Alcohol Consumption and the Risk of Colorectal Cancer for Mismatch Repair Gene Mutation Carriers

Study sample

This study involved carriers of heterozygous germline pathogenic mutations in MMR genes who had been recruited by the Colon Cancer Family Registry. Detailed descriptions of study design and recruitment have been published (21) and are available at http://coloncfr.org (21). Between 1997 and 2012, the Colon Cancer Family Registry recruited and interviewed probands who were either recently diagnosed with colorectal cancer that was reported to state or regional population cancer registries in the United States (Washington, California, Arizona, Minnesota, Colorado, New Hampshire, North Carolina, and Hawaii), Australia (Victoria), and Canada (Ontario); or were from multiple-case families referred to family-cancer clinics in the United States (Mayo Clinic, Rochester, MN; and Cleveland Clinic, Cleveland, OH), Australia (Melbourne, Adelaide, Perth, Brisbane, and Sydney), Canada (Ontario), and New Zealand (Auckland). Permission was obtained from probands to contact their affected and unaffected relatives and seek their enrolment in the Colon Cancer Family Registry. For families recruited via population-based registries, first-degree relatives of probands were recruited by all centers. Recruitment was extended to more distant relatives by some centers. For families recruited via family-cancer clinics, attempts were made to recruit also second-degree relatives of affected individuals [details provided by Newcomb and colleagues (21)]. Informed consent was obtained from all participants and the study protocol was approved by research ethics review board at each recruitment center.

Data collection

At the time of baseline recruitment, standardized questionnaires were used to collect self-reported information on demographics, lifestyle factors including alcohol consumption, personal and family history of cancer, cancer-screening history, and history of polyps, polypectomy, and other surgeries from all participants via personal interviews, telephone interviews, or mailed questionnaires. The questionnaires used by the Colon Cancer Family Registry centers are available at http://coloncfr.org/questionnaires. Pathology reports, medical records, cancer registry reports, and death certificates were consulted, when possible, to confirm reported cancer diagnoses and age at diagnosis. Attempts were made to obtain blood samples from all participants and tumor tissue samples from all colorectal cancer-affected participants.

MMR gene mutation testing

Testing for germline mutations in MLH1, MSH2, MSH6, and PMS2 was performed for all population-based probands who had a colorectal tumor that showed impaired MMR function, as evidenced by tumor microsatellite instability (MSI) or absence of MMR protein expression in immunohistochemical analysis. Testing was also performed for the colorectal cancer–affected participants from clinic-based families regardless of tumor MSI or MMR protein expression status. Sanger sequencing or denaturing high-performance liquid chromatography, followed by confirmatory DNA sequencing, was performed to screen for mutations in the MLH1, MSH2, and MSH6 genes. Large duplication and deletion mutations were detected by Multiplex Ligation Dependent Probe Amplification (MLPA) according to the manufacturer's instructions (MRC Holland; refs. 21–23). PMS2 mutation testing involved a modified protocol from Senter and colleagues (7), in which exons 1–5, 9, and 11–15 were amplified with 3 long-range PCRs followed by nested exon-specific PCR and sequencing. The remaining exons (7, 8, and 10) were amplified and sequenced directly from genomic DNA. Large-scale deletions in PMS2 were detected using the P008-A1 MLPHA kit (MRC Holland; ref. 24). Relatives of probands with a pathogenic MMR germline mutation (25) who provided a blood sample were tested for the specific mutation identified in the proband.

Definitions of exposure and outcome

The primary outcome was self-reported diagnosis of colorectal cancer. The primary exposure was grams of ethanol consumption per day from alcoholic beverages from age 20 years to age at colorectal cancer diagnosis or age at censoring.

At baseline, participants were asked to report the number of 12-oz servings of beer or alcoholic cider (included as beer in the remainder of this article), 4-oz servings of wine or 1-oz servings of sake (included as wine), and 1-oz servings of spirits that they had consumed daily or weekly in their 20s, 30s and 40s, and 50s and older, as well as the number of years they had consumed the alcoholic beverages at least once a week for 6 months or longer during each of these periods.

