The Physicians’ Health Study was a randomized, double-blind, placebo-controlled trial of aspirin and β-carotene among 22,071 U.S. male physicians aged 40 to 84 in 1982, without a history of heart disease, cancer, or major chronic diseases (13
). Participants were excluded if they had a diagnosis of myocardial infarction, stroke, or transient ischemic attack, cancer (except non-melanoma skin cancer), renal or liver disease, peptic ulcer, or gout or used vitamin A or β-carotene supplements. The aspirin component of the trial was terminated on January 25, 1988, because of a substantial reduction in the risk of myocardial infarction (13
). After the end of the aspirin trial, participants were offered a choice of either aspirin or placebo along with their randomly assigned β-carotene supplement or placebo. Blood samples were obtained from 14,916 of the participants during 1982 and 1984 (more than 70% of participants provided blood between September and November 1982) before randomization. Blood collection kits were sent to all participants with instructions to have their blood drawn into the EDTA tubes. These samples were centrifuged to collect both plasma and whole blood, sent to investigators, divided into aliquots, and stored at −82°C. The median interval (10th
percentile) between blood collection and freezing in this analysis was 1.00 (1.0–2.0) days. Samples used in this analysis were thawed only once. Information on height, weight, physical activity, alcohol intake, multivitamin use, and smoking habits was collected by self-administered questionnaires at baseline. The frequency of intake of specified food groups was obtained on the 18-week or 12-month questionnaires. Follow-up is >99% complete for mortality and morbidity. This study was approved by the Institutional Review Board of Brigham and Women’s Hospital.
Cases were identified through annual follow-up questionnaires and then confirmed by subsequent review of medical records by the end-point committee. Among those who provided baseline blood samples, 197 cases were ascertained through July 31, 2000. The majority of the cases were matched to two controls on age (±1 year for younger participants, ±5 years for older participants) and smoking status (never, past, current). However, for 23 cases, only one appropriate control was available. Thus a total of 371 controls were included in our study here; 407 of these blood samples were collected less than 8 h after their last meal (non-fasting).
Pyridoxal 5′-phosphate (PLP), an active form of vitamin B6
, was measured by an enzymatic procedure using radioactive tyrosine and the apo-enzyme tyrosine decarboxylase (14
). The mean intraassay coefficient of variation (CV) was 5%. Plasma vitamin B12
was measured with a commercially available radioassay kit (Ciba-Corning, MA). Plasma homocysteine and cysteine were measured by high-performance liquid chromatography with fluorescence detection (15
). Plasma folate was measured by the microbiological method (16
). All assays for folate, vitamins B6
, and homocysteine were conducted at the Jean Mayer U.S. Department of Agriculture Human Nutrition Research Center on Aging at Tufts University. Plasma TNF-αR2 and IL-6 were measured with enzyme-linked immunosorbent assays (R&D Systems, Minneapolis, MN), and plasma high-sensitive CRP was determined by a highly sensitive immunoturbidimetric assay (Denka Seiken, Niigata, Japan) at the Boston Children’s Hospital laboratory. The mean intraassay CVs were all <10%. Cases and their controls were analyzed in the same batch, and laboratory personnel were blinded to case, control, and quality-control status.
DNA was extracted from whole-blood samples, with a commercially available process based on the absorption of DNA to a silica membrane after lysis with a proprietary agent (Qiagen, Chatsworth, CA). The TaqMan assay was used to genotype methylenetetrahydrofolate reductase (MTHFR) C677T (17
We compared medians and proportions of the baseline risk factors between cases and controls. To test for differences between cases and controls, the paired t-test was used for loge-transformed plasma biomarker levels and the Mantel-Haenszel test for categorical variables. We examined the associations between plasma PLP and other biomarkers as well as intakes of food groups potentially related to colorectal cancer risk using partial Spearman correlations among the controls adjusted for age, smoking status, and lab assay batches.
To estimate the relative risks (RRs) and 95% confidence intervals (95% CIs), we categorized participants in quartiles or by the median of plasma PLP level among the controls and used a conditional logistic regression model to account for the matched case-control study design. To test for trend across quartiles, we assigned participants the median value of their quartile level. This variable was entered as a continuous term in the model, the coefficient for which was evaluated by the Wald test. In multivariate analyses, we adjusted for BMI (<23, 23–<25, 25–<27, ≥27 kg/m2), vigorous exercise (<2, ≥2 times/wk), multivitamin use (never, past, current use), fasting status (<8 hours, ≥ 8 hours), red meat consumption (quartiles), cold cereal intake (<2/wk, ≥2/wk), alcohol intake (<1/wk, 1/wk-<1/d, ≥1/d), and aspirin assignment (yes, no). In additional analyses, we adjusted for plasma levels of folate, homocysteine, vitamin B12, TNF-αR2, IL-6, and CRP to examine whether the association for vitamin B6 was independent of other one-carbon metabolites and inflammatory biomarkers. We examined whether the associations between plasma PLP level and colorectal cancer risk differed by alcohol intake (<1 or ≥1 drinks/d), cold cereal intake (<2 or ≥2 servings/wk), multivitamin use (never, ever), BMI (<25, ≥25 kg/m2), MTHFR C677T genotype (CC/CT, TT), aspirin assignment (yes, no), β-carotene assignment (yes, no), and plasma levels of folate, TNF-αR2, CRP, and IL-6 (two categories based on the median) using conditional logistic regression. To test the null hypothesis that there was no interaction by the potential effect modifiers, we used a Wald test of the cross-product term of PLP as a dichotomous variable, with the specific modifier variable modeled as a dichotomous variable. The study was analyzed with the SAS 9.1 statistical package (SAS Institute, Cary, NC). P<0.05 was considered statistically significant.