The hazard ratio (HR) (95% CI) was adjusted for participation in weight lifting (yes vs no) and the same set of variables as denoted in Table 2; was 0.83 (95% CI, 0.74-0.92) at 11 metabolic equivalent of task (MET)-hours/week, 0.68 (95% CI, 0.56-0.83) at 30 MET-hours/week, and 0.68 (95% CI, 0.56-0.82) at 42 MET-hours/week. P = .02 for the fit of the nonlinear vs linear model; P < .001 for overall significance of the curve.
eTable 1. HR and 95% CI of Cancers at Subsites of Upper Digestive Tract by Total Physical Activity
eTable 2. HR and 95% CI of Upper Digestive Tract Cancers by Total Physical Activity after Excluding Individuals Who Had Family History of Cancer or Underwent Physical Examination for Screening Purpose
eTable 3. HR and 95% CI for Joint Associations of Amount of Aerobic Exercise/Weight Lifting and Participation in Weight Lifting with Cancers at Digestive System Subsites
eTable 4. HR and 95% CI for Joint Associations of Amount of Walking/Jogging/Running and Intensity with Cancers at Digestive System Subsites
eFigure 1. HR and 95% CI for Total Physical Activity and Upper Digestive Tract Cancer Risk
eFigure 2. Relationship between Aerobic Physical Activity and Risk of Cancers at Digestive System Subsites
Customize your JAMA Network experience by selecting one or more topics from the list below.
Keum N, Bao Y, Smith-Warner SA, et al. Association of Physical Activity by Type and Intensity With Digestive System Cancer Risk. JAMA Oncol. 2016;2(9):1146–1153. doi:10.1001/jamaoncol.2016.0740
Is physical activity a common protective factor for digestive system cancers, and what is the optimal amount, type, and intensity of physical activity?
In this prospective cohort study of 43 479 men, physical activity was inversely associated with digestive system cancer risk. Aerobic exercise regardless of intensity was particularly beneficial, reaching a plateau in benefit at approximately 30 metabolic equivalent of task hours per week with 32% risk reduction.
The optimal exercise regime to prevent digestive system cancers may be to accumulate the equivalent of energy expenditure of 10 hours of walking at an average pace per week primarily through aerobic physical activity of any intensity.
Accumulating evidence indicates that common carcinogenic pathways may underlie digestive system cancers. Physical activity may influence these pathways. Yet, to our knowledge, no previous study has evaluated the role of physical activity in overall digestive system cancer risk.
To examine the association between physical activity and digestive system cancer risk, accounting for amount, type (aerobic vs resistance), and intensity of physical activity.
Design, Setting, and Participants
A prospective cohort study followed 43 479 men from the Health Professionals Follow-up Study from 1986 to 2012. At enrollment, the eligible participants were 40 years or older, were free of cancer, and reported physical activity. Follow-up rates exceeded 90% in each 2-year cycle.
The amount of total physical activity expressed in metabolic equivalent of task (MET)-hours/week.
Main Outcomes and Measures
Incident cancer of the digestive system encompassing the digestive tract (mouth, throat, esophagus, stomach, small intestine, and colorectum) and digestive accessory organs (pancreas, gallbladder, and liver).
Over 686 924 person-years, we documented 1370 incident digestive system cancers. Higher levels of physical activity were associated with lower digestive system cancer risk (hazard ratio [HR], 0.74 for ≥63.0 vs ≤8.9 MET-hours/week; 95% CI, 0.59-0.93; P value for trend = .003). The inverse association was more evident with digestive tract cancers (HR, 0.66 for ≥63.0 vs ≤8.9 MET-hours/week; 95% CI, 0.51-0.87) than with digestive accessary organ cancers. Aerobic exercise was particularly beneficial against digestive system cancers, with the optimal benefit observed at approximately 30 MET-hours/week (HR, 0.68; 95% CI, 0.56-0.83; P value for nonlinearity = .02). Moreover, as long as the same level of MET-hour score was achieved from aerobic exercise, the magnitude of risk reduction was similar regardless of intensity of aerobic exercise.
