Abstract
Background
Previous estimates of the colorectal cancer (CRC) burden attributed to behaviors have not considered joint effects, competing risk, or population subgroup differences.
Methods
We pooled data from seven prospective Australian cohort studies (n = 367 058) and linked them to national registries to identify CRCs and deaths. We estimated the strength of the associations between behaviors and CRC risk using a parametric piecewise constant hazards model, adjusting for age, sex, study, and other behaviors. Exposure prevalence was estimated from contemporary National Health Surveys. We calculated population attributable fractions for CRC preventable by changes to current behaviors, accounting for competing risk of death and risk factor interdependence. Statistical tests were two-sided.
Results
During the first 10 years of follow-up, there were 3471 incident CRCs. Overweight or obesity explained 11.1%, ever smoking explained 10.7% (current smoking 3.9%), and drinking more than two compared with two or fewer alcoholic drinks per day explained 5.8% of the CRC burden. Jointly, these factors were responsible for 24.9% (95% confidence interval [CI] = 19.7% to 29.9%) of the burden, higher for men (36.7%) than women (13.2%, Pdifference < .001). The burden attributed to these factors was also higher for those born in Australia (28.7%) than elsewhere (16.8%, Pdifference = .047). We observed modification of the smoking-attributable burden by alcohol consumption and educational attainment, and modification of the obesity-attributable burden by age group and birthplace.
Conclusions
We produced up-to-date estimates of the future CRC burden attributed to modifiable behaviors. We revealed novel differences between men and women, and other high–CRC burden subgroups that could potentially benefit most from programs that support behavioral change and early detection.
Australia has a high incidence of colorectal cancer (CRC) (1). Ever smoking, being physically inactive, being overweight or obese, and consuming processed meat and excessive alcohol are established to increase CRC risk (2,3). Risk may also be increased by consuming red meat and inadequate whole grains, dietary fiber, and dairy. These behaviors often co-occur in individuals (4–9), and the burden related to one risk factor may be mediated by others. Typically, disease burden estimates do not take into account the simultaneous effects of other factors or the interdependence of effects, nor do they account for competing risk of death (10,11). To maximize their accuracy and policy relevance, population attributable fractions (PAFs) are best estimated from prospective cohort studies (12) and up-to-date risk factor prevalence estimates representative of the population of interest.
We addressed this evidence gap by applying a comprehensive PAF method to an Australian cohort consortium and contemporaneous representative exposure prevalence data.
Methods
Study Population
We used individual-level data from the Australian cancer-PAF cohort consortium (13), which comprises the Melbourne Collaborative Cohort Study (MCCS) (14), Blue Mountains Eye Study (BMES) (15), Australian Longitudinal Study on Women’s Health (ALSWH) (16), Australian Diabetes, Obesity and Lifestyle Study (AusDiab) (17), North West Adelaide Health Study (NWAHS) (18), Concord Health and Ageing in Men Project (CHAMP) (19), and the 45 and Up Study (45&Up) (20). The combined cohort sample was 369 515 adult Australians. The analytic sample was 360 489 individuals, after excluding 2457 who enrolled in more than one cohort, 1885 who did not consent to record linkage, and 4684 with a history of CRC (Table 1).
Table 1.
Cohort data |
External prevalence data |
|||||||||
---|---|---|---|---|---|---|---|---|---|---|
Characteristic | MCCS | BMES | ALSWH | AusDiab | NWAHS | CHAMP | 45&Up | Pooled | NHS | NDSHS |
Baseline year(s) | 1990–1994 | 1992–1993 | 1996 | 1999–2000 | 1999–2003 | 2005–2007 | 2006–2009 | 1990–2008 | 2014–2015 | 2013 |
Population, No. | 41 328 | 3623 | 38 192 | 11 136 | 4012 | 1566 | 260 632 | 360 489 | 14 560 | 22 696 |
Incident CRC cases, No.* | 600 | 80 | 397 | 133 | 40 | 40 | 2181 | 3471 | ||
Deaths, No.* | 2095 | 682 | 2575 | 779 | 280 | 416 | 13 313 | 20 140 | ||
State/territory | VIC | NSW | All | All | SA | NSW | NSW | All | All | All |
Age at baseline, mean (range), y | 55 (27–76) | 66 (45–97) | 45† (18–75) | 51 (25–91) | 50 (18–90) | 77 (70–96) | 62 (45–>100) | 59 (18–>100) | 46 (18–85) | 46 (18–84) |
Women, % | 59 | 57 | 100 | 55 | 52 | 0 | 54 | 59 | 51 | 51 |
During the first 10 years of follow-up. 45&Up = 45 and Up Study; ALSWH = Australian Longitudinal Study on Women's Health; AusDiab = Australian Diabetes, Obesity and Lifestyle Study; BMES = Blue Mountains Eye Study; CHAMP = Concord Health and Ageing in Men Project; MCCS = Melbourne Collaborative Cohort Study; NDSHS = National Drug Strategy Household Survey; NHS = National Health Survey; NSW = New South Wales; NWAHS = North West Adelaide Health Study; SA = South Australia; VIC = Victoria.
The ALSWH recruited three cohorts age 18–23, 45–50, and 70–75 years, so the age distribution is not continuous.
We obtained the most recent available risk factor prevalence estimates from the 2014–2015 National Health Survey (21) and the 2013 National Drug Strategy Household Survey (Table 1; Supplementary Table 1, available online) (22). The Australian Institute of Health and Welfare ethics committee approved the study (EC2013/4/62).
Data Harmonization
We examined modifiable behaviors with convincing or probable evidence of a causal association with CRC, as judged by expert review panels (2,3), if they were measured in our cohort and the national health surveys. These exposures were smoking, physical inactivity, body fatness (approximated by body mass index [BMI]), and excessive alcohol consumption at baseline (cohort entry). For smoking, we examined status, time since quitting (lag time, in decades), and intensity for current smokers. We could not estimate PAF for consumption of red or processed meat or inadequate consumption of whole grains, dietary fiber or dairy, as data on these exposures were not measured by the health surveys. We harmonized the exposures across the cohort studies and health surveys, classifying them in accordance with current Australian recommendations for healthy living, that is, not smoking, doing at least 150 minutes of moderate physical activity or 75 minutes of vigorous physical activity per week, maintaining a healthy weight (BMI 18.5–25 kg/m2), and drinking two or fewer alcoholic drinks per day (13). We also harmonized country of birth, marital status, educational attainment, socioeconomic status (23), and residential location (rurality) (24) (Supplementary Table 1, available online).
Data Linkage
We matched the pooled cohort to the Australian Cancer Database (25) and National Death Index to identify cancers and deaths using probabilistic linkage (26). These records were available until December 31, 2012, providing eight to 22 years of follow-up (Table 1).
Statistical Methods
We classified incident primary invasive CRCs of epithelial cell origin according to International Classification of Diseases for Oncology codes (ICD-O; C18-20), with subclassification to the colon (C18.0–18.9) and rectum (C19–20).
