Meat intake, cooking methods, dietary carcinogens, and colorectal cancer risk: findings from the Colorectal Cancer Family Registry

Study subjects

The characteristics of the Colon CFR have been previously described 33. In this case–control study, we included incident cases with CRC (affected probands) who were recruited through population-based registries in three of the six centers of the Colon CFR: Cancer Care Ontario (CCO), University of Hawaii (UH), and the University of Southern California (USC) Consortium, which include the Arizona Cancer Center, Dartmouth College, the University of Colorado, the University of Minnesota, the University of North Carolina, and the University of Southern California. Individuals recruited at these three Colon CFR centers completed a quantitative food frequency questionnaire (FFQ) that included a meat cooking module, and therefore had suitable information for this study. We included, from these centers, all affected probands (cases) who had FFQ data available and who completed the interview ≤5 years after diagnosis, for a total of 3486 cases (1887 from CCO, 1153 from USC, 446 from UH). As controls we included unaffected siblings from the families of the probands (N = 1812; 977 from CCO, 499 from USC, 336 from UH), unaffected spouses of the probands (N = 141 from UH), and unaffected population-based controls (N = 1620 from CCO). These controls were included if they were unaffected at the time of diagnosis of the proband (for sibling and spouse controls) or unaffected at the time of recruitment (population-based controls). Details on ascertainment and eligibility criteria used by the CFR were summarized by Newcomb and colleagues 33. All participants signed a written informed consent approved by the Institutional Review Board of each institution, completed an in-person risk factor questionnaire (RFQ), a quantitative FFQ, and a subset of them donated blood or buccal samples. All cases were pathologically confirmed. Tumor tissue blocks were available from approximately 70% of the cases, providing data on tumor characteristics, such as mismatch repair (MMR) status.

Data collection

All participants answered a baseline RFQ, which was designed to be used by all colon CFR sites 33 and includes questions covering established and suspected risk factors for CRC, including medical history and medication use, reproductive history (for female participants), family history, physical activity, demographics, alcohol and tobacco use, and limited dietary factors. The method of administration of the questionnaire varied by center: face-to-face interview, Computer-Assisted Telephone Interview (CATI), or mail for self-administration. Manuals for administration were developed for further standardization. In addition to the risk factor core questionnaire, the Hawaii, Ontario, and USC registry centers also administered a detailed quantitative FFQ developed at the University of Hawaii Cancer Center (UHCC) 34, which included a meat/fish cooking module specifically developed to address the role of meat carcinogens. All questions were asked in reference to two years before the cancer diagnosis. For a subset of cases (N = 1119), data were available either for microsatellite instability (MSI) status or for immunohistochemistry (IHC) phenotype or both. We classified CRC patients as MMR-deficient if there was evidence of MSI-high status or abnormal IHC phenotype 35. Patients with MSI-stable/low status and normal IHC phenotypes were defined as MMR-proficient. If either MSI or IHC status was missing, the determination was made on the basis of the available phenotype. Cases with both MSI and IHC data missing were not assigned MMR status. We were also able to classify patients into sporadic or familial cancer based on Amsterdam criteria.

