Development of a prediction model for advanced age-related macular degeneration

点击翻译按钮获取中文摘要

PROJECT ABSTRACT Background Age-related macular degeneration (AMD) is a progressive disease, the advanced forms of which account for over 50% of legal blindness in the US. Vision impairment due to advanced AMD also significantly reduces quality of life and consumes a large portion of Medicare budget. Diminishing the modifiable risk factors for the progression of AMD could lead to significant clinical benefit and save health care cost. However, early identification and close follow-up of patients at high risk of developing advanced AMD are essential to allow for implementing strategies to delay progression of the disease to stages when vision is compromised. Recently, using the Age-Related Eye Disease Study (AREDS) dataset PI (Chiu) developed a prediction model for advanced AMD (c-index=0.877). However, validation analysis of this AREDS model in the Blue Mountains Eye Study (BMES) cohort indicated that it is necessary to use data from multiple cohorts to develop a prediction model with maximal generalizability. Objective Our objective is to use risk factor information provided in the patient history and clinical eye examinations to develop a widely applicable tool for the early prediction of advanced AMD. Methods Using methods extended from PI's (Chiu's) previous publications and pooled data of over 15,000 persons from four major cohorts, including the AREDS cohort (n=4,757 at baseline; data followed for 8 y will be used), the Beaver Dam Eye Study (BDES) cohort (n= 4,926 at baseline; data followed for 15 y will be used), the BMES cohort (n= 3,654 at baseline; data followed for 10 y will be used), and the Melbourne Visual Impairment Project (VIP) cohort (n= 3,271 at baseline; data followed for 5 y will be used), we will use logistic regression to model result-specific likelihood ratios of developing advanced AMD by 8 baseline demographic (n=5) and ocular (n=3) predictors. The quasi-likelihood under the independence model criterion (QIC) statistic will be used to determine the best model. Next, a composite scoring system (C score) derived from this regression analysis will be applied in the four cohorts individually to evaluate the accuracy and to depict the relationship of C score-advanced AMD risk at various times during follow-up (up to 15 y) by Kaplan-Meier estimators and Cox proportional-hazards regression using the Andersen-Gill estimators. Potential implications Our C scoring system will enhance our ability to delay progress of AMD from early stages to clinically relevant diseases. It will be useful to clinicians for communicating with patients and guiding prevention and treatment plans at very early stages of disease well in advance of vision- compromising manifestations, to researchers for increasing study power while reducing cost, and to policymakers for allocating Medicare resources.