Optimal Dosage Combinations while Protecting Patient Safety
Conference
66th ISI World Statistics Congress
Format: WSC 2027: IPS Abstract
Keywords: dose-response, optimal
Abstract
Optimal treatment combinations for a nonlinear objective function are
considered while controlling for multiple adverse outcomes. The objective
response is modeled using a second-order approximation, while adverse outcomes
define the tolerable dosage region. Model parameters are estimated within a
Bayesian framework using Markov chain Monte Carlo.
\textcolor{red}{The resulting constrained optimization problems are
solved numerically, and a representative dosage combination is selected
using the 5th percentile of the Euclidean distances of the posterior
optimal solutions from the origin.}
Simulation studies consider cases in which the unconstrained maximum
lies inside or outside the tolerable region. The method identifies
dosage combinations near the true maximum when it is feasible and
adjusts the recommendation when adverse-outcome constraints are active.
\textcolor{red}{In the two-treatment simulations, the recommended
doses satisfied the true adverse-outcome constraints across the
scenarios considered, and the numerical optimality verification
rate was 100\%.}
Additional simulations examine the effects of response variability
and objective-function curvature.
\textcolor{red}{An illustrative three-treatment example with pairwise
interactions further demonstrates the applicability of the approach
to higher-dimensional dosage combinations.}
The results provide a flexible approach for optimizing treatment
combinations under nonlinear dosage-response relationships.