By continuing to browse this site you agree to us using cookies as described in About Cookies
Notice: Wiley Online Library will be unavailable on Saturday 7th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 08.00 EDT / 13.00 BST / 17:30 IST / 20.00 SGT and Sunday 8th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 06.00 EDT / 11.00 BST / 15:30 IST / 18.00 SGT for essential maintenance. Apologies for the inconvenience.
Correspondence to: Christine L. Parr, Institute of Basic Medical Sciences, Department of Biostatistics, University of Oslo, P.O. Box 1122 Blindern, N-0317 Oslo, Norway, Tel.: +47-22851280, Fax: +47-22851313, E-mail: email@example.com
Red and processed meat intake is an established risk factor for colorectal cancer (CRC), but epidemiological evidence by subsite and sex is still limited. In the population-based Norwegian Women and Cancer cohort, we examined associations of meat intake with incident proximal colon, distal colon and rectal cancer, in 84,538 women who completed a validated food frequency questionnaire (FFQ) during 1996–1998 or 2003–2005 (baseline or exposure update) at age 41–70 years, with follow-up by register linkages through 2009. We also examined the effect of meat cooking methods in a subsample (n = 43,636). Multivariable hazard ratios (HRs) were estimated by Cox regression. There were 459 colon (242 proximal and 167 distal), and 215 rectal cancer cases with follow-up ≥ 1 (median 11.1) year. Processed meat intake ≥60 vs. <15 g/day was associated with significantly increased cancer risk in all subsites with HRs (95% confidence interval, CI) of 1.69 (1.05–2.72) for proximal colon, 2.13 (1.18–3.83) for distal colon and 1.71 (1.02–2.85) for rectal cancer. Regression calibration of continuous effects based on repeated 24-hr dietary recalls, indicated attenuation due to measurement errors in FFQ data, but corrected HRs were not statistically significant due to wider CIs. Our study did not support an association between CRC risk and intake of red meat, chicken, or meat cooking methods, but a high processed meat intake was associated with increased risk of proximal colon, distal colon and rectal cancer. The effect of processed meat was mainly driven by the intake of sausages.
European Prospective Investigation into Cancer and Nutrition
food frequency questionnaire
International Classification of Diseases, Seventh Edition
Norwegian Women and Cancer cohort study, 24-HDR: 24-hr dietary recall
Intake of red and processed meat has been established as a risk factor for colorectal cancer (CRC) in several,[1-6] but not all[7, 8] meta-analyses of prospective studies. The evidence was judged convincing in the 2007 World Cancer Research Fund/American Institute of Cancer Research (WCRF/AICR) report, and again as supportive of an association in 2011 by the WCRF/AICR Continuous Update Project, but the estimated increase in CRC risk per 100 g/day increase in red meat intake was reduced from 29 to 17%. For processed meat the figures reported per 50 g/day increase were similar in 2007 and 2011: 21 and 18%, respectively. In contrast to the Continuous Update Project, Alexander et al.[7, 8] concluded that summary associations of CRC risk were modified by sex and tumor site, with stronger associations in men than women, and colon than rectal tumors (red meat only). However, the epidemiologic evidence from cohort studies is still limited for subsites, in particular of the colon,[9-14] with few reports of sex-specific estimates.[9, 11, 14] Several mechanisms involving meat components have been proposed to explain the association of red and processed meat intake and CRC risk. Heterocyclic amines (HCAs), a class of potent mutagens and carcinogens in animal models are formed in meat and fish cooked at high temperatures, such as pan-frying and grilling/barbequing. Thus, meat cooking methods could play a role. Other mechanisms more specific to red and processed meats, involve heme iron and mutagenic N-nitroso compounds (NOC),[18, 19] and recently the involvement of bovine viruses has been hypothesized. Multiple lines of evidence, including epidemiologic, clinical and biologic,[21, 22] indicate that proximal and distal colon cancers could have distinct etiologies with different susceptibility to environmental factors, as indicated for colon and rectal cancers.
Meat consumption patterns have been found to vary distinctly cross Western Europe, with Norwegian women having a particularly high intake of processed meat. Using data from the Norwegian Women and Cancer study, a population-based cohort, the objective of the present study was to examine associations between meat intake and risk of proximal colon, distal colon and rectal cancer, taking into account additional information about meat cooking methods. We also estimate the effect of measurement errors in food frequency questionnaire (FFQ) data, on the associations.
