2438 - Common Analysis Approaches Can Indicate Apparent Association of Increasing Proton-LET With Radiation Associated Image Changes Where No Such Relationship Exists

Purpose/Objective(s): Contradictory evidence exists in the literature for whether a relationship between linear energy transfer (LET) and radiation-associated image changes (RAIC) can be observed following proton therapy. As LET optimization in proton therapy is increasingly considered, reconciling sources of disagreement between clinical studies that link LET to tissue toxicity is important to make an informed decision about whether and to what extent the dose distribution should be perturbed in order to further optimize LET. The aim of this study is to understand whether the analysis technique used influences if clinical evidence of variable proton relative biological effectiveness (RBE) is observed in retrospective studies of RAIC occurrence. This study evaluates the impact of statistical approaches for extracting clinical relationships between LET and RAIC by applying four distinct analysis approaches to a single, historically significant patient cohort.

Materials/Methods: A cohort of 34 pediatric ependymoma patients treated with passively scattered proton therapy which had been previously investigated and was the first dataset on which it was claimed that quantitative clinical evidence of spatially varying proton-RBE could be observed is reanalyzed in this study. Techniques previously described in the literature including: grouping voxels from all patients together and extracting necrotic proportions, probit regression, a dose-matching method, and mixed effects probit regression are used to extract clinical relationships between dose, LET, ventricular proximity, and RAIC.

Results: A naïve voxel grouping approach indicates that RBEs approaching 2 occur for track-averaged linear energy transfer (LET_t) in the range 4-5 keV/µm compared to 2-3 keV/µm protons. A dose-matching approach indicates only 5 of 14 patients who developed RAIC demonstrate evidence of enhanced LET_t in their RAICs. The LET_t coefficient of a mixed effect probit regression model for RAIC risk is -0.30 (p = 0.4. 95% CI: -0.96 to 0.36).

Conclusion: The analysis method used to extract relationships between LET (i.e., variable RBE) and RAIC has a profound effect on whether a correlation is observed. When drawing many thousands of voxels from a single patient, care must be taken to avoid pseudoreplication, a common data mishandling practice in which highly correlated measurements are erroneously treated as statistically independent. A statistically significant relationship between LET_t and increasing RAIC does not exist in the cohort studied.