RE: "INTERACTION BETWEEN TOBACCO AND ALCOHOL CONSUMPTION AND THE RISK OF CANCERS OF THE UPPER AERO-DIGESTIVE TRACT IN BRAZIL"

William J. Waddell and Paul S. Levy

Department of Pharmacology and Toxicology School of Medicine University of Louisville Louisville, KY 40292
Division of Epidemiology and Biostatistics School of Public Health University of Illinois at Chicago Chicago, IL 60612–7260


    INTRODUCTION
 TOP
 INTRODUCTION
 REFERENCES
 INTRODUCTION 
 REFERENCES 
 
Schlecht et al. (1Go) continue the same long-standing confusion in interpretation of the data on the relation of smoking and drinking as independent risk factors for cancer of the upper aero-digestive tract (UADT). They studied 784 patients with cancer of the oral cavity, pharynx, and larynx for smoking and drinking habits and other characteristics. Since virtually all of these cases both smoked and drank, the authors used statistical methods in an attempt to determine the effect of each of these two habits. This is the same method that has been used in many other, similar studies, leading some to conclude that drinking is an independent risk factor. However, this is very problematic since other biologic data show that, although potent carcinogens exist in tobacco smoke, alcohol is not a carcinogen. Consequently, the most often used explanation for an effect of alcohol is that it increases the permeability of the carcinogens from tobacco. This is entirely different from an implication that alcohol is an independent carcinogen.

Any conclusion from epidemiologic data alone that alcohol is a carcinogen must therefore examine the effect in never smokers. Of the 784 patients in the report by Schlecht et al., there were only 17 never smokers who were drinkers. Furthermore, only the group of seven patients who drank "more than" the equivalent of one beer per day had a statistically significant increase in cancer above controls (calculated at 18 g/day x 365 days x 37 years = 243 kg, average age of patients, 58 years, assuming that drinking started at age 21 years). The authors give no indication of how much more than the equivalent of one beer per day the individual patients consumed or of whether they had other risk factors associated with cancer of the UADT. Excessive consumption of alcohol (the alcoholic lifestyle) and other specific factors, such as poor dentition and diet, have been associated with cancer of the UADT. Consequently, potential confounding in the never smokers in the study by Schlecht et al. leaves serious doubts about the validity of their conclusion.

We are puzzled by the "crude" odds ratios shown for never smokers only in their table 6. These odds ratios are apparently based on the discordant data only, since the controls were matched individually two to one to the cases on 5-year age group, sex, and location. Examining the raw data on number of cases and controls at each level of drinking, we obtain odds ratios of 1.59, 0.68, and 1.78 for increasing classes of drinking (based on ordinal values of one, two, and three given to the classes). This visually and statistically (p = 0.48) indicates no evidence of trend and contrasts sharply with their odds ratios of 2.5, 2.0, and 7.7 (trend p = 0.022) apparently obtained from the discordant data only. Assuming that their computations are valid, we find it highly unusual that the matching design would change the point estimates so dramatically (e.g., from 1.78 to 7.7 for the highest drink class) and suspect the existence of some kind of artifact or interaction that was not explored in their analysis and may threaten the external validity of their findings.

Authors of studies of a potential association of smoking and drinking with cancer of the UADT complain of the "problem" of too few cases of drinkers who do not smoke. Different reports have dealt with this problem in different ways; some combine light drinkers or light smokers or select groups with "convenient" doses for comparison, etc., in order to be able to apply statistics. The fact speaks for itself—there are very few drinkers who do not smoke who develop cancer of the UADT. Even among those nonsmoking drinkers who do develop cancer, other potential confounders have not been eliminated. There are many moderate drinkers who do not smoke. It is a disservice to these individuals to conclude that alcohol is an independent risk factor when the evidence does not support this conclusion.


    REFERENCES
 TOP
 INTRODUCTION
 REFERENCES
 INTRODUCTION 
 REFERENCES 
 

  1. Schlecht NF, Franco EL, Pintos J, et al. Interaction between tobacco and alcohol consumption and the risk of cancers of the upper aero-digestive tract in Brazil. Am J Epidemiol 1999;150:1129–37.[Abstract]

 

TWO OF THE AUTHORS REPLY

Nicolas F. Schlecht and Eduardo L. Franco

Departments of Epidemiology and Oncology McGill University Montreal, Quebec, Canada H2W 1S6


    INTRODUCTION 
 TOP
 INTRODUCTION
 REFERENCES
 INTRODUCTION 
 REFERENCES 
 
Waddell and Levy (1Go) suggest that confounding bias may have been responsible for the apparent effect of alcohol that we observed on the risk of cancers of the upper aero-digestive tract (UADT) (2Go, 3Go). They argue that there is no biologic evidence to support the independent carcinogenic role of alcohol. We disagree with their interpretation, as shown below.

