DoseTailor for personalised tacrolimus dose optimisation across multicentre populations using interpretable AI

Abstract Tacrolimus, an immunosuppressant commonly used following organ transplantation, has a narrow therapeutic index (NTI) and a high risk of suboptimal drug concentrations, with over 50% of patients experiencing serious, potentially life-threatening adverse effects. To improve dosing precision, we developed an interpretable long short-term memory (LSTM) model trained on 14 years of data from 1774 patients across Spain and the United States. The LSTM model predicts tacrolimus blood concentrations using clinical data, achieving a mean absolute error below 2 ng/mL and outperforming current machine-learning and deep-learning baseline models. To translate this work into clinical practice, we launched dosetailor.com, a platform designed to support clinician-led tacrolimus dose optimisation for therapeutic drug monitoring using routinely collected patient information. Although currently focused on tacrolimus, the platform provides a foundation for clinician-in-the-loop dosing support across other high-risk, NTI drugs using healthcare data.