Material and Methods
The NOWAC cohort and linkages
NOWAC is a national, population-based cohort with over 172,000 participants recruited since 1991.[25, 26] The study has been approved by the Regional Committee for Medical Research Ethics and the Norwegian Data Inspectorate. All participants have given their informed consent. The unique 11-digit identity number of Norwegian citizens was used to link individuals from NOWAC to the population register at Statistics Norway for postal address and follow-up of vital status (alive, emigrated or dead), and to the Cancer Registry of Norway for cancer incidence until December 31, 2009. CRC was registered according to the International Classification of Diseases, Seventh Edition (ICD-7; codes 153.0–153.9 for colon cancer, and 154.0 for rectal cancer). Codes 153.0 (cecum, ascending colon), 153.1 (transverse colon, including hepatic and splenic flexures) and 153.6 (appendix) were defined as proximal, and codes 153.2 (descending colon) and 153.3 (sigmoid colon) as distal colon cancer in subsite analyses.
The current study included 95,906 women (161,822 invited, 59% response overall) who returned a mailed health- and lifestyle questionnaire incorporating a validated FFQ[27, 28] during 1996–1998 (either baseline or first exposure update) or during 2003–2005 (baseline) at age 41–70 years. Participants with FFQ data updated in 1998 (about 37,000) constitute the Norwegian subcohort in the European Prospective Investigation into Cancer and Nutrition (EPIC) study.
At start of follow-up (registration date of the returned baseline- or exposure update questionnaires) 91,836 women were alive without cancer (4,033 women had any type of prevalent cancer, 13 were dead and 22 emigrated, of whom 19 with unknown emigration date, and two had unknown vital status). After excluding those with a blank FFQ (n = 109), daily energy intake <2.5 MJ (n = 814) or >15 MJ (n =112), missing values for all meat intake frequencies (n = 1,543), or for one or more covariates (n = 4,087) except physical activity (missing was included as a separate category to preserve 72 CRC cases), a reported body mass index (BMI, kg/m2) <13.0 (n = 4) or >60 kg/m2 (n = 19), or follow-up time <1 year (n = 610), there were 84,538 women included (72% obtained during 1996–1998 and 28% during 2003–2005) in the analysis of meat intake, of whom 43,636 had data on cooking methods, and 220 had FFQ validation data used for calibration purposes in the current study. A subsample of 7,020 NOWAC women with two FFQ measurements of meat intake (baseline 1997 and update 2002) was used to investigate changes in meat intake over time.
Assessment of diet and cooking methods
The FFQ covered habitual, but not total diet, during the previous year with foods commonly consumed in Norway. Intake of red meat included “roast meat (beef, pork and mutton),” “chops” and “steak”; processed meat included “meatballs, hamburgers,” “sausages” and “sandwich meats, liver pâté”; dishes with meat included “casseroles, stew,” “pizza with meat” and “other meat dishes.” The only poultry meat included in the FFQ was “chicken” (one item). Amounts (g/day) were calculated from intake frequencies (maximum ≥4 times/day for sandwich meats and ≥2 times/week for other items) and collected portion sizes or standard servings/amounts (only used for steak, chicken and sandwich meats). Baseline and exposure update FFQs from 1996 to 1998 were generally shorter (66–78 items) than the baseline FFQs from 2003 to 2005 (99 items). Questions about meat intake were identical, except for an early FFQ version with separate questions for wiener and a more chunky “dinner sausage” that were later combined into “sausage.” Alcohol intake (g/day) was calculated from FFQ data on consumption frequencies (maximum ≥2 times/day) and standard amounts for “beer,” “wine,” “spirits” and also “fortified wines/liqueur” for participants included in 2003–2005. Intake of energy, nutrients and alcohol was calculated using values from the Norwegian Food Composition table. Most questionnaires from 1997 to 1998 (52% of all 95,906 questionnaires) included a section on cooking methods with questions about the preferred color of frying crust, the use of frying fat or pan residues, hereafter referred to as just “pan residues,” and the intake of fried or grilled “dark meat (steak, etc.),” “processed meat (meatballs, etc.),” “bacon,” “white meat (chicken, etc.)” and “fish” with frequency range never/rarely to ≥7 times/week (six alternatives) without portion sizes. In the FFQ validation study, the reference method was four nonconsecutive 24-hr dietary recalls (24-HDRs) collected per person over a year (2002–2003) by telephone interview and the computer program EPIC-SOFT.
