ESTRO 2024 - Abstract Book

S5048

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2024

Purpose/Objective:

Re-irradiation (re-RT) for non-small cell lung cancer is occurring more frequently 1 . It is associated with significant toxicity, with a 7% median rate of ≥grade 3 pneumonitis 2 . Dose constraints can guide the safe delivery of re-irradiation and reduce the risk of toxicity. Current re-irradiation dose constraints have been formed by expert consensus, based on limited data and clinical experience. These suggest a maximum cumulative mean lung dose (cMLD) of <22Gy and a maximum cumulative volume of lung receiving at least 20Gy (cV20Gy) of 30-40% 3 . We developed models of the risk of lung toxicity to allow better counselling of patients regarding the potential risks of re-RT and to refine the current dose constraints.

Material/Methods:

We performed a MEDLINE literature search to identify studies published between 01/01/1970 and 01/10/2018 that quoted cumulative doses to the lung (cV20 and/or cMLD) and associated grade 3 or above toxicity rates. Other variables collected were the interval between treatments and the use of concurrent chemotherapy. We performed logistic regression to identify significant predictors of toxicity using R (version 3.6.1, R core team, 2019). We calculated the cV20 and cMLD associated with a pre-specified toxicity rate of 20%. 95% confidence intervals (CI) were calculated using 2000 block bootstrapped samples. Assessment of model fit was performed by dividing the collected dataset into deciles and plotting the modelled toxicity rate against the observed rate. Model performance was assessed using a Pearson correlation coefficient.

Results:

We identified 15 studies that provided both cumulative dose and toxicity data, including 476 patients and 37 ≥grade 3 lung events. Median cMLD and cV20Gy were 14.5Gy (5.5 – 23.2) and 17.6% (3.8 – 38.4) respectively. On univariable logistic regression, cMLD, cV20Gy and concurrent chemotherapy were individually significant (all p-values <0.05). Multivariable modelling insignificantly improved toxicity prediction, and therefore dose-response analyses used the univariable models (see Figures 1 and 2 and Table 1). The maximum likelihood prediction of 20% toxicity rate occurred at a cMLD 19.3Gy (95% CI 17.6, 21.0Gy) and cV20Gy 28.4% (95% CI 25.6, 47.8%). The Pearson correlation coefficients for the cMLD and cV20Gy models were 0.81 (p<0.01) and 0.71 (p=0.02) respectively.

Model equation

cV20Gy

cMLD

Table 1. Equations describing the fitted cV20Gy and cMLD univariable models. x 1 = cV20Gy and x 2 = cMLD.