2101 - Parametric Response Mapping to Predict Pulmonary Toxicity from Thoracic Radiation

S. R. Miller¹, K. Suresh¹, D. Polan¹, C. Hadley¹, K. A. Vineberg², C. Lockhart², C. Amadi³, D. ODwyer⁴, W. Labaki⁴, C. Galban³, S. G. Allen¹, M. M. Matuszak¹, S. Jolly¹, and C. Matrosic¹; ¹Department of Radiation Oncology, University of Michigan, Ann Arbor, MI, ²University of Michigan, Ann Arbor, MI, ³Department of Radiology, University of Michigan, Ann Arbor, MI, ⁴Division of Pulmonary & Critical Care Medicine, Department of Internal Medicine, University of Michigan, Ann Arbor, MI

Purpose/Objective(s): Radiation to treat thoracic malignancies is efficacious but limited by pulmonary toxicity. Parametric Response Mapping (PRM) is an analytic algorithm utilizing parametric response mapping accounting for voxel-level changes between inspiratory and expiratory CT scans to characterize lung structure. We hypothesized PRM can identify patients at increased risk for pulmonary toxicity from radiation. Materials/

Methods: From 3/2022 – 5/2023 patients undergoing thoracic radiation received paired inspiratory and expiratory CT scans at time of simulation at treating physician discretion. CT PRM categorized lung tissue into Normal, Functional Low Density (FLD), Persistent Low Density (PLD), and Inspiration High Density (IHD). For each patient, dose distributions were deformably registered to the expiratory CT to calculate dosimetric parameters in equivalent dose in 2Gy fractions (EQD2). Patients were classified with total lung elevated FLD, PLD, and IHD using pre-determined cutoffs. Toxicity and survival outcomes were obtained via chart review. The primary endpoint was grade =2 radiation induced lung toxicity (RILT) using CTCAE v5.0. Multivariable logistic regression models were fit to examine the association between baseline lung classification and RILT. Bivariate association between toxicity and volume of each classification receiving various dose levels and percentage of each PRM classification within various isodose lines (IDLs) was assessed via the Wilcoxon rank sum test. Receiver operator curve (ROC) analyses were performed on each metric and optimal cutoffs for predicting RILT were selected based on Youden’s index.
Results: 98 patients were treated with 8.9 months median follow-up for toxicity with no baseline differences in stage, smoking status, lung disease diagnosis, or receipt of immunotherapy between patients with and without RILT (p=0.2). Most patients (61%) received SBRT (median, 50Gy/5Fx) with the remainder receiving conventional or hypofractionated radiation. 22 patients experienced grade =2 RILT. In multivariable analysis, adjusted for other baseline PRM categorizations, continuous increases in FLD, IHD, and PLD were not significantly associated with RILT; however elevated baseline IHD (OR 4.15, 95% CI 1.44-13.4, p=0.01) and PLD (OR 3.65, 95% CI 1.14-12.3, p=0.03) were. Optimal dosimetric (EQD2) cutoffs predicting RILT were %IHD of 5Gy IDL and 10Gy IDL >15%, 13Gy IDL >16% and 20Gy IDL >18%. In patients receiving SBRT, %IHD of 13Gy, 20Gy, 25Gy and 30Gy IDLs >15-18% were optimal and in patients treated with conventional fractionation IHD V5Gy > 42%, V13Gy > 25%, and V20Gy >18% were optimal to predict RILT.
Conclusion: PRM identified baseline elevated PLD and IHD as associated with increased risk of RILT as well as increased percentage of IHD in high dose regions, suggesting dose to dysfunctional lung may also drive adverse events.