Critique 138: Underreporting of alcohol intake affects the relation of alcohol to the risk of cancer — 23 April 2014

Klatsky AL, Udaltsova N, Li Y, Baer D, Tran HN, Friedman GD.  Moderate alcohol intake and cancer: the role of underreporting.  Cancer Causes Control 2014; on-line publication; DOI 10.1007/s10552-014-0372-8 2014.

 Authors’ Abstract

Purpose  There is compelling evidence that heavy alcohol drinking is related to increased risk of several cancer types, but the relationship of light–moderate drinking is less clear.  We explored the role of inferred underreporting among light–moderate drinkers on the association between alcohol intake and cancer risk.

Methods  In a cohort of 127,176 persons, we studied risk of any cancer, a composite of five alcohol-associated cancer types, and female breast cancer.  Alcohol intake was reported at baseline health examinations, and 14,880 persons were subsequently diagnosed with cancer.  Cox proportional hazard models were controlled for seven covariates.  Based on other computer-stored information about alcohol habits, we stratified subjects into 18.4 % (23,363) suspected of underreporting, 46.5 % (59,173) not suspected of underreporting, and 35.1 % (44,640) of unsure underreporting status.

Results  Persons reporting light–moderate drinking had increased cancer risk in this cohort.  For example, the hazard ratios (95 % confidence intervals) for risk of any cancer were 1.10 (1.04–1.17) at 1 drink per day and 1.15 (1.08–1.23) at 1–2 drinks per day.  Increased risk of cancer was concentrated in the stratum suspected of underreporting.  For example, among persons reporting 1–2 drinks per day risk of any cancer was 1.33 (1.21–1.45) among those suspected of underreporting, 0.98 (0.87–1.09) among those not suspected, and 1.20 (1.10–1.31) among those of unsure status.  These disparities were similar for the alcohol-related composite and for breast cancer.

Conclusions  We conclude that the apparent increased risk of cancer among light–moderate drinkers may be substantially due to underreporting of intake.

Potential Conflict of Interest:  The first author of this paper is a member of the International Scientific Forum on Alcohol Research.  He has not contributed to this review.  However, readers of our comments may wish to take this fact into account when interpreting the conclusions of the review. 

Forum Comments

Epidemiologists are often faced with adverse effects of alcohol among subjects reporting very low levels of consumption, levels that physiologically should not adversely affect health.  It is often assumed that at least some of the subjects reporting low levels of intake may be underreporting their actual intake, but heretofore it has been difficult to judge this. 

As reported in this paper, Klatsky and his associates, using the very large cohort of the Kaiser Permanente Medical group (more than 100,000 subjects), have developed an approach of estimating which subjects may be underreporting their consumption.  The approach is based on numerous reports over time for alcohol intake, liver enzymes, and the occurrence of medical conditions known to result from excessive alcohol consumption.  For example, subjects classified at likely underreporters had diagnoses such as alcoholic psychosis, alcohol dependence syndrome, alcohol abuse, alcoholic polyneuropathy, alcoholic gastritis, or alcoholic liver disease either prior to or following their reported “moderate” intake.  Based on the occurrence in the computer-stored data of more than one such diagnosis, there were 23,363 inferentially “Likely underreporters;” of these 13,995 (59.9 %) reported heavier drinking on another questionnaire, 14,717 (63.0 %) had an ARD and 5,349 (22.9 %) had both. 

Persons with at least two computer-stored examinations (index measurement and at least one other before or after) and no indicator of being likely underrerporters were classified as unlikely to be underreporting their alcohol intake; this group made up almost one-half of the total cohort.  Persons with only one computer-stored alcohol intake report were classified as “Unsure” with respect to underreporting.

A total of 14,880 subjects in the cohort developed cancer during follow up.  When the alcohol intake of moderate drinkers was related to the risk of cancer, there were large differences according to whether or not the subject had been suspected of being an underreporter of their alcohol intake.  For example, among subjects reporting 1-2 drinks/day, the likely underreporters showed an increased relative risk of any cancer [RR 1.33, 95% CI (1.21–1.45)], in comparison with non-drinkers, while those unlikely to be underreporting their intake showed no increase in risk of any cancer [RR 0.98, (95% CI 0.87-1.09)].  Those subjects with only one computer-stored alcohol intake report were categorized as “unsure” with respect to underreporting; they had an intermediate risk of developing cancer. 

Similar results were seen for alcohol-related cancer as for any cancer.  Also, the risk of breast cancer among women was less than one-half as high among moderate drinkers who were unlikely to be underreporters of their alcohol intake as among those considered likely to be underreporting their drinking.

For epidemiologic studies (such as the Kaiser-Permanente Study) that have extensive longitudinal data on all medical conditions, this approach could be an extremely valuable way of seeking the true relation between light to moderate alcohol intake and a variety of health outcomes.

