Zisman AL, Nickolov A, Brand RE, Gorchow A, Roy HK. Associations Between the Age at Diagnosis and Location of Colorectal Cancer and the Use of Alcohol and TobaccoImplications for Screening. Arch Intern Med. 2006;166(6):629-634. doi:10.1001/archinte.166.6.629
Copyright 2006 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2006
Individualizing recommendations for colorectal cancer (CRC) screening intervals and modalities requires accurate risk assessment. Although hereditary predisposition is commonly used, the effect of exogenous risk factors has remained largely unexplored. To address this, we analyzed the age at presentation and location of CRC in relation to alcohol and tobacco use.
We queried the IMPAC Medical Registry Services Cancer Information Resource File for CRCs diagnosed between June 1, 1993, and December 31, 2003. Subjects were classified as current, past, or never users of alcohol and tobacco. A logistic regression model for location of CRC and a linear regression model for age at diagnosis were constructed using these explanatory variables along with gender, race, and insurance status.
Our data set consisted of 161 172 patients with CRC. Current drinking, smoking, and smoking plus drinking were associated with younger ages at onset of CRC (adjusted age difference, 5.2, 5.2, and 7.8 years, respectively; P<.001 for all). A distal location of CRC was more likely to occur in current drinkers (odds ratio, 1.192; 95% confidence interval, 1.15-1.23) and smokers (odds ratio, 1.164; 95% confidence interval, 1.12-1.21). Colorectal cancer in men tended to occur earlier (adjusted age difference, 1.9 years; P<.001) and have a distal predominance (odds ratio, 1.42; P<.001) compared with women. The smoking but not the drinking effect size was greater in women than in men (adjusted age difference, 2.6 years; P<.001).
Alcohol use, tobacco use, and male gender were associated with earlier onset and a distal location of CRC. If confirmed, these factors should guide recommendations regarding initiation of CRC screening and, possibly, choice of techniques.
Colorectal cancer (CRC) remains the second leading cause of cancer deaths and in 2005 was projected to take the lives of 56 290 Americans.1 Given the insidious nature of CRC, with development of symptoms such as abdominal pain or bleeding typically signaling the presence of advanced incurable disease, screening the asymptomatic population is the only viable approach to curbing CRC fatalities. Two central issues in designing screening strategies are the timing of screening initiation and the choice of modality. Given the dramatic (almost logarithmic) rise in CRC incidence starting in the sixth decade of life, the onset of screening at age 50 years has generally been recommended for those deemed at “average risk.”2 However, CRC can occur before age 50 years. Indeed, in asymptomatic average-risk persons aged 40 to 50 years, the number of neoplastic lesions found on colonoscopic screening is not trivial (8.7% had tubular adenomas and 3.5% had advanced lesions).3 With regard to modality, flexible sigmoidoscopy every 5 years (coupled with annual fecal occult blood tests) and colonoscopy every 10 years have comparable cost-effectiveness and are viewed as equivalent options for screening.4 From an efficacy point of view, colonoscopy is superior to flexible sigmoidoscopy, because more than half of the patients with advanced proximal colonic lesions do not have a sentinel distal colonic polyp identifiable by flexible sigmoidoscopy.5 However, the advantages of flexible sigmoidoscopy include lower cost, decreased complication rates, and widespread availability, because it is generally performed by primary care physicians (as opposed to colonoscopy, which is generally performed by gastroenterologists or surgeons).
It is clear that generic “1-size-fits-all” guidelines are not optimal, and a more nuanced approach based on risk factors could represent an important advance. Tailoring screening based on inherited predisposition to CRC is commonly used. For instance, with familial adenomatous polyposis, annual flexible sigmoidoscopy starting at age 12 years is generally adequate, whereas in hereditary nonpolyposis, CRC, colonoscopy every 2 years after age 25 years (and annually after age 40 years) is recommended.6 For nonsyndromic patients with a family history of CRC, screening typically begins before age 50 years, with the method and exact age determined by the strength of the inherited risk.4
Clearly, heredity is an important means for designing screening strategies; however, approximately 75% of patients with CRC do not have a previously identified familial component.4 There are different nongenetic factors that modulate the risk of CRC, including obesity, lack of exercise, a high-fat diet, and alcohol and tobacco use.7 Indeed, most cases of CRC are believed to be attributable to these modifiable risk factors.8 Although there is compelling evidence that alcohol and tobacco use markedly increase the risk of CRC, these risk factors have typically not been factored into screening decisions. To assess whether alcohol and tobacco use should modify screening recommendations regarding age at initiation and modality, we performed a retrospective review of a large national CRC database.
