Anastasia Ivanova, Ph.D, University of North Carolina at Chapel Hill

1. In most psychiatric trials the efficacy of a treatment is measured as a score reduction (e.g. HAMD score) from baseline.
   For example, a standardized treatment mean of a new treatment can be -2 and placebo mean -1.6, yielding the standardized treatment effect (difference) of -0.4.
   The software below implements a one-sided test with negative value of the treatment effect as an input.
2. We assume that the marginal placebo mean in stage 2, μX, that is, the mean in the whole population, is equal to the marginal placebo mean in stage 1.
   Similarly for the marginal drug mean μY. We also assume that the correlation between placebo responses in stages 1 and 2 is non-negative.

Calculating required sample size
for a given power

Type I error rate, α (one-sided):

Power:

Allocation ratio,
a
:

Marginal treatment effect μY – μX in stages 1 and 2, θ1:
Conditional treatment effect in stage 2, θ2:
Marginal standard deviation in placebo group in both stages, σX:

Conditional standard deviation in the placebo group in stage 2, τx:

Marginal standard deviation in drug group in both stages, σY:
Proportion of placebo non-responders in stage 1, r:

Retention rate, s:

First-stage weight in the linear combination of treatment effects, w:

Calculating power
for a given total sample size

Type I error rate, α (one-sided):

Total sample size:

Allocation ratio,
a
:

Marginal treatment effect μY - μX in stages 1 and 2, θ1:
Conditional treatment effect in stage 2, θ2:
Marginal standard deviation in placebo group in both stages, σX:

Conditional standard deviation in the placebo group in stage 2, τx:

Marginal standard deviation in drug group in both stages, σY:
Proportion of placebo non-responders in stage 1, r:

Retention rate, s:

First-stage weight in the linear combination of treatment effects, w: