Objective: Autoimmune diseases are complex disorders with varied clinical manifestations, and their diagnosis often requires specialized tests not readily available in primary care. This study aimed to design and assessed machine learning (ML) algorithms capable of identifying potential cases of autoimmune diseases by using routinely available, low-cost clinical and laboratory parameters.Materials and Methods: A total of 1650 individuals from four family health centers were retrospectively evaluated. Commonly used primary care tests—including complete blood count, biochemical markers, metabolic indices, and anthropometric variables—were incorporated. Five ML approaches (Random Forest, Extreme Gradient Boosting (XGBoost), Support Vector Machine, Logistic Regression, and Deep Learning) were implemented and compared. To improve interpretability, SHapley Additive Explanations (SHAP) were employed to determine the relative importance of predictors.Results: XGBoost demonstrated the highest performance (Accuracy: 0.77, AUC: 0.83, F1 score: 0.77), whereas Random Forest and Deep Learning achieved moderately high accuracy. SHAP interpretation revealed that sex, systolic blood pressure, mean platelet volume, monocyte levels, triglyceride-glucose, waist-to-height ratio, and neutrophil-tolymphocyte ratio were the most impactful predictors.Conclusion: ML models using routinely available measures may support risk stratification for autoimmune diseases in primary care. SHAP improves transparency; however, external and prospective validation is needed before clinical use.
Hatır et al. (Tue,) studied this question.