We used data from the IMPAC Medical Registry Services Cancer Information Resource File between June 1, 1993, and December 31, 2003, to identify all cases of CRC (histological diagnosis of adenocarcinoma) reported to this registry during that decade. This database includes data from more than 350 teaching and community hospitals nationwide, including self-reported demographic factors such as alcohol and tobacco use. In the database, current alcohol or tobacco use is defined as use within the past year. Past substance users are those who quit at least 1 year before cancer diagnosis. Quantitative data on the extent of substance use are not available within this database.
A multivariate linear regression model was constructed with age as the outcome variable.9 We also used a multivariate logistic regression model with location of CRC (proximal vs distal) as the outcome variable.10 Tumors in the splenic flexure, descending colon, sigmoid, rectosigmoid, and rectum were assigned to the distal category, whereas others were designated as being proximal. Explanatory variables for both models were gender, alcohol history (current use, past use, or never used), tobacco history (current use, past use, or never used), race (Asian, black, other, or white), and insurance status (yes or no). Potential explanatory variables with P<.25 in univariate regressions were entered into the multivariate models. The initial multivariate models included all such explanatory variables and the interaction with alcohol and tobacco use. In addition, we modeled CRC age at diagnosis and location in 2 subsequent multivariate models adding all first-order interactions between alcohol use, tobacco use, and gender to assess gender-specific effects of alcohol and tobacco use. Nonsignificant variables and interactions were dropped 1 at a time using a backward elimination strategy. Regression coefficients and standard errors for variables without interactions in multivariate models were compared among the univariate and multivariate models to assess confounding among variables. P<.05 was considered statistically significant. All data were analyzed using SAS statistical software (SAS Institute, Cary, NC).
We identified 166 172 patients with CRC who were included in this analysis. The predominant histological diagnoses were adenocarcinoma (89%) and mucinous adenocarcinoma (11%). Fifty percent of the patients were male, and 9% were African American. The stage and grade data in Table 1 approximate national trends. Table 2 lists the unadjusted ages at CRC diagnosis associated with all potential explanatory variables, stratified by alcohol and tobacco use groups. Thirty-five percent of the patients had missing data on alcohol history, 27% on tobacco history, and 26% on insurance status; there was a trend toward more missing data for earlier periods.
From a statistical perspective, the models appeared to be robust with favorable diagnostics. As would be expected from the large number of CRC risk factors that were unable to be assessed, the predictive ability of the model was modest but in the range of findings typically seen with other discrete or binary models. For instance, the adjusted R2 for the age at diagnosis model was 0.090. The overall F test score was F17 = 501.4 (P<.001). The Hosmer-Lemeshow statistic for the logistic regression goodness of fit was χ27 = 9.25 (P = .24).
In the multivariate model, all initial explanatory variables, along with the alcohol and tobacco use interaction, were significant (P<.001 for all) (Table 3). In the table, age at diagnosis for the baseline case (never smoker, never drinker, white, insured, and male) is presented. After controlling for the other variables, alcohol users had a younger age at CRC diagnosis. For instance, in never smokers, current and past use of alcohol was associated with a stepwise younger age at CRC diagnosis (mean difference, 5.2; 95% confidence interval [CI], 4.9-5.5 years; and mean difference, 2.1; 95% CI, 1.3-2.9 years; respectively; P<.001 for both). Similarly, in never drinkers, current tobacco use was linked to lower adjusted age at diagnosis by 5.2 (95% CI, 4.9-5.5) years (P<.001). Current drinkers and smokers had a 7.8-year (95% CI, 7.5-8.1 years) younger age at diagnosis (P<.001). In sensitivity analyses for different scenario assumptions regarding missing information (eg, all were users or all were nonusers), all results remained statistically significant (data not shown).
We next investigated whether alcohol and tobacco use affect the location of CRC (Table 4). All other initial explanatory variables, as well as current alcohol and tobacco use, were significant (P<.001). Women were less likely to have distal CRC at initial examination than men (odds ratio [OR], 0.705; 95% CI, 0.686-0.723; P<.001). When adjusted for gender, race, and insurance status, current alcohol consumption was associated with an increased likelihood of distal CRC (OR, 1.192; 95% CI, 1.153-1.233; P<.001); the effect size was muted in past drinkers (OR, 1.093; 95% CI, 1.027-1.163; P = .005). Similarly, the probability of distal CRC was higher in current smokers (OR, 1.164; 95% CI, 1.116-1.214; P<.001), with the effect dissipating in past smokers (OR, 1.009; 95% CI, 0.975-1.044; P = .60).
In the next multivariate model, we added all first-order interactions between alcohol use, tobacco use, and gender to the analysis of age at CRC diagnosis. Alcohol use and gender interactions at P>.05 were dropped from the multivariate model. All other initial explanatory variables, as well as the alcohol and tobacco use and gender and tobacco use interactions were significant (P<.001 for all) (data not shown). We focused on current vs never users because these data are more robust (the past users group as defined in the database comprises a heterogeneous population). As indicated in Table 5, in the never drinker and current smoker group, the magnitude of the difference compared with the never drinker and never smoker group was significantly greater in women than in men (6.3 vs 3.7 years, P<.001). There was no significant gender difference in the younger age at diagnosis in alcohol consumers among never smokers (4.9 vs 4.9 years, P = .85). Furthermore, the gender differential in current drinkers and smokers appeared to be entirely attributable to tobacco consumption alone.