Assessment of other exposures
Dietary intake and other exposures were assessed from the same health- and lifestyle questionnaire. BMI was calculated from weight (kg) and height (cm) data. Questions about smoking history and current smoking were combined into a variable for smoking status. Overall current physical activity level (including both recreational and occupational) was recorded on a rating scale from 1 (very low) to 10 (very high), which has been validated against a combined heart rate and movement sensor according to the EPIC protocol. In a sample of NOWAC women, the scale range from 1 to 10 corresponded to mean values of 0.8 and 3.4 hr/day, respectively, of moderate to vigorous physical activity (i.e., >3 metabolic equivalent of task units) with a linear increase across categories (ptrend < 0.001).
Cox proportional-hazards regression models with age as time-scale were used to estimate hazard ratios (HRs) with 95% confidence intervals (CIs) for associations between meat intake or cooking method, and risk of CRC. Entry time was the age at start of follow-up, and exit time the age at diagnosis of any cancer, emigration, death, or end of follow-up in December 31, 2009, whichever occurred first. We estimated HRs for categories of meat intake (g/day) or cooking method, using the lowest category as the reference. For meat intake we used evenly spaced categories compatible with the case distribution: 10 g increments for red meat, except the lowest category (<5 g); 15 g for processed meat, 30 g for total red and processed meat, and 20 g for dishes with meat. For chicken the category levels were the intake frequencies (never/rarely, 1/month, 2–3/month and ≥1/week) converted to g/day using a standard portion size. The main multivariable models for meat intake or cooking methods were adjusted for continuous energy intake (MJ/day) and the following covariates selected a priori based on the literature[2, 3]: alcohol (g/day),[2, 3] fiber (g/day), calcium (mg/day) and BMI (kg/m2),[2, 3] all continuous, and categories of smoking status (never, ex and current) and current physical activity[2, 3] level (1–10 scale where 1–3 = low, 4–7 = moderate, 8–10 = high). Correlations between meat variables, and between meat variables and other covariates were assessed with Spearman's correlation coefficient, rs. Sensitivity analyses to test the robustness of the results on meat intake (Table 2) included additional adjustment for folate (continuous) or fruit and vegetables (both continuous), or household income; adjustment for fiber and calcium as nutrient densities (per 10 MJ); mutual adjustment for different meat groups; analysis of meatballs/hamburgers defined as red meat instead of processed meat; exclusion of 19 participants, including three CRC cases, with intake of sausage >86 g/day (natural threshold in the data distribution); and correction for measurement error as later described. Sensitivity analyses of the results on cooking methods (Table 3) included adjustment for folate, or fruit and vegetables in addition to the main model covariates. Trends in HRs were evaluated by fitting the Cox models with meat or cooking method as a continuous variable, using the category levels on an ordinal scale, and the Wald chi-square statistic. We also used the Wald statistic to test for heterogeneity between colon subsites.
Statistical interaction was evaluated with the likelihood ratio test, comparing multivariable models before and after adding a product term. Tests of interaction between categories of meat intake and follow-up time (≤6 or >6 years), or attained age (continuous) indicated no significant time-period effects or deviations from the proportional hazards assumption for the Cox model (p ≥ 0.08). We tested for interaction between meat group and cooking method in models for total CRC risk. To increase the power of the test, we created binary exposure variables for meat group intake (bottom two versus. higher categories) and cooking method (dark brown vs. medium/light brown frying crust; always/usually vs. never/sometimes use of pan residues; fried or grilled meat intake ≤3/month vs. ≥1/week for dark meat, ground meat, bacon and white meat; <2/week vs. ≥2/week for total meat). Based on the cross-tabulation of the binary variables, we created an exposure variable for the four different combinations of processed meat intake (<30 vs. ≥30 g/day) and use of pan residues (always/usually vs. never/sometimes).
