Sample size calculation for non-inferiority trial, using means rather than proportions

[Ka Wai Kam]

Dear all,

I am designing a non-inferiority trial that compares three groups of subjects receiving different intraocular lenses. The primary outcome that I wanted to look at was the binocular vision at intermediate distance postoperatively.

Nearly all the tutorials online talked about using artbin, and proportions as a way to calculate the sample size.
May I know if it is feasible to use means instead?

e.g.
Instead of using "proportions of eyes exceeding a certain level of vision"

can I calculate the sample size by
stating the previously reported "mean visual acuity" of the two study lenses with their respective SD?


Based on published results of other cohort studies
I gathered the below information

binocular intermediate VA
- lens 1 (control) : 0.24 +/= 0.14
- lens 2 (study lens) : 0.05 +/- 0.1

If I use normal t test

. power twomeans 0.24 0.05, sd1(0.14) sd2(0.1)

Performing iteration ...

Estimated sample sizes for a two-sample means test
Satterthwaite's t test assuming unequal variances
Ho: m2 = m1 versus Ha: m2 != m1

Study parameters:

alpha = 0.0500
power = 0.8000
delta = -0.1900
m1 = 0.2400
m2 = 0.0500
sd1 = 0.1400
sd2 = 0.1000

Estimated sample sizes:

N = 16
N per group = 8


Is this 8 per group - for 'superiority trial' rather than non-inferiority?

In my study design, there are TWO experimental groups - where I implant the study lens in dominant eye in one group, and non-dominant eye in one group.
- should I take into account of this third group during my sample size calculation? My original thought was to use the dominant eye vs control group only.

Thank you so much!

Kam

[Ian White]

Kam, this is probably too late for you, but here's a reply. Standard software such as -power twomeans- can be used to find sample sizes for non-inferiority trials with continuous outcome. -artbin- provides a specific non-inferiority option for trials with binary outcome because the calculation is a little different.

The calculation you posted seems to be about superiority. You don't mention a margin at all. The idea for NI trials with continuous outcome is this: if you expect means of (say) 0.05 in each group, and you want power to demonstrate non-inferiority with margin (say) 0.03, then your sample size is the same as for a superiority trial with expected difference 0.03. So you can use -power twomeans 0.03 0- with suitable sd options.

For a 3-arm trial, the approach is to work out what hypotheses you will test. Usually people want to test each of 2 groups against the 3rd, so they use a standard sample size calculation. You also need to decide whether you want to control the comparison-wise or overall type 1 error rate.