Development of gRADient: An Inclusive Tool for Objective Measurement of Radiation Acute Skin Toxicity across Skin Tones in Breast Cancer

Abstract

<h3>Purpose/Objective(s)</h3> Existing tools for grading radiation acute skin toxicity (RAST) primarily rely on subjective interpretation of erythema and desquamation, potentially limiting validity for patients of color (POC). Yet, more objective tools like in-lab colorimetry are not readily available for use in clinical practice. We report the development and preliminary testing of a novel digital photograph-based tool (gRADient) for RAST assessment in hypofractionated whole breast irradiation (hWBI) by: (1) validating the tool against colorimetry, and (2) testing its ability to detect skin changes over the course of hWBI across a range of skin tones. <h3>Materials/Methods</h3> Women ≥18 years undergoing unilateral hWBI to 40.5-42.6 Gy, +/- boost, were prospectively enrolled at 2 facilities (N=39). ≥2 physicians rated Fitzpatrick skin phenotype. Multiple photos of the treated and untreated (control) breast were taken weekly during hWBI (N=15 photos per time point). We mapped pixel-level RBG values to 2 corresponding color spaces: 3-item L*a*b* and Prism model (12 colors, shade, and tint on an increasing 0-1 scale). 13 patients also underwent in-lab colorimetry, collected in 4 quadrants of each breast. Patient PRO-CTCAE and Skindex-16, and provider CTCAEv.5 and RTOG tools were completed weekly. Spearman's rho assessed in-lab colorimetry vs. photo-based L*a*b*, and paired t-tests evaluated changes in Prism values between the treated and control breast at the first and last assessment. <h3>Results</h3> We evaluated 1,815 photos from 30 patients who underwent ≥3 photo sessions at the time of analysis. Patient-reported race was 43% White, 40% Black, and 17% other, corresponding to Fitzpatrick skin types of I-II (44%), III-IV (20%), and V (37%). For the colorimetry cohort, all 3 values of L*a*b* were significantly correlated for in-lab colorimetry vs. photos (all p<0.005). Across all photos, there was a significant increase in red, pink, and orange hues as well as in shading [mean ∆ of differences (dod) 0.023, 0.0004, 0.023, and 0.033, respectively, all p<0.05] for Prism in the treated vs. control breast over time. In exploratory analyses, a composite of these variables detected significant changes in time across all skin types (mean dod 0.047). While RTOG and CTCAE responses correlated with color space changes, more granular variations could be detected with gRADient. Inter-patient Skindex-16 results varied widely. <h3>Conclusion</h3> In a preliminary analysis, our novel photo-based tool yielded measurements that correlated with in-lab colorimetry, supporting its validity. Moreover, gRADient detected increases in red and pink (corresponding to erythema) as well as orange hues and shading (corresponding to increased pigmentation) across a range of skin tones. Next steps include continued prospective evaluation of gRADient, with the goal of building a web-based photo application to facilitate clinical use in RAST grading and patient education.