Conclusions and Relevance
Physical activity, as indicated by MET-hours/week, was inversely associated with the risk of digestive system cancers, particularly digestive tract cancers, in men. The optimal benefit was observed through aerobic exercise of any intensity at the equivalent of energy expenditure of approximately 10 hours/week of walking at average pace. Future studies are warranted to confirm our findings and to translate them into clinical and public health recommendation.
Digestive system cancers (DSCs) include cancers of the digestive tract (mouth, throat, esophagus, stomach, small intestine, colorectum) and cancers of digestive accessory organs (pancreas, gallbladder, liver). While individual DSCs are etiologically heterogeneous, accumulating evidence suggests the potential presence of common carcinogenic pathways underlying DSCs. The most substantiated is the proinflammatory pathway mediated by cyclooxygenase (COX)-2 enzyme. For instance, a functional polymorphism in the COX-2 gene was associated with an increased risk of DSCs, in particular.1 Furthermore, observational studies and randomized clinical trials consistently observed that aspirin, which inhibits COX-2, was notably protective against DSCs.2,3 Thus, factors that affect the COX-2 pathway may modify the risk of overall DSC. One potential candidate is physical activity (PA), which is shown to reduce the COX-2 proinflammatory pathway.4 In addition, the benefit of PA on colorectal cancer survival was confined to COX-2–positive cancers.5
Currently, the American Cancer Society advises that adults engage in at least 150 minutes of moderate PA, or 75 minutes of vigorous PA, or an equivalent combination per week for cancer prevention.6 However, this guideline was not specifically designed for DSCs, and potential interactions between domains of PA (overall energy expenditure, type, and intensity) remain to be investigated to help refine the recommendations. Considering that PA is a modifiable lifestyle factor and that DSCs accounted for an estimated 18% of cancer incidence and 26% of cancer deaths, excluding nonmelanoma skin cancer, in the United States in 2012,7 investigating the relationship between PA and DSC risk accounting for diverse domains of PA has an important public health implication.
The Health Professionals Follow-up Study is an ongoing cohort study that began in 1986, enrolling 51 529 US male health professionals ages 40 to 75 years.8 Participants completed a baseline questionnaire and biennial follow-up questionnaires, with follow-up rates exceeding 90% in each 2-year cycle. The study was approved by the human subjects committees at Brigham and Women's Hospital and Harvard T. H. Chan School of Public Health. All participants provided written informed consent; they did not receive compensation.
For this analysis, the baseline was defined as 1986 when PA was first assessed. At baseline, we excluded men having a prior cancer diagnosis (n = 2000) or missing data on PA (n = 216), body mass index (BMI) (n = 2090), or dietary information (n = 1595). To minimize potential bias from reverse causation, we also excluded men reporting difficulty with walking or stair climbing at any point during the follow-up since 1988 (n = 2149), the first year when this information was collected. The final analytic cohort included 43 479 men.
In 1986 and every 2 years thereafter, participants reported mean time spent per week in the previous year at each of the following 9 PAs: walking, jogging, running, bicycling including stationary machine, lap swimming, tennis, squash and/or racket ball, calisthenics and/or rowing, and outdoor work. Participants also reported their usual walking pace as easy, average, brisk, or very brisk and number of flights of stairs climbed per day. Questions on weight lifting were added in 1990 and asked every 2 years thereafter. Each PA was asked in 10 response categories ranging from none to 11 to 20 hours per week.
For each PA, weekly energy expenditure was estimated by multiplying the typical intensity expressed in metabolic equivalent of task (MET) (the ratio of metabolic rate during the activity to metabolic rate at rest)9 by the reported hours spent per week. For walking, we assigned 2.0, 3.0, 4.0, and 4.5 METs to casual, average, brisk, and very brisk pace, respectively, and assumed 3.0 METs for pace missing. Across PAs, the weekly MET-hour scores were summed to derive our primary exposure (total weekly energy expenditure attributable to overall discretionary PAs in units of MET-hours/week).
