laboratory variables (LDL-C, TG, glucose, and insulin) and the metabolic risk score, fasting sample weights provided by NHANES were used as recommended to reweight the data to be nationally representative.^{11} A significance level of 0.05 was chosen. Standardized regression coefficients (std. β) are presented; however, tests of mediation were performed using the unstandardized coefficients.