scholarly journals P8‐145: Successful immune checkpoint inhibition in an EGFR‐TKI‐refractory lung cancer with EGFR mutation and transformation

Respirology ◽  
2021 ◽  
Vol 26 (S3) ◽  
pp. 344-344
Keyword(s):  
Lung Cancer ◽  
Immune Checkpoint ◽  
Egfr Mutation ◽  
Checkpoint Inhibition ◽  
Immune Checkpoint Inhibition ◽  
Egfr Tki

Robust prediction of immune checkpoint inhibition therapy for non-small cell lung cancer.

2020 ◽  
Vol 38 (15_suppl) ◽  
pp. e15192-e15192
Author(s):  
Yiqun Zhang ◽  
Guodong Zhao ◽  
Ning He ◽  
Lingyu Wang ◽  
Jianfei Wang ◽  
...  
Keyword(s):  
Lung Cancer ◽  
Immune Checkpoint ◽  
Small Cell ◽  
Checkpoint Inhibition ◽  
Immune Checkpoint Inhibition ◽  
Multivariable Model ◽  
Small Cell Lung ◽  
Training Set ◽  
Mutational Status ◽  
Pan Cancer

e15192 Background: Immune checkpoint inhibition (ICI) has been validated and applied in advanced lung cancer patients. However Oonly part of patients response to ICI, thus optimizing ICI would be emergent. Based on public non-small cell lung cancer (NSCLC) cohorts’ data, we expected to build a multivariable model which would improve the efficacy of predicting ICIs response. Methods: 2 NSCLC and 1 pan-cancer cohorts treated with ICIs were collected from cBioPortal (luad_mskcc_2015, n = 203; nsclc_pd1_msk_2018, n = 33; tmb_mskcc_2018, n = 1661, pan-cancer). Another NSCLC cohort (cancer_cell_2018) was collected from MD. Hellmann’s study (Cancer Cell, 2018, n = 75). We incorporated features including tumor mutation burden (TMB), PD-L1 expression, Mutant-Allele Tumor Heterogeneity (MATH), and mutational status of cancer-related pathways. And we used Bayesian Regularization for Feed-Forward Neural Networks ( brnn) method to fit models of predicting clinical response. Results: The correlation analysis showed that MATH was independent of TMB (R = 0.141, p = 0.01252) and PD-L1 expression (p = 0.432, ANOVA test among high, median and low PD-L1 expression group) in 3 merged NSCLC cohorts. We randomly selected 70% of cancer_cell_2018 cohort as training set, and the AUC improved from 0.5702/0.5191/0.6728 to 0.6893/0.527/0.8554 (in cancer_cell_2018 validation set, luad_mskcc_2015, and nsclc_pd1_msk_2018 cohort, respectively) after incorporating MATH to TMB + PD-L1 model. Additionally, we identified that mutational status of DDR, HRR, MAPK, Wnt, JAK, PI3K, and NFKB pathway was significantly associated with overall survival in tmb_mskcc_2018 cohort. After incorporating the mutation status of these pathways, the model exhibited further improvement in prediction efficacy (AUC reached 0.8595/0.5634/0.7831). When using luad_mskcc_2015 cohort as training set, the model performance improvement was also observed after joining pathway mutation information into TMB + PD-L1 + MATH model (AUC improved from 0.5742/0.6339 to 0.7305/0.7 in predicting luad_mskcc_2015 and cancer_cell_2018 cohort). Conclusions: We built a multivariable model with robust performance in predicting ICI response in independent NSCLC cohorts. With the accumulation of larger datasets, further studies are warranted to refine the predictive performance of the approach.

Sign in / Sign up

Export Citation Format

Share Document