A regression calibration procedure implemented in the SAS macro %BLINPLUS was applied to correct Cox regression coefficients for measurement error in the FFQ variables, using the 24-HDR data as the reference. The linear regression coefficient from regressing the 24-HDR data on the FFQ data is equivalent to the correction factor (denoted by λ) for crude HRs, which can be corrected as HR1/λ. As only error in continuous variables can be corrected with this procedure, HRs for meat intake are presented as continuous effects per 50 or 100 g daily increase before and after calibration. Because of limited statistical power of the validation data, the calibrated HRs were only adjusted for energy intake. Energy appeared to be the strongest potential confounder for all meat types (rs of 0.17 for red meat, 0.34 for processed meat, 0.34 for dishes with meat and 0.16 for chicken) based on the correlation matrix with all covariates in the main dataset. We tested for non-linearity by comparing energy-adjusted models with and without an additional quadratic term for meat intake, using the likelihood ratio test.
All tests were two-sided with statistical significance set at p <0.05. Analyses were done in SAS 9.2.
Among the selected background characteristics of the study sample (Table 1), the association with meat intake was inverse for age, weakly positive for BMI (except chicken), and positive for intake of energy, alcohol (in particular red meat and chicken) and all nutrients (except fiber in red meat categories). The high red meat category (≥35 g/day, 5% of sample) was further characterized by the highest proportion of current smokers (44%), participants that prefer a dark brown frying crust (20%), always use pan residues (16%) and consume fried or grilled meat at least three times per week (57%). The high processed meat category (≥60 g/day, 12% of sample) was also characterized by a high proportion of current smokers (38%), and the highest median energy intake (8.01 MJ/day). Only processed meat intake was inversely associated with household income. The high chicken category (≥28 g/day, 20% of sample) had the highest proportion of high-income households (24%), the highest median intake of alcohol (2.6 g/day), vegetables (157 g/day) and fruit (206 g/day), and a low proportion of current smokers (26%) and individuals always using pan residues (8%). The association with current physical activity level was inverse for red meat and processed meat, and positive for dishes with meat and chicken, but overall weak. Compared with the total sample of 84,538 women, the validation subsample (n = 220) was similar, but slightly younger (mean age 47.4 vs. 51.5 years) and leaner (mean BMI 24.1 vs. 24.7 kg/m2) with fewer daily smokers (24% vs. 30%).
Table 1. Background characteristics according to lowest and highest category of meat intake in study participants aged 41–70 years from the Norwegian Women and Cancer cohort study (n = 84,538)
Red meat: roast (beef, pork, mutton), chops, steak. Processed meat: meatballs/hamburgers, sausages, sandwich meats/liver paté. Dishes with meat: casseroles/stew, pizza with meat, “other” meat dishes. The highest intake category in the analysis of colon and rectal cancers was ≥25 g/day for red meat, and ≥60 g/day for dishes with meat.
Adjusted for intake of energy (MJ/day), alcohol (g/day), fiber (g/day), calcium (mg/day), body mass index (kg/m2), smoking status (never, ex, current), and physical activity score categories (1–3 = low, 4–7 = moderate, 8–10 = high, missing).
Red meat (n = 83,997)
Processed meat (n = 84,210)
Red and processed meat (n = 83,753)
Dishes with meat (n = 83,262)
Chicken (n = 80,740)
Table 3. Hazard ratios (95% confidence interval) for risk of total colorectal cancer according to meat cooking methods in the Norwegian Women and Cancer cohort study (n = 43,636)
Adjusted for intake of energy (MJ/day), alcohol (g/day), fiber (g/day), calcium (mg/day), body mass index (kg/m2), smoking status (never, ex and current), and physical activity score categories (1–3 = low, 4–7 = moderate, 8–10 = high, missing).
Fried/grilled white meat (chicken etc.) (n = 40,373)
Fried/grilled all meat (times/week) (n = 41,145)
In the total sample, there were 674 cases of incident CRC, of which 459 (68%) occurred in the colon and 215 (32%) in the rectum during 834,583 person-years with a median follow-up time of 11.1 (range 1.0–13.6) years. Of all 459 colon cancers, 242 (53%) could be classified as proximal and 167 (36%) as distal. The 50 remaining cases (30 in the recto-sigmoid junction and 20 in unspecified sites, ICD-7 153.4 and 153.9, respectively) were censored at time of diagnosis in the analysis of proximal and distal sites. The mean age at diagnosis (all CRC) was 61.4 (range 42.2–82.6) years.
Compared with the lowest intake group, those in the highest category of processed meat intake had a 54% higher risk of colon cancer and 71% higher risk of rectal cancer (Table 2). The association with processed meat intake was stronger for distal colon than for proximal colon cancer, but the difference was not statistically significant (pheterogeneity = 0.55).
