Endocrine Abstracts

Background: Hashimoto’s thyroiditis (HT) is the most common autoimmune disorder and, also, the leading cause of hypothyroidism in iodine-sufficient areas. In recent years, a concept emerged, that thyroid autoimmunity could be associated with low-grade chronic inflammation, which may result in future cardiovascular comorbidities, independent of thyroid function. It is therefore essential to diagnose Hashimoto’s thyroiditis as early as possible and to test for thyroid function.

Methods: We recruited 129 participants, including 104 diagnosed HT patients and 25 non-HT controls. Secondly, we collected 12 factors and analyzed their significant differences between controls and HT patients; the clinical factors analyzed were age, family history of autoimmune thyroid disease, personal history of: breast cancer, surgically induced menopause, diabetes mellitus type 2, polycystic ovary syndrome. The paraclinical parameters evaluated were: anemia, hemoglobin, hematocrite values, hypertriglyceridemia, hypercholesterolemia, hyperuricemia, fasting hyperglycemia, abnormal liver function tests. We evaluated the following models: Decision Tree, K-Nearest Neighbor, Extreme Gradient Boost, and Support Vector Machine for classification and regression, as well as neural networks: Artificial Neural Network and Deep Neural Network.

Results: The best model for binary classification was K-Nearest Neighbor, with an accuracy of 85%, sensitivity of 78% and specificity of almost 100%. Concerning regression analysis, we obtained a Pearson coefficient of 97% and an R-squared value of 94% for the Deep Neural Network. Statistical indicators, designed for the regression part confirmed a family history of autoimmune disease, personal history of breast cancer, surgically induced menopause, anemia, hypertriglyceridemia, hyperuricemia, fasting hyperglycemia, and increased alanine aminotransferase values as significant risk factors for Hashimoto’s thyroiditis.

Conclusions: The proposed models of machine learning, combined with multiple factors, are efficient for HT diagnosis. These findings suggest screening for autoimmune thyroid disease in metabolic syndrome patients, in breast cancer patients and in women with surgically induced menopause.