Chen Yanxian, Li Mingge, Shang Xianwen, Li Guangyu, Zhao Ziwei, Li Pengju, Liu Yanjun, Xiong Ruilin, Lai Mengying, Wang Yueye, He Mingguang, Zhu Zhuoting
School of Optometry, The Hong Kong Polytechnic University, Kowloon, Hong Kong; Research Centre for SHARP Vision (RCSV), The Hong Kong Polytechnic University, Kowloon, Hong Kong.
Department of Ophthalmology, Shenzhen Nanshan People's Hospital, Shenzhen, China.
Photodiagnosis Photodyn Ther. 2025 Jun 16;54:104672. doi: 10.1016/j.pdpdt.2025.104672.
To evaluate the early predictors for achieving full myopia control with repeated low-level red light (RLRL) therapy based on two independent randomized clinical trials (RCTs).
Myopic children undergoing RLRL therapy from a multi-center RCT (training set) and a single-center RCT (validation set) were included. Full myopia control was defined as axial elongation <0.1mm/year. Variables included age, sex, baseline refraction, ocular parameters at baseline, 1 and 3 months (axial length [AL] and subfoveal choroidal thickness [sChT]), as well as their rates of change over the first 3 months. Four random forest models to predict full myopia control after 1 year and a logistic regression was used to estimate 2-year outcome.
A total of 148 children were analyzed. The proportions of 1-year full myopia control was 54.2 % of eyes in the training set and 55.0 % in the validation set. Random forest models incorporating the rate of change in AL and sChT showed high predictive accuracy (AUC: 0.97 to 0.98) in external validation. The rate of change in AL contributed the most for model accuracy. For 2-year control, the rate of AL change had an AUC of 0.99 while the rate of change in sChT achieved only 0.69.
The rate of change in AL during the first three months emerged as the most important predictor for treatment outcomes at both 1-year and 2-year, rather than the change in sChT. Early monitoring of AL changes could be a valuable tool for identifying children most likely to benefit from this intervention.
基于两项独立的随机临床试验（RCT），评估重复低强度红光（RLRL）治疗实现完全近视控制的早期预测因素。
纳入来自多中心RCT（训练集）和单中心RCT（验证集）接受RLRL治疗的近视儿童。完全近视控制定义为眼轴伸长<0.1mm/年。变量包括年龄、性别、基线屈光、基线、1个月和3个月时的眼部参数（眼轴长度[AL]和黄斑中心凹下脉络膜厚度[sChT]），以及它们在最初3个月内的变化率。使用四个随机森林模型预测1年后的完全近视控制情况，并采用逻辑回归估计2年的结果。
共分析了148名儿童。训练集中1年完全近视控制的眼睛比例为54.2%，验证集中为55.0%。纳入AL和sChT变化率的随机森林模型在外部验证中显示出较高的预测准确性（AUC：0.97至0.98）。AL变化率对模型准确性的贡献最大。对于2年控制，AL变化率的AUC为0.99，而sChT变化率仅为0.69。
前三个月内AL的变化率成为1年和2年治疗结果的最重要预测因素，而非sChT的变化。早期监测AL变化可能是识别最有可能从该干预中受益儿童的有价值工具。
