In allocating subjects to the GP and SiH lens groups, a block-randomization scheme was used, such that approximately equal numbers of subjects were enrolled in the two lens groups at any given time throughout the study. This scheme minimized the chance that a systematic bias would be introduced by shifts in observer criteria, change of personnel, or instrument drift during the course of the study.
After a thorough exploratory analysis, we constructed multivariate, linear, mixed-effects models of epithelial permeability, with a random effect for the pairs of eyes within subjects and fixed effects for lens type, study design factors, and both time-variant and time-invariant covariates. The types of explanatory variables examined included study design factors (baseline/P30D, PM/AM, patched/unpatched), demographic information, properties of ocular morphology and tear film, ocular response variables, and contact lens fitting and performance measures. In some cases, statistical power was gained by using a continuous variable (e.g., time awake before measurement), rather than the categorical factor (e.g., AM versus PM). Since both eyes were unpatched all day before the PM measurements, the time awake was the same for both eyes—approximately 500 minutes, on average. The patched and unpatched eyes were distinguished at the AM visit by their awake times of approximately 5 and 50 minutes on average, respectively. The means, standard deviations, and distributions of awake time were well matched in the GP and SiH groups and were not significantly different. The raw P dc readings were transformed by natural logarithm where necessary, to ensure the approximate normality required by statistical tests. Additive main effects and multiplicative interactions were examined as potential explanatory variables. The fitted models were evaluated by comparing F-test P values, examining regression diagnostic plots, comparing log-likelihood values between nested models, and comparing Akaike's Information Criterion between non-nested models.