Abstract
Purpose: :
Statistical methods to detect reduction in RNFL thickness resulting from glaucomatous progression may suffer from reduced specificity due to alpha error inflation upon repeat testing over long follow-up. The purpose of this study is to evaluate the performance of and compare two strategies for assessing progression: linear regression and confirmed change from baseline.
Methods: :
Data representing 15 semi-annual visits was generated (n=10,000) from a random normal distribution with variability based on that reported for the Average RNFL thickness measurement of the Cirrus HD-OCTTM. In addition, a slow progressive thinning (1.4 µm per year) was applied on top of this to simulate a population of progressors. The simulated, accumulating data was fit with linear regression at each visit, starting with visit 4, until statistical significance (p<0.05) occurred or the 15th visit was reached. Additionally, at each visit, a confirmed change from baseline analysis was performed between the two most recent visits and the two baseline visits (4 separate comparisons). A comparison was considered statistically significant if it exceeded the 95% confidence limits on the difference between two Cirrus average thickness measurements based on an in-house reproducibility study. If three of four comparisons met this criterion, then change from baseline was confirmed. This procedure was similarly repeated for each visit until a change was confirmed or the 15th visit was reached. Sensitivity and specificity of two methods was evaluated.
Results: :
The figures display how sensitivity and specificity change upon repeat testing for each method.
Conclusions: :
Compared to linear regression, the confirmed change from baseline strategy keeps specificity high across multiple visits. Sensitivity is reduced with confirmed change, but improves with additional visits.
Keywords: imaging/image analysis: clinical • clinical (human) or epidemiologic studies: biostatistics/epidemiology methodology • optic disc