Data were obtained from NHANES, a program of studies by the National Center for Health Statistics which assesses the health and nutritional status of the U.S. population through questionnaires, interviews, physical examinations, and laboratory tests (33). NHANES uses a stratified, multistage probability sampling design to attain a nationally representative sample of the non-institutionalized U.S. population. The sampling procedure begins with selecting primary sampling units (counties) which are then divided into segments (city blocks). A sample of households within each segment is randomly drawn and individuals within those households are randomly selected. Only the 2005-2006 NHANES cycle has concurrent accelerometer and cannabis use measures, so analyses were restricted to this time period. The sample was limited to participants aged 20-59 years (n = 3,409), since only respondents in this age range were given drug and alcohol use questionnaires. Participants were excluded if they did not have one or more adherent accelerometer wear days, defined as having at least 10 hours of wear time (n = 652; (34)); did not complete the Drug Use Questionnaire (n = 124); had incomplete cannabis use questions (n = 303); reported concurrent use of heroin, cocaine, or methamphetamines (n = 30); or self-reported or tested positive for pregnancy via urine test (n = 208). The final analytic sample contained 2,092 participants.
Measures of Cannabis Use
Using responses from the Drug Use Questionnaire (35), cannabis use was determined by responses to the following questions: “Have you ever, even once, used marijuana or hashish?” (yes, no). Individuals who responded “yes” were directed to the following question: “During the past 30 days, on how many days did you use marijuana or hashish?”. Current cannabis users were defined as those who used cannabis ³ 1 day in the past 30 days. All other participants were classified as non-current users.
Accelerometer measures of physical activity and sedentary behavior
Participants were requested to wear an Actigraph 7164 over their right hip for 7 days and to remove the device only for showering, swimming, and when in bed. The accelerometer data was processed via a computer script released by the National Cancer Institute, the full details of which are published elsewhere (36). Briefly, using the most common protocol for adults, accelerometer non-wear was characterized by 60 consecutive minutes with zero movement allowing for up to 2 minutes of movement below 50 counts per minute (cpm; (34)). As is commonly defined, an adherent day was designated as having at least 10 hours of wear time (34). Common acceleration cutpoints were then used to categorize different behaviors: minutes with less than 100 cpm were classified as sedentary, minutes between 100 and 1951 cpm were classified as light PA, and minutes above 1951 cpm were considered MVPA (37,38). The physical behavior metrics used in all analyses were computed as the average number of minutes in sedentary time, light PA, and MVPA over all adherent days.
Measures of self-reported physical activity
Participants self-reported their levels of PA through the Physical Activity Questionnaire (35). Engagement in MVPA was determined by the following questions: “Over the past 30 days, did you do moderate activities for at least 10 minutes that cause only light sweating or a slight to moderate increase in breathing or heart rate?” (yes, no); “Over the past 30 days, did you do any vigorous activities for at least 10 minutes that caused heavy sweating, or large increases in breathing or heart rate?” (yes, no). Responses to these questions were combined into one binary yes/no variable; if a respondent replied “yes” to one or both questions, they were categorized as having self-reported MVPA engagement. This method of measurement was chosen to replicate a previous study that used NHANES data to examine the cannabis-PA relationship (21). Self-reported light PA and SB were not assessed due to lack of appropriately corresponding questions in 2005-2006 NHANES.
Sociodemographic covariates included age, gender, race/ethnicity, income-to-poverty ratio, and education. Health-related covariates included body mass index (BMI), cigarette smoking status, and alcohol use. All covariates were obtained through self-report with the exception of BMI, which was measured by trained examiners. Age was reported in years and gender was indicated as male or female. Race/ethnicity was categorized as Non-Hispanic White, Non-Hispanic Black, Hispanic (inclusive of Mexican American and Other Hispanic), or Other (inclusive of Asian and multi-racial); non-Hispanic White was the reference level. Income-to-poverty ratio was calculated by dividing family income by the poverty threshold. Education was indicated as one of five categories: less than 9th grade, 9th-11th grade, high school graduate/GED, some college/AA degree, or college graduate and above (less than 9th grade was the reference level). Body mass index was calculated based on height and weight measurements. Cigarette smoking status was determined by the following questions: “Have you smoked at least 100 cigarettes in your entire life?” (yes, no) and “Do you now smoke cigarettes?” (every day, some days, not at all). Current cigarette smokers were classified as those who answered “yes” to smoking at least 100 cigarettes in their lifetime and now smoke “every day” or “some days.” Alcohol use was defined by the average number of days per week participants drank and was assessed using the following questions: “In your entire life, have you had at least 12 drinks of any type of alcoholic beverage?” (yes, no); “In the past 12 months, how often did you drink any type of alcoholic beverage? How many days per week, per month, or per year did you drink?” (responses recorded from 0-365). Responses were converted to days per week as applicable. Participants who have not had at least 12 alcoholic drinks in their lifetime or responded with “0” to frequency of drinking in the past 12 months were recorded as drinking an average of 0 days per week.
All statistical analyses were performed in R Version 3.6.1 (39) using the survey package (40), which accounts for NHANES sample weights, strata, and primary sampling units to make results more generalizable to the U.S. population. To determine statistical significance, a was set to 0.05.
Means and frequencies of covariates were calculated for the total sample and separately by cannabis use categories. Differences between these groups were tested using t-tests for continuous variables and c2 tests for categorical variables.
To account for differences in physical behavior measurements resulting from differences in the duration of wear time, SB, light PA, and MVPA were adjusted for accelerometer wear time via the commonly used (36,41) residuals method (42). Linear regression models assessed associations of cannabis use with SB, light PA, and MVPA as separate outcome variables using the following successively adjusted regression models: Model 1 (unadjusted), Model 2 (adjusted for age and gender), and Model 3 (adjusted for age, gender, ethnicity, income-to-poverty ratio, education, body mass index, cigarette smoking status, and alcohol use). Models 1 and 2 contained 2,092 observations while Model 3 contained 2,022 observations due to missing income-to-poverty ratio, BMI, or alcohol use data. To facilitate interpretation, regression results are reported as marginal means (40).
To assess potential effect modification, interaction terms for cannabis use and (a) age, (b) gender, and (c) cigarette smoking status were added to Model 3. Previous research has documented age and gender differences in SB time (43) and associations between cigarette smoking and cannabis dependence (44). Marginal means of SB, light PA, and MVPA were estimated for current and non-current cannabis users and differences were computed and reported separately for each level of the potential modifiers in Model 3. Age was dichotomized into younger or older than the median sample age (40 years) for the reporting of marginal means. Statistical significance of the between-level differences was obtained from the p-value of the interaction coefficients in the regression model.
To replicate a previous study (21), logistic regression models were calculated with self-reported MVPA as a binary outcome. Three logistic models were created using the same covariates and sample sizes as the linear accelerometry models (without interaction terms).
The non-interaction regression analyses were re-analyzed for a subsample of participants that had a minimum of four adherent accelerometer wear days (n = 1,618), an eligibility criterion thought to produce physical behavior estimates that more accurately represent usual weekly behavior patterns. Results were substantively similar, so only those derived from the full sample are detailed below.