The Power Derivative Principle, and Its Application to How and When to Perform a One-Shot Unblinded Reassessment Sample Size.

BACKGROUND

In adaptive two-group clinical trials, a current method is to perform a one-shot unblinded sample size reassessment. Whereas the interim unblinded look of the data is adjusted for inference by using the weighted Cui-Hung-Wang statistics, some questions remain: how and when to reassess the sample size?

METHODS

We define the Power Derivative Principle as follows: a sample size is optimal when the derivative of the power with respect to the sample size has reached some implicit value. Applied to two-group clinical trials, this Power Derivative Principle determines a new one-shot unblinded sample size reassessment rule (including the determination of futility bounds). A full Power Derivative Strategy induces furthermore an optimal information fraction for the interim analysis. The Power Derivative Strategy is then compared to adaptive design methods proposed in the literature and to group sequential strategies. For this comparison, we used, on the one hand, the very common information fraction f = 0.5 and, on the other hand, the information fraction found as being optimal with the Power Derivative Principle.

RESULTS

The optimal information fraction depends only on α-and β-risks. For usual values of these risks, the optimal information fraction value is very close to 0.9. Moreover, with this unexpected optimal value, reassessment methods become roughly comparable (it is definitely not the case when f=0.5).

CONCLUSIONS

Our results suggest that a sample size reassessment is more beneficial when considered close to the planned end of a trial, allowing a study with borderline interim results to be saved.