ESTRO 2024 - Abstract Book

S5055

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2024

If the overall survival data was in a Kaplan-Meier plot, the N eff was derived from life tables reconstituted using the method outlined by Guyot et al. 4

We conducted logistic regression to determine significant predictors of OS 2yr using R (version 3.6.1, R core team, 2019). Using the dose/tumour control model, we calculated the predicted dose associated with 30% and 50% rates of OS 2yr . 95% CIs 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 19 studies that had used the following re-RT techniques: SABR (11), conventional fractionation with photons (4), Carbon-ion (2), and protons (2). The studies had data from 675 patients, which was corrected to 506 to account for studies with less that 2 years of follow-up. There were 261 deaths, with median initial and retreatment doses of 65.9 (42.3 – 133) equivalent dose in 2-Gray fractions (EQD2 Gy) and 83 (40 – 126) EQD2 Gy respectively. The median interval between treatments was 17 (8 – 36) months and the median reported tumour volume was 79.5 (24 – 237) cm 3 . On univariable logistic regression, the initial dose, retreatment dose and interval had p-values <0.1. When these variables were combined in a multivariable model, only the retreatment dose was significant (p=0.04). The model expression therefore is:

where x 1 = retreatment dose in EQD2 Gy and is plotted in Figure 1.

This model predicted OS 2yr rates of 30% and 50% at retreatment doses of 49.8 EQD2 Gy (95% bootstrap CI 36.4 - 58.0) and 76.5 EQD2 Gy (95% bootstrap CI 70.8 - 82.7) respectively. The expected OS 2yr rate correlated well with the model predictions, with a Pearson correlation coefficient of 0.78 (p-value <0.01).