Assessment of meat intake and meat cooking practices

A validated FFQ was mailed to participants who had answered the RFQ and was returned after completion by 79.6% of study participants in three centers (CCO, USC, and UH). This FFQ obtained data on the usual consumption of more than 200 food items and more than 100 nutrients, including data on portion sizes and frequency of intake of many relevant meat items, including beef, poultry, lamb, pork, and fish as separate items and as part of mixed dishes; processed meats (meats that have been treated with preservatives such as nitrites or nitrites, or treated by curing, smoking, salting, etc.); and fish and processed fish 34. This FFQ has been validated in a random sample of 2000 Hawaii and California residents using 24-h recalls 34. These data (portion size, frequency, and total intake) were used to derive amount (g/day) of each of the individual meat types and summary variables, including: total nonprocessed red meat (beef, pork, veal, lamb, or game); total poultry (chicken, turkey); total processed red meat (bacon, hot dogs, luncheon meats, sausages); total processed poultry. We also examined some of the individual meat types, including: beef, pork, organ meat, sausages, and lunchmeats. In addition, this FFQ contains a meat/fish cooking module which includes questions on frequency of intake of eight meat items (beefsteak, hamburger, chicken, spam or ham, bacon, sausage, short ribs or spareribs, and fish), cooked by either of three different methods (pan-fried, oven-broiled, and grilled or BBQ), and with three levels of doneness (light, medium, dark). These variables were used to derive meat type/cooking method frequency (servings/day) variables for each combination of meat type and cooking method. We obtained estimated intake levels of individual HCAs (3,8-dimethyl-3H-imidazo[4,5-f]quinoxalin-2-amine, MeIQx; 3,4,8-trimethyl-3H-imidazo[4,5-f]quinoxalin-2-amine, DiMeIQx; and 2-amino-1-methyl-6-phenylimidazo(4,5-b)pyridine, PhIP) using the CHARRED database 36, as previously described 24. All intake variables were energy-adjusted using the nutrient density method (in g/1000 kcal/day) and were categorized into quintiles for individual meat types and meat-derived carcinogens and quartiles for meat cooking variables. We also considered the use of two types of marinades, which were asked separately for red meat or chicken. Two types of marinades were considered: teriyaki sauce or shoyu sauce (soy sauce) and BBQ sauce. We created categories of marinated red meat or poultry consumption (never, <1 per month, >1 per month).

Data analyses

In our analyses, we excluded participants with extreme daily caloric intake levels (<600 or >6000 kcal/day). 203 observations were excluded: 136 cases (3.9%) and 67 controls (1.9%). We also excluded two participants with any missing covariate information. Final study population included affected probands (cases, N = 3350), unaffected siblings (N = 1759), unaffected spouses (N = 138), and unaffected and unrelated population-based controls (N = 1607). For nutrient density meat variables, estimated meat mutagen variables, and meat cooking variables from the FFQ, odds ratios (OR) and 95% confidence intervals (CI) were estimated using generalized estimating equations with robust variance estimates to predict binary outcome. This approach allowed for the nonindependence of members of the same family 37. All models were adjusted for age (years, continuous), BMI (<25, 25.0–29.9, ≥30 kg/m²), sex, race (NHW, non-Hispanic white; AA, African-American; Asian and others), saturated fat (g/1000 kcal/day), dietary fiber (g/1000 kcal/day), center, vegetables (g/1000 kcal/day), physical activity (h/week, continuous), and total energy intake (kcal/day, continuous). We further adjusted for potential confounding factors including income (categorical), education (categorical), smoking (current/past/never), NSAID use (current/past/never), medical history of diabetes, ulcerative colitis, Crohn's disease, irritable bowel syndrome, diverticulitis, alcohol (g/1000 kcal/day), dietary folate (g/1000 kcal/day), fruit (g/1000 kcal/day). However, results for the fully adjusted model are not shown because risk estimates did not change by >10%. Risk estimates for quintile/quartile categories of meat variables were estimated for: CRC; colon cancer (ICD-O-2 C180-C188); rectal cancer (ICD-O-2 C199, C209); MMR-proficient cancers; and MMR-deficient cancers. For all analyses, tests for trend were performed by assigning median values to each quartile/quintile category and modeling these categories as a continuous variable. Subtype-specific estimates for tumor site (colon vs. rectal cancer) or MMR status (proficient vs. deficient) were computed using all controls. Heterogeneity of risk estimates between colon versus rectal cancers and MMR-proficient versus MMR-deficient cancers was tested using case-only analyses. For analyses of effect modification by use of marinades of the association between total red meat and chicken intake, we conducted interaction tests using logistic regression and likelihood ratio tests, using energy-adjusted variables for nonprocessed red meat and nonprocessed poultry using the nutrient density method (in g/1000 kcal/day) that were categorized into quintiles, and categorical variables for the frequency of use of the two types of marinades (BBQ sauce and shoyu sauce). These interaction models were adjusted for age (years, continuous), BMI (<25, 25.0–29.9, ≥30 kg/m²), sex, race (NHW, non-Hispanic white; AA, African-American; Asian and others), saturated fat (g/1000 kcal/day), dietary fiber (g/1000 kcal/day), center, vegetables (g/1000 kcal/day), physical activity (h/week, continuous), and total energy intake (kcal/day, continuous). All statistical tests were two-sided and all analyses were done using Stata S/E 11.0 (STATA Corporation, College Station, TX).

