Combinations of modifiable lifestyle behaviours in relation to colorectal cancer risk in Alberta's Tomorrow Project

Nearly one in two Canadians will be diagnosed with cancer in their lifetime and one in four will die of the disease. Colorectal cancer (CRC) is the third most common cancer in Canada and is responsible for a large portion of cancer-related mortality. Non-modifiable CRC risk factors include increasing age, male sex, family history in a first-degree relative, inflammatory bowel disease, and a personal history of colorectal adenomas. Several modifiable lifestyle behaviours are consistently associated with CRC, including obesity, inadequate physical activity, high consumption of red and processed meat, low consumption of fruits and vegetables, alcohol consumption, and tobacco smoking. An estimated 43% of CRC cases in Canada are attributable to modifiable lifestyle exposures. Despite sufficient or probable evidence of causal links, many cases continue to be attributed to these behaviours, and trends suggest increases in obesity, alcohol consumption, and poor diet, with some decreases in smoking and physical inactivity. Individuals often engage in multiple behaviours that co-occur, making prevention messaging challenging. For example, smokers are more likely to consume alcohol regularly, and highly physically active individuals may also consume more alcohol. Some paradoxical co-occurrences (e.g., high physical activity and smoking) also exist. Given limited prevention resources, targeting subgroups that share behaviour patterns may be more cost-effective and impactful. This necessitates research on common patterns of behaviours and their relationships with disease risk. Latent class analysis (LCA) can identify homogeneous subgroups based on multiple variables within heterogeneous populations, reducing data into a parsimonious set of classes with unique behavioural profiles. No prior study has characterized clustering of CRC-specific risk behaviours in Canada or linked LCA-derived classes to incident CRC. The aim of this study was to identify distinct groups of individuals with unique patterns of lifestyle-related CRC risk factors in Alberta's Tomorrow Project (ATP) and to determine how these groups were associated with CRC risk. The study also explored the impact of sex and family history of CRC on behavioural patterns and subsequent CRC risk, to illustrate the utility of an LCA-based approach for surveillance and intervention design.

Previous LCA studies have characterized clustering of modifiable behavioural risk factors in specific populations such as veterans, college students, adolescents, and children, as well as general populations in the United States and United Kingdom, and primary care patients. However, none have focused on Canada, CRC-specific risk behaviours, or examined how LCA-defined classes relate prospectively to disease development. Evidence shows behaviours often co-occur (e.g., smoking with alcohol use; high physical activity with higher alcohol use; smokers tending to have lower BMI), and paradoxical combinations are observed. These patterns highlight the need for analytic approaches like LCA to inform multifactorial prevention strategies and resource allocation. The present study addresses these gaps by applying LCA to CRC-relevant behaviours in a large Canadian cohort and linking the resulting classes to incident CRC risk.

