ESTRO 2023 - Abstract Book

S1910

Digital Posters

ESTRO 2023

completion, and nadir ALC was defined after RT start. To estimate FLL, as Ellsworth et al. (2020), we fitted an exponential

curve to ALCs plotted against radiation fractions prior to 16th Fx for each patient:

(x: fraction

number, a and b: fitting parameters). FLL was calculated as . As endpoints, nadir ALC was evaluated for ≥ Grade 3 and Grade 4 lymphopenia (G3+RIL and G4RIL) (CTCAE v.5.0). Univariate logistic regression analyses (UVA) associated dose volume histogram (DVH) and non-DVH parameters with endpoints. Factors with p-value <0.2 and best Akaike information criterion (AIC) among DVH parameters for each organ at risk in UVA were included in multivariate logistic regression analyses (MVA), and forward stepwise variable selections were used based on p-value <0.05. To access effects of FLL on models, we also performed MVA without FLL. Receiver operating characteristic analyses were performed to evaluate model discriminations using area under the curve (AUC). Results Baseline ALC was 1614 [/µl] (IQR:1352 to 2066) and incidence rates were 83% (n=70) for G3+RIL and 30% (n=25) for G4RIL. By exponential curve fitting to the early lymphocyte changes, FLL was found to be a median of 7.1% (IQR: 5.1% to 8.9%). With factors selected from UVA (Table 1), MVA revealed that FLL was significantly associated with both G3+ and G4RIL (Table 2), resulting in better model quality (lower AIC) with slightly higher AUC compared to models without FLL: AUC of 0.96 and 0.92 with and without FLL for G3+RIL and 0.92 and 0.90 for G4RIL respectively. In addition to FLL, lower baseline ALC as well as higher body V10 [cc] and lungs V20 [%] was associated with increasing G3+ and G4RIL risks respectively. For G4RIL, prescribed dose difference remained significantly associated.