Associations between Dietary Patterns and Incident Colorectal Cancer in 114,443 Individuals from the UK Biobank: A Prospective Cohort Study

Abstract

Background: Diet-disease association studies increasingly use dietary patterns (DP) to account for the complexity of the exposure. We assessed if a DP associated with type 2 diabetes mellitus, cardiovascular disease, and all-cause mortality is also associated with colorectal cancer.

Methods: We used reduced rank regression on 24-hour recall data to identify DPs, explaining the maximum variation in four nutrient-response variables: energy density, saturated fatty acids, free sugars, and fiber density. Cox proportional hazards models examined prospective associations between DP adherence (coded in a continuous scale as z-scores as well as in quintiles) and incident colorectal cancer. Subgroup analyses were conducted for tumor site, age, and sex.

Results: After exclusions, 1,089 colorectal cancer cases occurred in 114,443 participants over a median follow-up of 8.0 years. DP1 was characterized by increased intake of chocolate and confectionery; butter; low-fiber bread; red and processed meats; and alcohol, as well as low intake of fruits, vegetables, and high-fiber cereals. After accounting for confounders, including body mass, there were positive linear associations between DP1 and incident overall colorectal cancer (HR of quintile 5 vs. 1, 1.34; 95% confidence interval, 1.16-1.53, Ptrend = 0.005) and rectal cancer (HR of quintile 5 vs. 1, 1.58; 95% confidence interval, 1.27-1.96, Ptrend = 0.009) but not for proximal or distal colon cancers. No DP2-colorectal cancer association was observed.

Conclusions: A DP previously associated with cardiometabolic disease is also associated with incident colorectal cancer, especially rectal cancers.

Impact: These consistent associations of particular food groups with both cardiometabolic disease and this diet-related cancer strengthen the evidence base for holistic population dietary guidelines to prevent ill-health.

Figure 1.

Sequential stratification/adjustments to the model comparing the risk of incident colorectal cancer associated with DP1 and DP2. Top, sequential changes to the HRs for quintiles 5 versus 1 for DP1 and DP2. Bottom, sequential changes to the HRs for DP1 and DP2 z-scores in continuous form. The X-axis represents HRs on the log-scale. CIs were obtained using the floating absolute risk method (34). χ² and P values were calculated by likelihood ratio tests to assess the heterogeneity in the associations with sequential adjustments for covariates. Behavioral factors included smoking status, physical activity level (MET-hours/week), and total energy intake (ln-kJ). SES* comprises TDI and educational attainment. Models were stratified by covariates violating the proportional hazards assumption: physical activity, family history of colorectal cancer, education, and BMI. SES, socioeconomic status.

Figure 2.

Shape plot of the HRs and 95% CIs denoting incident colorectal cancer risk by DP1 quintiles in the fully adjusted model; DP1 quintile 1 is shown as the reference. CIs were obtained using the floating absolute risk method (34).

Figure 3.

HRs (95% CIs) of continuous form. A, DP1 z-scores and B, DP2 z-scores for the risk of incident colorectal cancer. HRs (solid black line) and 95% CIs (gray shading) were derived from spline regression models to evaluate for potential nonlinear associations between DP1 scores and incident colorectal cancer. The reference point of the DP z-scores was set to the fifth percentile values. X-axis represents DP z-scores.

Figure 4.

DP1 associations with colorectal cancer by anatomic subsite. Top, HRs of quintile 5 versus quintile 1. Bottom, HRs of DP1 z-scores in continuous form. Unspecified (n = 68) and overlapping (n = 4) colorectal cancers were excluded as they could not be localized to an anatomic subsite. Test for trend down using LRTs. LRT, likelihood ratio test.

Figure 5.

DP1 sensitivity analyses. The x-axis comprises HRs and associated 95% CIs and is on the log-scale. CIs presented were obtained using the floating absolute risk method. WebQ, Oxford WebQ 24-hour dietary assessment tool.