Data source: Baseline data from Alberta's Tomorrow Project (ATP), a prospective cohort of adults aged 35–69 years without prior cancer in Alberta, Canada. Enrollment began in 2000. A total of 31,121 enrolled and completed the Health and Lifestyle Questionnaire (HLQ). Of these, 26,538 completed both the Canadian Diet History Questionnaire-I (DHQ) and the Past Year Total Physical Activity Questionnaire (PYTPAQ). For this analysis, 26,460 participants with complete data on relevant CRC risk factors and demographics were included; 78 participants with missing data were excluded. Follow-up extended to December 2017 via linkage with the Alberta Cancer Registry (ACR). Exposures of interest: Established modifiable CRC risk factors were assessed and categorized into three levels using guidelines or distribution-based cut points: - Tobacco smoking: never, former, current. - Body mass index (BMI): normal (<25 kg/m2), overweight (25–<30 kg/m2), obese (≥30 kg/m2) per WHO. - Alcohol consumption: abstainer; low risk (≤1 drink/day for women, ≤2/day for men); high risk (>guideline), per WCRF/AICR recommendations. - Recreational physical activity: low (<150 min/week MVPA), moderate (150–300 min/week), high (>300 min/week) per Canadian/international guidelines. - Fruits and vegetables: low (did not meet recommendation for either), moderate (met recommendation for either fruit or vegetable), high (met recommendation for both; >4 servings of each per day). - Red meat: low (<3 servings/week), moderate (3–6/week), high (>6/week). - Processed meat: low (<1 serving/week), moderate (1–2/week), high (>2/week). Latent class analysis (LCA): Exploratory LCA was conducted using PROC LCA in SAS to identify mutually exclusive classes based on the above behaviours. Sex-stratified LCAs were also conducted, incorporating family history of CRC as an indicator in class formation for sex-specific models. Model selection started with 2 classes, increasing sequentially. Fit assessed by AIC, BIC, and sample size–adjusted BIC, along with class interpretability, class size (avoiding very small classes), and presence of a distinctly low-risk class. The final selections were 7 classes overall, 6 classes for men, and 7 classes for women. Posterior probabilities were computed and individuals assigned to the class with maximum posterior probability; mean/median maximum posterior probabilities were examined to assess assignment certainty. Relative probabilities of each behaviour category within classes were standardized and visualized via heat maps. Outcome ascertainment and statistical analysis: Incident primary CRC cases were identified through ACR linkage to December 2017. Time scale in Cox models was attained age; entry at exact age at baseline, exit at CRC diagnosis, diagnosis of another cancer (censored), or end of follow-up. Proportional hazards models estimated hazard ratios (HRs) for each class compared to the lowest-risk class, adjusting for sex (in overall model), ethnicity (white/other), family history of CRC (yes/no; except where included in class formation for sex-specific LCA), highest education (high school or less; some post–high school; post-secondary degree), and household income ($0–49,999; $50,000–99,999; ≥$100,000). Proportional hazards assumption checked using cumulative sums of martingale-based residuals. Two-tailed P<0.05 was considered significant; no multiple-comparison correction due to limited power. Sensitivity analyses: (1) Excluded participants with follow-up <1 year and <2 years to reduce reverse causation; (2) Re-specified exposures with more precise measures: smoking as non-smoker, <15 pack-years, ≥15 pack-years; obesity using combined BMI and waist circumference (normal; overweight with normal waist; overweight with high waist or obese). New LCAs and corresponding risk models were conducted under these specifications. Ethics: All participants provided informed consent including data linkage to ACR. Ethics approvals were obtained from the Alberta Cancer Board Research Ethics Committee and the University of Calgary Conjoint Health Research Ethics Board. Methods followed relevant guidelines and STROBE reporting standards.

- Cohort and follow-up: 26,460 participants (9,892 men; 16,568 women) with 267 CRC cases (119 men; 148 women). Median follow-up: 13.23 years. - Baseline predictors (mutually adjusted Cox): Compared to normal BMI, overweight HR=1.49 (95% CI: 1.07–2.07) and obese HR=1.82 (1.29–2.56). Current smokers vs never HR=1.63 (1.16–2.29). Household income $50,000–$99,999 HR=0.76 (0.57–1.00) and ≥$100,000 HR=0.68 (0.48–0.97) vs <$50,000. Family history of CRC HR=1.65 (1.17–2.31). Sex not significantly associated (men vs women HR=1.23; 0.94–1.62). Among men, high-risk alcohol, high BMI, current smoking, and family history were significant; among women, current smoking was significant. - Model selection: Overall LCA supported a 7-class solution (AIC=3242.97; BIC=4094.70; sample-size adjusted BIC=3764.19). Sex-specific models selected 6 classes in men and 7 in women based on fit and interpretability. - Overall LCA classes and CRC risk (reference: low-risk Class 2, 10.2% prevalence; tended to be never smokers with normal BMI but low fruit/vegetable intake): • Class 7 (8.1%): current smokers, high-risk drinkers, inactive — HR=2.87 (1.43–5.77). • Class 1 (13.9%): high meat consumption, low fruit/vegetable (poor diet) — HR=2.48 (1.27–4.83). • Class 5 (16.4%): obese BMI, alcohol abstainers — HR=2.46 (1.28–4.70). • Class 4 (20.5%): good diet but inactive and overweight/obese — HR=2.34 (1.23–4.45). • Class 6 (11.8%): former smokers, high-risk drinkers, high physical activity, overweight — HR=1.73 (0.85–3.49). • Class 3 (19.1%): low meat, high physical activity, normal BMI, current/former smokers — HR=1.56 (0.79–3.06). - Sex-specific LCA and CRC risk: • Men (reference: Class 5, 11.0%; never smokers, low relative family history): - Class 1 (9.5%): current smokers, high-risk drinkers, moderate family history — HR=3.89 (1.42–10.66). - Class 4 (19.1%): poor diet with overweight/obese BMI — HR=3.15 (1.22–8.13). - Class 6 (14.5%): abstain from alcohol, low meat, physically inactive — HR=2.34 (0.87–6.28). - Class 3 (27.6%): former smokers, high BMI, high relative family history — HR=2.26 (0.88–5.80). - Class 2 (18.3%): high relative family history, highly active, low meat — HR=1.53 (0.55–4.26). • Women (reference: Class 2, 24.8%; low meat, highly active, normal BMI): - Class 5 (7.0%): former smokers with obese BMI — HR=2.19 (1.20–3.99). - Class 1 (15.0%): current smokers and high-risk drinkers — HR=1.99 (1.14–3.47). - Class 7 (11.3%): highly active, poor diet, high-risk drinkers — HR=1.66 (0.84–3.27). - Class 6 (18.8%): physically inactive, otherwise relatively healthy — HR=1.54 (0.91–2.59). - Class 3 (13.6%): non-smokers, low meat, low relative family history — HR=1.35 (0.75–2.43). - Class 4 (9.6%): highest relative family history, poor diet, high BMI — HR=1.21 (0.58–2.52). - Sensitivity analyses: Excluding <1 or <2 years of follow-up yielded similar estimates for overall and women, with slightly stronger effects in men. Using refined smoking and adiposity measures yielded similar class structures overall and in men; among women, a new high-risk class emerged (heavy smokers, high red meat intake, low physical activity, obese BMI and high waist circumference) with HR=3.05 (1.64–5.68) versus the low-risk class. - Overall, multiple distinct moderate-risk profiles (approximately 2–3× risk) were identified rather than a single extreme high-risk profile. Physical activity appeared in profiles with non-significantly elevated risks, suggesting potential offsetting of other risks.

