Multiple Factors of Thyroid Nodule Occurrence in the Elderly-based on Logistic Regression Curves and Random Forest Models

Abstract

This study aims to ascertain the pivotal factors associated with thyroid nodules. To this end, an array of variables was meticulously examined, encompassing age, gender, lifestyle, diet, thyroid markers, and genetics, with a particular emphasis on the elderly population. Two distinct machine learning algorithms were employed for the analysis: logistic regression and a random forest model. The random forest model performs better in predicting thyroid nodule counts, with an accuracy of 0.97 and an AUC of 0.86, substantiating its efficacy in medical decision-making. The ordered multi-categorical regression model also demonstrates high accuracy in predicting nodule counts. The integration of these models is expected to improve diagnostic and therapeutic accuracy. The present study shows a high correlation between bilateral thyroid length, width, thickness, iodine concentration, and gender with the incidence of thyroid nodules. In terms of disease, hypertension and hyperlipidemia have also shown some correlation with thyroid nodules. This study offers a more reliable predictor of the probability of diagnosing thyroid nodules and provides clinicians with more personalized prevention strategies for older adults, thus helping clinicians to make effective diagnoses and treatments. Future studies should explore how to expand the sample size, train and integrate more models to further enhance the management of thyroid nodules, improve patients' quality of life, and prevent nodule enlargement and malignancy.