We defined follow-up as the time from baseline to the date of CRC diagnosis, death, or end of follow-up, whichever occurred first. We estimated the strength of the association between the behaviors and CRC and death using a parametric piecewise constant hazards model (27) and expressed them as hazard ratios (HR) with 95% confidence intervals (CIs). We restricted the analyses to the first 10 years of follow-up to generate comparable estimates across the cohorts and tested heterogeneity among the cohort-specific hazard ratios using the asymptotic DerSimonian and Laird Q statistic (28). In sensitivity analyses, we excluded the first 12 months of follow-up to evaluate the potential impact of reverse causality. We also adjusted our risk estimates by processed meat and red meat consumption, measured in the two largest cohorts, and CRC family history and screening, collected in the largest cohort.
We predefined two main effects models. The first model included age, sex, study, and each behavior separately. The second model included age, sex, study, and all behaviors statistically significantly associated with CRC. We computed the corresponding exposure prevalence (PR) estimates from the health surveys. We then combined the strength of the association and exposure prevalence estimates to calculate the PAF point estimates using our recently developed PAF formula (12). This formula defines PAF for cancer incidence as the expected excess cancer incidence during the follow-up time due to certain modifiable risk factors, while accounting for censoring due to death. This is done by comparing the probabilities that an individual is alive and disease-free given the original and modified risk factor values. The asymptotic variance estimate of PAF was obtained using the delta method, and two-sided 95% confidence intervals for PAFs were calculated by applying a symmetrizing complementary logarithmic transformation of PAF.
We calculated PAFs both for the individual and joint contributions of behaviors to the CRC burden. We evaluated scenarios in which the exposure was completely eliminated or only reduced. For example, we evaluated the scenario in which current or former smokers (eventually) had the same CRC risk as never smokers, and also the scenario in which current smokers of 20 or more cigarettes per day consumed instead fewer than 20 cigarettes per day. We estimated the number of Australian CRC cases that could be prevented by multiplying the PAF estimates by the projected numbers of CRCs over the next 10 years (2017–2026) (29).
We tested for potential effect modification of PAFs by other behaviors and sociodemographic factors. This was performed by including an interaction term between the risk factor and the potential effect-modifying factor in the model and by calculating the 95% confidence interval of the difference of the PAF estimates between the categories of the effect-modifying factor (30). The PAF difference between subgroups was deemed statistically significant if the P value for the difference was less than .05. We also calculated PAFs by CRC topography and using the traditional PAF method (11).
We carried out all statistical analyses using SAS 9.4 (SAS Institute, Inc., Cary, NC) and our publicly available PAF program based on SAS macros (31). All statistical tests were two-sided.
Results
We observed 3471 incident CRCs and 20 140 deaths during the first 10 years of follow-up (Table 1).
CRC Behavioral Risk Factors
We found no heterogeneity among the cohort-specific hazard ratios of CRC in relation to behaviors (Supplementary Table 2, available online).
Taking men and women together, CRC risk was positively associated with smoking, overweight and obesity (BMI ≥25 kg/m2), and excessive alcohol consumption but was not associated with physical inactivity in the multivariable-adjusted model (Table 2). The risk for former smokers was virtually identical to that for current smokers, and for former smokers the risk was elevated up to 40 years after cessation. In terms of current smoking frequency, only those who smoked 20 or more cigarettes per day were at increased risk. The strength of these associations was modest at most and did not materially change after excluding the first 12 months of follow-up or in subset analyses adjusted for consumption of processed meat or red meat, or CRC family history or screening (data not shown). The associations were also largely unchanged after adjustment for country of birth, educational attainment, socioeconomic status, and residential location (Supplementary Table 3, available online). The findings were broadly similar for colon and rectal cancers (Supplementary Tables 4 and 5, available online).
Table 2.
Behavioral risk factor | Prevalence, % |
HR (95% CI)* |
P† | ||||
---|---|---|---|---|---|---|---|
All | Men | Women | All | Men | Women | ||
Smoking status | |||||||
1. Never smoker | 53 | 45 | 60 | 1 | 1 | 1 | |
2. Former smoker | 31 | 36 | 27 | 1.23 (1.14 to 1.33) | 1.25 (1.12 to 1.39) | 1.22 (1.09 to 1.36) | |
3. Current smoker | 16 | 19 | 13 | 1.26 (1.10 to 1.44) | 1.30 (1.07 to 1.57) | 1.23 (1.02 to 1.49) | |
PAF (2–3 → 1) | 10.7 (6.8 to 14.4) | 13.0 (6.7 to 18.9) | 8.4 (3.8 to 12.7) | .22 | |||
PAF (2 → 1) | 6.8 (4.3 to 9.3) | 8.0 (4.1 to 11.8) | 5.5 (2.3 to 8.6) | .33 | |||
PAF (3 → 1) | 3.9 (1.4 to 6.3) | 5.0 (1.0 to 8.9) | 2.8 (0.01 to 5.6) | .37 | |||
Current smoking frequency | |||||||
1. Never | 56 | 52 | 61 | 1 | 1 | 1 | |
Former smoker, who quit: | |||||||
2. 40+ y ago | 2 | 3 | 1 | 0.98 (0.82 to 1.16) | 0.99 (0.81 to 1.22) | 0.98 (0.71 to 1.36) | |
3. 30–39 y ago | 3 | 4 | 3 | 1.15 (1.00 to 1.33) | 1.13 (0.94 to 1.35) | 1.22 (0.96 to 1.54) | |
4. 20–29 y ago | 5 | 5 | 5 | 1.20 (1.06 to 1.36) | 1.22 (1.03 to 1.43) | 1.19 (0.98 to 1.45) | |
5. 10–19 y ago | 6 | 6 | 6 | 1.30 (1.15 to 1.47) | 1.38 (1.17 to 1.63) | 1.19 (0.97 to 1.45) | |
6. <10 y ago | 10 | 10 | 10 | 1.46 (1.28 to 1.66) | 1.52 (1.27 to 1.81) | 1.40 (1.15 to 1.70) | |
Current smoker, cigs/d | |||||||
7. 0–19 | 14 | 16 | 11 | 1.11 (0.91 to 1.36) | 1.20 (0.89 to 1.60) | 1.06 (0.81 to 1.37) | |
8. ≥20 | 4 | 5 | 4 | 1.45 (1.21 to 1.74) | 1.43 (1.22 to 1.83) | 1.51 (1.17 to 1.96) | |
PAF (8 → 7) | 1.3 (0.03 to 2.6) | 1.0 (–1.0 to 3.1) | 1.5 (–0.01 to 3.0) | .73 | |||
BMI, kg/m2 | |||||||
1. <18.5 | 2 | 1 | 2 | 1.04 (0.78 to 1.40) | 1.28 (0.74 to 2.22) | 0.93 (0.66 to 1.33) | |
2. 18.5–24.9 | 35 | 28 | 42 | 1 | 1 | 1 | |
3. 25.0–29.9 | 36 | 42 | 29 | 1.11 (1.02 to 1.20) | 1.27 (1.13 to 1.44) | 0.97 (0.87 to 1.09) | |
4. ≥30.0 | 28 | 28 | 27 | 1.30 (1.19 to 1.44) | 1.49 (1.29 to 1.72) | 1.18 (1.03 to 1.34) | |
PAF (3–4 → 2) | 11.1 (6.4 to 15.6) | 20.4 (13.1 to 27.1) | 4.0 (–1.9 to 9.5) | <.001 | |||
PAF (3 → 2) | 3.4 (0.6 to 6.2) | 9.2 (4.7 to 13.5) | –0.7 (–3.9 to 2.4) | <.001 | |||
PAF (4 → 2) | 7.7 (4.8 to 10.4) | 11.2 (7.1 to 15.1) | 4.7 (0.9 to 8.4) | .02 | |||
PAF (4 → 3) | 5.0 (2.1 to 7.7) | 4.9 (0.8 to 8.9) | 5.4 (1.4 to 9.2) | .88 | |||
Alcohol consumption, drinks/d | |||||||
1. ≤2 | 81 | 71 | 90 | 1 | 1 | 1 | |
2. >2 | 19 | 29 | 10 | 1.32 (1.20 to 1.46) | 1.37 (1.22 to 1.53) | 1.15 (0.94 to 1.42) | |
PAF (2 → 1) | 5.8 (3.7 to 7.9) | 9.0 (5.6 to 12.3) | 1.5 (–0.8 to 3.7) | <.001 | |||
Physical activity, min/wk | |||||||
1. ≥150 | 26 | 31 | 21 | 1 | 1 | 1 | |
2. <150 | 74 | 69 | 79 | 1.02 (0.93 to 1.12) | 0.98 (0.87 to 1.11) | 1.07 (0.93 to 1.23) | |
Joint behaviors | |||||||
PAF: ever smoking, BMI ≥25 kg/m2, and >2 alcoholic drinks/d | 24.9 (19.7 to 29.9) | 36.7 (29.2 to 43.4) | 13.2 (6.0 to 19.8) | <.001 | |||
PAF: current smoking, BMI ≥25 kg/m2, and >2 alcoholic drinks/d | 19.4 (14.3 to 24.2) | 31.1 (23.7 to 37.7) | 8.1 (1.3 to 14.3) | <.001 |
Adjusted for age, sex, study, smoking, body mass index, and alcohol consumption. Some percentages do not add up to 100 because of rounding. BMI = body mass index; CI = confidence interval; CRC = colorectal cancer; HR = hazard ratio; PAF = population attributable fraction.
