Purchase this article with an account.
Benjamin D Sullivan; Statistical Arbiters of Clinical Trial Success in Dry Eye. Invest. Ophthalmol. Vis. Sci. 2014;55(13):1974.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
The purpose of this study was to provide a formal statistical basis to understand clinical trial outcomes when using various biomarkers in dry eye disease.
The model used the TearLab CVS patient database of bilateral tear osmolarity, corneal/conjunctival staining, TBUT, Schirmers, MGD, and OSDI measures. Dry eye subjects were randomized into two groups of N=50, with resulting composite severities of 0.40 and 0.41. A target severity was chosen for placebo (0.36), low effect (0.30), and high effect (0.27) compounds. To estimate therapeutic efficacy, patients with either a uniform or Gaussian probability centered around the target severity were randomly selected to populate treatment groups. The use of composite severity as a coordinate provided an unbiased model free from assumptions on any specific biomarker. Results were compared to a retrospective analysis of clinical trial data to test the model performance.
The uniform low effect model returned significant reductions compared to placebo in osmolarity (309.3 vs. 317.9 mOsm/L, p < 0.005) and severity (0.32 vs 0.37, p < 0.017), and almost achieved significance in OSDI (15.6 vs. 24.4, p < 0.057). TBUT, staining, and MGD grading were not significant (p = 0.314, 0.345, 0.134 respectively). The uniform high effect model demonstrated significant reductions across all signs (p < 0.001 for all). However, when selections were repeated 50 times for each effect size, the average p value across all signs was 0.39 for the low effect and 0.15 for the high effect, with none of the signs averaging < 0.05, even though in individual trials each marker would occasionally reach significance. Of particular interest, the primary determinant of clinical trial success was the homogeneity of response. Reducing the standard deviation of the Gaussian model resulted in consistent significance in both low and high effect sizes. The retrospective analysis of the literature suggests that this model partly explains the superiority of tear osmolarity as a marker for therapy, as osmolarity was the only sign to reduce its variability (e.g., improve homogeneity) following therapy in a variety of trials.
The key driver in clinical trial success is the homogeneity of the patient response, suggesting that qualifying patients with a single etiology and relying upon biomarkers that reduce variability following treatment is paramount.
This PDF is available to Subscribers Only