Learning pharmacometric model structures from data

Pharmacometric models are mathematical models aiming to describe the relationship between pharmaceutical therapy and patient response. A central aspect of pharmacometric models is prediction of individual responses to therapy based on covariates (i.e. patient characteristics). Such individual predictions constitute the foundation of precision medicine with the ultimate goal of optimal therapy in each patient. The covariates have historically been limited to patient demographics such as weight, height and sex and the covariate modelling has focused on which covariates that are relevant for prediction of response. The appropriate mathematical structure of the covariate model has received less attention.
Expanding knowledge on an individual patient level due to collection of additional data such as genetic and life-style data offers opportunity to improve individual predictions by pharmacometric models. This project aims to develop methodology capable of handling such increasingly complex data. The project further has a focus on developing methods for identification of mathematical structures of covariate-response relationships based on machine learning.