This is the first report, to our knowledge, to demonstrate that alcohol and tobacco use is associated with a younger age at CRC presentation. The magnitude of effect suggests clinical applicability for the timing of screening initiation. Furthermore, alcohol and tobacco use along with male gender increased the probability of distal cancers, possibly having relevance to the choice of screening modality. These observations were possible because of the immense database size, which mitigates the masking from the variability in factors related to population (eg, increased susceptibility from the presence of polymorphisms in the alcohol and tobacco metabolizing systems) and use (dosage, duration, and type of alcohol and tobacco use). Indeed, it is conceivable that there may be particular subgroups that are extraordinarily susceptible to the CRC promotional effects of alcohol or tobacco.
Previous studies7,11 support the biological plausibility of our observations. Smoking is a well-established CRC risk factor, associated with a 2-fold increase in risk and implicated in 12% of all CRC deaths.12 Results of most studies indicate that alcohol use causes a similar increase in CRC risk.13 Although previous studies have assessed the effects of alcohol and tobacco use on tumor location, the results have been discordant.14 Subsite predilection can be affected by the molecular subtype of CRCs, microsatellite stable (MSI-stable) or microsatellite unstable (MSI-high). Microsatellite unstable tumors compose approximately 20% of all CRCs and are generally proximal.15 Slattery et al16 suggested that smoking selectivity predisposed subjects to MSI-high CRCs, which agrees with the findings of a subsequent study17 among patients with hereditary nonpolyposis CRC. On the other hand, Diergaarde et al18 failed to detect an increase in MSI-high status among smokers with CRC. Similarly, although some investigators suggested that alcohol consumption increases the tumorigenesis of MSI-high colon tumors,19 a subsequent study20 demonstrated that alcohol use also increased the tumorigenesis in an MSI-stable experimental model. In addition, a recent meta-analysis13 of 8 cohort studies disclosed an increase in proximal, distal, and rectal cancers with alcohol use, although the magnitude of increased risk seemed to be greater in the distal colon rather than the proximal colon. Therefore, our large database analysis herein convincingly demonstrates that alcohol and tobacco users have a distal predilection for CRC.
From a screening perspective, a potentially more important observation is the age at onset of CRC. Alcohol and tobacco use was associated with a younger age at diagnosis, with evidence of an interaction as indicated by subjects who used both agents. This magnitude was dramatic, with the mean age of patients with CRC who did not smoke or drink being 71.3 years, while the mean age of current drinkers and smokers was 62.6 years. To keep this in perspective, the age at CRC diagnosis among subjects with alcohol and tobacco use was comparable to that reported for hereditary nonpolyposis CRC mutation carriers (61 years),21 a risk group who should undergo initial colonoscopic screening at an early age.6 The effect of alcohol or tobacco use in past users was minimal, especially compared with that in current users, suggesting that stopping drinking and smoking may mitigate the risk.
Although smoking was associated with a younger age at diagnosis among both genders, the effect was greater in women. This is consistent with reports that women are more susceptible to the carcinogenic effect of smoking in other malignancies. For instance, in subjects who smoked more than 40 pack-years, the ORs of developing lung adenocarcinoma, bladder cancer, and oral cancer in men were 13.8, 2.5, and 2.1, respectively, whereas for women they were markedly higher, 32.7, 4.5, and 5.4, respectively.22 Whether this can be attributed to social factors (rate of inhalation or type of cigarette smoked) or to biological factors (hormone related or genetic variability) is unclear. It is somewhat surprising not to see this gender-specific effect with alcohol use because it is believed that women are more susceptible to alcoholic liver disease. However, the mechanisms for alcohol-induced liver disease and CRC may be different.
