3468 - Serial MRI-Informed Proliferation-Saturation Index Modeling for Risk-Adapted Radiotherapy in HPV-Positive Oropharyngeal Cancer
Presenter(s)
W. Floyd1, L. C. Okereke2, B. Yilmaz2, M. M. Badawy1, A. Lawless2, L. McCullum3, C. Dede4, L. Shbita5, Y. I. Mohamed6, S. Thrower7, M. El-Jammal1, H. Enderling6, and C. D. Fuller1; 1Division of Radiation Oncology, The University of Texas MD Anderson Cancer Center, Houston, TX, 2MD Anderson Cancer Center, Houston, TX, 3The University of Texas MD Anderson Cancer Center UTHealth Houston Graduate School of Biomedical Sciences, Houston, TX, 4The Department of Radiation Oncology, The University of Texas MD Anderson Cancer Center, Houston, TX, 5The University of Texas MD Anderson Cancer Center, Houston, TX, 6Department of Radiation Oncology, The University of Texas MD Anderson Cancer Center, Houston, TX, 7Department of Radiation Physics, The University of Texas MD Anderson Cancer Center, Houston, TX
Purpose/Objective(s): HPV-positive OPSCC is the most common cancer of the head and neck in the United States and has seen a rapid increase in incidence over the past two decades. Definitive chemoradiation achieves excellent control but can cause substantial acute and late adverse events. Recent phase III randomized controlled trials have demonstrated poor outcomes for blanket de-escalation in HPV+ OPSCC, highlighting the need for personalized approaches to radiotherapy in this cohort. We hypothesized that tumor volumetric change during radiotherapy can be mechanistically modeled with the Proliferation–Saturation Index (PSI) framework to predict recurrence and long-term outcomes, thereby enabling response-adaptive radiotherapy dose adjustment.
Materials/Methods: Twenty patients with HPV-positive OPSCC treated using IMRT, VMAT, or Proton radiotherapy (RT) were included. Data inputs include MRI-derived gross tumor volumes (GTV) and radiotherapy (RT) dose and fractionation schedule. The proliferation saturation index (PSI) model previously introduced in Prokopiou et. al. Rad. Onc. 2015, was used to model the dynamics of tumor growth and RT response. The model was calibrated using pre-treatment and early on-treatment tumor volumes measurements following the Bayesian framework to account for uncertainty in data. Patient specific growth and radiosensitivity parameters distributions were obtained. The model was then used to simulate and predict later on- treatment tumor volume dynamics and follow up outcomes. For each patient, the Pearson Correlation Coefficient (PCC) of model-estimated and observed tumor volumes were calculated for all timepoints to evaluate model performance.
Results: Using serial MRI-derived gross tumor volumes (GTV) from N=20 patients, The PSI model was trained using the first 2, 3, or 4 weeks of on-treatment volumes and evaluated across each patient’s remaining observed course (during RT and available post-RT imaging). Three- and four-week training best reproduced individual trajectories [median PCC= 0.92] and showed the strongest recurrence separation. Given similar performance, 3-week training was selected to maximize clinical lead time. In evaluable patients with sufficient follow-up imaging for prediction (n=17), 3-week PSI predictions yielded only 1 false-positive recurrence calls and correctly identified 80 percent of all recurrences (4/5) during median 5-year follow-up.
Conclusion: MRI-based dynamic PSI modeling can leverage early on-treatment response dynamics to forecast long-term outcomes in HPV-positive OPSCC, supporting risk-adapted RT beyond blanket deintensification. These findings provide a quantitative foundation for a phase II trial in development testing MRI/PSI-guided, dose-adaptive RT to reduce treatment intensity for favorable responders while preserving cure for higher-risk disease.