Abstract
The proper analysis of the data generated by studies of carcinogenic risks of drinking alcoholic beverages would be the application of models from the relatively new approach of meta‐analysis. In this study, 441 articles were generated by a 1992 MEDLINE search of the key words “alcohol drinking” and “cancer.” Of these, only 29 met the criteria for a formal meta‐analysis. For these 29 research reports, the 95% confidence limits for the odds ratio were 1.28 and 1.15, suggesting a weak association between drinking and cancer. This conclusion was rendered even less decisive by the following problems in the studies analyzed: 1) absence of comparable measures of either dosages or drinking patterns; 2) absence of comparable methods of data analysis; 3) absence of comparable measures of other population characteristics; and 4) widely varying results from study to study. For example, the 95% confidence limits for the odds radio of the 16 European studies were 1.14 and 0.98, indicating not even a reliable directional difference between drinking and nondrinking populations. Although the World Health Organization International Agency for Research on Cancer concluded in 1987 that alcoholic beverages are carcinogenic, the scientific literature extant in 1992 provides only very weak support for that finding. There is a need for multiple nonexperimental investigations using methods that will produce results sufficiently comparable to justify the application of the statistical models being generated for the meta‐analysis of important questions not subject to direct experimentation. The major implication of this study is that cancer education regarding the use of alcoholic beverages should include more than a simple tally of research results reported in the scientific literature.