Diabetic retinopathy (DR) is a severe retinal disorder that can lead to vision loss; however, its underlying mechanism has not been fully understood. Previous studies have taken advantage of Optical Coherence Tomography (OCT) and shown that the thickness of individual retinal layers is affected in patients with DR. We propose a density function-based statistical framework to analyze the thickness data obtained through OCT images, and to compare the predictive power of various retinal layers to assess the severity of DR. Using a Riemannian-geometric framework, we construct novel features that capture variation in the distribution of pixel-wise retinal layer thicknesses. Using these features as covariates, we quantify the predictive power of each retinal layer to distinguish between different categories of severity in DR. We also formulate a permutation-based hypothesis test that tests for differences between averages of any two groups of density functions. Our results indicate considerable differences in retinal layer structuring based on the severity of DR, and some of these layers could serve as potential imaging biomarkers.