(www.improvingmedicalstatistics.com)
home
author contact
web
site dedication
|
|
|
......................m......................... Specific guide to this web site for:
Additional Topics: Large Randomized Clinical Trials
|
Analysis of contemporary women and alcohol ingestion regarding pregnancy outcome would not give as reliable of results as the study of women from previous decades. The issue of whether alcohol ingestion adversely affects birth weight in nonsmokers would be more effectively evaluated in women in the United States from earlier decades. In the 1970s and early 1980s, there was not a pervasive impression among the public that even small amounts of alcohol should be avoided by pregnant women. If women were studied in a similar fashion in contemporary times, a more biased data set would be obtained. Women that follow a presumably unhealthy practice of one type during pregnancy are potentially more likely to follow other unhealthy practices in regards to infant outcome. Since any alcohol ingestion during pregnancy is now popularly thought to be unhealthy, a contemporary pregnant women who drank would also be more likely to have other behaviors which have negative effects for the pregnancy. Since a less biased set of data is available for analysis, alcohol ingestion during earlier decades is the preferable, less biased data set for analysis. The converse of this type of influence was recently demonstrated in the studies of estrogen use in post menopausal women. Nonrandomized studies comparing women who were taking estrogen to those who did not take this medication suggested that women who took estrogen had a substantially lower risk of cardiovascular events. Randomized trials did not show this at all. Why the difference? Women who were taking estrogens previously tended to have other behaviors and characteristics that were associated with low risk which gave the incorrect impression that the estrogen itself was beneficial. Epidemiology studies are quite valuable, but are always at risk for this type of problem despite the best efforts of the statisticians to factor out these influences. The question of whether a relationship is a causative one or simply an association remains. Epidemiology studies (descriptions of prevalence in populations) suggest relationships and subsequent experimental studies with randomized trials can help determine if those relationships are casually related. An example of a superbly done epidemiology study is the Framingham study which has studied a particular population for many decades and helped show that cholesterol levels were associated with heart disease and stroke. It took subsequent randomized clinical trials to prove that lowering a significantly elevated cholesterol can lead to a reduction in heart attack and stroke. Obviously, a randomized trial of alcohol use in pregnancy is neither appropriate or feasible. Hence, epidemiology studies, though less reliable, are the basis for many of the recommendations regarding alcohol and pregnancy. If the raw data from the large study by Mills et al1 still exists it would be quite interesting if this data was given to several different groups of physicians and statisticians to reanalyze the relation of low level alcohol ingestion in nonsmokers to birth weight and pregnancy outcomes.
|