Optimizing Levothyroxine Dose Adjustment After Thyroidectomy With a Decision Tree.

Area of Science:

• Endocrinology
• Medical Informatics
• Pharmacology

Background:

• Post-thyroidectomy patients require Levothyroxine (LT4) therapy for euthyroidism.
• Achieving optimal LT4 dosage can be a lengthy process, often taking years.
• Current LT4 dose adjustments rely on clinical estimation, lacking a standardized algorithm.

Purpose of the Study:

• To develop a decision tree model for estimating Levothyroxine (LT4) dose adjustments.
• To reduce the time required for patients to reach euthyroidism after thyroidectomy.
• To provide a data-driven tool to aid clinicians in LT4 management.

Main Methods:

• Retrospective cohort analysis of 320 patients undergoing total or completion thyroidectomy.
• Application of the Classification and Regression Tree (CART) algorithm to build predictive models.
• Utilized patient characteristics and thyroid-stimulating hormone (TSH) values for dose estimation.

Main Results:

• The most accurate decision tree model achieved a mean absolute error of 13.0 μg in LT4 dose adjustment prediction.
• The decision tree's accuracy (75%) in predicting the correct 12.5 μg dose increment surpassed a naïve system (17.2 μg MAE) and was comparable to expert providers (11.7 μg MAE).
• The model correctly predicted dose adjustments within the smallest increment 75% of the time, compared to 72% for expert providers.

Conclusions:

• A decision tree model demonstrates predictive accuracy for LT4 dose adjustments comparable to expert clinicians.
• This algorithmic approach can assist providers, particularly those less experienced with LT4 management.
• The decision tree has the potential to expedite the achievement of euthyroidism and improve patient outcomes.