2024 World Conference on Lung Cancer (WCLC) - Posters-P2.11A.12 Machine Learning-Based Clinicogenomic Prediction of Response to PD-(L)1 Inhibition in KRAS Altered Non-Small Cell Lung Cancer

P2.11A.12 Machine Learning-Based Clinicogenomic Prediction of Response to PD-(L)1 Inhibition in KRAS Altered Non-Small Cell Lung Cancer

Back to course

Pdf Summary

The study explores the use of machine learning models to predict the response to PD-(L)1 inhibitor therapy in patients with KRAS-altered non-small cell lung cancer (NSCLC). It aims to address the complex interplay between biomarkers that affect immune checkpoint inhibitor (ICI) sensitivity and those associated with clinical features, ultimately guiding treatment selection.<br /><br />Researchers targeted patients with untreated EGFR/ALK-negative metastatic NSCLC who received ICI therapy alone or in combination with chemotherapy across three major cancer centers. The study utilized targeted gene panel sequencing to identify relevant genetic mutations and assessed the efficacy of different treatment regimens.<br /><br />For KRAS-altered NSCLC, researchers compared the outcomes of ICI monotherapy (ICI-mono) versus combination therapy with chemotherapy (ICI-chemo), considering partial or complete response, progression-free survival (PFS), and overall survival (OS) as primary outcomes. Inverse propensity weighting was employed to adjust for various clinical and demographic variables.<br /><br />Machine learning models were developed to predict responses, employing both logistic regression and Cox models to identify biomarkers that independently predict success with ICI therapy. Models were fine-tuned using cross-validation techniques to ensure robust predictions.<br /><br />Additionally, a random forest double machine learning (RFDML) model estimated the causal effect of adding chemotherapy, recommending combination therapy when a significant probability of enhanced response was predicted. SHAP values helped identify variables causally linked to positive outcomes with ICI-chemo.<br /><br />Subgroup analyses revealed nuanced differences in effectiveness between KRAS-altered and KRAS-wildtype cases, with KRAS-altered patients showing significantly increased OS when treated, as reflected in the hazard ratios.<br /><br />Overall, this research provides a framework to potentially improve personalized treatment strategies in lung cancer by integrating clinical, genomic, and machine learning data, aiming to enhance patient outcomes through more informed therapeutic choices.

Asset Subtitle

Natalie Vokes

Meta Tag

Speaker Natalie Vokes

Topic Metastatic NSCLC – Immunotherapy

Keywords

machine learning

PD-(L)1 inhibitor

KRAS-altered NSCLC

immune checkpoint inhibitor

targeted gene panel sequencing

ICI monotherapy

combination therapy

inverse propensity weighting

random forest double machine learning

personalized treatment strategies