P value for difference in PAF estimates between men and women.
The hazard ratios for risk factors stratified by each other are shown in Table 3. Overall, 40% of individuals (50% men, 29% women) had at least two of the three behavioral risk factors.
Table 3.
Effect modifier | Subgroup 1 |
Subgroup 2 |
Subgroup 3 |
P† | |||
---|---|---|---|---|---|---|---|
Behavioral risk factor | PR, % | HR (95% CI)* | PR, % | HR (95% CI)* | PR, % | HR (95% CI)* | |
Smoking status | Never smoker | Former smoker | Current smoker | ||||
BMI, kg/m2 | |||||||
1. <18.5 | 2 | 0.89 (0.57 to 1.39) | 1 | 1.19 (0.73 to 1.93) | 2 | 1.22 (0.62 to 2.41) | |
2. 18.5–24.9 | 39 | 1 | 27 | 1 | 38 | 1 | |
3. 25.0–29.9 | 34 | 1.16 (1.04 to 1.30) | 39 | 1.04 (0.92 to 1.18) | 34 | 1.12 (0.85 to 1.47) | |
4. ≥30 | 25 | 1.40 (1.22 to 1.60) | 34 | 1.25 (1.08 to 1.44) | 26 | 1.02 (0.72 to 1.46) | |
PAF (3, 4 → 2) | 13.5 (7.4 to 19.2) | 9.1 (0.6 to 16.8) | 4.4 (–12.2 to 18.5) | P1-2 = .38, P1-3 = .27, P2-3 = .59 | |||
PAF (4 → 3) | 5.1 (1.4 to 8.7) | 6.3 (1.3 to 11.0) | –2.3 (–12.1 to 6.6) | P1-2 = .71, P1-3 = .15, P2-3 = .11 | |||
Alcohol consumption, drinks/d | |||||||
1. ≤2 | 90 | 1 | 77 | 1 | 69 | 1 | |
2. >2 | 10 | 1.15 (0.96 to 1.38) | 23 | 1.37 (1.21 to 1.55) | 31 | 1.55 (1.19 to 2.03) | |
PAF (2 → 1) | 1.6 (–0.6 to 3.6) | 8.1 (4.7 to 11.5) | 15.0 (4.8 to 24.1) | P1-2 = .001, P1-3 < .001, P2-3 = .19 | |||
BMI, kg/m2 | 18.5–24.9 | 25.0–29.9 | ≥30 | ||||
Smoking status | |||||||
1. Never | 58 | 1 | 51 | 1 | 47 | 1 | |
2. Former | 24 | 1.31 (1.15 to 1.49) | 34 | 1.18 (1.05 to 1.33) | 38 | 1.19 (1.01 to 1.38) | |
3. Current | 17 | 1.34 (1.08 to 1.66) | 15 | 1.30 (1.05 to 1.61) | 15 | 0.99 (0.72 to 1.36) | |
PAF (2, 3 → 1) | 12.4 (6.4 to 18.1) | 10.2 (3.9 to 16.0) | 6.5 (–2.3 to 14.5) | P1-2 = .60, P1-3 = .25, P2-3 = .48 | |||
PAF (3 → 1) | 5.5 (1.1 to 9.8) | 4.4 (0.4 to 8.2) | –0.2 (–4.9 to 4.3) | P1-2 = .69, P1-3 = .07, P2-3 = .13 | |||
Alcohol consumption, drinks/d | |||||||
1. ≤2 | 84 | 1 | 80 | 1 | 82 | 1 | |
2. >2 | 15 | 1.34 (1.13 to 1.59) | 20 | 1.34 (1.17 to 1.54) | 18 | 1.27 (1.04 to 1.55) | |
PAF (2 → 1) | 5.5 (2.0 to 8.9) | 6.8 (3.3 to 10.1) | 4.9 (0.4 to 9.5) | P1-2 = .61, P1-3 = .84, P2-3 = .51 | |||
Alcohol consumption, drinks/d | ≤2 | >2 | |||||
Smoking status | |||||||
1. Never | 58 | 1 | 29 | 1 | |||
2. Former | 29 | 1.21 (1.11 to 1.31) | 42 | 1.43 (1.17 to 1.74) | |||
3. Current | 13 | 1.18 (1.01 to 1.39) | 29 | 1.59 (1.20 to 2.11) | |||
PAF (2, 3 → 1) | 8.0 (4.3 to 11.6) | 25.8 (12.8 to 36.8) | P1-2 = .005 | ||||
PAF (3 → 1) | 2.3 (–0.04 to 4.5) | 12.4 (4.3 to 19.8) | P1-2 = .01 | ||||
BMI, kg/m2 | |||||||
1. <18.5 | 2 | 1.05 (0.77 to 1.43) | 1 | 1.02 (0.42 to 2.48) | |||
2. 18.5–24.9 | 36 | 1 | 31 | 1 | |||
3. 25.0–29.9 | 35 | 1.11 (1.01 to 1.21) | 40 | 1.09 (0.89 to 1.32) | |||
4. ≥30 | 28 | 1.32 (1.19 to 1.47) | 29 | 1.22 (0.96 to 1.54) | |||
PAF (3, 4 → 2) | 11.5 (6.4 to 16.3) | 8.8 (–4.1 to 20.1) | P1-2 = .69 | ||||
PAF (4 → 3) | 5.3 (2.2 to 8.4) | 3.4 (–3.4 to 9.7) | P1-2 = .60 |
Adjusted for age, sex, study, smoking, BMI, and alcohol consumption. Some percentages do not add up to 100 because of rounding. BMI = body mass index; CI = confidence interval; CRC = colorectal cancer; HR = hazard ratio; PAF = population attributable fraction; PR = prevalence.