Results were generally robust to the different sensitivity analyses. However, the association between processed meat intake (≥60 vs. <15 g/day) and rectal cancer in the main multivariable model lost significance after additional adjustment for all other meat groups (HR = 1.45, 95% CI: 0.82–2.55, ptrend = 0.53), or household income (HR = 1.48, 95% CI: 0.87–2.49, ptrend = 0.47). HRs for processed meat intake ≥60 g/day without meatballs/hamburgers were slightly lower and non-significant, but generally compatible with the estimates (i.e., within 95% CIs) in Table 2. Significantly higher risk of total CRC (30%), total colon cancer (34%) and proximal colon cancer (69%) was found for the highest versus lowest quintile of intake (≥36 vs. <13 g/day), results not shown. The continuous effect of processed meat per 50 g/day increase with meatballs/hamburgers (Fig. 1) was similar to that without (HR = 1.25, 95% CI: 1.00–1.57). No significant associations between risk of CRC and red meat intake was found after adding meatballs/hamburgers to red meat, analyzing quintiles, higher intake categories up to ≥50 g/day, and continuous effects.
Analysis of individual items indicated that sausages contributed most to the observed associations with processed meat: multivariable HRs for total CRC were 1.29 (95% CI: 0.99–1.67, ptrend = 0.09) for sausage (≥25 vs. <5 g/day), 1.19 (95% CI: 0.71–1.98, ptrend = 0.75) for meatballs/hamburgers (≥25 vs.<5 g/day) and 1.13 (95% CI: 0.86–1.48, ptrend = 0.36) for sandwich meats/liver pâté (≥35 vs. <5 g/day). The result for sausage was similar after exclusion of participants with the highest intake (>86 g/day) with a HR of 1.25 (95% CI: 0.96–1.63, ptrend = 0.14).
Before calibration for measurement error the multivariable and energy-adjusted HRs were similar for both processed meat (Fig. 1) and red meat (Fig. 2). Energy-adjusted HRs for processed meat were strengthened after calibration, indicating some attenuation due to measurement errors, but were not statistically significant due to wider CIs (Fig. 1). Tests of nonlinearity in meat intake were not significant.
The validation data indicated some under-reporting of red meat and processed meat and over-reporting of dishes with meat and chicken in the FFQ compared with the 24-HDRs with rs of about 0.2 for red meat, 0.3 for processed meat and chicken, and 0.1 for dishes with meat, and correction factors λ (95% CI) of 0.62 (0.26–0.98) for red meat, 0.42 (0.18–0.66) for processed meat, 0.33 (0.07–0.58) for dishes with meat, and 0.35 (0.16–0.53) for chicken.
No significant effects were found of meat cooking methods or intake frequencies of fried or grilled meat on the risk of CRC (Table 3). However, the use of pan residues significantly modified the association between total CRC risk and processed meat intake (pinteraction = 0.02). Always/usually use of pan residues was associated with significantly increased CRC risk in those having a low intake (<30 g/day), but not in those having a high intake (≥30 g/day) of processed meat with HRs (95% CI) of 1.39 (1.03–1.88) and 0.99 (0.70–1.38), respectively, compared with the reference group (low intake and never/sometimes use of pan residues).
In this large, population-based cohort of Norwegian women, we found that processed meat intake ≥60 g/day was associated with a significantly increased risk of cancer in all colorectal subsites, compared with intake <15 g/day. The risk increase was higher for distal (113%) than proximal (69%) colon cancer, and higher for rectal (71%) than total colon (54%) cancer, which included 50 additional cases excluded from the analysis of proximal and distal sites. The overall risk increase associated with a 50 g/day increase in processed meat intake was significant for total CRC (21%) and proximal colon cancer (45%). The positive association between processed meat consumption and CRC risk was mainly driven by the intake of sausages. Our study did not support an association between CRC risk and intake of red meat, dishes with meat, chicken, or meat cooking methods. Regression calibration indicated attenuation of most observed effects, but calibrated HRs were not significant due to wider CIs.
