PERSONALIZED RADIOTHERAPY IN HPV+ OROPHARYNGEAL CANCER
Methods: Clinical trials which are currently underway are exploring de-escalated therapy for HPV+ oropharyngeal patients. We model the choice of prescription dose using an Influence Diagram and, in place of a traditional utility node, calculate QALYs with a Markov Cohort Simulation. We calibrate the outcome model against current clinical results. We use the model to explore a range of possible outcomes of a clinical trial comparing overall survival using a lower dose (experimental arm) with a standard-of-care arm. Value of Information Analysis is used to estimate losses in terms of QALYs when a single policy is applied to a heterogeneous cohort, over a range of uncertainties from the possible trial results. Varied parameters include utilities, probabilities of overall survival and toxicities, with the latter affected by tumor size, location and patient anatomy.
Results: We explore VOI metrics for a range of possible trial outcomes. In particular, with current knowledge entered into the model the population-EVPI is 514 QALYs, which rises to 964 QALYs for the possible trial result that dose de-escalation reduces overall survival by less than 7%. The population-EVPPI for achievable doses to critical structures (heterogeneity related to patient/tumor geometry) rises from less than 10 QALYs with current knowledge, to 459 QALYs with the trial result entered into the model, reflecting the increased likelihood to change the decision based upon patient/tumor geometry when the reduction in overall survival is small.
Conclusion: A one-size fits all approach to radiotherapy prescription dose is not optimal when multiple treatment options offer competitive levels of overall survival and toxicity. We find that information from a planning CT (available before treatment begins) can be used to improve the decision. Assessing preferences for individual patients can recover additional losses, but the process is expensive and is not likely to be optimal from a cost-benefit perspective with current assessment techniques. A decision model can highlight which parameters contribute to most to decision uncertainty and need to be parametrized by a clinical trial.