Nonprocessed red meat and poultry and CRC risk

When ignoring cooking methods, we did not find statistically significant trends for any of the three specific nonprocessed red meats (beef, pork, organ meat) considered or for total nonprocessed red meat. Similarly, we did not observe statistically significant heterogeneity of effects by tumor localization or MMR status. Estimates of associations for CRC did not change for any of the abovementioned variables when restricting the analyses to those individuals likely to have sporadic CRC, given their tumor characteristics and absence of strong family history of CRC as defined the Amsterdam criteria (data not shown). We also conducted analyses restricting comparisons of cases to unrelated population-based controls, who were available from one CFR center only (Table S1). We observed similar risk estimates for all meat variables except for pork, which showed a positive association with CRC risk (fifth quintile vs. first quintile OR = 1.2; 95% CI = 1.0–1.6; P_trend = 0.028), although this trend was compatible with chance when applying a Bonferroni correction (Table S1). For all nonprocessed poultry combined, we did not see evidence of an association between diets high in poultry and CRC risk (Table 2).

We next considered the modifying effect of frequency of intake of two different types of marinades (BBQ sauce and shoyu) on the association between nonprocessed red meat and nonprocessed poultry with CRC risk. We observed evidence of effect modification of the association between nonprocessed red meat and CRC risk by frequency of use of shoyu marinade. Specifically, we observed that a positive association between increasing levels of nonprocessed red meat and CRC risk was restricted to individuals who reported never eating shoyu marinade (fifth quintile vs. first quintile OR = 1.3; 95% CI = 1.1–1.6; P_trend = 0.007). In contrast, no associations were observed among individuals with low (once a month or less) or high (more than once a month) use of shoyu (test of heterogeneity of trends P = 0.008) (Table S2). We also observed some evidence that the frequency of intake of BBQ sauce modified the association between high nonprocessed red meat intake and CRC risk. Specifically, we observed that this association was slightly stronger among individuals who reported using BBQ sauce marinade more than once a month (fifth quintile vs. first quintile OR = 1.5; 95% CI = 0.9–2.4; P_trend = 0.01). Although positive associations were observed between some levels of intake of nonprocessed red meat and CRC risk among individuals who reported never using BBQ sauce, the trend was not statistically significant. No comparable trend was observed among individuals who reported using BBQ sauce less than once per month. However, the test of heterogeneity of trends did not reach statistical significance (P = 0.084) (Table S2). No differences in trends were observed for nonprocessed poultry for use of BBQ sauce or shoyu marinade.

Processed red meat and poultry and CRC risk

For total processed red meat, no statistically significant associations with CRC were observed, and there was no statistically significant heterogeneity between colon and rectal tumors, or by MMR status. There was no statistically significant trend between diets high in sausages and lunchmeats and CRC risk; however, we observed for these foods a statistically significant positive association among MMR-proficient CRC tumors (fifth quintile vs. first quintile OR = 1.3; 95% CI = 1.0–1.7; P_trend = 0.029), which was compatible with chance when considering a Bonferroni correction, but the test for heterogeneity was not statistically significant (P = 0.069) (Table 2). Similarly, there was no association between diets high in processed poultry and CRC risk, but we observed a significant positive association for these foods with among MMR-proficient CRC tumors (fifth quintile vs. first quintile OR = 1.3; 95% CI = 1.0–1.6; P_trend = 0.046), which was also compatible with chance when considering a Bonferroni correction, and no statistically significant test of heterogeneity (P = 0.166) (Table 2).

