Response to Corrrespondence by Nelson

Extract

We thank Dr. Nelson for his interest and we appreciate his concern. Contrary to randomized trials, however, no consensus exists concerning the use of meta-analysis in observational studies. There are two major problems: 1) the different metrics used and the different sets of confounders accounted for in the various studies; and 2) the inability to adjust for quality of the studies (the weights generally used are far from adequate). When the overall effect relies on large numbers, as is the case with respect to lactation and cancer risk, small deviations from the null are likely to be significant as well as meaningless.