P value for difference in PAF estimates between subgroups.
Competing Risk of Death
Smoking, underweight, and obesity (BMI ≥30 kg/m2) increased the risk of death, whereas excessive alcohol consumption and overweight (BMI 25–<30 kg/m2) were inversely associated (Supplementary Table 6, available online).
CRC Burden
Individual Behaviors
We found that the CRC burden for men and women attributable to ever smoking was 10.7%, and that attributable to current smoking was 3.9%; given the extended excess risk for former smokers, these correspond to the burdens avoidable over a 50-year time frame (Table 2). Of the burden for current smokers, 1.3% could be prevented if those smoking 20 or more cigarettes per day were to smoke fewer than 20 cigarettes per day.
Our estimate for the CRC burden attributable to overweight or obesity (BMI ≥25 kg/m2) was 11.1%, of which obesity explained 7.7% (Table 2). Our modeling predicts that 5.0%, or up to 9100 CRCs in Australia in the next 10 years, would be prevented if all obese individuals were overweight.
Excessive alcohol consumption contributed 5.8% of the burden (Table 2), or up to 10 600 preventable CRCs over the next 10 years. The CRC burden attributable to physical inactivity was not statistically significant (data not shown).
Joint Behaviors
We estimated that ever smoking, BMI of 25 kg/m2 or greater, and excessive alcohol consumption jointly explain 24.9% of the future CRC burden for men and women (Table 2). Quitting smoking, reaching a healthy weight, and not drinking excessively could reduce 19.4% or up to 35 400 CRC cases over the next 10 years.
Population Subgroups
We found PAF effect modification by sex. The burdens attributable to a BMI of 25 kg/m2 or greater and excessive alcohol consumption were higher for men than women, as were the joint contributions of smoking, excess body weight, and excess alcohol (Table 2).
We found PAF effect modification between smoking and alcohol (Table 3). The CRC burden attributable to smoking and to excessive alcohol was higher in the presence of the other exposure.
The PAF attributable to obesity was higher for those younger than 75 years than those 75 years or older. The burden attributable to a BMI of 25 kg/m2 or greater, and the burden attributable to the joint effects of ever smoking, BMI of 25 kg/m2 or greater, and excessive alcohol consumption, was also higher for Australian-born participants compared with migrants (Table 4). We found PAF effect modification between smoking (ever or current) and education, with excess smoking-related risk and burden of CRC for people of high, but not low, educational attainment. We found no trends in CRC burden attributable to smoking, a BMI of 25 kg/m2 or greater, or excess alcohol by socioeconomic status, and no variation in relation to residential location or marital status (data not shown).
Table 4.
Effect modifier | Subgroup 1 |
Subgroup 2 |
Subgroup 3 |
P† | |||
---|---|---|---|---|---|---|---|
Behavioral risk factor | PR, % | HR (95% CI)* | PR, % | HR (95% CI)* | PR, % | HR (95% CI)* | |
Age group, y | <65 | 65–74 | ≥75 | ||||
Smoking status | |||||||
1. Never | 53 | 1 | 47 | 1 | 52 | 1 | |
2. Former | 29 | 1.33 (1.18 to 1.51) | 44 | 1.19 (1.06 to 1.34) | 44 | 1.14 (0.98 to 1.33) | |
3. Current | 18 | 1.29 (1.07 to 1.55) | 9 | 1.23 (0.98 to 1.54) | 4 | 1.44 (0.94 to 2.20) | |
PAF (2, 3 → 1) | 13.4 (7.3 to 19.0) | 9.9 (3.6 to 15.8) | 7.5 (–0.3 to 14.7) | P1-2 = .42, P1-3 = .22, P2-3 = .62 | |||
PAF (3 → 1) | 4.8 (1.0 to 8.4) | 1.9 (–0.3 to 4.1) | 1.5 (–0.6 to 3.6) | P1-2 = .20, P1-3 = .14, P2-3 = .80 | |||
BMI, kg/m2 | |||||||
1. <18.5 | 2 | 1.15 (0.63 to 2.10) | 0 | 1.20 (0.77 to 1.88) | 2 | 0.89 (0.53 to 1.49) | |
2. 18.5–24.9 | 37 | 1 | 25 | 1 | 29 | 1 | |
3. 25.0–29.9 | 34 | 1.11 (0.97 to 1.28) | 39 | 0.99 (0.87 to 1.12) | 42 | 1.21 (1.03 to 1.42) | |
4. ≥30 | 27 | 1.33 (1.15 to 1.55) | 35 | 1.24 (1.07 to 1.44) | 27 | 1.11 (0.88 to 1.40) | |
PAF (3, 4 → 2) | 11.6 (3.7 to 18.8) | 7.3 (–1.6 to 15.5) | 10.8 (0.3 to 20.3) | P1-2 = .47, P1-3 = .91, P2-3 = .60 | |||
PAF (4 → 3) | 5.4 (1.1 to 9.6) | 8.4 (3.0 to 13.5) | –2.3 (–8.8 to 3.7) | P1-2 = .39, P1-3 = .04, P2-3 = .01 | |||
Alcohol consumption, drinks/d | |||||||
1. ≤2 | 82 | 1 | 81 | 1 | 89 | 1 | |
2. >2 | 18 | 1.33 (1.15 to 1.54) | 19 | 1.28 (1.10 to 1.49) | 11 | 1.25 (0.995 to 1.57) | |
PAF (2 → 1) | 6.1 (2.7 to 9.3) | 5.4 (1.8 to 8.9) | 2.9 (–0.3 to 6.0) | P1-2 = .78, P1-3 = .17, P2-3 = .29 | |||
Joint behaviors | |||||||
PAF: ever smoking, BMI ≥25 kg/m2, and >2 alcoholic drinks/d | 27.7 (19.3 to 35.1) | 20.8 (11.3 to 29.4) | 19.8 (8.1 to 30.0) | P1-2 = .26, P1-3 = .25, P2-3 = .88 | |||
Educational attainment | Low | Intermediate | High | ||||
Smoking status | |||||||
1. Never | 49 | 1 | 46 | 1 | 67 | 1 | |
2. Former | 31 | 1.12 (1.01 to 1.25) | 36 | 1.27 (1.10 to 1.47) | 26 | 1.48 (1.23 to 1.77) | |
3. Current | 20 | 1.07 (0.90 to 1.28) | 19 | 1.32 (1.00 to 1.73) | 7 | 2.19 (1.58 to 3.03) | |
PAF (2, 3 → 1) | 5.3 (–0.5 to 10.8) | 13.8 (5.3 to 21.6) | 17.8 (10.7 to 24.3) | P1-2 = .09, P1-3 = .01, P2-3 = .47 | |||
PAF (3 → 1) | 1.4 (–2.4 to 5.2) | 5.3 (–0.4 to 10.7) | 7.2 (3.2 to 10.9) | P1-2 = .26, P1-3 = .04, P2-3 = .58 | |||
BMI, kg/m2 | |||||||
1. <18.5 | 2 | 1.01 (0.69 to 1.49) | 1 | 1.12 (0.61 to 2.05) | 2 | 1.11 (0.52 to 2.36) | |
2. 18.5–24.9 | 34 | 1 | 31 | 1 | 42 | 1 | |
3. 25.0–29.9 | 34 | 1.04 (0.93 to 1.16) | 37 | 1.24 (1.06 to 1.45) | 36 | 1.15 (0.95 to 1.40) | |
4. ≥30 | 30 | 1.26 (1.11 to 1.42) | 31 | 1.40 (1.16 to 1.69) | 21 | 1.20 (0.93 to 1.55) | |
PAF (3, 4 → 2) | 8.3 (1.7 to 14.3) | 17.8 (8.0 to 26.5) | 9.1 (–1.6 to 18.7) | P1-2 = .09, P1-3 = .89, P2-3 = .21 | |||
PAF (4 → 3) | 6.1 (2.1 to 9.9) | 4.1 (–2.1 to 9.9) | 1.0 (–4.9 to 6.5) | P1-2 = .58, P1-3 = .14, P2-3 = .46 | |||
