Associations of Leisure-Time Physical Activity and Television Viewing with Life Expectancy Cancer-Free at Age 50: The ARIC Study

Study population

We used data from the ARIC study, a prospective cohort of 15,792 mostly White and African-American adults. Participants, ages 44 to 66 years at visit 1, were enrolled from four geographic areas in the United States (Forsyth County, North Carolina; Jackson, Mississippi; Washington County, Maryland; and Minneapolis, Minnesota). ARIC study cohort members have participated in interviews, clinical examinations, and annual (semi-annual from 2012) telephone follow-up interviews over seven examination visits (visit 1 1987–1989, visit 2 1990–1992, visit 3 1993–1995, visit 4 1996–1998, visit 5 2011–2013, visit 6 2016–2017, and visit 7 2018–2019; ref. 23). All participants provided informed written consent at each study visit. The ARIC study was approved by the Institutional Review Boards at all participating institutions.

For this analysis, we excluded participants who did not consent to non-CVD research (n = 149), those with a history of any type of cancer at baseline (n = 902), and those who experienced cancer (colorectal, lung, prostate, or breast) or death within the first year of follow-up to reduce the likelihood of reverse causation (n = 130). Due to small numbers of race/ethnic groups at ARIC sites, we also excluded Asian and American Indian/Alaskan Indian participants (Forsyth County n = 21, Minnesota n = 15, Washington County n = 12) and African-American participants in Minnesota (n = 22) and Washington County (n = 33). After exclusions, 14,508 (92%) participants were included in analysis for colorectal and lung models, 6,582 men for prostate models, and 7,849 postmenopausal women for postmenopausal breast models (77 women were not included in breast cancer models because they died or developed breast cancer before reaching menopause).

Physical activity and sedentary behavior

Participants reported engaging in sport or exercise activities in the past year with the Baecke questionnaire at visits 1 and 3 (24, 25). Participants could report up to four leisure-time physical activities (LTPA). For each activity, participants were asked to report the number of hours/week and months/year they engaged in the activity. A metabolic equivalent of task (MET) value was assigned to each activity based on the compendium of physical activities (26). Activities with a MET ≥3 were classified as moderate-to-vigorous intensity (27). MET hours per week (MET-h/week) spent in moderate-to-vigorous LTPA at visits 1 and 3 was categorized as none, <median (0.1–<13.2 MET-h/week), and ≥median (13.2+ MET-h/week). The cut point was based on the visit 1 median value of MET-h/week among those reporting any LTPA. In a supplemental analysis, to further enhance the translation of our results, we classified participants as meeting the 2018 U.S. physical activity guidelines of participating in at least 150 minutes/week of moderate intensity activity, or 75 minutes/week of vigorous intensity activity, or an equivalent combination (27). We classified participants as no LTPA participation, LTPA less than guidelines (0.1−<7.5 MET-h/week), met guidelines (7.5–<15.0 MET-h/week), and doubled guidelines (15.0+ MET-h/week).

Sedentary behavior, any waking behavior that expends little energy expenditure (≤1.5 MET) while in a sitting or reclining posture (28), was estimated with a question on TV viewing. Participants reported at visit 1 how often they viewed TV and were classified as never/seldom, sometimes, and often/very often, due to small numbers in some categories.

Cancer and all-cause mortality outcomes

We focused on four leading types of cancer—colorectal, lung, prostate, and postmenopausal breast—which accounted for 59% of cancer diagnoses in ARIC. First primary invasive cancer diagnoses were ascertained by linkage with statewide cancer registries from 1987 to December 31, 2012, and supplemented by abstraction of medical records and hospital discharge codes. In addition to the cancer registry linkage, participants who self-reported a cancer diagnosis at an annual/semi-annual interview or at an ARIC study visit were contacted for further information and their medical records were abstracted to confirm a cancer diagnosis and tumor characteristics. Some of the state cancer registries were not complete or established at the start of the ARIC study, and for this time period, cancer cases were identified by surveillance of hospital discharge summaries in the ARIC study regions supplemented with medical record abstraction (29). The ARIC Cancer Coordinating Center adjudicated all potential cases of colorectal, lung, prostate, and postmenopausal breast cancers (29). For the purpose of exclusions, participants were considered to have prevalent cancer at visit 1 if they said a doctor told them that they had cancer of any type.

Deaths were ascertained from visit 1 until December 31, 2012, through active surveillance of vital status and by linkage with the National Death Index (23, 30).

Covariate assessment

Confounders were selected on the basis of a priori knowledge of the relationship of physical activity and sedentary behavior with cancer incidence and all-cause mortality (5, 21, 31–35). Covariates measured at visit 1 included age (continuous), gender (male, female), race by ARIC center (African-American Forsyth county, White Forsyth County, African-American Jackson, White Minneapolis, White Washington County), education (≤high school, vocational school, some college/college degree, graduate/professional), daily servings of red meat intake (continuous), and smoking pack-years (continuous; Supplementary Materials and Methods). Time-varying covariates that were assessed at visits 1 to 4 included: smoking status (current smoker, past smoker, never smoker), alcohol intake (not current drinker, ≤100 g, >100 g), body mass index (BMI; underweight/normal <25.0 kg/m², overweight 25.0–<30 kg/m², obese ≥30.0 kg/m²), diabetes (yes, no), and in women, use of menopausal hormone therapy (MHT; never users, former users of any MHT type, current users of unopposed estrogen, current users of estrogen plus progestin) and postmenopausal status (premenopausal, postmenopausal) (Supplementary Materials and Methods).

