Nonparametric and Survival Methods in Ophthalmology

摘要

ABSTRACT Ophthalmic data is of necessity bivariate. Important information is lost when eye-specific outcome and exposure data are collapsed into person-specific scores. This necessitates adjustment to standard inferential methods to account for clustering. For example, mixed effects regression models are commonly used to model normally distributed longitudinal data, but require modification when clustering exists both among fellow eyes and repeat visits for an individual. However, many ocular measures are not normally distributed and nonparametric methods of longitudinal analysis are needed. We also consider nonparametric methods in the context of confounding by eye-specific covariates where a subject may be in different strata defined by confounders for the left and right eye. These are the goals of specific aim 1. Secondly, there have been major advances in risk prediction for AMD with the discovery of important genetic predictors. However, commonly used measures of discrimination and calibration of risk prediction rules require adjustment for correlated data. Furthermore, risk factors may vary by stage of maculopathy. This is the goal of specific aim 2. In specific aim 3, we seek to use empirical Bayes methods to better predict disease course for individual RP patients, where the number of follow-up visits and duration of follow-up differs for individual patients. In specific aim 3, we propose innovative techniques for disseminating information on correlated data methods to the ophthalmic community including periodic newsletters to NEI clinical trial investigators, giving education courses at ARVO and writing review papers on correlated data methods for ophthalmic journals.