By type of PA, aerobic PAs included walking at least at an average pace, jogging, running, bicycling, swimming, tennis, squash and/or racket ball, calisthenics and/or rowing, and stair climbing; resistance PA was marked by weight lifting.
The questionnaire was validated against 4 single-week activity diaries administered across 4 different seasons (correlation, 0.65) and against resting heart rate (correlation, –0.45). The test-retest correlation in a 2-year span was 0.41.10
A validated semiquantitative food frequency questionnaires listing more than 130 food items11,12 was administered every 4 years to assess dietary intake over the past year. Biennial questionnaires collected information on potential cancer risk factors including age, race, smoking status, history of diabetes mellitus (DM), family history of cancer, screening physical examination, endoscopy, aspirin use, and multivitamin use. We calculated BMI using self-reported height and weight. Missing values were handled by carrying forward information from the previous cycle or using missing category, but the proportion of missing was negligible for most covariates (<1%).
Our primary end point was incident DSC as defined herein, which was reported through biennial follow-up questionnaires through 2012. Physicians blinded to the participants’ exposure status reviewed medical records to confirm self-reported cancer diagnosis.
Hazard ratios (HRs) and 95% CIs were estimated with Cox proportional hazards models using age as the underlying time scale.13 Person-time of follow-up was accrued from the return date of the baseline questionnaire (1990 questionnaire for analyses involving weight lifting) until the date of cancer diagnosis, death from any cause, or analysis end (2012) as determined by availability of data ready for analysis, whichever came first. To reduce potential bias from reverse causation, we added a 2-year latency period between PA and cancer incidence. Potential confounders in multivariable analyses were determined a priori from established and suspected risk factors for DSC. To better represent the long-term average and to minimize random within-person variation, values for PA and potential confounders were updated using cumulative average whenever new information was obtained from the follow-up questionnaires.
For categorical analysis of total PA, we created 5 categories (≤8.9, 9.0-20.9, 21.0-41.9, 42.0- 62.9, and ≥63.0 MET-hours/week). The cutoffs were chosen in multiples of 3 MET-hours per week (equivalent to 1 hour/week of walking at an average pace) so that each category represents PA in terms of the most popular and safest type of exercise in a realistic and achievable range (≤0.3, 0.4-0.9, 1.0-1.9, 2.0-2.9, and ≥3.0 hours per day of walking at an average pace), which facilitates translation of our findings into public health recommendations. A potential linear trend was examined by assigning each exposure category the median value and by treating this variable as a continuous variable. Because obesity and DM may be potential intermediates,14-17 our primary multivariable analysis did not control for them. To estimate the degree to which the association of PA with cancer incidence was explained by obesity and DM, we ran another multivariable model including these variables as well. Potential heterogeneity in the relationship by digestive tract cancers vs digestive accessory organ cancers was checked using competing Cox proportional hazards model with a data augmentation method.18 As sensitivity analyses to examine potential influence of residual confounding, we ran the primary multivariable analyses across strata of potential confounders after excluding health-conscious individuals as indicated by having family history of cancer or undergoing screening physical examination. There was no evidence of departure from the proportional hazard assumption, given P > .05 from the Wald test performed for an added interaction term between continuous PA and continuous age.
Given the same amount of energy expenditure from PA, the effect of PA on DSCs may vary by PA type (aerobic vs resistance). Thus, we examined the joint association of amount and type, using weekly MET-hour scores from aerobic PA or weight lifting and participation in weight lifting.
