Identification of important symptoms and diagnostic hypothyroidism patients using machine learning algorithms

Abstract

Background: Hypothyroidism is one of the most common endocrine diseases. It is, however, usually challenging for physicians to diagnose due to non-specific symptoms. The usual procedure for diagnosis of Hypothyroidism is a blood test. In recent years, machine learning algorithms have proved to be powerful tools in medicine due to their diagnostic accuracy. In this study, we aim to predict and identify the most important symptoms of Hypothyroidism using machine learning algorithms. Method: In this cross-sectional, single-center study, 1296 individuals who visited an endocrinologist for the first time with symptoms of Hypothyroidism were studied, 676 of whom were identified as patients through thyroid-stimulating hormone (TSH) testing. The outcome was binary (with Hypothyroidism /without Hypothyroidism). In a comparative analysis, random forest (RF), decision tree (DT), and logistic regression (LR) methods were used to diagnose primary Hypothyroidism. Results: Symptoms such as tiredness, unusual cold feeling, yellow skin (jaundice), cold hands and feet, numbness of hands, loss of appetite, and weight Hypothyroidism gain were recognized as the most important symptoms in identifying Hypothyroidism. Among the studied algorithms, RF had the best performance in identifying these symptoms (accuracy = 0.83, kappa = 0.46, sensitivity = 0.88, specificity = 0.88). Conclusions: The findings suggest that machine learning methods can identify Hypothyroidism patients who show relatively simple symptoms with acceptable accuracy without the need for a blood test. Greater familiarity and utilization of such methods by physicians may, therefore, reduce the expense and stress burden of clinical testing.