Circulating Vitamin D and Colorectal Cancer Risk: An International Pooling Project of 17 Cohorts

Abstract

Background: Experimental and epidemiological studies suggest a protective role for vitamin D in colorectal carcinogenesis, but evidence is inconclusive. Circulating 25-hydroxyvitamin D (25(OH)D) concentrations that minimize risk are unknown. Current Institute of Medicine (IOM) vitamin D guidance is based solely on bone health.

Methods: We pooled participant-level data from 17 cohorts, comprising 5706 colorectal cancer case participants and 7107 control participants with a wide range of circulating 25(OH)D concentrations. For 30.1% of participants, 25(OH)D was newly measured. Previously measured 25(OH)D was calibrated to the same assay to permit estimating risk by absolute concentrations. Study-specific relative risks (RRs) for prediagnostic season-standardized 25(OH)D concentrations were calculated using conditional logistic regression and pooled using random effects models.

Results: Compared with the lower range of sufficiency for bone health (50-<62.5 nmol/L), deficient 25(OH)D (<30 nmol/L) was associated with 31% higher colorectal cancer risk (RR = 1.31, 95% confidence interval [CI] = 1.05 to 1.62); 25(OH)D above sufficiency (75-<87.5 and 87.5-<100 nmol/L) was associated with 19% (RR = 0.81, 95% CI = 0.67 to 0.99) and 27% (RR = 0.73, 95% CI = 0.59 to 0.91) lower risk, respectively. At 25(OH)D of 100 nmol/L or greater, risk did not continue to decline and was not statistically significantly reduced (RR = 0.91, 95% CI = 0.67 to 1.24, 3.5% of control participants). Associations were minimally affected when adjusting for body mass index, physical activity, or other risk factors. For each 25 nmol/L increment in circulating 25(OH)D, colorectal cancer risk was 19% lower in women (RR = 0.81, 95% CI = 0.75 to 0.87) and 7% lower in men (RR = 0.93, 95% CI = 0.86 to 1.00) (two-sided Pheterogeneity by sex = .008). Associations were inverse in all subgroups, including colorectal subsite, geographic region, and season of blood collection.

Conclusions: Higher circulating 25(OH)D was related to a statistically significant, substantially lower colorectal cancer risk in women and non-statistically significant lower risk in men. Optimal 25(OH)D concentrations for colorectal cancer risk reduction, 75-100 nmol/L, appear higher than current IOM recommendations.

Figure 1.

Pooled multivariable relative risks (RRs; indicated by open symbols) and 95% confidence intervals (CIs; indicated by vertical lines) of colorectal cancer according to categories of season-standardized circulating 25-hydroxyvitamin D (25(OH)D) concentrations (A) overall, (B) in women, and (C) in men. 25(OH)D categories correspond to Institute of Medicine (IOM) recommendations for bone health: deficiency was defined as <30 nmol/L; insufficiency includes 30–<40 and 40–<50 nmol/L; sufficiency includes 50–<62.5 (referent) and 62.5–<75 nmol/L; and beyond sufficiency includes 75–<87.5, 87.5–<100, and ≥100 nmol/L. The RR (95% CI) for each category is plotted at the median concentration of 25(OH)D among control participants in that category. The RRs for 75–<100 nmol/L, relative to 50–<62.5 nmol/L, are 0.78 (95% CI = 0.67 to 0.92) overall, 0.67 (95% CI = 0.54 to 0.83) for women, and 0.90 (95% CI = 0.68 to 1.18) for men (not shown in figure). Models were conditioned on study-specific matching factors, including date of blood draw and age, and were additionally adjusted for body mass index, physical activity, race, family history of colorectal cancer, alcohol consumption, smoking status, aspirin and/or nonsteroidal anti-inflammatory drug use, and, in women, menopausal status and menopausal hormone therapy. See the Supplementary Methods (available online) for individual studies excluded in specific 25(OH)D categories due to low numbers of participants. Two-sided P_trend value was calculated using a continuous variable based on the median 25(OH)D in each category. 25(OH)D = 25-hydroxyvitamin D; CI = confidence interval; IOM = Institute of Medicine.

Figure 2.

Study-specific and pooled multivariable relative risks (RRs; indicated by solid squares and diamonds, respectively) and 95% confidence intervals (CIs; indicated by horizontal lines) of colorectal cancer per 25 nmol/L increment in season-standardized circulating 25-hydroxyvitamin D (25(OH)D). The size of the square is proportional to the inverse of the variance of the study-specific RR. Full cohort names are listed in Table 1. An RR of 1.0, marked by the vertical line, indicates no association. Models were conditioned on study-specific matching factors, including date of blood draw and age, and were additionally adjusted for body mass index, physical activity, race, family history of colorectal cancer, alcohol consumption, smoking status, aspirin and/or nonsteroidal anti-inflammatory drug use, and, in women, menopausal status and menopausal hormone therapy. Two-sided P_{heterogeneity} value for between-studies heterogeneity was calculated using the Q statistic. CI = confidence interval; RR = relative risk.

Figure 3.

Pooled multivariable relative risks (RRs; indicated by solid diamonds) and 95% confidence intervals (95% CIs; indicated by horizontal lines) of colorectal cancer per 25 nmol/L increment in season-standardized circulating 25-hydroxyvitamin D (25(OH)D) stratified by demographic, lifestyle, and other factors and tumor characteristics. An RR of 1.0, marked by the vertical line, indicates no association. Summer- and winter-specific RRs were not standardized by season. Conditional models were used for stratified analyses by sex, age at diagnosis, region, 25(OH)D data, tumor stage, colorectal and colon subsite, and time to diagnosis. These models were conditioned on study-specific matching factors, including date of blood draw and age, and were additionally adjusted for body mass index, physical activity, race, family history of colorectal cancer, alcohol consumption, smoking status, aspirin and/or nonsteroidal anti-inflammatory drug use, and, in women, menopausal status and menopausal hormone therapy, except in models stratified by that variable. Unconditional models, which were used for the remaining stratified analyses, were adjusted for study-specific matching factors including date of blood draw and age, and the covariates listed above. See the Supplementary Methods (available online) for individual studies excluded in specific strata due to low numbers of participants (n < 25). P_{heterogeneity} value for between-studies heterogeneity was calculated using the Q statistic, except for race. Due to small numbers of nonwhites within most individual cohorts, analyses stratified by race were conducted using aggregated data, adjusted for study. P_{heterogeneity} value for heterogeneity across strata, except for race, tumor stage, and subsite, was calculated using meta-regression. Statistical significance for interaction by race was assessed using a Wald test. Evaluation of common effects by tumor stage and subsite was assessed using a contrast test. All statistical tests are two-sided. CI = confidence interval; RR = relative risk.