Validation of polygenic risk score in age-related macular degeneration prognosis: a pilot study

Abstract

Purpose : Age-related macular degeneration (AMD) is a major cause of blindness, necessitating improved prognosis and personalized treatment strategies. This study aimed to investigate the potential of combining genetic analysis with imaging techniques to develop more accurate and effective predictors for AMD outcomes.

Methods : A total of 118 patients were recruited voluntarily at Moorfields Eye Hospital. Their genotype, demographics and Optical Coherence Tomography (OCT) B-scans were collected to establish the dataset for the study. A polygenic risk score (PRS) was derived from 157 variants. The associations of PRS with age, treatment interval and tissue volume were analyzed statistically with linear regression. To explore the effect of multimodal integration with imaging data, the model was extended with imaging features extracted from passing OCT B-scans into LeNet-5 model in treatment burden classification and into NEO model in dry retina classification. In order to fuse multimodal information, imaging features were concatenated with PRS variants at the final fully connected layer of LeNet-5 model, while XGBoost was employed to take NEO imaging features and PRS as features for prediction. The extended multimodal models as well as their unimodal counterparts were trained and evaluated using the dataset with 10-fold cross validation. Their performance was compared in terms of accuracy and AUC ROC score.

Results : The study found significant associations between PRS and clinical parameters such as age, treatment burden and retina fluid volume (Table 1), though the one with treatment burden lost significance after controlling for age. Moreover, predictive models using both PRS and imaging features showed higher accuracy and AUC ROC score (Table 2). This indicated the effectiveness and predictive power of PRS when used alone or in a multimodal setting, where imaging features demonstrated a complementary effect with PRS.

Conclusions : The integration of genetic information with multimodal imaging techniques in AMD research offers promise for improved prognosis of the disease. This study highlights the significance of genetic factors in AMD development and progression and supports the potential for personalized medicine approaches.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.

 

Table 1. Associations between PRS and clinical parameters.

 

Table 2. Performance of models with PRS only, imaging features only and both as model features.