ESTRO 2024 - Abstract Book

S5081

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2024

models for prognostic prediction were generated using nomograms. Overall survival was evaluated using the Kaplan Meier method and survival rates were compared using the log-rank test.

Results:

For the clinical model, performance status and surgical resection extent were identified as significant prognostic factors from the clinical data. The C-index for this model was 0.547 (95%CI: 0.469-0.625). The 1 and 2-year overall survival rate for the favorable prognosis group, according to the clinical model, was 88% and 64% respectively, compared to 80% and 47% respectively for the unfavorable prognosis group, showing no significant difference. In contrast, for the radiomics model, 17 features from the radiomic data were selected as key prognostic factors in the Lasso-Cox analysis. This model demonstrated a high predictive accuracy with a C-index of 0.888 (95%CI: 0.860-0.916). According to the radiomics model, the 1 and 2-year overall survival rate was 100% and 94%, respectively, for the favorable prognosis group and 64% and 7%, respectively, for the unfavorable prognosis group, showing a significantly better prognosis (p<0.001).

Conclusion:

We successfully constructed a prognostic prediction model for high-grade glioma patients after radical radiotherapy using a radiomics analysis. The radiomics model showed markedly improved accuracy compared to the conventional clinical factors.

Keywords: High grade glioma, Radiomics, Prognostic factors

1756

Proffered Paper

Novel causal inference (natural experiment) method for estimating dose-effect from complex outcomes

Marcel van Herk, Azadeh Abravan, Alan McWilliam, Eliana Vasquez Osorio

University of Manchester, Radiotherapy Related Research, Manchester, United Kingdom

Purpose/Objective:

Estimating dose-effect relationships is challenging due to the presence of confounding variables. For instance, in lung cancer, overall survival is a complex outcome depending on the disease but also on toxicity of normal tissues, particularly the heart [1], which is under-reported. However, dose-effect relationships for the heart are not yet established in detail. The main factors driving dose to organs at risk (OARs) are tumour size and location. Through these variables OAR dose is indirectly associated to survival independent of toxicity (Fig. 1). To estimate a dose-effect relationship for toxicity we must carefully correct for such correlations.