Commentary on Parish <i>et al</i> . (2017): What is the best exposure for estimating alcohol‐attributable burden of disease?

Triangulation of survey data on alcohol use with more reliable per capita data is a common procedure for estimating true alcohol exposure. Improved methodology for such triangulation will also help to establish comparability between surveys, and to improve comparative risk assessments for burden of disease. However, additional research is needed to loosen the assumption of under-reporting of all survey respondents by a fixed proportion. Survey data usually underestimate true consumption in a population, and triangulation with per-capita data is a common procedure for estimating true alcohol exposure or alcohol-attributable harm 1, 2. Parish and colleagues presented a new methodology for such triangulations 3, which will not only improve comparability between consumption level of surveys, but also lead to more reliable comparative risk assessments for burden of disease. To measure alcohol use in a population, aggregate statistics based on sales or on production, export and import have been shown to yield reliable and, in most cases, much higher levels of use compared to surveys 4, 5. However, the level of coverage varies widely, from as low as 12% in some South African surveys 6 to close to 90% in New Zealand surveys 7. This leads to problems of comparability between surveys within and between countries, with estimating much lower consumption levels based on surveys with low coverage compared to surveys with high coverage. Triangulations such as the one by Parish and colleagues 3 can be used for more valid comparisons, by standardizing to per capita consumption or an arbitrary proportion thereof 1, 8. A more difficult question is to determine the best level of adult per-capita consumption to be used for alcohol-attributable burden calculations as part of comparative risk assessments. Undercoverage of surveys results from three main mechanisms: (a) individual under-reporting; (b) use of assessment instruments which make it difficult to capture variability and thus will result in omission of heavy drinking occasions (e.g. questions on usual drinking 9); and (c) sampling frames which exclude the heaviest drinkers and differential non-response 10. Individual under-reporting may also occur in medical cohort studies to determine risk relations so, if one uses 100% of adult per-capita consumption for triangulation, it may lead to overestimation of harm. The assessment methodology in medical epidemiology is often different, and has been shown to be associated with less underestimation (e.g. 48 hours’ dietary protocol 11; see also 12). Moreover, for clinical cases with very heavy consumption, there is not necessarily an under-reporting of alcohol 13, 14. Finally, as epidemiological studies do not have to be representative to estimate risk relations, and as the whole variation of exposure can be covered by meta-analyses of different studies, including studies with samples of heavy or very heavy drinkers, the problem of sampling frame is irrelevant. Taken together, these considerations led the World Health Organization (WHO) Technical Advisory Group on Substance Use Epidemiology to recommend an assumption of 80% coverage for triangulation. Such a recommendation could also be implemented easily using the Parish approach. Up to now, all triangulation approaches suffer from one crucial assumption, that all survey respondents under-report their alcohol consumption by a fixed proportion. This assumption is certainly false in this rigidity. Future research will need to differentiate levels of underestimation by different types of responders with different drinking patterns. In this context it should be noted that, for different kinds of risk relations, precision in different parts of the alcohol use exposure is important. For example, for diseases with relatively flat and almost linear curvature of the relative risk curve such as breast cancer 15, estimating mild to moderate levels of drinking correctly becomes crucial, whereas for exponential curves such as liver cirrhosis 16, estimating the heavy drinking levels is more important (for an overview of curvatures see 17, 18). Thus, flexible triangulation techniques such as the Parish methodology 3 will become more important. Unfortunately for these estimates, the causal impact of alcohol use on disease and mortality is not limited to the dimension of level of use 17. Patterns of drinking, especially irregular heavy drinking occasions, play a crucial role, and current triangulation methods have not yet created solutions for capturing this dimension comparatively. In summary, while the contribution of Parish and colleagues 3 will enlarge our methodological arsenal, many more such contributions are needed to improve population distributions of alcohol exposure and attributable harm. Given the importance of alcohol as a risk factor for burden of disease and mortality 19, the scarcity of such research is surprising. None. Part of this work was supported by WHO Collaboration Centre for Addiction and Mental Health.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.