Interpretable machine learning model based on blood parameters for screening high myopia.

Area of Science:

• Ophthalmology
• Biomedical Informatics
• Clinical Diagnostics

Background:

• High myopia (HM) poses a significant public health challenge.
• Current screening methods for HM can be inconvenient and costly.
• There is a need for accessible and cost-effective HM screening tools.

Purpose of the Study:

• To develop an interpretable machine learning (ML) model for high myopia (HM) screening.
• To utilize routine blood parameters as a basis for the ML model.
• To establish a convenient and cost-effective alternative to traditional HM screening methods.

Main Methods:

• A cross-sectional study included 313 participants (158 HM, 155 non-HM).
• Feature selection was performed using univariate analysis and Lasso regression.
• Eight ML algorithms were trained and validated, with Extreme Gradient Boosting (XGBoost) identified as optimal. Performance was assessed using AUC, accuracy, sensitivity, specificity, and DCA. SHAP analysis was used for feature importance.

Main Results:

• Eight key blood parameters and demographic factors were identified: direct bilirubin (DBIL), total bilirubin (TBIL), albumin (ALB), alkaline phosphatase (ALP), age, glucose (GLU), creatinine (CREA), and uric acid (UA).
• The optimal XGBoost model achieved high performance with AUC values of 0.954 (training), 0.822 (validation), and 0.898 (test) in fivefold cross-validation.
• SHAP analysis confirmed the model's interpretability and clinical utility, demonstrating good calibration and decision-making utility via DCA.

Conclusions:

• A blood-based ML model is feasible for screening high myopia (HM).
• This interpretable model offers an accessible and cost-effective tool for early HM detection.
• The model shows promise for implementation in resource-limited settings.