Abstract
Purpose :
To understand and quantify the growth rates of geographic atrophy (GA) in age-related macular degeneration using statistical methods to analyse inherent uncertainties.
Methods :
Epistemic uncertainties arise from incomplete knowledge due to limitations in measurement devices, insufficient data or subjective human evaluation. This study identified epistemic uncertainties and investigated modelling the growth of GA using statistical approaches. Fundus autofluorescence (FAF) images were acquired along with progression data (e.g., total area of growth) using Spectralis HRA+OCT instrumentation and its associated RegionFinder software. Eyes with ≥3 visits and good quality FAF were used in the study. We describe sources of epistemic uncertainty associated with GA progression and, to address model uncertainty, tested five regression models to explain underlying growth and progression. Various statistical tests, including the coefficient of determination (r2) and the uncertainty metric (U = 1 - r2) were used. The tested models included linear, exponential, power, logarithmic and quadratic.
Results :
A total of 81 cases with bilateral GA and 531 FAF images (range: 3-17 images per eye) over an average of 65 months (range: 23-119 months) were collected. The mean baseline lesion size was 2.62 mm2 (range: 0.11-20.69 mm2). Among the five regression models tested, the linear model was the most promising approach to predict the progression of GA. It had the smallest average uncertainty (U = 0.025) and highest average r2 (0.92). The hypothesis test for the significance of the correlation coefficient, r, supported the applicability of the model (p = 0.01). The linear model was a balance between statistical performance and physical assumptions, within the domain of application, and was easy to use, interpret and implement. Clinical assumptions suggest that progression will eventually taper off due to the limited space in which GA lesions can grow within the retina. It is suggested that the linear model identified is the linear portion of a growth model, such as a logistic function. Additional early-onset and end-stage data could expose the entirety of the GA growth process.
Conclusions :
Statistical analysis of uncertainty suggests that the linear model provides an effective and practical representation of the rate and progression of GA, based on available data from patients in clinical presentations.
This is a 2020 ARVO Annual Meeting abstract.