We conducted a case-control study nested within the biorepository subcohort of the Multiethnic Cohort (MEC) (7
). The baseline cohort comprised over 215,000 people between the ages of 45 and 75 years who returned a questionnaire by mail between 1993 and 1996. The biorepository subcohort (n
= 67,594) included White, Hawaiian, and Japanese American participants in Hawaii and African American, Japanese American, and Latino participants in Los Angeles who agreed to donate a biospecimen, largely between 2001 and 2006. Fasting blood was drawn in heparinized tubes, processed, and separated; the 0.5 mL aliquots of plasma for this study were stored in the vapor phase of liquid nitrogen (−150°C). The MEC and this study were approved by the Institutional Review Boards of the University of Southern California and the University of Hawaii. All participants provided informed consent at time of blood collection.
Cases were diagnosed with primary colorectal cancer after contributing a biospecimen but before the end of 2006. Diagnoses were identified by linkage with population-based cancer Surveillance, Epidemiology and End Results Program registries in Hawaii, Los Angeles County, and the California State Cancer Registry. Deaths were identified by linkage with the state vital statistics databases and the National Death Index. Two controls per case were randomly selected from a pool of subcohort members alive and not diagnosed with colorectal cancer at the age of the diagnosis of the case, and matched to the case by area (Hawaii, Los Angeles), sex, race/ethnicity, birth year (± 1 year), date (± 6 months) and time of blood draw (± 2 hours), and hours fasting before blood draw (8-<10, ≥10).
The concentration of 25(OH)D was determined at Heartland Assays (Ames, IA) using a direct, competitive chemiluminescence immunoassay using the DiaSorin LIAISON platform2 (DiaSorin, Stillwater, MN) (9
). The antibody used is co-specific for 25(OH)D3
. The samples from each matched case-control set were analyzed together within laboratory batches in random order. Laboratory personnel were blind to case status. Based on 27 quality control samples, the intra-batch coefficient of variation (CV) was 6.7% and the intraclass correlation coefficient (ICC) was 0.83. Based on 24 quality control samples, the inter-batch CV was 5.0% and the ICC was 0.82.
The baseline questionnaire, completed between 1993 and 1996, assessed ethnicity/race, weight and height, smoking, physical activity, and diet (7
). Body mass index (BMI) was calculated as weight divided by the square of height (kg/m2
). Volume of physical activity was calculated as metabolic equivalents (METs) spent in moderate activity, strenuous sports, and vigorous work (10
). Diet over the past year was assessed with a food frequency questionnaire from which food nutrient intake was calculated using a food composition table maintained at the Cancer Research Center of Hawaii (7
). Total nutrient intake was the sum from both food and supplement sources.
Of the 234 cases who had a fasting plasma sample available, one had a missing 25(OH)D result and three had missing covariate information. Of the 460 controls matched to the remaining cases, six did not have a sample available, one had a missing 25(OH)D result, and 19 had missing information about diet or BMI. One additional case was not included because both matched controls had been excluded. Thus, 205 cases with two matched controls and 24 cases with one matched control were included in the analyses (229 cases, 434 controls).
The odds ratios (OR) for the association between plasma 25(OH)D and colorectal cancer risk were estimated using conditional logistic regression. Using season-specific quintiles (11
) affected our results minimally so our results use quintiles based on the overall distribution in controls. Included in the models were the matching factors, as well as age at blood draw and hours fasting as continuous variables to control for any residual differences. We evaluated as potential confounders first degree family history of colorectal cancer, BMI, physical activity, pack-years of smoking, years of education, alcohol intake, ever use of nonsteroidal anti-inflammatory drugs for two or more years, processed red meat intake, fruit and vegetable intake, and season (December to June, July to November). Family history, BMI, and intake of processed red meat were retained as confounders because only these factors changed at least one of the ORs associated with 25(OH)D more than 3% when added to the models. Trend tests were calculated by using the logarithm (base 2) of 25(OH)D levels. A square term of log-transformed 25(OH)D levels was added to the model to investigate non-linearity. Supplementary analyses included the investigation of associations by subsite (colon: ICD-O2 C18.0 to C18.9 and rectum: ICD-O2 C19.9 and 20.9) and ethnicity. P
-values to test for the significance of effect modification were derived from Wald tests of the interaction terms. The analyses were done using SAS version 9.1 (SAS Institute, Cary, NC). All statistical tests were two-sided with a 0.05 level of significance.