Room: Stars at Night Ballroom 2-3
In this talk we present three examples of how statistical modeling can be used to individualize and adapt Radiation Therapy (RT) treatment plans. In the first part of the talk we consider the setting of competing efficacy (e.g. tumor control) and one or more toxicity outcomes. The goal is to select an RT dose that maximizes the expected utility defined as a weighted combination of the probabilities of efficacy and toxicity. The probabilities are estimated from statistical models which include dose and patient covariates. We discuss two approaches for selecting the tradeoff parameters. The methods are implemented in an R Shiny App and illustrated using an ongoing Phase II trial based on these methods. In the second part of the talk we consider settings with a single survival type outcome in which the goal is to estimate an optimal dosing rule which will maximize expected survival times. Several modeling approaches are compared including simple parametric models as well as tree based ensemble methods. In the last part of the talk we describe ongoing work to estimate optimal Dynamic Treatment Regimes to allow two or more stages of RT in which some dose is given followed by assessment of some intermediate outcome (e.g. liver function) followed by additional RT. An optimal DTR would give the optimal dose of RT for the first stage using baseline covariates and then the optimal dose for the second stage using baseline and mid-treatment covariates.