83
Case Study:
Impact of Interim Analyses
Increased maximal sample size actually afforded better efficiency on average
–Pocock boundary shape function: lower ASN over range of alternatives examined
•This improved behavior despite the 36.7% increase in maximal sample size
–Worst case behavior
•O’Brien-Fleming: never more than N= 1773
•Pocock continues past 1755 only if MLE for treatment effect is between -0.0357 and -0.0488
»Always less than 16.01% chance, which occurs when the difference in mortality is -0.0422