Alcohol consumption, drinks/d | |||||||
1. ≤2 | 83 | 1 | 79 | 1 | 86 | 1 | |
2. >2 | 17 | 1.38 (1.20 to 1.58) | 21 | 1.29 (1.08 to 1.54) | 14 | 1.23 (0.99 to 1.52) | |
PAF (2 → 1) | 6.3 (3.4 to 9.2) | 6.1 (1.5 to 10.5) | 3.6 (–0.5 to 7.4) | P1-2 = .95, P1-3 = .27, P2-3 = .39 | |||
Joint behaviors | |||||||
PAF: ever smoking, BMI ≥25 kg/m2, and >2 alcoholic drinks/d | 18.4 (10.5 to 25.6) | 33.3 (22.8 to 42.3) | 27.5 (16.9 to 36.7) | P1-2 = .02, P1-3 = .15, P2-3 = .41 | |||
Country of birth | Australia | Elsewhere | |||||
Smoking status | |||||||
1. Never | 50 | 1 | 58 | 1 | |||
2. Former | 32 | 1.26 (1.15 to 1.38) | 30 | 1.17 (0.999 to 1.37) | |||
3. Current | 18 | 1.28 (1.09 to 1.51) | 12 | 1.33 (1.03 to 1.72) | |||
PAF (2, 3 → 1) | 12.3 (7.4 to 16.6) | 9.0 (1.7 to 15.7) | P1-2 = .45 | ||||
PAF (3 → 1) | 4.5 (1.3 to 7.7) | 4.0 (0.1 to 7.8) | P1-2 = .83 | ||||
BMI, kg/m2 | |||||||
1. <18.5 | 2 | 1.11 (0.79 to 1.55) | 2 | 0.81 (0.40 to 1.63) | |||
2. 18.5–24.9 | 34 | 1 | 38 | 1 | |||
3. 25.0–29.9 | 34 | 1.12 (1.02 to 1.24) | 38 | 1.05 (0.89 to 1.24) | |||
4. ≥30 | 30 | 1.41 (1.26 to 1.57) | 23 | 1.06 (0.87 to 1.40) | |||
PAF (3, 4 → 2) | 14.4 (8.9 to 19.6) | 3.2 (–6.8 to 12.3) | P1-2 = .045 | ||||
PAF (4 → 3) | 7.4 (3.8 to 10.8) | 0.4 (–4.3 to 4.8) | P1-2 = .02 | ||||
Alcohol consumption, drinks/d | |||||||
1. ≤2 | 80 | 1 | 88 | 1 | |||
2. >2 | 20 | 1.27 (1.14 to 1.42) | 12 | 1.49 (1.22 to 1.81) | |||
PAF (2 → 1) | 5.6 (2.8 to 8.3) | 6.0 (2.6 to 9.2) | P1-2 = .87 | ||||
Joint behaviors | |||||||
PAF: ever smoking, BMI ≥25 kg/m2, and >2 alcoholic drinks/d | 28.7 (22.6 to 34.4) | 16.8 (5.9 to 26.5) | P1-2 = .047 |
Adjusted for age, sex, study, smoking, BMI, and alcohol consumption. Some percentages do not add up to 100 because of rounding. BMI = body mass index; CI = confidence interval; CRC = colorectal cancer; HR = hazard ratio; PAF = population attributable fraction; PR = prevalence.
P value for difference in PAF estimates between subgroups.
Colon and Rectal Cancer Burden
PAF estimates for colon cancer and rectal cancer were broadly similar to those for all CRCs (Supplementary Tables 4 and 5, available online).
Traditional PAF Method
PAF estimates using our method and the traditional approach were largely similar (Supplementary Table 7, available online).
Discussion
One-quarter of Australian CRCs are attributable to the combined effects of ever smoking, being overweight or obese, and drinking excessive alcohol. We showed that adopting healthy living recommendations with respect to these behaviors is likely to produce a marked reduction of the CRC burden for men but a relatively modest reduction for women. We found that the burden of CRC attributed to smoking persists for four decades after quitting, reinforcing the importance of preventing smoking initiation, in addition to measures encouraging smoking cessation. Nevertheless, our data indicate that the future CRC burden would be markedly lower if current and former smokers did not drink excessive alcohol. We also identified other inequalities in the population-level burden of CRC that may guide cancer control activities.
Our CRC PAF estimates are not directly comparable to previous estimates because they are based on different populations, time periods, and analytical approaches, including Australian PAF estimates for 2010: 6.4% for ever smoking, 9.0% for overweight and obesity, 9.0% for excessive alcohol consumption, and 4.8% for physical inactivity (32).
Based on current exposure prevalence, and consistent with prior large-scale cohort studies, we found that tobacco smoking (33–35), excess body weight (35–38), and excessive alcohol consumption (33,35,39,40) each contributed statistically significantly to the burden of CRC or colon cancer. Smoking is a major modifiable risk factor for CRC, with cases attributed to smoking up to 40 years after stopping, in accordance with previous lag time estimates (41,42). Critically, smokers are less likely than nonsmokers to participate in CRC screening (43–46). Our data add to the compelling case for ongoing and new tobacco control initiatives and programs that promote CRC screening participation by smokers.
To our knowledge, we are the first to formally test differences of PAF estimates by subgroup, including by sex. We found a higher CRC burden attributable to overweight or obesity and excessive alcohol for men compared with women. These differences in burden appeared to be due to differences in both exposure prevalence and magnitude of the risk. This is consistent with prior evidence showing a higher CRC risk associated with excess BMI for men compared with women (47,48). Although prior PAF estimates for men and women have not been compared statistically, they align with our findings (35,38). Sex hormones (49–51) and related differences in body fat distribution, in particular abdominal adiposity (52), appear likely to contribute to the sex disparity. Together with the global trajectory of increasing BMI (53), our findings make a case to support men, in particular, achieving and maintaining a healthy weight to prevent CRC. Regarding excessive alcohol consumption, it has long been appreciated that men drink more alcohol than women, and meta-analyses have identified higher CRC risk associated with moderate or heavy consumption for men compared with women (54,55). Increasing recognition of the contribution of alcohol to the cancer burden has led to calls for multifaceted public health strategies designed to prevent people from starting drinking and to reduce high-risk consumption (56). Our results suggest that these efforts may need to be especially targeted to current and former smokers.
