Machine Learning-Based Prediction of Hashimoto Thyroiditis Development Risk

Abstract

Introduction: Hashimoto Thyroiditis (HT) is a prevalent autoimmune disorder impacting thyroid function. Early detection allows for timely intervention and improved patient outcomes. Traditional diagnostic methods rely on clinical presentation and antibody testing, lacking a robust risk prediction tool. Objective: To develop a high-precision machine learning (ML) model for predicting the risk of HT development. Method: Data patients were acquired from PubMed. A binary classifier was constructed through data pre-processing, feature selection, and exploration of various ML models. Hyperparameter optimization and performance evaluation metrics (AUC-ROC, AUC-PR, sensitivity, specificity, precision, F1 score) were employed. Results: Out of a total of 9,173 individuals, 400 subjects within this cohort exhibited normal thyroid function, while 436 individuals were diagnosed with HT. The mean patient age was 45 years, and 90% were female. The best performing model achieved an AUC-ROC of 0.87 and AUC-PR of 0.85, indicating high predictive accuracy. Additionally, sensitivity, specificity, precision, and F1 score reached 85%, 90%, 80%, and 83% respectively, demonstrating the model's effectiveness in identifying individuals at risk of HT development. Hyperparameter tuning was optimized using a Random Search approach. Conclusion: This study demonstrates the feasibility of utilizing ML for accurate prediction of HT risk. The high performance metrics achieved highlight the potential for this approach to become a valuable clinical tool for early identification and risk stratification of patients susceptible to HT. Keywords: Hashimoto Thyroiditis, Machine Learning, Risk Prediction, Algorithms.