Fig. 2: Multivariate HRs and 95% CIs for early and late-onset colorectal cancer incidence in relation to lifestyle factors by anatomic sub-site.

Multivariable Cox regression models was constructed using age as the underlying time variable stratified by sex and cohort. For smoking, alcohol consumption and physical activity, the above models were mutually adjusted, and adjusted for age at recruitment, country, height, body mass index, diabetes, hypertension and ever use of postmenopausal hormone replacement therapy and physical activity. Multivariable model for diabetes and hypertension was mutually adjusted and adjusted further for age at recruitment, country, height, body mass index, smoking, alcohol and physical activity. For multivariable joint Cox models for anatomical subsite, similar variable adjustments were made as above. EOCRC early-onset colorectal cancer, LOCRC later-onset colorectal cancer, P-het: heterogeneity p value between subsites.