Abstract
Purpose:
A general strategy for showing non-inferiority of a test product, to an existing approved product, is to demonstrate the following efficacy criteria: the upper limit of a 95% confidence interval (CI) around the difference in IOP: test product - approved product is: a) < 1.5 mm Hg at all defined time points b) < 1.0 mm Hg at a majority of time points Historically, 1) the correlation among time points within a subject/eye and 2) the requirement of a majority of time points having an upper 95% CI < 1 mm Hg have not been overtly used when determining sample size and power. We propose a method to account for both of these points.
Methods:
Multivariate normal distributions with varying assumptions around the sample size, mean differences, variances, and correlations were used; where correlations ranged from 40% to 70% as estimated from historic data. 500,000 random samples were created and 95% CIs calculated. Power was calculated as the proportion of random samples showing the upper limit of the 95% CI < 1.5 mm Hg, with a majority < 1.0 mm Hg.
Results:
Power calculations from common methods either over estimate the required sample size (calculations assuming all 95% CIs must be < 1.0 mm Hg) or under estimate the required sample size (calculations assuming only that all 95% CIs < 1.5 mm Hg). The amount of the over estimation and under estimation depend on the assumed correlation among time points; the higher the correlation the less the over estimation and the more the under estimation of sample size.
Conclusions:
When determining power and sample size for primary open angle glaucoma / ocular hypertension clinical trials, neglecting to account for the complete efficacy criteria and the correlation among time points will result in the trial requiring too many subjects/eyes or having too low power.
Keywords: 459 clinical (human) or epidemiologic studies: biostatistics/epidemiology methodology •
568 intraocular pressure