Given the clustering of unhealthy behaviors in individuals and the complex interelated pathophysiologic changes associated with coexisting behaviors, burden estimates stratified by other behaviors are essential to addressing residual confounding and reverse causation (57). Using this approach, we revealed that the burden attributable to ever smoking was exacerbated by excessive alcohol consumption, and vice-versa. The only previous study found no evidence of interaction based on a small number of CRCs (33). In support of our observations, strong interactions between smoking and alcohol use have been identified by risk factor studies for cancers of the upper aerodigestive tract (58–60). Adjustment for screening did not affect estimates for smoking or alcohol, in either the main or interaction analyses.
Subgroups that bear the greatest future burden have the most to gain from effective strategies aimed at modifying unhealthy behaviors and encouraging early detection. The only sociodemographic characteristic we found to mediate the CRC burden attributable to smoking was educational attainment, and it was driven by a lack of association between CRC risk and smoking for those with low educational attainment. Two European cohorts observed a positive association between CRC or colon cancer risk and education level (61,62), but only one was robust to full adjustment (61). This finding requires confirmation and investigation of the underlying cause. Interestingly, we saw no PAF modification by socioeconomic status.
In addition to sex, other sociodemographic factors mediated the CRC burden attributable to excess body weight. PAFs were higher for those younger than 75 years compared with those 75 years or older and the Australian-born compared with migrants; again, the differences were driven by differences in risk rather than prevalence. A higher incidence of CRC in Australian-born people compared with migrants has been documented (63) but not attributed to differences in BMI. Our results add to the existing strong case for continued public health campaigns promoting the health benefits of avoiding weight gain, and they could be used to guide personalized cancer prevention programs.
Physical activity was not associated with CRC risk in our cohort, although we had reduced statistical power to examine this at most modest protective association, with only four cohorts having harmonizable data. The World Cancer Research Fund classifies the evidence supporting this association as convincing but also acknowledges moderate heterogeneity between studies and no association with rectal cancer (2).
Our study was strengthened by several design features. First, we matched the strength of association estimates from prospective cohort data with contemporary, representative exposure prevalence estimates. Second, we used a PAF method that accounted for the simultaneous effects of risk factors and their modification both on CRC incidence and death, and generated 95% confidence intervals for the PAF estimates. Together, these features maximized the accuracy and generalizability of our PAF estimates, and they allowed us to perform risk factor interaction and population subgroup analyses that identified statistically significant differences in CRC burden. The high prevalence of coexisting harmful behaviors across industrialized countries (6–9) supports the generalizability of this analytical approach.
As one established and several probable lifestyle risk factors were not measured by the national health surveys, we could not assess their contribution to CRC burden. Reassuringly, risk estimates for smoking, BMI, and alcohol remained statistically significant when we adjusted for processed and red meat consumption in the two cohorts that ascertained these exposures. Although we pooled multiple prospective cohorts with individual-level data, our power remained limited for some analyses, particularly those assessing effect modification. We also acknowledge that we only considered exposures measured at baseline, and they may have changed during follow-up. Relatively large changes in the population-level prevalence of BMI have been observed over time (13), and such changes are likely to bias our risk and PAF estimates toward the null. It is also worth noting that PAF estimation assumes immediate risk reduction following the hypothetical exposure modification. In reality, risk reduction is likely to be gradual, as we demonstrated for smoking and CRC risk, but there are no data on CRC risk reduction after healthy changes in BMI and alcohol consumption.
In summary, we have shown that a large proportion of CRC is potentially preventable by behavior modification, particularly for men. We used a novel PAF method to generate precise estimates of the behavioral factors contributing to the future burden of CRC, the highest burden behavior combinations, and the highest burden subgroups. This information can inform both general and targeted education, public policy, health literacy, and health promotion campaigns aimed at reducing cancer incidence and maximizing early detection.
Funding
This work was supported by the Australian National Health and Medical Research Council (NHMRC; ID1060991). The Australian NHMRC also supported Dr. Laaksonen (ID1053642), Prof. Canfell (ID1082989), Prof. Banks (ID1042717), Prof. Shaw (ID1079438), and Prof. Magliano (ID1118161). Dr. Laaksonen was additionally supported by the Cancer Institute of New South Wales (ID13/ECF/1-07). Ms. Arriaga was supported by an Australian Postgraduate Award and a Translational Cancer Research Network PhD Scholarship Top-up Award.
Notes
Affiliations of authors: Centre for Big Data Research in Health (CMV, PH, MEA, MAL) and Prince of Wales Clinical School (KC, BAA), University of New South Wales, Sydney, Australia; Cancer Epidemiology and Intelligence Division, Cancer Council Victoria, Melbourne, Australia (RJM, GGG); Centre for Epidemiology and Biostatistics, Melbourne School of Population and Global Health, The University of Melbourne, Melbourne, Australia (RJM, GGG); Cancer Research Division, Cancer Council New South Wales, Sydney, Australia (KC); School of Public Health (KC, VH, RGC), School of Life and Environmental Sciences, Charles Perkins Centre (VH), and Centre for Vision Research, Westmead Institute for Medical Research (PM), University of Sydney, Sydney, Australia; ANZAC Research Institute, University of Sydney and Concord Hospital, Sydney, Australia (RGC); Research Centre for Gender, Health and Ageing, University of Newcastle, Newcastle, Australia (JEB); ANU College of Medicine, Biology and Environment, Australian National University, Canberra, Australia (EB); Adelaide Medical School, University of Adelaide, Adelaide, Australia (AWT, TKG); Clinical Diabetes Laboratory (JES) and Diabetes and Population Health Laboratory (DJM), Baker Heart and Diabetes Institute, Melbourne, Australia; Cancer Voices South Australia, Adelaide, Australia (JM).
The authors would like to thank the participating cohort studies and surveys and their participants for the data for this cohort consortium. Specific details of funding and data sources for the 45 and Up Study and the Australian Longitudinal Study on Women’s Health are available at www.saxinstitute.org.au/our-work/45-up-study/for-partners/ and www.alswh.org. Cohort recruitment for the Melbourne Collaborative Cohort Study (MCCS) was funded by Cancer Council Victoria (http://www.cancervic.org.au/) and VicHealth (https://www.vichealth.vic.gov.au/). The MCCS was further supported by Australian National Health and Medical Research Council grants 209057 and 396414 and by infrastructure provided by Cancer Council Victoria. The MCCS was made possible by the contribution of many people, including the original investigators, the teams that recruited the participants and continue working on follow-up, and the many thousands of Melbourne residents who continue to participate in the study. The CHAMP study is funded by the NHMRC (ID301916) and the Ageing and Alzheimer’s Institute. We acknowledge the assistance of the Data Linkage Unit at the Australian Institute of Health and Welfare for undertaking the data linkage to the Australian Cancer Database and the National Death Index.
Author contributions: concept and design: MAL, CMV, KC, RJM; development of methodology: ML; acquisition of data: GGG, RJM, EB, PM, RGC, DJM, JEB, AWT, JES, TKG, VH, MAL, CMV; analysis and interpretation of data: all authors; writing, review, and/or revision of manuscript: all authors; administrative, technical, or material support: N/A.
The funders 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; or the decision to submit the manuscript for publication.
