Investigative Ophthalmology & Visual Science Cover Image for Volume 65, Issue 7
June 2024
Volume 65, Issue 7
Open Access
ARVO Annual Meeting Abstract  |   June 2024
Machine Learning (ML) Models for Predicting Development of Geographic Atrophy (GA) and Scar in the Comparison of Age-related Macular Degeneration (AMD) Treatment Trials (CATT)
Author Affiliations & Notes
  • Rajat Chandra
    University of Pennsylvania Perelman School of Medicine, Philadelphia, Pennsylvania, United States
  • Gui-Shuang Ying
    University of Pennsylvania Perelman School of Medicine, Philadelphia, Pennsylvania, United States
    Scheie Eye Institute, Philadelphia, Pennsylvania, United States
  • Footnotes
    Commercial Relationships   Rajat Chandra Sumitovant Biopharma, Code C (Consultant/Contractor), Roivant Sciences, Code I (Personal Financial Interest); Gui-Shuang Ying None
  • Footnotes
    Support  Supported by National Eye Institute Grant P30 EY01583 and Research to Prevent Blindness
Investigative Ophthalmology & Visual Science June 2024, Vol.65, 228. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Rajat Chandra, Gui-Shuang Ying; Machine Learning (ML) Models for Predicting Development of Geographic Atrophy (GA) and Scar in the Comparison of Age-related Macular Degeneration (AMD) Treatment Trials (CATT). Invest. Ophthalmol. Vis. Sci. 2024;65(7):228.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Purpose : GA and scar are the leading causes of visual acuity loss in eyes treated with anti-vascular endothelial growth factor (anti-VEGF) for neovascular AMD (nAMD). Predicting the development of GA and scar has important implications for assessing risk of visual acuity loss. This project is to evaluate ML models for predicting incidence of GA and scar in 2 years following anti-VEGF treatment for nAMD patients in the CATT.

Methods : This is a secondary analysis of publicly available data from CATT at https://hyperprod.cceb.med.upenn.edu/catt/catt_index.php. CATT randomized 1185 patients to ranibizumab or bevacizumab treatment for 2 years in eyes with nAMD (one study eye per participant). Four ML models [support-vector machine (SVM), random forest (RF), extreme gradient boosting (XGBoost), and multi-layer perceptron (MLP) neural network] were evaluated for predicting incidence of GA and scar in 2 years separately and combined. These ML models used baseline demographic, clinical and image features in fundus photos, fluorescein angiogram, and optical coherence tomography (OCT). Model performances were assessed by the area under the receiver operating characteristic (ROC) curve (AUC). The CATT data were randomly split into a training dataset (80%) and testing dataset for final validation (20%). The importance of each feature for these predictions was quantified by the decrease in model’s AUC after shuffling the feature.

Results : Among eyes eligible for the prediction of GA or scar, 18% (187/1021) developed GA, 45% (477/1054) developed scar, and 59% (576/979) developed either GA or scar in 2 years. The four ML models achieved AUCs of 0.62-0.70, 0.74-0.77, and 0.73-0.76 in cross-validation in the training dataset, and AUCs of 0.66-0.74, 0.70-0.75, and 0.73-0.76 in final validation in the testing dataset for predicting GA, scar, and either GA or scar, respectively (Table 1). The nAMD lesion characteristics as well as OCT fluid and thickness measurements were among the most important features (Figure 1).

Conclusions : In the CATT, ML models using baseline data were able to predict GA and scar development reasonably well in eyes treated with anti-VEGF for nAMD over a 2-year period. ML models can be helpful for identifying which nAMD patients are at greater risk for vision loss during anti-VEGF treatment for nAMD.

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

 

 

×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×