Machine-learning models to predict myopia in children and adolescents.

Area of Science:

• Ophthalmology
• Biomedical Engineering
• Data Science

Background:

• Myopia prevalence is increasing globally, posing a significant public health challenge.
• Early prediction and intervention are crucial for managing myopia progression and its associated complications.

Purpose of the Study:

• To evaluate the efficacy of machine learning algorithms in predicting myopia among elementary school students.
• To identify key ocular, environmental, behavioral, and genetic factors influencing myopia development.

Main Methods:

• A case-control study involving 2,947 elementary school students in Shenzhen, China.
• Utilized stratified cluster random sampling for participant selection.
• Developed and compared five machine learning models: Random Forest (RF), Decision Tree (DT), Extreme Gradient Boosting Trees (XGBoost), Support Vector Machine (SVM), and Logistic Regression (LR).
• Models were trained using myopia screening data, ocular biological parameters, and questionnaire responses.

Main Results:

• Myopia prevalence was 47.2% in the study population.
• All five machine learning models demonstrated strong predictive performance with Area Under the Curve (AUC) values above 0.75.
• SVM, LR, RF, and XGBoost showed superior predictive performance (AUCs ranging from 0.815 to 0.846) compared to DT (AUC = 0.791).
• Key predictors identified included axial length, age, sex, maternal myopia, and infant feeding patterns. Axial length was the most significant risk factor (OR = 8.203).

Conclusions:

• Machine learning models offer a robust approach for myopia prediction in children.
• Accurate identification of myopia risk factors can inform targeted prevention and control strategies.
• These predictive models can aid in early intervention for high-risk individuals, potentially mitigating myopia progression.