Computer-assisted dosing recommendation framework, feasibility study and proof of concept implementation for tacrolimus

An overwhelming number of medicinal products are marketed with the same fixed dose for every patient. Adapting the dose for each patient should result in superior efficacy and safety, at least in theory. The gold standard of dosing individualization is model-informed precision dosing (MIPD). An extensive dataset of many patients is used to identify a population pharmacokinetic/pharmacodynamic model (popPK/PD). This is composed of a mathematical model predicting outcomes (drug concentration, biomarkers, or clinical outcome) over time, the variability of parameters for that model between individuals, and any predictive covariates for these individually variable parameters. By then using observations of an individual patient, the model parameter values for that individual patient can be estimated. These parameter values are subsequently used to accurately predict future outcomes for a candidate dose, and select the most optimal future dose: model-informed precision dosing.

This thesis aims to pave the road towards MIPD by exploring two key aspects. First, we show how to predict the effect of precision dosing in silico. Similar to how model-informed drug development has rationalized the drug development process, quantifying the effect of MIPD allows us to make well-informed choices, optimize investment into high-value opportunities for clinical improvement, develop better models, design better dosing strategies, and design better trials to show benefit. Second, we simplify building MIPD software tools through reusable software. Such a reusable software framework and accompanying scientific methodology reduces MIPD implementation time and cost. To show these goals are achieved, we apply this methodology to clinical use cases: infliximab induction therapy for ulcerative colitis patients, and tacrolimus immunosuppressive therapy for kidney transplant recipients.

We first developed the tdmore software package, integrating pharmacometric models with individual parameter estimation and future dose optimization. The mathematical routines are accompanied by debugging tools, likelihood profile visualization, and population simulation. This software package and simulation methodology was applied to investigate precision dosing of infliximab induction therapy in ulcerative colitis. In silico, a gradual reduction in outcome variability was predicted when moving from fixed dosing to covariate-based MIPD and concentration-based MIPD. Surprisingly, average mean dose per patient was predicted to increase, without an associated improvement to mean outcome. This important negative case-study predicted that precision dosing, contrary to popular belief, may not be appropriate for infliximab induction therapy, at least in the proposed implementation. This software and simulation methodology was also applied to tacrolimus dosing for kidney transplant recipients. We showed how early simulation of predictive performance can inform modeling decisions, leading us to discard covariates in favor of the base model, as well as implementing a new estimation technique to account for parameter drift. Population simulation of MIPD showed a clinically relevant improvement in probability of target attainment, speed of target attainment and -for patients not in target- distance to target window. Clinical trial simulation informed the clinical team to expand enrollment from 100 to 200 patients, reducing the probability of an expensive but ultimately inconclusive clinical study. Both use cases demonstrate the overarching goal of rational MIPD development. Based on these simulations, a precision dosing tool was developed. First, general-purpose user interface components were developed. These were combined with an automated exchange of patient data with the electronic patient record database, and extensive business rules for automated conversion of clinical data to pharmacometric data. This system is undergoing clinical trial testing at Leuven University Hospitals since April 2021.

In conclusion, this thesis has advanced the domain of MIPD. To the best of our knowledge, this is the first scientific work proposing a clear roadmap to perform informed development of MIPD: we showed how to develop a model fit-for-purpose, how to simulate whether MIPD would outperform standard of care, and which clinical trial can demonstrate this. From an engineering point of view, we greatly simplified the transformation from pharmacometric model to precision dosing tool. Finally, we showed clinical results: our work allowed future infliximab MIPD efforts to refocus their aims, and showed improved tacrolimus target attainment in silico. We developed a tacrolimus precision dosing tool in a cost-effective manner, and designed a prospective randomized clinical trial which is currently ongoing at University Hospitals Leuven.