Prediction and associated factors of fatigue in hypothyroidism using explainable machine learning models

Problem considered Fatigue is a prevalent yet under-recognized symptom among individuals with hypothyroidism, influenced by multiple interacting biological and behavioral factors. Traditional statistical methods may overlook nonlinear and complex relationships between predictors and fatigue severity. Identifying key associated factors through advanced analytical approaches could support more comprehensive management strategies. Methods A cross-sectional study was conducted among 315 adults with hypothyroidism attending an endocrine outpatient clinic between April and August 2025. Fatigue was measured using the Fatigue Severity Scale (FSS; mean score ≥4). Potential predictors included demographic, clinical, and lifestyle characteristics. After data preprocessing (scaling, encoding, imputation), three models, logistic regression, support vector machine (SVM), and random forest (RF), were trained using stratified training–testing splits. Model performance was evaluated using AUC, accuracy, precision, recall, F1-score, Brier score, and calibration metrics. Feature importance and SHAP values were applied for model interpretability. Results Fatigue prevalence was 47.9%. The RF model showed the best discrimination (AUC 0.88, 95% CI 0.80–0.95; accuracy 0.81, 95% CI 0.71–0.90) with acceptable calibration. Major predictors included physical activity, age, electronic device use, BMI, diet, smoking status, and education, while TSH and hemodynamic measures contributed minimally. Conclusion Explainable machine learning effectively identified key behavioral and clinical factors associated with fatigue in hypothyroidism. Findings highlight the dominant role of modifiable lifestyle factors, suggesting that management should extend beyond thyroid hormone replacement. External validation is recommended before clinical integration.