2669 - Testing Adaptive Radiation Fractionation in Head and Neck Cancer Using Digital Twins of Clinical Trials

Purpose/Objective(s): Current radiation therapy (RT) practice typically uses fixed RT fractionation schedules. We hypothesize that by using a model of tumor regrowth calibrated to historical survival data, it will be possible to identify patient-specific adaptive RT fractionation schedules that minimize the number of RT fractions, while maximizing disease control.

Materials/Methods: Post-RT tumor regrowth and disease recurrence were modeled using the following differential equation model: dB/dt = ?B(B/e_v – 1), where B is the viable tumor burden, ? is the tumor regrowth rate, and e_V is a minimum tumor burden threshold for tumor viability. Locoregional control (LRC) for an individual patient was modeled using a recurrence detection threshold, ?_d, where LRC = 1 (i.e. locoregional control) when B < ?_d and LRC = 0 (i.e. locoregional failure) when B = ?_d. This model was calibrated in two steps: (1) Digital cohorts of 250 in silico patients were sampled from tumor volume dynamics and RT response parameter distributions derived from previously published fitting results for a cohort of n = 39 head and neck cancer patients. Post-RT tumor burden distributions were obtained by simulating both standard (2 Gy q.d.) and hyperfractionated (1.2 Gy b.i.d.) RT; (2) The tumor regrowth/recurrence model was calibrated using LRC data abstracted from the results of RTOG 9003. The calibrated model was used to identify the subset of in silico patients that had LRC after hyperfractionation but not after standard fractionation and to find optimal personalized fractionation schedules. The calibrated regrowth/recurrence parameter values were validated using LRC data abstracted from other trials that tested different RT schedules, doses and systemic therapies (i.e. HN002, RTOG 1016, and GORTEC 99-02).

Results: Out of 256 in silico patients, 8-12% were identified to have LRC with hyperfractionated RT but not with standard fractionation (averaged over 10 independently sampled digital cohorts). We tested various treatment schedules on this subset and found that by using upfront hyperfractionated RT followed by standard fractionation yielded the same LRC rate as a full course of hyperfractionation, with a reduced number of RT fractions. When calculating personalized RT fractionation schedules, we found that on average patients needed 3.0 +/- 1.5 weeks of upfront hyperfractionation to receive the same survival benefit as hyperfractionation alone. This resulted in estimate of a potential of 29.8 RT fractions that could be spared on average with such personalized fractionation schedules.

Conclusion: This is the first demonstration of math model of post-RT tumor regrowth/recurrence calibrated to survival data. Additionally, the results of this in sliico study suggest that it may be possible to tailor RT fractionation schedules for individual patients to reduce the number of fractions while maximizing LRC. These modeling results require further prospective validation before clinical implementation.