AI-Driven Tacrolimus Dosing in Transplant Care: Cohort Study

Area of Science:

• Transplantation Medicine
• Pharmacokinetics
• Artificial Intelligence in Healthcare

Background:

• Tacrolimus is crucial for immunosuppression in solid organ transplantation but has a narrow therapeutic range.
• Precise tacrolimus dosing is challenging post-transplant due to patient, donor, and drug-related factors.
• Maintaining optimal drug levels is essential to prevent graft rejection and toxicity.

Purpose of the Study:

• To design and validate a machine learning model for predicting next-day tacrolimus trough concentrations (C0).
• To guide tacrolimus dosing strategies and prevent persistent under- or overdosing in transplant recipients.
• To leverage artificial intelligence for personalized immunosuppressive therapy management.

Main Methods:

• Developed a long short-term memory (LSTM) model using retrospective data from 1597 kidney and liver transplant recipients.
• Included predictors such as transplant type, demographics, comorbidities, vital signs, labs, diet, and medications.
• Evaluated model performance using mean absolute error and a 3-class classification for dosing accuracy, and generated dose recommendations.

Main Results:

• The LSTM model achieved a mean absolute error of 1.880 ng/mL in predicting next-day tacrolimus C0.
• Key predictors included recent tacrolimus C0, dosage, transplant type, diet, and interacting drugs.
• The model demonstrated a microaverage F1-score of 0.653 for classifying underdosing, therapeutic dosing, and overdosing.

Conclusions:

• This study presents one of the largest AI applications for tacrolimus dosing, with an effective LSTM model for C0 prediction.
• The model shows potential for guiding accurate tacrolimus dose recommendations in transplant care.
• Prospective studies are necessary to confirm the model's real-world efficacy in dose adjustments.