Some statistical issues

The non-continuous feature of the measures of drinking habit has limited studies on the alcohol-health relationship to categorical analyses. The categories selected, the number of categories, and the distance between the categories were determined by the available data, the sample sizes and number of outcome events. The clear demonstration of a U- or J-shaped relationship requires a much larger amount of data than is generally appreciated. Some studies have combined all cardiovascular diseases or all types of stroke as the endpoint. This is often necessary because of the small numbers of cases for some of the diagnostic endpoints, but might result in partial cancellation of positive and negative associations. Furthermore, as the result of a loss of power in the categorical analysis, many reports drew conclusions based on visual observation on the trends in relative risks by drinking categories and ignored the results of formal statistical tests. Another drawback of the categorical approach is the inability to clearly define the alcohol intake level related to minimum risk. On the other hand, treating alcohol intake as a continuous variable and fitting a quadratic function forces the estimated relationship between alcohol and outcome to be symmetric. Over- or underestimation of the optimal level will occur if the true relationship is asymmetric.

In the study of the alcohol-health relationship, it is sometimes difficult to differentiate a potential confounder from a mediator of a causal outcome. Examples of these factors are blood pressure, lipoproteins, hemostasis, perceived health status, and other indicators of current health. Control strategies in a data analysis initially should not include risk factors that might serve as potential mediators lying in the pathway between alcohol intake and outcome. Failing to control for confounders may inflate the potential benefit of alcohol intake. On the other hand, improper control for mediators may spuriously deflate the potential benefit of alcohol intake. If, for example, alcohol use increases hemorrhagic stroke risk through alcohol-induced hypertension alone, then control for blood pressure would produce a non-significant association between alcohol consumption and stroke.

Alcohol No More

Alcohol No More

Do you love a drink from time to time? A lot of us do, often when socializing with acquaintances and loved ones. Drinking may be beneficial or harmful, depending upon your age and health status, and, naturally, how much you drink.

Get My Free Ebook

Post a comment