They point out that we included few cases of nonsmokers who drank and conclude that there was lack of evidence because statistical significance was shown only for the highest category of consumption (>243 kg alcohol). However, this ignores the significant dose-response trend with alcohol consumption and the elevated odds ratios across all consumption categories. They also suggest that our point estimates were affected by residual confounding due to omission of covariates in our models. We argue that this is highly unlikely, since we adjusted all analyses for confounders that were selected empirically on the basis of a 5 percent change in the estimates for alcohol, arguably a very conservative threshold. Using this low cutpoint to select covariates, we screened more than 100 candidate exposures, including tens of dietary and dental health variables (the latter reported earlier (4Go)). We are not aware of any study of this kind that has controlled for confounding with such zeal. Furthermore, with a mean consumption of 7.4 drinks per week in the highest category, it is unlikely that the contribution of nonsmoking subjects with an "alcoholic lifestyle" is important.

Waddell and Levy mistakenly ignored the conditional assumptions in the matched design of our study and calculated unconditional odds ratios for alcohol consumption using our frequency data. They then used the resulting underestimated odds ratios to buttress their claim of no effect for alcohol. In a multicenter study involving populations from diverse geographic areas, there are differentials in UADT cancer risk for many exposures that vary markedly according to the matching factors that we used (city, gender, age, and period). To avoid bias, one must consider the net influence of these relations via conditional logistic regression. To illustrate this point, we calculated the odds ratios of UADT cancer for tobacco and alcohol based on tertiles (tobacco: 0–10, 11–45, and >45 pack-years; alcohol: 0–10, 11–530, and >530 kg). Without taking matching into account, we obtained estimates of 1.0 (reference), 2.9, and 4.5 for tobacco and 1.0 (reference), 0.9, and 2.0 for alcohol. However, when matching is preserved by conditional analysis, the odds ratios are 1.0, 4.2, and 10.9 for tobacco and 1.0, 1.7, and 7.3 for alcohol. Even with adjustment for the matching factors while ignoring matching, the odd ratios remain severely underestimated, stressing the importance of conserving the matching introduced in the study.

Although not mentioned in the letter by Waddell and Levy (1Go), differential misclassification of smoking could be a concern. However, this is unlikely to have happened to an appreciable extent because our hospital-based controls were recruited and interviewed under conditions similar to those of cases (2Go). Compared with North American and Western European populations, our subjects were not as aware of the hazards of smoking and excessive drinking.

In conclusion, we stand by our original findings (2Go), since the arguments in the letter by Waddell and Levy either are incorrect or are based on untenable assumptions. Admittedly, our study did not distinguish between types of alcohol, which prevents us from reaching firm conclusions about the role of specific alcoholic beverages. We refer to other studies that have attempted to address this problem (5GoGo–7Go). Detailed investigations are needed to elucidate the role of alcohol in UADT cancers, particularly comparing populations with respect to the impact of changes in consumption and cancer incidence.


    REFERENCES 
 TOP
 INTRODUCTION
 REFERENCES
 INTRODUCTION 
 REFERENCES 
 

  1. Waddell WJ, Levy PS. Re: "Interaction between tobacco and alcohol consumption and the risk of cancers of the upper aero-digestive tract in Brazil." (Letter). Am J Epidemiol 2000;152:193.[Free Full Text]
  2. Schlecht NF, Franco EL, Pintos J, et al. Interaction between tobacco and alcohol consumption and the risk of cancers of the upper aero-digestive tract in Brazil. Am J Epidemiol 1999;150:1129–37.[Abstract]
  3. Franco EL, Schlecht NF. Authors' response to "Invited commentary: more evidence of increased risks of cancer among alcohol drinkers." Am J Epidemiol 1999;150:1141.[ISI]
  4. Velly AM, Franco EL, Schlecht N, et al. Relationship between dental factors and risk of upper aerodigestive tract cancer. Oral Oncol 1998;34:284–91.[ISI][Medline]
  5. Mashberg A, Garfinkel L, Harris S. Alcohol as a primary risk factor in oral squamous carcinoma. CA Cancer J Clin 1981;31:146–55.[Abstract]
  6. Rothman KJ, Cann CI, Fried MP. Carcinogenicity of dark liquor. Am J Public Health 1989;79:1516–20.[Abstract]
  7. Hsu TC. Alcohol: a cocarcinogen in head and neck malignancies. In: Newell GR, Hong WK, eds. The biology and prevention of aerodigestive tract cancers. New York, NY: Plenum Press, 1992;2:35–7.