For aerobic PA, the dominant type of PA in our study population, we ran more detailed analyses. First, the dose-response relationship was examined with restricted cubic splines.19 To reduce the influence of extreme values, we excluded individuals above the 95th percentiles of aerobic PA level. Three knots were fixed at the 5th, 50th, and 95th percentiles of the remaining aerobic PA level and reference was set to 1 MET-hours/week. The potential for nonlinearity was evaluated using the likelihood ratio test comparing the fit of 2 nested multivariable models: linear and nonlinear models. Second, to explore if achieving a level of energy expenditure through intense exercise was more beneficial, we examined the joint association of amount and intensity of aerobic PA. In this analysis, activities included were restricted to walking at least at an average pace, jogging, and running, which represent similar styles of aerobic PA with different intensity. Amount was estimated by weekly MET-hour scores from the 3 PAs; intensity was estimated by average MET of the 3 PAs weighted by hours spent for each PA. We dichotomized the intensity score at 4.5, with 4.5 or lower representing mostly walking inclusive of all walking paces and higher than 4.5 representing more intense aerobic PA.
For all the joint analyses, a relevant MET-hour score lower than 3, irrespective of type or intensity of PA, was set as the reference because it represents a negligible level of PA. All the statistical tests were 2-sided, and P < .05 was considered statistically significant. Analyses were performed using SAS statistical software (version 9.3; SAS Institute Inc).
The 43 479 men contributed 686 924 person-years from 1986 to 2012 and 1370 incident cases of DSCs (1070 digestive tract cancers and 300 digestive accessory organ cancers) from 1988 to 2012. Across all levels of total PA, men accumulated total MET-hours/week primarily through aerobic PA. More physically active men engaged in more intense aerobic PA and more weight lifting. They also tended to be older, leaner, never smokers, and nondiabetics; to have a family history of cancer; to undergo physical examinations for screening; to take aspirin and multivitamins; and to consume higher levels of total calories, alcohol, whole grains, fruits, and vegetables (Table 1).
Total PA was inversely associated with DSC risk, which was largely driven by digestive tract cancers (Table 2). The age-adjusted and multivariable results were similar, with the multivariable HR comparing 63.0 or more vs 8.9 or less MET-hours/week being 0.74 (95% CI, 0.59-0.93; P value for trend = .003) for DSCs and 0.66 (95% CI, 0.51-0.87; P value for trend = .003) for digestive tract cancers. The corresponding results for digestive accessory organ cancers were 1.02 (95% CI, 0.66-1.58; P value for trend = .46), but heterogeneity in the linear association across digestive tract cancers and digestive accessory organ cancers was not statistically significant (P value for heterogeneity = .47). The results were not altered appreciably after additional adjustments for BMI and DM. Within digestive tract cancers, a strong association was observed for cancers of the upper digestive tract from mouth to small intestine (HR for ≥63.0 vs ≤8.9 MET-hours/week, 0.50; 95% CI, 0.30-0.83; P value for trend = .02) and persisted independent of BMI and DM. An inverse relationship was consistently observed for individual cancers of the digestive tract (see eTable 1 in the Supplement), in stratified analyses by potential confounders (see eFigure 1 in the Supplement) and after excluding those with family history of cancer or screening physical examination (see eTable 2 in the Supplement), although its statistical significance was sensitive to case numbers in each analysis. For colorectal cancer, statistical significance of an inverse trend was lost when BMI was additionally adjusted for, but further adjustment for DM had little influence on the results.
By type of PA comparing aerobic PA vs weight lifting (Table 3), the greatest reduction in DSC risk occurred in the most active group exclusively engaging in aerobic PA (HR, 0.54; 95% CI, 0.41-0.72). The dose-response curve controlled for weight lifting and other confounders suggested that most of the benefit on DSC risk occurred up to approximately 30 MET-hours/week of aerobic PA, with little incremental benefit thereafter (P value for nonlinearity = .02) (Figure). Compared with the risk at 1 MET-hours/week, the risk reduced by 32% at 30 MET-hours/week (HR, 0.68; 95% CI, 0.56-0.83). Regarding intensity of aerobic PA, as long as the same MET-hour score was achieved, the benefit was similar irrespective of whether the activity was walking at least at an average pace, jogging, or running (Table 4). The dominant benefit of aerobic PA, nonlinearity of curve, and irrelevance of intensity (see eTable 3, eFigure 2, and eTable 4 in the Supplement, respectively) were more consistently observed with digestive tract cancers than with digestive accessory organ cancers.
