June 2017
Volume 58, Issue 8
ARVO Annual Meeting Abstract  |   June 2017
A systematic approach to statistical evaluation of geographic atrophy clinical studies
Author Affiliations & Notes
  • Thomas Kim
    Allergan, Irvine, California, United States
  • Alan Mitchell
    Allergan, Irvine, California, United States
  • Jeen Liu
    Allergan, Irvine, California, United States
  • Kevin Kerr
    Allergan, Irvine, California, United States
  • Francisco J Lopez
    Allergan, Irvine, California, United States
  • Susan Schneider
    Allergan, Irvine, California, United States
  • Footnotes
    Commercial Relationships   Thomas Kim, Allergan (E); Alan Mitchell, Allergan (E); Jeen Liu, Allergan (E); Kevin Kerr, Allergan (E); Francisco Lopez, Allergan (E); Susan Schneider, Allergan (E)
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science June 2017, Vol.58, 2341. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Thomas Kim, Alan Mitchell, Jeen Liu, Kevin Kerr, Francisco J Lopez, Susan Schneider; A systematic approach to statistical evaluation of geographic atrophy clinical studies. Invest. Ophthalmol. Vis. Sci. 2017;58(8):2341.

      Download citation file:

      © ARVO (1962-2015); The Authors (2016-present)

  • Supplements

Purpose : Various methods have been proposed for the analysis of geographic atrophy (GA) lesion area in clinical studies. GA lesion area skewness and small sample size present challenges during analysis of these datasets. We explored various approaches to statistical analysis of GA lesion area and propose a measure called effective radius (ER) to evaluate progression of the lesions.

Methods : Analyses were based on a phase 2 study (NCT00658619) with 113 bilateral GA patients. Patients were randomized to sham procedure, 132μg or 264μg Brimonidine Drug Delivery System. ER was defined as the square root of GA lesion area divided by π and represents an integrated radius for all the lesions. Skewness of the data was assessed at each time point after scaling within each treatment group. Departures from normality were assessed using the Shapiro-Wilk’s test (SW). Changes from baseline (CFB) were estimated using various models. An ANCOVA model estimated CFB at Month 12. Mixed-effects models with time as a categorical or continuous variable were also fit to the data. For the categorical time models, fixed effects included baseline GA lesion area (BGLA), time, treatment, and time by treatment, as well as BGLA by time as interactions; a random subject effect was included. For the continuous time models, BGLA and the treatment by time interaction were fixed effects and BGLA and time included as random effects.

Results : GA lesion areas showed asymmetric and non-normal distributions at all time points. Skewness ranged from 0.523 to 0.889 and SW p-values from < 0.001 to 0.003. Transformation to ER reduced skewness and SW p-values showed no evidence of non-normality. Evaluation of model residual distribution via QQ-plots suggests no changes after transformation, despite the improvement in skewness. Boxplots of the residuals by time showed no differences or patterns in the spread suggesting appropriate fit of the models.

Conclusions : Models for raw GA lesion area and ER fit the data equally well. Although statistical properties improved after ER transformation, models seem to account well for the skewness seen in raw data. The potential advantage of ER transformation is partially offset by the difficulty in interpreting the novel measure. Larger datasets may provide better discrimination of the potential advantages of transformations.

This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.


This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.