2024 World Conference on Lung Cancer (WCLC) - Posters-P2.11A.27 Generalizability of Radiomics Based Progression Risk Models in Immunotherapy Treated Mnsclc Subjects

Pdf Summary

The research paper focuses on the development and testing of a radiomics-based model to predict progression risk in patients with metastatic non-small cell lung cancer (mNSCLC) undergoing immunotherapy. Radiomics, which involves extracting quantitative features from medical images, holds potential for improving prognostication in this domain. However, the challenge remains in ensuring the model's generalizability across different clinical settings.<br /><br />The methodology involved using pre-treatment CT scans of mNSCLC patients from a single institution (Discovery cohort) to build the model. Radiomics features were extracted from the largest lung tumor, and the eight most predictive features were identified using a LASSO Cox regression. The risk model was trained using a survival random forest algorithm with 5-fold cross-validation on the progression-free survival (PFS) data. To validate its predictiveness, the model was tested on a chemotherapy-treated cohort (Chemo cohort) and an independent cohort from publicly available retrospective data.<br /><br />The paper reports successful generalization of the PFS risk model, which was initially trained on an nMSCLC immunotherapy cohort. It was found to be effective not only in the external immunotherapy cohort but also in the chemotherapy cohort from the same institution. The model appears to predict generalizable prognostic features rather than those specifically related to the therapy type administered. Statistical analysis indicated hazard ratios (HR) and p-values suggestive of significant model predictions.<br /><br />This work underlines the potential of radiomics to enhance prognostic capabilities in mNSCLC, while also highlighting the challenge and importance of ensuring generalizability for clinical application. The model's ability to predict progression risk consistently across different treatment contexts adds to its potential utility in clinical practice.

Asset Subtitle

Jacob Gordon

Meta Tag

Speaker Jacob Gordon

Topic Metastatic NSCLC – Immunotherapy

Keywords

radiomics

metastatic non-small cell lung cancer

immunotherapy

prognostication

CT scans

LASSO Cox regression

survival random forest

progression-free survival

generalizability

clinical application