The fragility of cardiovascular clinical trial results

Clinical trials that have their prospective analysis plan altered are difficult to interpret.

After providing 4 examples of problematic trial results that have had their findings reversed, the necessity of a fixed research protocol is developed. Investigators generally wish to extend the results from their research sample to the larger population; however, this delicate extension is complicated by the presence of sampling error. No computational or statistical tools can remove sampling error--the most that researchers can do is to provide to the medical and regulatory communities a measure of the distorting effect that sampling error can produce. Investigators accomplish this by providing an estimate of how likely it is that the population produced a misleading sample for them to study. However, studies in which the data determine the analysis plan damage these estimators. When they are damaged, these estimators produce untrustworthy assessments of the degree to which the study results reflect the population findings.

The way to avoid these complications is to design the experiment carefully, then carefully execute the experiment as it was designed.