Machine learning web application for predicting varicose veins utilizing global prevalence data

Area of Science:

• Epidemiology
• Medical Informatics
• Machine Learning

Background:

• Varicose veins represent a significant global health concern with multifactorial origins.
• Understanding the lifetime likelihood of developing varicose veins is crucial for public health initiatives.
• Existing research often lacks comprehensive predictive models integrating diverse epidemiological factors.

Purpose of the Study:

• To develop and deploy a web-based machine learning (ML) model for predicting the lifetime probability of developing varicose veins.
• To utilize global disease prevalence data and demographic/environmental factors for predictive modeling.
• To create a non-discriminatory predictive baseline for future epidemiological studies.

Main Methods:

• A systematic review provided data from 81 studies on varicose vein prevalence.
• A neural network regression model was trained using TensorFlow.js, incorporating predictors such as mean age, BMI, gender distribution, and regional gravity field.
• The model was standardized and deployed as a web-based application.

Main Results:

• The ML model achieved a test loss of 0.49 and a mean absolute error (MAE) of 0.56.
• Predictions showed up to a 6.7% difference between predicted and true disease probabilities.
• Age demonstrated the strongest correlation (0.78) with predicted varicose vein likelihood, followed by gravity anomaly (0.30), BMI (0.27), and gender (0.15).

Conclusions:

• A web-based ML model was successfully developed to predict varicose vein development risk.
• The model leverages literature-reported data, offering a valuable tool for epidemiological research.
• The predictive model provides a non-discriminatory baseline for understanding disease prevalence.