This prospective LCA-based analysis identified seven distinct behavioural profiles related to CRC risk, revealing that no single class with universally poor behaviours dominated risk. Instead, four classes exhibited approximately two- to threefold higher CRC risk compared with a low-risk class, each characterized by different combinations of behaviours (e.g., smoking plus high-risk drinking and inactivity; poor diet; obesity with alcohol abstention; inactivity with overweight despite good diet). These findings suggest that multiple moderate-risk behaviours in combination elevate CRC risk more than any single behaviour (except obesity), emphasizing potential synergistic effects and the value of multifactorial interventions. Classes that included high physical activity tended not to have significantly elevated risks, supporting a protective role for activity that may mitigate other adverse behaviours. Sex-stratified analyses highlighted both shared and sex-specific patterns. In both men and women, profile types featuring current smoking with high-risk alcohol use and former smoking with obesity were common and associated with elevated CRC risk, indicating priority targets for intervention. Men exhibited higher relative risks across classes than women, and family history more strongly influenced men’s class structure, with the highest family-history group among men tending to engage in healthier behaviours. This suggests that family history may motivate greater risk mitigation among men, informing tailored prevention strategies. These results demonstrate the utility of LCA to identify actionable population subgroups for targeted, multifactorial prevention. For example, interventions combining smoking cessation with physical activity promotion could simultaneously reduce smoking-related risk while minimizing cessation-related weight gain, thereby avoiding unintended increases in BMI-related risk. The approach offers a more holistic assessment of co-occurring behaviours than single-factor or interaction analyses and provides a framework for surveillance and program design.

The study established distinct lifestyle behaviour profiles in a large Canadian cohort and quantified their associations with incident CRC, identifying several moderate-risk combinations rather than a single extreme high-risk group. Physical activity may offset risks associated with other adverse behaviours, and obesity remains a prominent single-behaviour driver of risk. Sex-specific patterns and the influence of family history underscore the need for tailored prevention. Targeting combinations of co-occurring behaviours with multifactorial interventions could yield greater population-level CRC prevention than single-behaviour strategies. Future research should replicate these findings in larger and more diverse cohorts, incorporate time-varying and lifetime exposure histories, refine behavioural categorization (e.g., pack-years, adiposity distribution), and extend the approach to other cancers and chronic diseases to inform comprehensive intervention programs.

- Self-reported behaviours may be subject to social desirability bias and measurement error. - Behavioural assessments were taken at baseline only and may not reflect lifetime or time-varying exposures; changes during follow-up were not captured. - Use of guideline-based cut-offs may misclassify extreme behaviours, though they aid comparability and interpretability. - Latent classes are not fully homogeneous; within-class heterogeneity and potential misclassification could bias risk estimates, especially with a relatively low number of CRC cases. - No data on personal history of colorectal polyps, a CRC risk factor, potentially confounding associations. - Modest sample size (26,460; 267 cases) limits precision and the ability to define more granular classes or conduct extensive stratified analyses. - Generalizability may be limited: ATP participants tend to be older, have higher socioeconomic status, higher BMI, less smoking, and higher alcohol consumption than the broader Alberta population. - Model selection involves trade-offs between fit indices and interpretability; although sensitivity analyses supported robustness, replication is needed.