The authors declare no potential conflicts of interest.
Supplementary Material
References
- 1. Arnold M, Sierra MS, Laversanne M, Soerjomataram I, Jemal A, Bray F.. Global patterns and trends in colorectal cancer incidence and mortality. Gut. 2017;66(4):683–691. [DOI] [PubMed] [Google Scholar]
- 2. World Cancer Research Fund International/American Institute for Cancer Research. Continuous update project report: Diet, nutrition, physical activity and colorectal cancer. wcrf.org/colorectal-cancer-2017. Accessed February 9, 2018.
- 3. International Agency for Research on Cancer. Personal Habits and Indoor Combustions. Lyon: IARC; 2012. IARC Monographs on the Evaluation of Carcinogenic Risks to Humans; Vol. 100 (E). [PMC free article] [PubMed]
- 4. Loef M, Walach H.. The combined effects of healthy lifestyle behaviors on all cause mortality: A systematic review and meta-analysis. Prev Med. 2012;55(3):163–170. [DOI] [PubMed] [Google Scholar]
- 5. Geller K, Lippke S, Nigg CR.. Future directions of multiple behavior change research. J Behav Med. 2017;40(1):194–202. [DOI] [PubMed] [Google Scholar]
- 6. Schuit AJ, van Loon AJ, Tijhuis M, Ocke M.. Clustering of lifestyle risk factors in a general adult population. Prev Med. 2002;35(3):219–224. [DOI] [PubMed] [Google Scholar]
- 7. Poortinga W. The prevalence and clustering of four major lifestyle risk factors in an English adult population. Prev Med. 2007;44(2):124–128. [DOI] [PubMed] [Google Scholar]
- 8. Chou KL. The prevalence and clustering of four major lifestyle risk factors in Hong Kong Chinese older adults. J Aging Health. 2008;20(7):788–803. [DOI] [PubMed] [Google Scholar]
- 9. Pronk NP, Anderson LH, Crain AL.. Meeting recommendations for multiple healthy lifestyle factors. Prevalence, clustering, and predictors among adolescent, adult, and senior health plan members. Am J Prev Med. 2004;27(2):25–33. [DOI] [PubMed] [Google Scholar]
- 10. Walter SD. The estimation and interpretation of attributable risk in health research. Biometrics. 1976;32(4):829–849. [PubMed] [Google Scholar]
- 11. Levin ML. The occurrence of lung cancer in man. Acta Unio Int Contra Cancrum. 1953;9(3):531–541. [PubMed] [Google Scholar]
- 12. Laaksonen MA, Härkänen T, Knekt P, Virtala E, Oja H.. Estimation of population attributable fraction (PAF) for disease occurrence in a cohort study design. Stat Med. 2010;29(7–8):860–874. [DOI] [PubMed] [Google Scholar]
- 13. Arriaga ME, Vajdic CM, Canfell K, et al. The burden of cancer attributable to modifiable risk factors: The Australian cancer-PAF cohort consortium. BMJ Open. 2017;7(6):e016178.. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Giles GG, English DR.. The Melbourne Collaborative Cohort Study. Lyon: IARC Scientific Publications; 2002. [PubMed] [Google Scholar]
- 15. Mitchell P, Smith W, Attebo K, Healey PR.. Prevalence of open-angle glaucoma in Australia. The Blue Mountains Eye Study. Ophthalmology. 1996;103(10):1661–1669. [DOI] [PubMed] [Google Scholar]
- 16. Lee C, Dobson AJ, Brown WJ, et al. Cohort Profile: The Australian Longitudinal Study on Women's Health. Int J Epidemiol. 2005;34(5):987–991. [DOI] [PubMed] [Google Scholar]
- 17. Dunstan DW, Zimmet PZ, Welborn TA, et al. The Australian Diabetes, Obesity and Lifestyle Study (AusDiab)—methods and response rates. Diabetes Res Clin Pract. 2002;57(2):119–129. [DOI] [PubMed] [Google Scholar]
- 18. Grant JF, Taylor AW, Ruffin RE, et al. Cohort profile: The North West Adelaide Health Study (NWAHS). Int J Epidemiol. 2009;38(6):1479–1486. [DOI] [PubMed] [Google Scholar]
- 19. Cumming RG, Handelsman D, Seibel MJ, et al. Cohort profile: The Concord Health and Ageing in Men Project (CHAMP). Int J Epidemiol. 2009;38(2):374–378. [DOI] [PubMed] [Google Scholar]
- 20. Banks E, Redman S, Jorm L, et al. Cohort profile: The 45 and up study. Int J Epidemiol. 2008;37(5):941–947. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Australian Bureau of Statistics. National Health Survey (2014-15), Expanded Confidentialised Unit Record File (CURF), Remote Access Data Laboratory (RADL). Findings based on use of ABS Microdata. Accessed December 22, 2017.
- 22. Australian Institute of Health and Welfare. National Drug Strategy Household Survey, 2013 [computer file]. Canberra: Australian Data Archive, The Australian National University; 2015. [Google Scholar]
- 23. Australian Bureau of Statistics. Census of Population and Housing: Socio-Economic Indexes for Areas (SEIFA), Australia, 2011. 2033.0.55.001. Canberra: ABS; 2013.
- 24. Australian Bureau of Statistics. Main Structure and Greater Capital City Statistical Areas, July 2011.1270.0.55.001 Canberra: ABS; 2010. Australian Statistical Geography Standard; Vol. 1.
- 25. Bray F, Colombet M, Mery L, Pineros M, Znaor A, Zanetti R, eds. Cancer Incidence in Five Continents. Vol. XI (electronic version). Lyon: International Agency for Research on Cancer; 2017. [Google Scholar]
- 26. Jaro MA. Probabilistic linkage of large public health data files. Stat Med. 1995;14(5–7):491–498. [DOI] [PubMed] [Google Scholar]
- 27. Friedman M. Piecewise exponential models for survival data with covariates. Ann Stat. 1982;10(1):101–113. [Google Scholar]
- 28. DerSimonian R, Laird N.. Meta-analysis in clinical trials. Control Clin Trials. 1986;7(3):177–188. [DOI] [PubMed] [Google Scholar]
- 29. Australian Institute of Health and Welfare. Cancer Incidence Projections: Australia, 2011 to 2020. Cancer Series No. 66. Cat. No. CAN 62. Canberra: AIHW; 2012.
