An outcome study was planned to show that a newly approved cardiovascular drug was more effective in preventing cardiovascular events than a competitor drug which had been on the market for a long time. The intention was to use the time to the first occurrence of one of the three cardiovascular events, non-fatal myocardial infarction (MI), non-fatal stroke, and cardiovascular death, as the primary variable. It turned out that the number of events needed to significantly show a relative risk reduction of 20% with 90% power was approximately 700 using a one-sided log-rank test at a 5% significance level. Since it was thought from a market perspective that the study duration would be too long to collect 700 events, a group sequential design was proposed. The idea was to perform 3 interim analysis, after approximately 25%, 50%, and 75% of the total events needed, and hoping that one of the interim analyses would show a significantly better effect for the new drug and then stop the study.. To calculate the maximum number of events needed using a Lan-DeMets(O’Brien-Fleming) approach for calculating the significance levels at each interim analysis Open
the File menu and choose New Calculation. Choose Lan-DeMets(O’Brien-Fleming)from the Boundaries for Rejecting H0: and click OK.
Set the values as shown and click Total No of Events.
The Total No of Events which will be 715.44, rounded to 716. Hence, it will only need 16 more events to get a chance to stop the study earlier with a positive result. Moreover even if the chance to stop already at the first interim analysis is small (0.012), the probability to stop at either the second or third interim analysis is 0.317 + 0.386 = 0.703 if the true relative risk reduction is 20%.
If it turns out that at the first interim analysis that the Hazard Ratio is 1.1, i.e. the new drug seemed to be worse than the competitor, it would be interesting to calculate the chance of still proving superiority for the new drug. To calculate the conditional power Open
the File menu and choose New Calculation. Set the parameter values shown below and click Conditional Power.
Similarly,
by choosing Predictive Power, it will be seen that the predictive power
is somewhat lower, 0.56005 |