Model-Informed Precision Dosing Using Machine Learning for Levothyroxine in General Practice: Development, Validation...

Area of Science:

• Pharmacology
• Artificial Intelligence
• Clinical Medicine

Background:

• Levothyroxine is a widely prescribed medication with challenging dosing due to individual variability and a narrow therapeutic window.
• Current dosing practices in primary care lack advanced decision support, leading to potential under or overdosage.
• A gap exists between AI developers and clinicians, hindering the adoption of healthcare algorithms.

Purpose of the Study:

• To develop, validate, and clinically simulate the first model-informed precision dosing (MIPD) application for levothyroxine in primary care.
• To assess the safety, feasibility, and clinical impact of MIPD compared to traditional dosing methods.
• To improve the accuracy of initial levothyroxine dosage selection by general practitioners.

Main Methods:

• A multiclass random forest model was trained and validated on a national primary care database (n=19,004) to predict optimal levothyroxine dosing classes.
• Key predictive features identified included TSH, fT4, weight, and age.
• A clinical simulation study involved 51 general practitioners prescribing levothyroxine for 20 cases, with and without MIPD support.

Main Results:

• The MIPD model achieved a weighted AUC of 0.71 for predicting dosing classes, demonstrating effectiveness even in subclinical hypothyroidism.
• MIPD significantly reduced overdosage rates (30.5% to 23.9%) and magnitude (median 50 to 37.5 μg).
• The use of MIPD increased the prescription of optimal starting dosages (18.3% to 30.2%) and GPs considered lab results more frequently.

Conclusions:

• The developed MIPD application is the first of its kind for levothyroxine in primary care, demonstrating clinical relevance and safety.
• MIPD effectively assists general practitioners in selecting safer and more optimal starting levothyroxine dosages.
• The study highlights the potential of AI-driven decision support to enhance precision medicine and improve patient outcomes in routine clinical practice.