STUDYING THE LONG-TERM NATURAL HISTORY OF DISEASES USING A SHAPE-RESTRICTED NONPARAMETRIC TEMPORAL REALIGNMENT METHOD

Understanding the natural course of chronic diseases is a fundamen-tal challenge in clinical science, given that disease onset is often unknown and long-term follow-up is impractical. To address this, statistical methods (i.e., disease progression modeling with temporal realignment) have been introduced to infer the long-term natural history of diseases using short-duration follow-up data. While promising for disease management and treatment development, existing approaches require improvements and extensions to the general framework and a deeper investigation into their statistical properties to ensure a more general and statistically reliable solution. To this end, we introduce Bernstein Polynomial Temporal Realignment (BPTR), a shape-restricted nonparametric regression method that relaxes restrictive as-sumptions made by some existing approaches. BPTR employs a hierarchical framework to capture individual-level variability, allowing simultaneous estimation of: (i) the individual-specific disease progression rate, (ii) the true timing of disease onset for an individual, and (iii) the monotonic function describing disease progression over time. Additionally, covariates are incor-porated to better account for variability in progression rates and disease onset times. Through simulations we find that BPTR can accurately estimate key model parameters across increasingly complex data generating settings. We then apply BPTR to a longitudinal dataset of individuals with geographic atrophy secondary to nonexudative age-related macular degeneration, provid-ing new insights into the long-term progression properties of the disease. The method is implemented in the R package BPTR and provides a flexible tool for analyzing the natural course of a chronic disease.

Duke Scholars

Author Liangbo Shen Ophthalmology, Vitreoretinal Diseases & Surgery