2023 World Conference on Lung Cancer (Posters)-P1.08. CoxMoE: A Deep Learning Model Predicts Survival and Response in T790M Positive NSCLC Patients Treated with Third-Generation EGFR-TKI

P1.08. CoxMoE: A Deep Learning Model Predicts Survival and Response in T790M Positive NSCLC Patients Treated with Third-Generation EGFR-TKI - PDF(Slides)

Back to course

Pdf Summary

Researchers have developed a deep learning model called CoxMoE that can predict survival and response in patients with T790M positive non-small cell lung cancer (NSCLC) treated with third-generation EGFR-TKI. Patient selection is a crucial aspect of clinical trials, and the success of a trial can be hampered by enrollment obstacles. The most common genotype variation for EGFR-TKI resistance is the EGFR T790M mutation. However, relying solely on this mutation for prognosis prediction is not sufficient, as some patients may fail to respond or develop resistance to EGFR-TKIs.<br /><br />CoxMoE utilizes routine laboratory tests and pharmacokinetic parameters to predict the efficacy of third-generation EGFR-TKIs and complement the current genotype detection methods. The model showed good performance in predicting response rates, with an average area under the curve (AUC) of 0.83 in the training cohort. For predicting progression-free survival (PFS), CoxMoE achieved an average C-index of 0.65 in the training cohort.<br /><br />The researchers applied CoxMoE to two different clinical trial cohorts and stratified the patients into high-risk and low-risk groups based on the risk scores calculated by the model. Significant differences in PFS were observed between the high-risk and low-risk groups in both cohorts, indicating the potential of CoxMoE in predicting patient outcomes.<br /><br />The model's performance was further explained using Shapley values, which highlighted the importance of certain features, such as APTT and Ccr, in predicting response rates and PFS. CoxMoE is a gated network with expert networks that contribute to its predictive capabilities.<br /><br />In conclusion, CoxMoE is a deep learning model that can predict survival and response in T790M positive NSCLC patients treated with third-generation EGFR-TKIs. By utilizing routine laboratory tests and pharmacokinetic parameters, it offers a non-invasive way to complement current genotype detection methods and improve patient selection for clinical trials.

Asset Subtitle

Ning Lou

Meta Tag

Speaker Ning Lou

Topic Tumor Biology: Translational Biology - Drug Resistance

Keywords

CoxMoE

deep learning model

survival prediction

response prediction

T790M positive NSCLC

third-generation EGFR-TKI

patient selection

clinical trials

genotype detection

pharmacokinetic parameters