Supplementary Material for: The British-Israeli Project for Algorithm-Based Management of Age-related Macular Degeneration: Deep Learning Integration for Real- World Data Management and Analysis

Purpose: To describe the development of an integrative dataset, combining clinical and optical coherence tomography (OCT) imaging data by applying a deep learning algorithm for automated, objective, and comprehensive quantification of OCT scans to two large real-world datasets of eyes with neovascular age-related macular degeneration (nAMD). We further report baseline characteristics of the study population, focusing on demographics, clinical parameters, and quantitative retinal morphological features.
Methods: This retrospective study analyzed data from 5,207 eyes of 4,265 nAMD patients treated at two centers in the UK and Israel. Longitudinal clinical data and OCT scans were. A deep learning algorithm (NOATM, Notal Ltd.) was used to quantify retinal fluid volumes and morphological features. Baseline characteristics were compared between the cohorts.
Results: The dataset included 134,340 visual acuity measurements, 79,457 OCT scans, and 73,218 anti-VEGF injections. Median follow-up was 4.54 years (UK) and 3.12 years (Israel). Baseline visual acuity differed significantly between cohorts due to varying treatment criteria. Fluid distribution patterns were similar, with most eyes showing combined intraretinal and subretinal fluid. Age-related trends in fluid volumes were observed. Weak correlations were found between baseline OCT measurements and visual acuity.
Conclusions: This study demonstrates the feasibility of integrating large-scale clinical and imaging data for automated analysis in nAMD. The comprehensive baseline characterization provides insights into real-world presentations and lays the groundwork for enabling personalized decision-making and optimizing outcomes based on individual patient profiles and fluid distribution patterns.

目的：描述整合数据集（integrative dataset）的构建过程——通过应用深度学习算法（deep learning algorithm）对两组大型新生血管性年龄相关性黄斑变性（neovascular age-related macular degeneration, nAMD）患者眼部的真实世界数据集（real-world dataset）进行光学相干断层扫描（optical coherence tomography, OCT）扫描的自动化、客观且全面量化（automated, objective, and comprehensive quantification），从而融合临床数据与OCT成像数据。我们进一步报告研究人群的基线特征（baseline characteristics），重点关注人口统计学特征（demographics）、临床参数（clinical parameters）及定量视网膜形态学特征（quantitative retinal morphological features）。

方法：本回顾性研究（retrospective study）分析了英国和以色列两个中心收治的4265例nAMD患者共5207只眼的数据，涵盖纵向临床数据（longitudinal clinical data）与OCT扫描。采用深度学习算法（NOATM，Notal有限公司）量化视网膜液体积（retinal fluid volume）及形态学特征，并比较两组队列（cohort）的基线特征。

结果：该数据集包含134340次视力测量值（visual acuity measurement)、79457次OCT扫描及73218次抗VEGF注射（anti-VEGF injection）记录。英国队列的中位随访时间（median follow-up time）为4.54年，以色列队列为3.12年。由于治疗标准（treatment criterion）不同，两组队列的基线视力（baseline visual acuity）存在显著差异。液体分布模式（fluid distribution pattern）相似，多数患眼表现为视网膜内液（intraretinal fluid）与视网膜下液（subretinal fluid）并存。观察到液体积随年龄变化的趋势，且基线OCT测量值与视力之间存在弱相关性（weak correlation）。

结论：本研究证实了整合大规模临床与成像数据以实现nAMD自动化分析的可行性（feasibility）。全面的基线特征描述为理解nAMD的真实世界表现（real-world presentation）提供了见解，并为基于患者个体特征（individual patient profile）及液体分布模式的个性化决策制定（personalized decision-making）与预后优化（outcome optimization）奠定了基础。