In men, a higher level of total PA as indicated by MET-hours/week was associated with a lower risk of DSCs, particularly digestive tract cancers. By type of PA, aerobic PA rather than resistance PA underlay the inverse association, with its benefit plateauing after about 30 MET-hours/week. Moreover, as long as the same level of MET-hour score was achieved from aerobic PA, the magnitude of risk reduction was similar regardless of intensity of aerobic PA.
Several mechanisms may explain our findings. Physical activity is associated with reduced levels of circulating insulin and bioavailable insulin-like growth factor 1, the 2 major mitogenic hormones implicated in carcinogenesis.20 Physical activity is also known to enhance anticancer immune function and antioxidant defenses.20 Some mechanisms may be more specific to digestive tract cancers. COX-2 is overexpressed in several digestive tract cancers,21-24 and PA reduces COX-2 mediated inflammatory response.4 In addition, PA reduces exposure of the digestive tract to carcinogens by stimulating gastrointestinal tract motility and thereby reducing gastrointestinal tract transit time.20 Finally, a higher predicted score of plasma 25-hydroxy-vitamin D level was largely associated with a reduced incidence for digestive tract cancers but not for other cancers in our cohort.25 Physical activity, particularly outdoor PA, may reduce digestive tract cancer risk by improving vitamin D status through sun exposure.
Despite the biological plausibility, our current understanding on the relationship between PA and DSC risk is largely limited. The World Cancer Research Fund/American Institute for Cancer Research expert panel26 judged that the evidence for a causal effect of PA is convincing only for colon cancer. Our study provides additional insight into digestive tract cancers that an inverse association of PA with upper digestive tract cancers was strong and independent of BMI and DM, whereas that with colorectal cancer was partially explained by BMI. The differential degree to which BMI explains the associations is in part attributable to the fact that excess adiposity is an established risk factor for colorectal cancer,16 while it forms divergent relationships with upper digestive tract cancers.16,27 Alternatively, in light of the fact that primary exposure to foods containing carcinogens or pro-oxidants occurs along the upper digestive tract, reduced gastrointestinal tract transit time and enhanced antioxidant defense promoted by PA20 might be particularly relevant to upper digestive tract cancers. Indeed, in our stratified analyses, an inverse association was pronounced among unhealthy eaters who are more likely vulnerable to carcinogenic and oxidative damage (see eFigure 1 in the Supplement). A recent meta-analysis also found an inverse relation of PA with gastroesophageal cancers independent of adiposity.28
Aerobic and resistance exercises represent 2 major types of PA. While both forms help regulate blood levels of glucose and lipids, aerobic PA is particularly effective at improving cardiorespiratory fitness, whereas resistance PA is at increasing muscular strength.29 To date, the relative importance of aerobic vs resistance PA in relation to cancer risk has not been a focus of study, but our findings suggest that those with low overall PA prioritize aerobic PA over resistance PA to achieve a greater reduction in DSC risk. As a potential explanation, cardiorespiratory fitness, which has been inversely associated with DSC mortality,30 may be the relevant parameter. In addition, aerobic PA may be more beneficial to gastrointestinal tract motility31 and antioxidant defenses,20 important determinants of DSC risk.
Our finding that intensity of aerobic PA does not matter as much as amount of aerobic MET-hour scores achieved supports the PA guideline for cancer prevention.6 The 2 exercise regimes recommended by the American Cancer Society, while based on different intensity and duration of PA, translate into a recommendation to accumulate roughly at least 11 MET-hours/week.6 Our dose-response curve estimates that individuals meeting the guideline through aerobic PA can reduce DSC risk at least by 17%, with the optimal benefit of 32% risk reduction achieved at around 30 MET-hours/week. Given the irrelevance of intensity of aerobic exercise, walking about 90 minutes per day at an average pace, an exercise regime that can be readily incorporated into everyday life, may be sufficient to reap the optimal benefit of PA against DSCs.
