Machine learning model for tacrolimus dosing: Can we validate and compare to existing options?

To the editor: The study published by Choshi et al. is directed at a difficult clinical challenge: achieving targeted tacrolimus concentrations during the early post-lung transplant time period. Transplant clinicians are concerned about the tension between acute rejection and nephrotoxic risks related to post-operative tacrolimus concentrations. Yet, achieving goal levels during this vulnerable period is frequently elusive. The prediction model developed by Choshi et al., using prior tacrolimus levels and doses in a machine-learning algorithm to predict subsequent levels, is therefore of clear clinical interest. To best vet this model for clinical use, we have several questions regarding model specification, how others may externally validate their work, and comparison to existing approaches.

National Institutes of Health

Michael G S Shashaty, Gary E Weissman, Marc H Scheetz, Todd A Miano