Researchers propose a new adaptive trial design called the PK/PD‑informed Regimen Optimization Platform (PROP) to choose drug dosing regimens more intelligently in phase II studies. PK/PD stands for pharmacokinetics and pharmacodynamics. Pharmacokinetics (PK) describes how drug concentrations change in the body over time. Pharmacodynamics (PD) describes the drug’s biological effect. The idea is to judge regimens by their predicted exposure and activity, not by dose alone.

The PROP design builds a population PK/PD model from patient data to predict both individual and average drug exposure and biological activity over time. Acute and cumulative harms (toxicities) are modeled with a discrete‑time “time‑to‑event” approach that uses those exposure predictions. Efficacy is evaluated using Bayesian model averaging, which combines an exposure‑driven model and a biomarker‑driven model and gives more weight to the model that fits the data. The platform supports decisions typical of adaptive trials: graduating promising regimens, stopping arms for futility or safety, and adding new regimens to the trial.

At a high level PROP links each regimen’s dosing history to predicted drug levels and then to outcomes. This lets the trial borrow information across regimens that produce similar exposure even if the nominal doses or schedules differ. The model‑averaging step helps the design choose whether exposure or a measured biomarker better explains patient benefit in the accumulating data. Time‑to‑event models simply mean the design looks at when toxicities or efficacy events happen, not only whether they happen.

The authors tested PROP in simulation studies motivated by intensive‑care treatment of severe influenza. Across six simulated scenarios, the PK/PD‑informed design generally made better graduation and futility decisions than dose‑based alternatives. It also reduced inappropriate graduations, helped add promising regimens that dose‑based rules missed, and produced more accurate estimates of regimen‑specific toxicity and arm‑specific efficacy. When the simulated data followed one type of efficacy mechanism, the model‑averaging framework tended to favor that mechanism.