Because all patients were diagnosed as having cancer, this study cannot evaluate the relationship between alcohol or tobacco use and CRC incidence per se but rather only investigates whether use of these agents serves as a marker for younger age at onset. Therefore, correlation rather than causation is the key factor. Although the literature is replete with data supporting the biological plausibility of this relationship, we need to be circumspect because of the numerous potential confounders.22 Because the database used herein is not linked to medical records, we were unable to assess many of the well-established CRC risk factors such as diet, exercise, and obesity. We were able to adjust for race in our model, given our observation that African Americans tend to develop CRC 3 years earlier than white subjects (in agreement with the literature).23 We were also able to control for insurance status. The effect of alcohol and tobacco use on age at presentation was unaffected by these factors (eg, the changes in age at presentation for current drinkers and smokers were −8.0 years in the unadjusted model and −7.8 years when adjusted for race and insurance status). However, many other factors, including family history, socioeconomic status, and educational achievement, could not be addressed. As previously discussed, genetic predisposition could lead to a younger age at CRC presentation. On the other hand, for this factor to bias our results, we would have to propose that patients with a genetic predisposition to CRC were more likely to smoke, which is inconsistent with the literature.24 A further assuaging concern regarding lead-time bias is our observation that CRCs in drinkers and smokers were not found at earlier stages than in nonusers (ORs for having stage 3 or 4 disease at initial examination for drinkers and smokers vs never users were 1.037 vs 1.039).25 Socioeconomic status and education could also potentially represent a confounder. Health prevention behavior is different in those with lower SES and education and may affect the age at CRC diagnosis. Generally, the SES and educational achievement among drinkers and smokers tend to be lower than those among nondrinkers and nonsmokers. Therefore, given the potential for confounding, conclusions from our report should be limited to stating that drinking and smoking are markers for younger age at CRC presentation.
Other limitations include the heterogeneous nature of the past user group, which likely included patients who had modest use of alcohol and tobacco and stopped several decades ago, as well as those who were heavy users and quit more recently (>1 year before diagnosis). Therefore, our major analysis focuses on current vs never users, which may be a more rigorous parameter. The lack of the quantitation of alcohol and tobacco use prevents us from assessing a dose-response effect on colon carcinogenesis. Moreover, self-reporting of the alcohol measure is somewhat crude and may be an underrepresentation, implying that the effect size may be greater than that reported. In addition, although approximately one third of patients lacked alcohol and tobacco use data, we doubt that this nonreporting biased our results because the general public perception is that alcohol and tobacco use are not significant CRC risk factors (as opposed to lung and liver cancer, respectively). Furthermore, sensitivity analyses showed that, with any assumptions regarding the drinking and smoking status of the nonresponders (ie, those who did not respond to the question), the results remained statistically significant (data not shown).
Finally, our findings regarding the younger age at diagnosis among smokers may reflect the role of competing causes of mortality.26 If this explanation were correct, all cancers should manifest a younger age at onset among smokers irrespective of the ability of smoking to induce malignant neoplasms. To assess this, we analyzed the database for the relationship between smoking and prostate cancer, as most studies27- 29 have shown that smoking does not increase the incidence of this malignancy. The age at diagnosis of prostate cancer in current smokers (67.4 years) was marginally older than that in never smokers (65.5 years) and was younger than that in past smokers (68.7 years) (data not shown). Therefore, there was no evidence of a clear relationship between smoking and age at occurrence of prostate cancer, arguing that competing causes of mortality are not major confounders for our CRC study.
Our central finding that alcohol and tobacco use identifies a subgroup at risk for earlier CRC presentation could have important clinical implications. One could argue for initiation of screening earlier than age 50 years, because the principle of colonoscopic screening lies in the interdiction of adenomatous polyps before evolution into frank CRC (approximately 10-20 years). However, given the numerous other factors (genetic and environmental) that affect CRC risk, nondrinkers and nonsmokers cannot necessarily be assigned to a low-risk group. This finding is highlighted by the small amount of variance explained by our model, indicating that additional factors not considered in our analysis were important determinants of the age at presentation of CRC.
Our report provides compelling evidence that modifiable risk factors such as alcohol and tobacco use are associated with a younger age at diagnosis and a predilection for distal CRCs. These data underscore the need for cessation of alcohol and tobacco use as an integral part of a CRC prevention program. If replicated in other data sets, our findings would argue that screening should be initiated at a younger age in alcohol and tobacco users. Moreover, these factors help define a subgroup at risk for proximal CRC (nondrinking and nonsmoking women) among whom total colonoscopy rather than flexible sigmoidoscopy would be more appropriate. In the future, we envision the development of risk scores with exogenous (eg, alcohol and tobacco use, age, body mass index, diet, and calcium consumption) and hereditary factors to tailor an individual's CRC screening program.
Correspondence: Hemant K. Roy, MD, Department of Internal Medicine, Evanston Northwestern Healthcare Research Institute, Feinberg School of Medicine, Northwestern University, 1001 University Pl, Evanston, IL 60201 (firstname.lastname@example.org).
Accepted for Publication: October 25, 2005.
Author Contributions: Dr Roy had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Financial Disclosure: None.
Funding/Support: This study was supported in part by research grant 1U01-CA111257-01 from the National Institutes of Health, Bethesda, Md.
Previous Presentations: This study was presented in abstract form at the Digestive Disease Week and the 106th Annual Meeting of the American Gastroenterological Association; May 15-19, 2005; Chicago, Ill; and at the American College of Gastroenterology 70th Annual Scientific Meeting; October 31, 2005; Honolulu, Hawaii.