For the analysis on postmenopausal breast cancer, we included only postmenopausal women. Women were considered postmenopausal at the ARIC visit in which they reported the following: (i) having both ovaries removed; (ii) if both ovaries had not been removed, they were not taking hormones, and had not had a hysterectomy, then they were considered menopausal when they reported reaching menopause or having no periods in the past two years; or (iii) if both ovaries had not been removed but women reported taking hormones, had a hysterectomy, or did not know their menopause status, then they were considered menopausal when they reached the average cohort race- and smoking status–specific age when menopause was reached (White: never 48.3, former 47.3, current 46.6; African American: never 47.8, former 47.0, current 45.6 years).

Statistical analysis

We used multistate Markov survival models to estimate how participants moved between health (state 1), disease (state 2), and all-cause mortality (state 3) states (Fig. 1; Supplementary Materials and Methods and Supplementary Table S1). This model had three possible transitions—health to disease (T1), health to all-cause mortality (T2), and disease to all-cause mortality (T3). All participants started in the health state, free of cancer, and could move to the disease state (T1) or to death from any cause (T2). Participants who developed cancer could move from the disease state to the death state (T3). Separate models were specified with different disease states: colorectal, lung, prostate, and postmenopausal breast cancer. Transition-specific (T1, T2, T3) hazard ratios (HR) and 95% confidence intervals (CI) were estimated with the msm R package (36, 37). The time scale for all models was age; participants started contributing time to the study at the age they entered at visit 1. For breast cancer analyses, women started contributing time at the age of their first visit in which they were considered postmenopausal. Follow-up continued until death or end of study (December 31, 2012). For time-varying covariates, information from the visit closest in time preceding each type of transition was used for the specific transition. Exposures and covariates were used on all transitions (T1, T2, T3) and did not vary by type of transition.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

Three-state multistate models used for analysis. Four sets of models were estimated separately for each cancer site: colorectal (A), lung (B), prostate (C), and postmenopausal breast (D) cancer. In each type of model all participants started in the health state (free of cancer) and could move to state 2 if they developed the cancer under analysis or could move to state 3, death from any cause. Those who developed cancer could move from the cancer state to all-cause mortality from the cancer state. The arrows indicate transitions between each state and are labeled T1 (transition from state 1 to state 2), T2 (transition from state 1 to state 3), and T3 (transition from state 2 to state 3). Once a participant entered a state they remained in that state until they experienced a transition or were censored at the end of study follow-up. The number of participants who experienced each type of transition for each of the four models is in parentheses. T, transition.

Life expectancy, the expected average number of remaining years of life in health and disease states conditional on reaching age 50 and regardless of health status at age 50, was estimated with the R package Estimating Life Expectancies in Continuous Time (ELECT; ref. 38). The ELECT package uses model parameters estimated from the multistate models. Total life expectancy was divided into life expectancy with and without disease. CIs were estimated using 1,000 bootstrap samples. We focused on life expectancy cancer-free, which was derived from the T1 and T2 transitions from the multistate models and accounted for time from age 50 until incident cancer diagnosis, death, or end of study, whichever occurred first. We estimated life expectancy at age 50 to compare with studies on physical activity and life expectancy disease-free that all estimated life expectancy at age 50 (10–13, 15–18, 20). Across models, life expectancies were estimated at each level of LTPA (no LTPA, LTPA <median, and LTPA ≥median) and TV viewing (often/very often, sometimes, seldom/never) while specifying other covariates used in analytic models to the baseline covariate distribution levels of the analytic cohort.

First, we specified models that separately examined LTPA and TV, and then specified models that included both LTPA and TV. For all models, we adjusted for age, gender, race by ARIC center, education, smoking, and alcohol intake. We conducted sensitivity analyses (Supplementary Materials and Methods) that included further adjustment for BMI (all models), MHT (breast cancer LTPA model), red meat intake (colorectal LTPA model), diabetes (colorectal LTPA model), smoking pack-years (lung LTPA model), restriction to never smokers [colorectal, prostate, and breast cancer LTPA models, lung cancer models would not converge possibly due to the small number of lung cancer cases among never smokers (n = 10)], and exclusion of cases that occurred within 5 years of baseline (lung LTPA model).

Missing exposure and covariate data that occurred at each visit were imputed with multiple imputation by chained equations (Supplementary Materials and Methods; refs. 39, 40). All analyses were carried out with SAS Version 9.4 and R Version 3.3.2. Analyses were approved by the Institutional Review Board at the University of North Carolina at Chapel Hill.