Abstract
Aim
This study aimed to develop a web-based machine learning (ML) model to predict the lifetime likelihood of developing varicose veins using global disease prevalence data.
Methods
We utilized data from a systematic review, registered under PROSPERO (CRD42021279513), which included 81 studies on varicose vein prevalence across various geographic regions. The data used to build the ML model included disease prevalence as the outcome (%), along with the following predictors: mean age, gender distribution (%), mean body mass index (BMI) of the study cohort, and the mean gravity field of the study region (mGal), representing variations in Earth’s underground mass distribution that influence blood and fluid redistribution in the human body, affecting disease prevalence. After standardizing the outcome and predictors, the model was trained using neural network regression implemented with the TensorFlow.js library and deployed as a web-based ML application.
Results
After 406 epochs of training, and upon achieving a validation loss (mean squared error) of 0.9, training was stopped due to no further improvement. The achieved test loss was 0.49, and the mean absolute error (MAE) was 0.56, corresponding to an up to 6.7% difference between the predicted and true disease probabilities (calculated as MAE x σ, where σ is the standard deviation of the mean disease prevalence = 0.56 x 11.9 = 6.7). The likelihood of developing varicose veins, as predicted by the model, showed the strongest correlation with age (0.78), followed by gravity anomaly (0.30), BMI (0.27), and gender (0.15).
Conclusion
This study summarizes research on the prevalence of varicose veins by developing a web-based ML model to predict an individual’s likelihood of developing the disease. Using data reported in the literature, the ML algorithm provides a non-discriminatory predictive baseline, offering a valuable tool for future investigations into disease epidemiology.
Get full access to this article
View all access options for this article.