The question of whether or not a study is convincing is related to the issue of comparability of treatment groups. When groups are not comparable with respect to prognostic factors, adjusted analyses can be performed. However, difficulties may arise with historical controls. Given that we must adjust, we therefore must have the information needed to do the adjustment. We may know on which variables we want to adjust, but we may have trouble obtaining the data for the historical controls. This may not be a problem for a major institution that is in the business of conducting such studies, but sometimes the needed information just is not there. Often it is not possible to go back and obtain missing data when dealing with the past. Either a needed result was never obtained, or the determination was made but the data are now lost.A different and even more difficult problem relates to the large number of unmeasured or unknown prognostic factors ... in spite of all the advances we have made in learning about the natural history of the disease, there is still so much we do not know. The imbalances in these factors are what can hurt us in nonrandomized studies, whereas randomization assures us that the groups will be alike on the average. Time trends are a very important aspect here. The patient population is changing, and these changes may be quite subtle. Changes in details of treatment and changes in supportive care may also introduce subtle and unconscious biases.
展开▼