D Some theory on equivalence trials
D.1 Type I error in equivalence trials
Assume we have a continuous outcome with variance σ2 and two treatment groups of size nA and nB, respectively. Now consider the difference ˉd of the mean outcomes as treatment effect, so ˉd∼N(Δ,σ2λ2) where the true treatment difference is Δ and λ=√1/nA+1/nB.
Equivalence is established if the γ⋅100 % confidence interval for the treatment difference is within the pre-specified interval of equivalence (−δ,δ), i.e. ˉd±z(1+γ)/2σλ⊂(−δ,δ). With z=z(1+γ)/2 we obtain the equivalent requirement ˉd∈(−ξ,ξ) where ξ=δ−z(1+γ)/2σλ
to establish equivalence. This happens with probability
Pr(ˉd∈(−ξ,ξ))=Φ(ξ−Δλσ)−Φ(−ξ−Δλσ)=Φ(δ−Δλσ−z(1+γ)/2)−Φ(−δ−Δλσ+z(1+γ)/2).
Now evaluate (D.1) at Δ=δ to obtain the Type I error rate (more precisely an upper bound on the Type I error rate)
α=Φ(−z(1+γ)/2)−Φ(z(1+γ)/2−2δλσ).
Likewise, the power 1−β is derived from (D.1) with Δ=0:
1−β=2Φ(δλσ−z(1+γ)/2)−1 so 1−β/2=Φ(δλσ−z(1+γ)/2).
Note that 1−β/2=Φ(z1−β/2), and therefore
δλσ=z(1+γ)/2+z1−β/2.
With (D.2) we obtain
α≈Φ(−z(1+γ)/2)=1−(1+γ)/2=(1−γ)/2
for relatively small β (say β≤30 %), since the second term in (D.2)
Φ(z(1+γ)/2−2δλσ)=Φ(z(1+γ)/2−2(z(1+γ)/2+z1−β/2))=Φ(−z(1+γ)/2−2z1−β/2)
is then very close to zero. Note that (D.3) differs from superiority trials, where α=1−γ.
D.2 Sample size calulations in equivalence trials
Assume equal group sizes, i.e. n=nA=nB, then λ=√2/n.
With (D.3) we then obtain
n=2σ2(z(1+γ)/2+z1−β/2)2δ2=2σ2(z1−α+z1−β/2)2δ2
as the required sample size in each group for Type I error rate α and power 1−β.