Specific comments on this study:  Forum members were in agreement that this paper makes an important contribution to our ability to interpret associations between reported alcohol intake and the risk of cancer.  It is based on a very large cohort, followed over 18 years, with essentially full ascertainment of the development of cancer.  The particular attribute of the study that is most important is the recording of a large number of diagnoses known to relate to heavy alcohol consumption (e.g., alcohol dependence syndrome, alcohol abuse, alcoholic polyneuropathy, alcoholic gastritis, or alcoholic liver disease).  Further, for many subjects, measurements of liver enzymes provide an objective index of alcohol misuse.  Thus, subjects categorized as likely to be underreporters of alcohol intake had other reports in the database supporting excessive alcohol use.

The authors of this report have previously reported that the supposedly “light-to-moderate drinkers” in their cohort who had evidence of being “likely underreporters” of alcohol intake more frequently were found to have hypertension that those considered to be unlikely underreporters (Klatsky et al, Higher prevalence of systemic HTN among moderate alcohol drikers: exploring the role of under-reporting.  J Stud Alcohol 2006;7:21-428).  The fact that this current application of the same approach also separates subjects for their risk of cancer strongly supports the merits of the procedure.  Indeed, it could be important for other epidemiologic studies with similar data on alcohol-related conditions to evaluate their moderate drinkers using such an approach.

Stated Forum reviewer Finkel: “It has seemed to me for some time that there has been a rush to judgment in many quarters incriminating moderate drinking as a cause or promoter of numerous cancers.  There likely also has been too eager an acceptance of likely trivial statistical suggestions of drinking protecting from various cancers.  The truths we seek are obscured by confounders, and thus conflicted and confusing.  This paper renders a real service by beginning to dispel some of the murk.  It sharpens the focus, enhancing understanding that simple errors have led to major misconceptions.  In addition to illuminating in part the complex engendering of cancers, this correction should give pause to those who too eagerly embrace the results they desire.”

Reviewer Waterhouse agreed: “My observation is that the triggers for being a likely underreporter suggest substantial or severe underreporting in that group.  This leads me to conclude that it is probable that the unlikely group may also include modest underreporting (as suggested by the authors).  The scope and severity of the problem as revealed here strongly indicates that further investigation into this phenomenon is needed.”

Forum member Goldfinger adds: “This paper, and its common sense deductions regarding association of disease with mild to moderate alcohol consumption, demands caution when trying to ascribe causation of serious medical complications to a lifestyle habit that may not be true, and in fact may be otherwise beneficial.”

Reviewer Skovenborg stated:  “Self-reports of alcohol consumption may be called into question because of the discrepancy between alcoholic beverage sales data and survey reports of alcohol consumption.  In general, comparison studies in the alcohol literature have shown that self-reported alcohol consumption accounts for only 40-60% of alcoholic beverages sold as measured by sales and tax data (Midanik L. The validity of self-reported alcohol consumption and alcohol problems: a literature review. Br J Addict 1982;77:357-82).

“Some groups of drinkers, defined either by demographics or drinking level, may bias their reports differently.  Underreporting of alcohol intake may be explained by response errors, e.g., difficulties in recall of drinking practices and culturally determined socially desirable answers.  To advise the public on ‘sensible’ limits of alcohol intake, methods are needed that properly rank individuals according to alcohol intake and also assess correctly the absolute level of intake (Feunekes GIJ et al. Alcohol intake assessment: The sober facts. Am J Epidemiol 1999;150:105-112).”

 Forum member De Gaetano wondered if the two groups (underreporting and unlikely to underreport) would still appear to be protected against cardiovascular events and total mortality.  If so this would suggest that (1) real moderate consumption protects against cardiovascular disease, an effect not counterbalanced by increased cancer risk; and (2) false moderate consumption (due to underreporting) increases cancer risk, but the effect may still be counterbalanced by reduced cardiovascular risk.” 

Reviewer Ellison adds: “These reasons may explain  the results of Le Strat and Gorwood (Le Strat Y, Gorwood P.  Hazardous drinking is associated with a lower risk of coronary heart disease: results from a national representative sample.  Am J Addict 2011;20:257-263. doi: 10.1111/j.1521-0391.2011.00125.x), in which the authors conclude that alcohol may be cardioprotective not only in individuals who drink moderately but also in those who drink amounts traditionally considered to be hazardous.  However, the method used in that study to diagnose coronary heart disease (CHD) raised concerns: only 1% of subjects reported having had myocardial infarction in the past year, the primary “hard” endpoint for CHD, whereas most reported angina pectoris, a ‘softer’ endpoint for CHD.   Further, subjects who quit drinking due to illness or those with hazardous drinking who died earlier than healthy subjects may have confounded results.  Further, the definition of ‘hazardous drinking’ in that study was too inclusive, including some people who might better be classified as moderate drinkers.”  That paper was reviewed by our Forum, available at www.bu.edu/alcohol-forum/critique-040).