- 30. Laaksonen MA, Knekt P, Rissanen H, et al. The relative importance of modifiable potential risk factors of type 2 diabetes: A meta-analysis of two cohorts. Eur J Epidemiol. 2010;25(2):115–124. [DOI] [PubMed] [Google Scholar]
- 31. Laaksonen MA, Virtala E, Knekt P, Oja H, Härkänen T.. SAS macros for calculation of population attributable fraction in a cohort study design. J Stat Soft. 2011;43(7):1–25. [Google Scholar]
- 32. Whiteman DC, Webb PM, Green AC, et al. Cancers in Australia in 2010 attributable to modifiable factors: Summary and conclusions. Aust N Z J Public Health. 2015;39(5):477–484. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Otani T, Iwasaki M, Yamamoto S, et al. Alcohol consumption, smoking, and subsequent risk of colorectal cancer in middle-aged and elderly Japanese men and women: Japan Public Health Center-based prospective study. Cancer Epidemiol Biomarkers Prev. 2003;12(12):1492–1500. [PubMed] [Google Scholar]
- 34. Agudo A, Bonet C, Travier N, et al. Impact of cigarette smoking on cancer risk in the European prospective investigation into cancer and nutrition study. J Clin Oncol. 2012;30(36):4550–4557. [DOI] [PubMed] [Google Scholar]
- 35. Aleksandrova K, Pischon T, Jenab M, et al. Combined impact of healthy lifestyle factors on colorectal cancer: A large European cohort study. BMC Med. 2014;12(1):168.. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36. Otani T, Iwasaki M, Inoue M.. Body mass index, body height, and subsequent risk of colorectal cancer in middle-aged and elderly Japanese men and women: Japan public health center-based prospective study. Cancer Causes Control. 2005;16(7):839–850. [DOI] [PubMed] [Google Scholar]
- 37. Thygesen LC, Gronbaek M, Johansen C, Fuchs CS, Willett WC, Giovannucci E.. Prospective weight change and colon cancer risk in male US health professionals. Int J Cancer. 2008;123(5):1160–1165. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38. Matsuo K, Mizoue T, Tanaka K, et al. Association between body mass index and the colorectal cancer risk in Japan: Pooled analysis of population-based cohort studies in Japan. Ann Oncol. 2012;23(2):479–490. [DOI] [PubMed] [Google Scholar]
- 39. Mizoue T, Inoue M, Wakai K, et al. Alcohol drinking and colorectal cancer in Japanese: A pooled analysis of results from five cohort studies. Am J Epidemiol. 2008;167(12):1397–1406. [DOI] [PubMed] [Google Scholar]
- 40. Schutze M, Boeing H, Pischon T, et al. Alcohol attributable burden of incidence of cancer in eight European countries based on results from prospective cohort study. BMJ. 2011;342:d1584.. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41. Giovannucci E, Rimm EB, Stampfer MJ, et al. A prospective study of cigarette smoking and risk of colorectal adenoma and colorectal cancer in U.S. men. J Natl Cancer Inst. 1994;86(3):183–191. [DOI] [PubMed] [Google Scholar]
- 42. Liang PS, Chen TY, Giovannucci E.. Cigarette smoking and colorectal cancer incidence and mortality: Systematic review and meta-analysis. Int J Cancer. 2009;124(10):2406–2415. [DOI] [PubMed] [Google Scholar]
- 43. He E, Lew JB, Egger S, et al. Factors associated with participation in colorectal cancer screening in Australia: Results from the 45 and Up Study cohort. Prev Med. 2018;106:185–193. [DOI] [PubMed] [Google Scholar]
- 44. Carlos RC, Underwood W 3rd, Fendrick AM, Bernstein SJ.. Behavioral associations between prostate and colon cancer screening. J Am Coll Surg. 2005;200(2):216–223. [DOI] [PubMed] [Google Scholar]
- 45. Blanks RG, Benson VS, Alison R, et al. Nationwide bowel cancer screening programme in England: Cohort study of lifestyle factors affecting participation and outcomes in women. Br J Cancer. 2015;112(9):1562–1567. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 46. Artama M, Heinävaara S, Sarkeala T, Prättälä R, Pukkala E, Malila N.. Determinants of non-participation in a mass screening program for colorectal cancer in Finland. Acta Oncol. 2016;55(7):870–874. [DOI] [PubMed] [Google Scholar]
- 47. Ma Y, Yang Y, Wang F, et al. Obesity and risk of colorectal cancer: A systematic review of prospective studies. PLoS One. 2013;8(1):e53916.. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 48. Kyrgiou M, Kalliala I, Markozannes G, et al. Adiposity and cancer at major anatomical sites: Umbrella review of the literature. BMJ. 2017;356:j477.. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 49. Lin JH, Zhang SM, Rexrode KM, et al. Association between sex hormones and colorectal cancer risk in men and women. Clin Gastroenterol Hepatol. 2013;11(4):419–424.e1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 50. Murphy N, Strickler HD, Stanczyk FZ, et al. A prospective evaluation of endogenous sex hormone levels and colorectal cancer risk in postmenopausal women. J Natl Cancer Inst. 2015;107(10):djv210. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 51. Koo JH, Leong RW.. Sex differences in epidemiological, clinical and pathological characteristics of colorectal cancer. J Gastroenterol Hepatol. 2010;25(1):33–42. [DOI] [PubMed] [Google Scholar]
- 52. Song M, Hu FB, Spiegelman D, et al. Long-term status and change of body fat distribution, and risk of colorectal cancer: A prospective cohort study. Int J Epidemiol. 2016;45(3):871–883. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 53. Ng M, Fleming T, Robinson M, et al. Global, regional, and national prevalence of overweight and obesity in children and adults during 1980-2013: A systematic analysis for the Global Burden of Disease Study 2013. Lancet. 2014;384(9945):766–781. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 54. Fedirko V, Tramacere I, Bagnardi V, et al. Alcohol drinking and colorectal cancer risk: An overall and dose-response meta-analysis of published studies. Ann Oncol. 2011;22(9):1958–1972. [DOI] [PubMed] [Google Scholar]
- 55. Cai S, Li Y, Ding Y, Chen K, Jin M.. Alcohol drinking and the risk of colorectal cancer death: A meta-analysis. Eur J Cancer Prev. 2014;23(6):532–539. [DOI] [PubMed] [Google Scholar]
- 56. LoConte NK, Brewster AM, Kaur JS, Merrill JK, Alberg AJ.. Alcohol and cancer: A statement of the American Society of Clinical Oncology. J Clin Oncol. 2018;36(1):83–93. [DOI] [PubMed] [Google Scholar]
- 57. Song M, Giovannucci E.. Estimating the influence of obesity on cancer risk: Stratification by smoking is critical. J Clin Oncol. 2016;34(27):3237–3239. [DOI] [PubMed] [Google Scholar]
- 58. Taylor B, Rehm J.. When risk factors combine: The interaction between alcohol and smoking for aerodigestive cancer, coronary heart disease, and traffic and fire injury. Addict Behav. 2006;31(9):1522–1535. [DOI] [PubMed] [Google Scholar]
- 59. Hashibe M, Brennan P, Chuang SC, et al. Interaction between tobacco and alcohol use and the risk of head and neck cancer: Pooled analysis in the International Head and Neck Cancer Epidemiology Consortium. Cancer Epidemiol Biomarkers Prev. 2009;18(2):541–550. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 60. Flanders WD, Rothman KJ.. Interaction of alcohol and tobacco in laryngeal cancer. Am J Epidemiol. 1982;115(3):371–379. [DOI] [PubMed] [Google Scholar]
- 61. Leufkens AM, Van Duijnhoven FJ, Boshuizen HC, et al. Educational level and risk of colorectal cancer in EPIC with specific reference to tumor location. Int J Cancer. 2012;130(3):622–630. [DOI] [PubMed] [Google Scholar]
- 62. van Loon AJ, van den Brandt PA, Golbohm RA.. Socioeconomic status and colon cancer incidence: A prospective cohort study. Br J Cancer. 1995;71(4):882–887. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 63. Kune S, Kune GA, Watson L.. The Melbourne colorectal cancer study: Incidence findings by age, sex, site, migrants and religion. Int J Epidemiol. 1986;15(4):483–493. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.