There are several limitations in our study. First, we might have underestimated the true benefit of PA against DSC because of measurement error in self-reported PA. However, our use of the cumulative average of PA assessed at multiple time points reduced the degree of measurement error. Second, our total MET-hour score might have underestimated overall PA because the questionnaire included select activities. Nevertheless, previous studies from our cohort wherein self-reported PA predicted diverse disease outcomes32-34 provide qualitative evidence for the capability of our questionnaire to discriminate PA levels. Third, lower MET-hours/week for weight lifting than for aerobic exercise may have contributed to greater benefit observed with aerobic exercise. However, the degree of weight lifting in our cohort was sufficient enough to predict other diseases35,36 and to form even a stronger association than aerobic exercise with waist circumference maintenance.32 Finally, owing to heterogeneity in the magnitude of the association by cancer site, our findings are not generalizable to a population with a markedly different distribution of DSC incidence.
Nevertheless, our study has strengths. We examined potential interactions between diverse domains of PA in relation to DSC risk. By cumulatively updating PA, we reflected a long-term average, which is likely more relevant to the carcinogenic process spanning several decades. While our cohort consisting of primarily white male health professionals may limit the representativeness of the findings, our estimates are less susceptible to confounding by health-conscious lifestyles and socioeconomic status. Furthermore, because risk factors vary across DSCs, they form a weaker association with a composite end point, which reduces the degree of confounding in our analyses. Also, given the large effect size, the observed association from our multivariable analyses is unlikely to be entirely attributable to confounding.
Physical activity as indicated by MET-hour score may be inversely associated with the risk of DSCs, particularly digestive tract cancers, in men, with the optimal benefit observed through aerobic exercise of any intensity at the equivalent of energy expenditure of approximately 10 hours of walking at an average pace per week. Future studies are warranted to further investigate etiological mechanisms underlying the optimal dose, type, and intensity of exercise and to evaluate preventive potential of exercise against overall DSCs.
Corresponding Author: NaNa Keum, ScD, Harvard T. H. Chan School of Public Health Department of Nutrition, Building 2, Third Floor, 665 Huntington Ave, Boston, MA 02115 (firstname.lastname@example.org).
Published Online: May 19, 2016. doi:10.1001/jamaoncol.2016.0740
Correction: This article was corrected on March 7, 2019, to correct omissions in the Conflict of Interest Disclosures.
Accepted for Publication: March 1, 2016.
Author Contributions: Dr Keum had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Keum, Fuchs, Giovannucci.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Keum.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Keum, Bao, Orav.
Obtained funding: Fuchs, Giovannucci.
Study supervision: Fuchs, Giovannucci.
Conflict of Interest Disclosures: Dr Fuchs has been a consultant and/or a scientific advisor for Eli Lilly, Entrinsic Health, Pfizer, Merck, Sanofi, Roche, Genentech, Merrimack Pharmaceuticals, Dicerna, Bayer, Celgene, Agios, Gilead Sciences, Five Prime Therapeutics, Taiho, and KEW. No other disclosures were reported.
Funding/Support: This work was supported by National Institutes of Health (NIH) grant UM1 CA167552. Dr Bao is funded by NIH grants U54 CA155626, P30 DK046200, and KL2 TR001100.
Role of the Funder/Sponsor: The NIH had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Additional Contributions: We thank the participants and staff of the HSPF for their valuable contributions. Staff are paid, and participants have no conflict of interest. We also thank the following state cancer registries for their help: Alabama, Arizona, Arkansas, California, Colorado, Connecticut, Delaware, Florida, Georgia, Idaho, Illinois, Indiana, Iowa, Kentucky, Louisiana, Maine, Maryland, Massachusetts, Michigan, Nebraska, New Hampshire, New Jersey, New York, North Carolina, North Dakota, Ohio, Oklahoma, Oregon, Pennsylvania, Rhode Island, South Carolina, Tennessee, Texas, Virginia, Washington, and Wyoming.