Estimating the effect of the run-in on the power of the Physicians' Health Study.