SEARCH

SEARCH BY CITATION

Keywords:

  • Alcohol;
  • epidemiology;
  • per capita consumption;
  • social costs;
  • survey;
  • validity

In this issue, Rey and colleagues [1] contribute to our understanding of estimation of alcohol-attributable harm by presenting a number of sensitivity analyses on their estimates. Their analyses show that the source of exposure estimates and the choice of approach—aggregated using all-cause mortality versus disaggregated using cause of death by disease—have a major impact on the results. In fact, estimates of alcohol-attributable mortality in France varied between 24 000 deaths avoided and 20 000 deaths caused, depending upon which assumptions were chosen. Does this mean that outcomes of such studies are arbitrary, and depend mainly upon the methodological preferences of the author?

Such a conclusion would be wrong, and reasons will be given in the following. First, an aggregated approach based on all-cause mortality would not be acceptable for any serious epidemiological assessment such as [2] or [3]. The main reason is that the all-cause mortality meta-analyses upon which to base such estimates would be built mainly on elderly cohorts with very specific mortality profiles, most importantly with a high proportion of ischaemic heart disease deaths [4]. Such a profile does not correspond with any mortality profile of any given country, neither for age nor for distribution of current causes of death, which is related and more important. There is no protective effect for alcohol in younger cohorts, as the beneficial effects on ischaemic disease and diabetes occur later in life ([5]; for a meta-analysis see [4]). Younger age groups with different causes of death are heavily under-represented in any all-cause mortality meta-analyses [6], including the one used by Rey and colleagues [1]. In addition, most underlying epidemiological cohort studies were not based upon representative samples at baseline, but on ease of tracking people after several years [7], and thus large parts of the population such as mobile or underprivileged people are missing almost completely in medical cohorts. However, these subpopulations and their mortality patterns are part of the overall mortality picture for any given country. Thus, the only reliable way to reflect the causes of burden of disease in a country at any given time is to base the calculations on the actual mortality and morbidity profiles of this country and the cause-specific risk relations.

The second major question concerns exposure information. Adult per capita consumption is arguably the best indicator for overall alcohol consumption in a country [8]. However, this information cannot be used to conduct estimates by sex and age, but has to be complemented by survey information [9,10]. The problem is that the answers from surveys added up over respondents amount to 20–80% of adult per capita consumption in most countries [10,11]. This has two implications: first, and most importantly, adult per capita should be used to standardize surveys. Currently, surveys are often used to estimate level of drinking in a country, even when coverage of real consumption is low. For example, the national surveys in Canada have coverage rates of less than 35%, and obviously cannot and should not be used to estimate level of drinking in this country without additional calculations (such as in [12]). The problems become even more complicated if drinking is compared between countries based on surveys: clearly, such comparisons make sense only when standardized against a valid external standard of real consumption. Who would dare to compare statistically and interpret the results of the aforementioned Canadian surveys with a US survey with 60% coverage rate, without trying to adjust for real consumption? Standardization of surveys should become a standard in our interpretation of surveys as standardization of mortality rates in other field of epidemiology [7].

Such standardization is necessary, irrespective which level of per capita consumption is chosen for standardization for cost or burden of disease studies. For these studies, it may be argued that part of the undercoverage is due to individual under-reporting, and as this under-reporting also occurs in surveys on risk relations, up-estimation of surveys to 100% of adult per capita estimation may lead to overestimation of the impact of alcohol. Unfortunately, undercoverage also results from incomplete sampling frames in surveys (leaving out high-volume drinkers systematically, such as the homeless or institutionalized) and from non-response [8]; both errors should be taken into consideration when survey and per capita results are triangulated. So, for the final standardization for cost studies, values such as 80% or 90% of per capita consumption may be most realistic.

In summary, it is surprising how many surveys on alcohol are published with relatively little methodological research being carried out on understanding their systematic biases (for exceptions see [13]). More research is necessary to understand the sources of undercoverage and to determine how to best triangulate survey and per capita consumption for studies such as that by Rey et al.[1]. However, many improvements to understanding and standardizing surveys have been made [10,13,14], and should be used routinely. As well, comparisons of surveys results without any kind of standardization in case they have not shown to have similar coverage rates should no longer be acceptable.

Declaration of interest

  1. Top of page
  2. Declaration of interest
  3. Acknowledgements
  4. References

The author had contracts with the pharmaceutical industry and received funds to participate in meetings by the alcohol industry.

Acknowledgements

  1. Top of page
  2. Declaration of interest
  3. Acknowledgements
  4. References

Support to CAMH for salary of scientists and infrastructure has been provided by the Ontario Ministry of Health and Long Term Care. The contents of this paper are solely the responsibility of the author and do not necessarily represent the official views of the NIAAA or NIH or those of the Ministry of Health and Long Term Care.

References

  1. Top of page
  2. Declaration of interest
  3. Acknowledgements
  4. References