In general, similar trends and estimates were observed when restricting analyses to cases and population-based controls (Table S1). Moreover, estimates of associations for CRC did not change for either processed meat variable when restricting analyses to those individuals without familial history of CRC, as defined by the Amsterdam criteria (data not shown).

Pan-fried meats and CRC risk

We investigated the association between CRC risk and frequency of intake of any of the following meats cooked by pan-frying (defined as “cooked in a preheated frying pan or griddle”): beefsteak, hamburger, chicken, sausage, spam or ham, or bacon. We observed a statistically significant positive association between intake of beefsteak and CRC risk (P_trend < 0.0002), which was not compatible with chance after Bonferroni correction (Table 3). Even though this association seemed stronger for colon (P_trend = 0.0005) than rectal cancer (P_trend = 0.09), the test of heterogeneity was not statistically significant. Moreover, this association with beefsteak seemed stronger among MMR-deficient cases (P_trend = 0.002; heterogeneity P = 0.059). We also noted positive associations between intake of pan-fried sausage and spam or ham and CRC risk, which were statistically significant, albeit compatible with chance when considering a Bonferroni correction (P_trend = 0.04 and P_trend = 0.048, respectively). However, the association with spam or ham was stronger when analyses were restricted to MMR-proficient cases, and not compatible with chance when considering a Bonferroni correction (P_trend = 0.0001; heterogeneity P = 0.026). Further consideration of level of doneness for each of these variables did not show appreciable differences between light- or medium-brown versus dark-brown meats. Restricting analyses to cases and population-based controls showed similar trends and estimates (Table S1). Similarly, restricting analyses to cases without family history of CRC (Amsterdam criteria), resulted in similar estimates (data not shown).

Oven-broiled and grilled meats and CRC risk

We investigated the association between CRC risk and any of the following meats cooked by oven-broiling, defined as “cooked at the broil setting”: beefsteak, hamburgers, chicken, and short ribs or spareribs. We observed a positive association between frequent intake of oven-broiled short ribs or spareribs and CRC risk (fourth quartile vs. first quartile OR = 1.2; 95% CI = 1.0–1.5; P_trend = 0.0002), which remained of borderline statistical significance after Bonferroni correction, without evidence of heterogeneity by tumor site (Table 4). This association was restricted to MMR-deficient cases (fourth quartile vs. first quartile OR = 1.9; 95% CI = 1.2–3.0; P_trend = 0.003; tests of heterogeneity P = 0.05) (Table 4). Similar trends and estimates were observed when restricting analyses to cases and population-based controls (Table S1).

We also investigated the association between CRC risk and any of the following meats cooked by grilling or barbecuing, defined as “cooked over charcoal or on an electric or gas grill”: beefsteak, hamburgers, chicken, sausage, and short ribs or spareribs. We did not observe any positive associations with CRC risk for any of these variables (Table 5). However, we observed a statistically significant inverse association between higher frequency of intake of grilled or BBQ hamburgers and CRC risk (fourth quartile vs. first quartile OR = 0.8; 95% CI = 0.7–0.9; P_trend = 0.002) (Table 5). This association was of borderline statistical significance after Bonferroni correction. Similar estimates were observed when restricting analyses to cases and population-based controls (Table S1).

HCA and CRC risk

We considered possible associations between CRC risk and imputed estimates of dietary intake for three different HCAs known to occur in meats: PhiP, MeIQx, and DiMeIQX. We only observed a statistically significant positive trend of association between increasing levels of DiMeIQx and MMR-deficient CRC (fifth quintile vs. first quintile OR = 1.2; 95% CI = 0.8–1.8; P_trend = 0.042; P for heterogeneity by MMR status = 0.17) (Table S3). However, this association was compatible with chance after Bonferroni correction. Similar trends were observed for PhIP and MeIQx, but neither were statistically significant. Slightly stronger evidence of a positive association for DiMeIQx and MMR-deficient CRC was observed when restricting analyses to cases and population-based controls (Table S1).