Physical activity construct validity and intensity levels are presented in eTable 1 . In short, CPS II, CLUE II, and WHS had 7–8 line items querying the average weekly time spent performing the following activities over the prior year: walking, jogging/running, swimming, tennis/racquetball, bicycling, aerobics and dance. The physical activity questionnaires were adapted from the Nurses’ Health Study questionnaire, which has shown correlation coefficients ranging from 0.79–0.83 compared to recalls and from 0.59–0.62 compared to diaries20 (link).
NIH-AARP and WLHS used physical activity questionnaires that have not formally been validated, but have shown expected associations between physical activity and mortality in previous studies 12 (link),21 (link), and USRT has shown expected inverse associations with breast cancer 15 (link). The NIH-AARP Study included a single line item for all moderate- or vigorous-intensity leisure time physical activities with categorical responses measured in hours per week (h/wk). The WLHS questionnaire included separate line items about hours per day in leisure-time physical activity such as walking, horseback riding, or in strenuous activities, and the U.S. Radiologic Technologists study had separate line items for h/wk spent walking for exercise and exercising strenuously. For all six studies, we calculated energy expended per activity by multiplying the estimated MET value 22 (link) (multiple of resting metabolic rate) by the number of h/wk and summed across activities to estimate overall leisure-time physical activity energy expenditure in MET h/wk.
We used standardized categories to harmonize data between cohorts as follows: race/ethnicity (black, white, other), education (did not finish high school, finished high school, post-high school training, some college, finished college, missing), smoking status (never, former, current, missing), history of cancer (yes, no/missing), history of heart disease (yes, no/missing), alcohol consumption (0, >0-<15, 15-<30, 30+ grams/day), marital status (married, divorced, widowed, unmarried, missing) and BMI (<18.5, 18.5–25, 25-<30, 30-<35, 35+ kg/m2). We imputed the value for alcohol using the median value because non-drinkers and true missing values were grouped differently between studies. In subsequent analysis we tested associations using a missing category for alcohol instead of the imputed value and found no change in our physical activity results (all hazard ratios were within 0.02 of previous estimates). Questionnaires did not distinguish between “missing” and “no” for history of heart disease and cancer history; thus individuals were dichotomized into groups of yes or missing/no. Missing data was <5% for all covariates. We performed analyses calculating follow-up time in two ways: first, using age at study entry to age at death or end of follow up and second, calculating time from baseline questionnaire to date of death or end of follow-up. Because results did not differ from analyses using age as the time metric or using follow-up time and adjusting for age, in further analyses we used the latter method and adjusted for continuous age. The National Death Index, death certificates, or medical records were used to ascertain date of death (eTable 1 ).
NIH-AARP and WLHS used physical activity questionnaires that have not formally been validated, but have shown expected associations between physical activity and mortality in previous studies 12 (link),21 (link), and USRT has shown expected inverse associations with breast cancer 15 (link). The NIH-AARP Study included a single line item for all moderate- or vigorous-intensity leisure time physical activities with categorical responses measured in hours per week (h/wk). The WLHS questionnaire included separate line items about hours per day in leisure-time physical activity such as walking, horseback riding, or in strenuous activities, and the U.S. Radiologic Technologists study had separate line items for h/wk spent walking for exercise and exercising strenuously. For all six studies, we calculated energy expended per activity by multiplying the estimated MET value 22 (link) (multiple of resting metabolic rate) by the number of h/wk and summed across activities to estimate overall leisure-time physical activity energy expenditure in MET h/wk.
We used standardized categories to harmonize data between cohorts as follows: race/ethnicity (black, white, other), education (did not finish high school, finished high school, post-high school training, some college, finished college, missing), smoking status (never, former, current, missing), history of cancer (yes, no/missing), history of heart disease (yes, no/missing), alcohol consumption (0, >0-<15, 15-<30, 30+ grams/day), marital status (married, divorced, widowed, unmarried, missing) and BMI (<18.5, 18.5–25, 25-<30, 30-<35, 35+ kg/m2). We imputed the value for alcohol using the median value because non-drinkers and true missing values were grouped differently between studies. In subsequent analysis we tested associations using a missing category for alcohol instead of the imputed value and found no change in our physical activity results (all hazard ratios were within 0.02 of previous estimates). Questionnaires did not distinguish between “missing” and “no” for history of heart disease and cancer history; thus individuals were dichotomized into groups of yes or missing/no. Missing data was <5% for all covariates. We performed analyses calculating follow-up time in two ways: first, using age at study entry to age at death or end of follow up and second, calculating time from baseline questionnaire to date of death or end of follow-up. Because results did not differ from analyses using age as the time metric or using follow-up time and adjusting for age, in further analyses we used the latter method and adjusted for continuous age. The National Death Index, death certificates, or medical records were used to ascertain date of death (