Forum member Lanzmann-Petithory had some interesting observations.  “It is a very elegant idea of Klatsky et al that the apparent increased risk of cancer among light–moderate drinkers may be substantially due to underreporting of intake.  In the Nancy cohort study initiated by Serge Renaud, we had an opportunity to evaluate two approaches for estimating alcohol intake: a self-administered recall questionnaire and a seven-day prospective nutritional survey validated by a dietician.  While these data have not yet been published, our colleague Bernard Herbeth from INSERM Nancy found a better correlation coefficient with of GGT with alcohol consumption obtained by seven-day survey (0.451) than with the self-administered questionnaire based on recall of intake over the preceding year (0.392) (p<0.001).  Some data from this analysis are shown below:

As shown in the above table, there was a strong correlation between the recall questionnaire and the estimate of alcohol from the 7-day nutritional survey (percent giving estimates in the same category by the two methods are shown in bold).  However, a total of 57.1% of those with an estimate from the survey of 88-200 g/day reported less   on the recall questionnaire.” 

Forum Summary

Epidemiologists are often faced with reported adverse health effects of alcohol among subjects reporting very low levels of consumption, levels that physiologically should not cause diseases such as cancer.  It is often assumed that at least some of the subjects reporting low levels of intake may be underreporting their actual intake, but heretofore it has been difficult to judge this. 

In the present study, the authors used each subject’s computer-based data on conditions and diseases (including laboratory and social factors), related to alcohol misuse, collected over many years to identify subjects reporting “light-to-moderate” alcohol intake who were likely, or unlikely, to be underreporting their intake at a baseline examination.  Overall, 18.4% of subjects were suspected of being underreporters while 46.5% had adequate data that suggested that they were not underreporters.  (The remaining 35.1% of subjects had inadequate data stored in the computer to be classified.)  

Among their cohort of more than 100,000 subjects, 14,880 developed cancer during an average follow-up period of 18 years.  There were 23,363 subjects who reported that they were light (up to 1 drink/day) or moderate (1 to 2 drinks/day) drinkers.  In all comparisons, subjects suspected of being underreporters of their alcohol intake had a higher risk of cancer than those not categorized as being underreporters.  For example, among subjects reporting 1 to 2 drinks/day, in comparison with non-drinkers, the risk ratio of any type of cancer among those considered to not be underreporters was 0.98 (95% CI 0.87-1.09); in other words, no effect on cancer risk from alcohol.  For those categorized underreporters, however, the relative risk of any cancer was 1.33 (95% CI 1.21–1.45).  Similar results were seen for alcohol-related cancers as for any cancer.  Also, the risk of breast cancer among women was less than one-half as high among moderate drinkers who were unlikely to be underreporters of their alcohol intake as among those considered likely to be underreporting their drinking.

Adjusting for confounding in prospective studies is an ever-present challenge for epidemiologists.  One factor that has often been considered probable, but difficult to adjust for, is the underreporting of actual alcohol intake by subjects.  Being able to adjust for such underreporting would greatly improve studies relating alcohol consumption to the risk of cancer (and other chronic diseases).  For epidemiologic studies that have extensive longitudinal data on all medical conditions (such as the Kaiser-Permanente Study, the basis for these analyses), the approach described in this paper could be an extremely valuable way of seeking the true relation between light-to-moderate alcohol intake and cancer as well as with a variety of other health outcomes. Based on the present study, an increase in the risk of cancer among light-to-moderate drinkers may occur primarily among those who underreport their alcohol intake.

                                             *                    *                      *

Comments on this critique by the International Scientific Forum on Alcohol Research were provided by the following members:

Giovanni de Gaetano, MD, PhD, Department of Epidemiology and Prevention, IRCCS Istituto Neurologico Mediterraneo NEUROMED, Pozzilli, Italy

R. Curtis Ellison, MD, Section of Preventive Medicine & Epidemiology, Boston University School of Medicine, Boston, MA, USA

Harvey Finkel, MD, Hematology/Oncology, Boston University Medical Center, Boston, MA, USA

Tedd Goldfinger, DO, FACC, Desert Cardiology of Tucson Heart Center, Dept. of Cardiology, University of Arizona School of Medicine, Tucson, Arizona, USA

Dominique Lanzmann-Petithory,MD, PhD, Nutrition/Cardiology, Praticien Hospitalier Hôpital Emile Roux, Paris, France

Erik Skovenborg, MD, Scandinavian Medical Alcohol Board, Practitioner, Aarhus, Denmark

Arne Svilaas, MD, PhD, general practice and lipidology, Oslo University Hospital, Oslo, Norway

Fulvio Ursini, MD, Dept. of Biological Chemistry, University of Padova, Padova, Italy

David Van Velden, MD, Dept. of Pathology, Stellenbosch University, Stellenbosch, South Africa

Andrew L. Waterhouse, PhD, Marvin Sands Professor, Department of Viticulture and Enology, University of California, Davis; Davis, CA, USA