Rehm et al. Respond to "Never, or Hardly Ever?"

Abstract

In this issue of the Journal, Dr. Klatsky (1) raised a couple of open questions with respect to defining the best control group for alcohol epidemiology in his insightful commentary regarding our finding (2) that more than half of the people stating lifetime abstention in a representative US panel survey had elsewhere reported drinking before. Here, we respond to his comments, trying to further illuminate the best way to define this control group. We completely agree with Dr. Klatsky that the exact wording of the questions plays an important role. Much work in alcohol epidemiology has been done on defining the best way to elicit drinking reports (3, 4). Work on the wording of the introduction to these questions usually dealing with distinguishing whether the respondent is a lifetime or a “current” abstainer has been less methodological. However, as laid out in our paper (2), defining this control group has important implications for alcohol epidemiology; thus, the wording of such queries should receive the same methodological attention as other measures of exposure. Therefore, we believe that the exact wording should be reported (1). Dr. Klatsky's commentary (1) deals mainly with the relation between alcohol consumption and chronic disease. For this outcome, an adequate control group would be people for whom alcohol exposure could not reasonably have a biologic impact, that is, a mixture of lifetime abstention and very low levels of infrequent drinking (1, 2). However, almost half of the mortality and morbidity burden from alcohol stems from injuries (5), with different dimensions of alcohol consumption being relevant, especially amount of intake before the event. The best control group here is no drinking before the event or, in cohort studies, its best correlate. Even moderate drinking has some effects on psychomotor abilities (6), and its risk would then be determined by the overall frequency of different kinds of drinking occasions, each associated with a specific relative risk of injury (7). While moderate drinking is already associated with an elevated risk compared with abstention, the relative risks tend to increase exponentially with increasing intake (refer, for example, to Borkenstein et al. (8)). Finally, we could not agree more with Dr. Klatsky's contention (1) that other measurement errors, such as underreporting, could have an even more important effect on epidemiologic indicators of public health importance, such as alcohol attributable fractions (2, 9). We see 3 consequences for future work here: first, even for complex indicators, we should always give confidence intervals and conduct sensitivity analyses (e.g., for confidence intervals around attributable fractions, refer to Natarajan et al. (10)). Second, alcohol epidemiology should start correcting for regression dilution bias based on measurement error of exposure, as is now standard in epidemiologic research on other risk factors (11). Third, triangulation of different data sources will help avoid some of the main problems of measurement error, such as underreporting (12). Overall, we hope that the above-mentioned steps will provide some guidance for future research in alcohol epidemiology and will help reduce the effects of measurement errors.