In the most recent dose–response meta-analysis of prospective studies of meat intake and CRC incidence, Chan et al. reported a pooled relative risk increase of 18% for total CRC per 50 g/day increase in processed meat intake (men and women combined), which is very similar to our estimated 21%. More importantly, our study contributes to the limited evidence on colon subsites. Summary estimates by Chan et al. based on two[4, 10] (dose-response) and five[9-13] (highest vs. lowest) studies only, suggested a larger increase in distal than proximal colon cancer risk with increasing processed meat intake. Their findings are consistent with our results, although the test of heterogeneity (proximal vs. distal) was nonsignificant. By contrast, a very large study published after Chan et al. based on the Korean National Health Insurance Corporation, found a larger effect of meat intake on proximal than distal colon cancer, lending some support to our result that risk increases in both segments of the colon.
To our knowledge, estimates of colon cancer risk by subsites in women have only been reported in two cohort studies; the Swedish mammography cohort, and the Korean National Health Insurance Corporation. In Swedish women, the risk of distal colon cancer was significantly increased for red, but not processed meat. These findings were reported for a higher intake of red meat (≥94 vs. <50 g/day) and lower intake of processed meat (≥32 vs. <12 g/day) than in our study. There was no evidence of an association with proximal colon cancer, whereas in Korean women, the risk of proximal colon cancer was significantly increased for meat intake (not specified) ≥4 versus ≤ 1 times/week. Moreover, the risk increase was higher in women than in men (HR = 1.7 vs. 1.3), suggesting a sex difference. For distal colon cancer, the risk estimates were similar (HR = 1.3), but only significant in men. This is probably the first study to present results by colon subsites in men and women from the same population, but the risk estimates are based on a rather crude measure of meat intake.
Potential explanations for the finding that processed, but not red meat increased CRC risk in the current study, include differences in intake levels, one or more components specific of processed meats, residual confounding by health- and lifestyle factors, or a combination. EPIC data based on a sample of single 24-HDRs from the entire cohort, including about 1,800 NOWAC women (data collected during 1999–2000), showed that the Norwegian/NOWAC women had the highest mean intake of total processed meat (about 45 g/day), the main source being heated sausages (intake about 20 g/day), while the mean intake of fresh red meat (about 30 g/day) was at the low end of the population range for women in all EPIC countries. Thus, the level of red meat intake in NOWAC may be too low to significantly affect CRC risk. Supporting this explanation, there was no elevated risk of CRC in red meat intake categories <40 g/day within EPIC, or of distal colon cancer for red meat intake <60 g/day in a cubic splines regression model in the Swedish mammography cohort.
In epidemiological studies, red meat usually refers to beef, lamb and pork. Processed meat is a more heterogeneous food group without a precise definition, but typically includes sausages, hamburgers, smoked, cured and salted meat and canned meat. Intake of sausages appeared to be more strongly associated with CRC risk than other processed meat items in our study. This is supported by a previous Norwegian cohort study where colon cancer risk was increased for sausage intake (poached or fried) ≥ 5 vs. < 1 times/month in women, but based on few cases with the highest intake. In the PLCO cancer screening trial, a greater intake of bacon and sausage was associated with increased colon adenoma risk, while the total intake of processed meat was not, also pointing to a stronger effect of specific processed food items. Sausages made by the main Norwegian manufacturer (i.e., Nortura), typically contain a mixture of pork and beef with a meat content around 50–60% (label information, Gilde sausages). Thus, sausages is a source of red meat and heme iron, but also other substances, such as nitrates/nitrites, which are commonly added to cured and other processed meats for preservation and color purposes. Heme iron may promote CRC by catalyzing the endogenous formation of carcinogenic NOCs, as well as cytotoxic and genotoxic lipid peroxidation end products. It has been speculated that nitrosylheme formed in processed meat due to the presence of nitrates/nitrites, is more toxic than fresh meat myoglobin. A randomized dietary intervention study found that meats cured with nitrite had the same effect as fresh red meat on endogenous nitrosation, but showed increased oxidative DNA damage. In Norway, the use of nitrates/nitrites and other food additives has been regulated by the EU since 1995, but added levels could vary between manufacturers and countries. Results from the NIH-AARP cohort indicate the involvement of multiple meat components in colorectal carcinogenesis, including heme iron, nitrite/nitrite and HCAs, but also potential collinearity problems in the analyses, as reported correlations were >0.8 for red meat and heme iron, and >0.9 for processed meat and both nitrates and nitrites in meat. A previous EPIC study showed that frying and boiling were the most common methods for cooking processed meat among NOWAC women. Sausages may be fried, but are frequently just heated in water, and may not be an important source of HCAs. In Norway, sausages are a common fast-food. As previously pointed out, a high intake of processed meat may just be an indicator of inappropriate diet and lifestyle factors associated with increased CRC risk,[37, 39] or low socioeconomic status. In the current study, only processed meat was inversely associated with household income.
