Integrating Machine Learning and Traditional Survival Analysis to Identify Key Predictors of Foveal Involvement in...

Area of Science:

• Ophthalmology
• Medical Imaging
• Machine Learning

Background:

• Age-related macular degeneration (AMD) is a leading cause of vision loss.
• Geographic atrophy (GA), a late stage of AMD, can lead to irreversible vision impairment.
• Foveal involvement in GA significantly impacts visual function.

Purpose of the Study:

• To investigate the incidence of foveal involvement in geographic atrophy (GA) secondary to age-related macular degeneration (AMD).
• To utilize machine learning to identify and assess the importance of risk factors for foveal involvement in GA.
• To understand the temporal progression and predictors of foveal involvement in GA patients.

Main Methods:

• Retrospective, longitudinal cohort study including 167 eyes with foveal-sparing GA.
• Fundus autofluorescence and optical coherence tomography (OCT) were used to measure GA characteristics and foveal involvement.
• Kaplan-Meier curves, Cox regression, and Random Survival Forests (RSF) were employed to analyze foveal survival and risk factors.

Main Results:

• The median foveal survival time was 45 months.
• Incidences of foveal involvement were 26% at 24 months and 67% at 60 months.
• Key risk factors identified included GA proximity to the fovea, worse baseline visual acuity, and thinner outer nuclear layer. RSF analysis confirmed these predictors.

Conclusions:

• Anatomic and functional factors significantly impact the risk of foveal involvement in GA.
• Identifying patients at risk for foveal involvement can facilitate tailored preventive strategies.
• These findings contribute to a better understanding of GA progression and management.