SEARCH

SEARCH BY CITATION

Addiction recently published a commentary by Poikolainen [1] on a cohort study by Harriss et al. [2] which supported the hypothesis that errors in the definition of ‘abstainers’ were responsible for apparent cardioprotective effects of alcohol, especially for male drinkers. Shaper et al. [3] first proposed that many prospective studies classified erroneously both former and occasional drinkers as ‘abstainers’. However, Poikolainen [1] also criticized our meta-analyses of the literature [4,5] which supported the Shaper et al. hypothesis and suggested the possibility of gender differences in susceptibility to protection.

Poikolainen [1] concluded:

As usual in epidemiological research, scientists try to eliminate bias, confounding . . . Nevertheless, the protective effect of coronary heart disease incidence and all-cause mortality has remained. All but one meta-analysis agree on this point. The deviant one (1) [referenced here as [4]] found no protection, but was shown to have errors in the selection of studies and interpretation of findings (2–4) [referenced here as [6–8]].

We suggest that this conclusion suffers from at least two problems:

  • 1
    It relied selectively on three of eight invited commentaries on our study [4] (see http://www.informaworld.com/smpp/title~content=g773385479~db=all) and overlooked our response to commentaries. The neglected five commentators on our research—Drs Andreasson, Holder, Rodgers et al., Romelsjö and Shaper—all endorsed efforts to understand more clearly potential misclassification error and confounding in these studies.
  • 2
    Poikolainen [1] cites two objections to our study. First, it is alleged that there were mistakes in our selection of studies deemed to contain error—a position advanced by Klatsky [6] and Mukamal [7]. In order to identify probable misclassification error, we paid close attention to the precise wording of alcohol use questions found in all the studies examined in order to ensure that the functional meaning of specific drinking categories was very clear [4,9]. We posit that the more ‘liberal’ definitions of drinking status applied by these studies (e.g. wording such as: ‘do you rarely/never drink?’ or ‘never or almost never drink’) contributed in large part to the possible mistaken conclusion that abstinence increases the risk of coronary heart disease.

This domain of epidemiological research may be characterized as lacking in systematic attempts to address problems of confounding and bias (e.g. [10]), an important fact which Dr Poikolainen overlooks. Critical work which gives careful consideration to such issues warrants equally thoughtful consideration. In our view, it is entirely plausible that one well-designed study (e.g. [2]) may provide more insight into the true relationship between exposure and outcome than almost 30 years of systematically flawed and confounded studies.

References

  1. Top of page
  2. References