Multiple machine learning models for the prediction of an early response to radioactive iodine therapy in hyperthyroidism: ablative dose concept

Abstract

BACKGROUND: The aim of this novel study was to explore the possibility of the development of a machine learning (ML) model that can predict early response to radioactive iodine (RAI) therapy in patients with hyperthyroidism on the basis of demographic, clinical, laboratory, and imaging features. METHODS: A total of 914 patients who received RAI for the treatment of hyperthyroidism between January 2000 and December 2023 were included in the study. The early positive response, defined as an accomplishment of hypothyroidism (thyroid-stimulating hormone (TSH) > 4 mIU/L) in the 6-month period following RAI therapy, was accomplished in 336 patients (36.76%) (hypothyroidism group), whereas hypothyroidism was not accomplished in 578 patients (63.24%) (control group). The training data consisted of 90% of the sample (822 patients), whereas the test set contained 92 patients. Multiple ML models, including k-nearest neighbors, random forest, various gradient boosting algorithms, and neural networks, were trained via AutoGluon. RESULTS: The neural network model performed the best, with a balanced accuracy of 68.81%, sensitivity of 61.76%, specificity of 75.86%, PPV of 60.0%, NPV of 77.19%, and F1 score of 60.87%, with the area under the curve (AUC) of 0.706 (95% confidence interval [CI]: 0.593-0.809; receiver operating characteristic [ROC] curve analysis). In the development of the top-performing model, baseline TSH, age, 24-h radioactive iodine uptake, dose, and duration of previous treatment with antithyroid drugs (ATDs) were the most important features. CONCLUSIONS: We developed models for the prediction of the response to RAI in patients with hyperthyroidism. However, these models require further development before they can be applied in clinical practice. By introducing more variables and increasing the sample size, the model performance can increase further.

Ähnliche Arbeiten