Sequential Updating of a New Dynamic Pharmacokinetic Model for Caffeine in Premature Neonates

Bayesian sequential integration within a preclinical pharmacokinetic and pharmacodynamic modeling framework: Lessons learned

“…Compared with the stochastic particle algorithm described by Micallef et al, our method allows to incorporate multiple observations at each integration step: when using particle filters, a very limited amount of data is typically added at each step; if a higher amount of data is included at once, the phenomenon called “particle depletion" can occur, ie, one or few particles have a much higher probability, whereas all other particles are not retained after the filtering step . Moreover, our method allows to implement prior downweighting in potential scenarios of incompatibility between previous and current trial, in terms of design or trial conduct .…”

“…Although the integration of incoming information recursively over time is well established in the field of computer science under the name of recursive Bayesian estimation, the technique has—to our knowledge—hardly ever been applied in the framework of a complex PK‐PD modeling approach . Micallef et al proposed a method to sequentially update the parameters of a PK model for caffeine in premature neonates using a stochastic particle algorithm. In our work, the Bayesian integration is performed by setting the hyperparameters of the prior distributions of a trial based on the posterior distributions resulting from the previous trial.…”