Grams of ethanol in alcoholic beverages (14 g in 12-oz serving of beer, 11.2 g in 4-oz serving of wine, 3.5 g in 1-oz serving of sake, and 9.3 g in 1-oz serving of spirits) were calculated on the basis of average alcohol content in each drink (5% in beer, 12% in wine, 15% in sake, and 40% in spirits), density of ethanol (0.79 g/mL), and converting United States/British Imperial measurements of volume to metric (1-oz equivalent to 29.57 mL).

Grams of ethanol consumption per day from alcoholic beverages when drinking beer, wine, and spirits were calculated on the basis of self-reported number of alcoholic beverages consumed and years of alcohol consumption from age 20 years to age at colorectal cancer diagnosis or age at censoring. Abstainers were defined as carriers who had not consumed alcohol for at least once a week for 6 months or longer throughout this period.

Statistical analysis

Cox proportional regression analyses, with age as the time scale, were used to estimate the association between alcohol consumption and the risk of colorectal cancer for MMR gene mutation carriers. Time at risk started at age 20 years and ended at age of first colorectal cancer diagnosis, any other cancer diagnosis, polypectomy, or age at baseline interview, whichever occurred first. Observation time ended at age of diagnosis of any other cancer, because subsequent cancer treatment and surveillance could have altered later colorectal cancer risk and MMR gene mutations carriers might have changed their behavior following cancer diagnosis.

Because colorectal cancer cases from multiple-cancer families and cases with early-onset colorectal cancer were preferentially tested for MMR gene mutations, selection of carriers was not random with respect to their disease status. This non-random ascertainment was adjusted for in the analysis by applying probability weights to carriers based on the weighted cohort approach developed by Antoniou and colleagues (26). Age-specific incidences of colorectal cancer for MMR gene mutation carriers (27) were used to calculate statistical weights for colorectal cancer–affected and -unaffected carriers for each age-stratum so the proportion of affected carriers in each age-stratum was equal to the proportion of affected carriers in the general population.

Variables that were considered as potential confounders are listed in Table 1. When applicable, all variables, including the alcohol consumption variables, were treated as time-varying covariates. We were not able to generate time-varying variables for consumption of aspirin or ibuprofen, and multivitamin, calcium, and folic acid supplements because we did not have information on carriers' age at first exposure to regular consumption of these medications and supplements. For investigating the associations between consumption of specific types of alcoholic beverages with colorectal cancer risk, after confirming that these variables were not strongly correlated (all correlation coefficients were ≤0.25) and running collinearity diagnostic tests, consumption of the other two sources of ethanol were included as potential confounders in the multivariable models. The proportional hazards assumption was tested with the Schoenfeld and scaled Schoenfeld residuals (28). The confounding variables that did not meet the proportional hazards assumption were stratified for in the final models. The overall model fit was assessed using Cox–Snell residuals as the time variable and plotting them against the Nelson–Aalen cumulative hazard function (29). For continuous variables, deviation from linearity was tested using the likelihood ratio test in comparison between models with the original variable and models that included the original and quadric transformation of the variable. To test for interaction, the difference in the log-likelihood ratio was assessed after adding a cross-product term between each exposure variable and the potential effect modifiers identified a priori. The numbers of missing values for all variables are reported in Tables 1–2. All univariable analyses were complete case analyses. In the multivariable models, all confounder variables were fitted as categorical variables with their missing values (all <8%) coded as an additional category.

We estimated associations by each colorectal cancer site, sex and cancer site, smoking status, body mass index (BMI) at age 20 years, and history of having received sigmoidoscopy or colonoscopy. The following additional analyses were also conducted: analyses with colorectal polyp or cancer as the outcome; and analyses restricted to carriers who received a colorectal cancer diagnosis or were censored within 5 years before interview to reduce survival bias. To take into account the potential correlation in risk between family members, we used the Huber–White robust variance estimation by clustering on family membership (30, 31). All statistical tests were two-sided. All statistical analyses were performed using STATA statistical software, version 13.0 (Stata Corp LP).