Efficacy of machine learning to identify clinical factors influencing levothyroxine dosage after total thyroidectomy

Abstract

BackgroundWe employed Machine Learning (ML) to evaluate potential additional clinical factors influencing replacement dosage requirements of levothyroxine. MethodThis was a retrospective study of patients who underwent total or completion thyroidectomy with benign pathology. Patients who achieved an euthyroid state were included in three different ML models. ResultsOf the 487 patients included, mean age was 54.1 ± 14.1 years, 86.0% were females, 39.0% were White, 53.0% Black, 2.7% Hispanic, 1.4% Asian, and 3.9% Other. The Extreme Gradient Boosting (XGBoost) model achieved the highest accuracy at 61.0% in predicting adequate dosage compared to 47.0% based on 1.6 mcg/kg/day (p < 0.05). The Poisson regression indicated non-Caucasian race (p < 0.05), routine alcohol use (estimate = 0.03, p = 0.02), and osteoarthritis (estimate = −0.10, p < 0.001) in addition to known factors such as age (estimate = −0.003, p < 0.001), sex (female, estimate = −0.06, p < 0.001), and weight (estimate = 0.01, p < 0.001) were associated with the dosing of levothyroxine. ConclusionsAlong with weight, sex, age, and BMI, ML algorithms indicated that race, ethnicity, lifestyle and comorbidity factors also may impact levothyroxine dosing in post-thyroidectomy patients with benign conditions.