Validation of a Composite Scoring System for the Prediction of Advanced Age-related Macular Degeneration

Abstract

 

Purpose:

 

Using 5 demographic and 3 ophthalmic predictors in a logistic regression to model result-specific likelihood ratios in the American Age-Related Eye Disease Study (AREDS) dataset (8-y follow-up), we devised a composite (C) scoring system (C_AREDS scoring system: C_AREDS = -4.48 + ΣC_{j, AREDS}) that permits the prediction of advanced age-related macular degeneration (AMD) (AREDS model in Table 2). The objective of this study is to validate this AREDS model in the Australian Blue Mountains Eye Study (BMES) cohort (10-y follow-up).

 

Methods:

 

To evaluate the performance of the C_AREDS scoring system, we first assigned C_AREDS score to all eligible eyes in the BMES cohort (Fig 1). Then, at various C score cutoff points each eye was classified into "positive" or "negative" according to its C_AREDS score. Next, through frequency table analysis of this classification vs. the actual status of progression to advanced AMD, we calculated the empirical sensitivities/specificities and measured the area under the receiver operating characteristic (ROC) curve (c-index_BMES/AREDS). To explore the deviation of the BMES dataset from the AREDS model, we developed a prediction model (C_BMES scoring system) using the BMES dataset as the training sample and compared it to the C_AREDS scoring system (Table 2).

 

Results:

 

The c-index_BMES/AREDS (= 0.921) implies an excellent performance of the C_AREDS scoring system in the BMES cohort. However, the empirical sensitivities/specificities deviate from the expected values (Table 1). Although both C_AREDS (c-index_AREDS = 0.877) and C_BMES (C_BMES = -16.26 + ΣC_{j, BMES}, c-index_BMES = 0.953; Fig 2) scoring systems have excellent internal validity, further analysis implies that the deviation is due to the differences in baseline risk factor profile and duration of follow-up between the two cohorts.

 

Conclusions:

 

These results suggest that, while the methodology of the C scoring system is valid, it will be optimal to use long-term follow-up data from multiple cohorts in order to develop a widely applicable model for the prediction of advanced AMD.  

 

Keywords: age-related macular degeneration • clinical (human) or epidemiologic studies: risk factor assessment • clinical research methodology