This study develops a comprehensive machine-learning framework to predict myopic progression and the future development of high myopia, a severe form of nearsightedness associated with irreversible vision-threatening complications. Rather than introducing a new deep-learning architecture, the scientific contribution lies in building a robust, clinically interpretable prediction system using established statistical learning methods and large-scale real-world data.
The model employs multivariate linear regression to estimate individual myopia progression trajectories and logistic regression to classify the risk of developing high myopia. This choice prioritizes transparency and clinical interpretability, allowing clinicians to understand how specific variables contribute to risk rather than relying on black-box predictions. The framework was trained and validated on 612,530 medical records from 227,543 patients across five independent cohorts, spanning multiple major urban populations in China, including Guangzhou, Shanghai, and Beijing. This scale enables reliable generalization, addressing a major limitation of many AI-based ophthalmology studies.
A key scientific advance is the model’s risk stratification capability. The analysis identified two critical, data-driven risk markers: onset of myopia before age 8 and an annual progression rate exceeding 1.0 diopter. These markers distinguish children at high risk of developing severe myopia from those with slower or later-onset progression, enabling personalized monitoring and intervention strategies.
The impact of this work is its contribution to preventive pediatric eye care. By combining interpretability, large-scale validation, and actionable risk thresholds, the study provides a practical framework for shifting myopia management from reactive correction to proactive, risk-based intervention, with potential to reduce the long-term burden of high myopia at the population level.
Li, J., Zeng, S., Li, Z., Xu, J., Sun, Z., Zhao, J., Li, M., Zou, Z., Guan, T., Zeng, J., Liu, Z., Xiao, W., Wei, R., Miao, H., Ziyar, I., Huang, J., Gao, Y., Zeng, Y., Zhou, X. T., & Zhang, K. (2024). Accurate prediction of myopic progression and high myopia by machine learning. Precision clinical medicine, 7(1), pbae005. doi: 10.1093/pcmedi/pbae005
