Machine learning algorithms indicated that race, ethnicity, lifestyle and comorbidity factors also may impact levothyroxine dosing in post-thyroidectomy patients with benign conditions, beyond the known factors of weight, sex, age, and BMI.
Key Findings
Results
The XGBoost machine learning model achieved higher accuracy than the standard weight-based dosing formula in predicting adequate levothyroxine dosage.
XGBoost model achieved 61.0% accuracy in predicting adequate dosage
Standard formula of 1.6 mcg/kg/day achieved only 47.0% accuracy
Difference was statistically significant (p < 0.05)
Study included 487 patients who underwent total or completion thyroidectomy with benign pathology and achieved euthyroid state
Results
Non-Caucasian race was associated with levothyroxine dosing requirements in Poisson regression analysis.
Non-Caucasian race was significantly associated with dosing (p < 0.05)
Study population was 39.0% White, 53.0% Black, 2.7% Hispanic, 1.4% Asian, and 3.9% Other
Race was identified as a factor beyond the known predictors of age, sex, and weight
Results
Routine alcohol use was positively associated with levothyroxine dosage requirements.
Routine alcohol use had a positive estimate of 0.03 (p = 0.02) in Poisson regression
This suggests patients with routine alcohol use required higher levothyroxine doses
Alcohol use was identified as a novel lifestyle factor influencing dosing
Results
Osteoarthritis was negatively associated with levothyroxine dosage requirements.
Osteoarthritis had an estimate of -0.10 (p < 0.001) in Poisson regression
This suggests patients with osteoarthritis required lower levothyroxine doses
Osteoarthritis was identified as a novel comorbidity factor influencing dosing
Results
Known clinical factors including age, sex, and weight were confirmed to be significantly associated with levothyroxine dosing in Poisson regression.
Age was negatively associated with dosing (estimate = -0.003, p < 0.001), indicating older patients require lower doses
Female sex was negatively associated with dosing (estimate = -0.06, p < 0.001), indicating females require lower doses
Weight was positively associated with dosing (estimate = 0.01, p < 0.001), indicating heavier patients require higher doses
Methods
The study population consisted predominantly of female and Black patients who underwent total or completion thyroidectomy for benign pathology.
Zheng H, Lai V, Lu J, Hu D, Kang J, Burman K, et al.. (2023). Efficacy of machine learning to identify clinical factors influencing levothyroxine dosage after total thyroidectomy.. American journal of surgery. https://doi.org/10.1016/j.amjsurg.2022.11.025