A possible explanation for the finding that processed meat was associated with a higher risk of distal than proximal colon cancer could be that various meat components or environmental factors are acting at different locations or along different pathways within the colorectum, which seems plausible in light of known anatomical, physiological and molecular differences between segments of the large intestine.[21, 22]
Out study did not support an association between intake frequencies of fried/grilled meat, the preferred color of frying crust, or the use of pan residues and CRC risk. We have no clear explanation for the increased risk of CRC for always/usually use of pan residues found only in combination with low processed meat intake (<30 g/day), and it could be a chance finding. Questions related to meat cooking are commonly used as markers of HCA exposure in epidemiological studies. Zheng et al. concluded that most studies provide some evidence of a positive association of well-done meat or HCA intake with risk of CRC of adenomas, based on a review of publications since 1996, predominantly case-control studies. Our questions may not have adequately captured HCA intake, as the formation is dependent on many factors, temperature and time being most important, but also cooking method and type of meat. More detailed meat cooking FFQs have been developed specifically to estimate the intake of HCAs[42, 43] in conjunction with mutagen databases, such as CHARRED (www.charred.cancer.gov) for meats consumed in the United States. This type of FFQ has been used in several large US cohort studies[13, 44, 45] and a German subcohort of EPIC with inconsistent results. Meat and HCA intake was associated with higher risk of CRC in the NIH-AARP cohort, but not in the Multiethnic Cohort Study. In the German cohort, risk of adenoma was increased for a specific HCA (PhIP), but not the HCAs found to have an effect in the NIH-AARP cohort (i.e., MeIQx and DiMeIQx). The Health Professionals Follow-up Study indicated higher risk of colorectal adenomas for an overall measure of meat-derived mutagenicity, but not HCAs, suggesting the presence of other active mutagens. Other factors that could contribute to the discrepant findings include the presence of anti-carcinogens in the diet, which may counteract the negative effects of HCAs,[46, 47] genetic variations in enzymes responsible for the bioactivation or detoxification of HCAs,[48, 49] and measurement errors in HCA exposure.
Study strengths and limitations
The strengths of the current study include a relatively large sample size with sufficient statistical power to analyze colon subsites in a population based cohort with nearly complete long-term follow-up through register linkages, and with high external validity. HRs were estimated in participants with at least one year of follow-up to control for any undiagnosed cancers and related dietary changes at the time of FFQ completion (i.e., reverse causality). The FFQ used in the study has been validated against multiple 24-HDRs which we used to correct for measurement error in meat and energy intake. However, there is a potential for misclassification of women with a high intake of red meat, processed meat or chicken from mixed dishes (e.g., pizza and casseroles), as we could not disaggregate the meat content, and of women with substantial changes in diet or other CRC risk factors over time, as the analysis was based on a single exposure measurement. However, the subsample of about 7,000 NOWAC women with two FFQ measurements, indicated modest changes in meat intake with mean decreases of 1 g/day for red meat and 3 g/day for processed meat, while the intake of chicken increased by 3 g/day over a 5-year-period. Last, residual confounding due to incomplete covariate adjustment cannot be ruled out and we could not assess the effect of potential confounders such as type two diabetes, family history of CRC, or use of non-steroidal anti-inflammatory drugs. The calibration procedure also has some limitations. The FFQ and 24-HDR methods have a potential for correlated reporting errors, which we could not correct for. Also, four 24-HDRs may be insufficient to capture the usual intake of different meat groups, indicated by many participants having null intake in all 24-HDRs, but not in the FFQ. A linear trend in HRs was only evident for processed meat, but we estimated continuous effects for all meat groups because implemented regression calibration procedures cannot handle the categorical variables used in the main analysis.
In conclusion, processed meat intake was associated with increased risk of cancer in all colorectal subsites in this population of Norwegian women with a high processed meat intake, and the effect was mainly driven by the intake of sausages. The risk increase was higher for distal than proximal colon cancer, and higher for rectal than total colon cancer. Various meat components could be responsible for the effect and act at different locations within the colorectum. Our results support current recommendations to limit processed meat intake.