(1. Randomization of the first patient. At least six examples are known to the author, and there are undoubtedly others. (2. Factorially structured treatments. The advantages of conducting two trials in the same patients when there is no suggestion of interaction, and documenting interaction if it exists must be weighed against the potential loss of power and the inconvenience of missing values. (3. RCT control groups as a source of unnatural history. Such a small percentage of apparently suitable patients are actually randomized in most clinical trials that be control group can never the considered representative of that disease seen by practicing physicians. (4. Bias in random and nonrandom treatment assignments. Differences in risk factors documented before randomization can be as much responsible for differences in outcome as the treatments under study. (5. Time lags between innovation, RCTs, and practice. These have varied from a very few years (coronary bypass operations) to 154 years (treatment of scurvy).