Deep learning to estimate durable clinical benefit and prognosis from patients with non-small cell lung cancer treated with PD-1/PD-L1 blockade

Abstract

Different biomarkers based on genomics variants have been used to predict the response of patients treated with PD-1/programmed death receptor 1 ligand (PD-L1) blockade. We aimed to use deep-learning algorithm to estimate clinical benefit in patients with non-small-cell lung cancer (NSCLC) before immunotherapy. Peripheral blood samples or tumor tissues of 915 patients from three independent centers were profiled by whole-exome sequencing or next-generation sequencing. Based on convolutional neural network (CNN) and three conventional machine learning (cML) methods, we used multi-panels to train the models for predicting the durable clinical benefit (DCB) and combined them to develop a nomogram model for predicting prognosis. In the three cohorts, the CNN achieved the highest area under the curve of predicting DCB among cML, PD-L1 expression, and tumor mutational burden (area under the curve [AUC] = 0.965, 95% confidence interval [CI]: 0.949–0.978, P< 0.001; AUC =0.965, 95% CI: 0.940–0.989, P< 0.001; AUC = 0.959, 95% CI: 0.942–0.976, P< 0.001, respectively). Patients with CNN-high had longer progression-free survival (PFS) and overall survival (OS) than patients with CNN-low in the three cohorts. Subgroup analysis confirmed the efficient predictive ability of CNN. Combining three cML methods (CNN, SVM, and RF) yielded a robust comprehensive nomogram for predicting PFS and OS in the three cohorts (each P< 0.001). The proposed deep-learning method based on mutational genes revealed the potential value of clinical benefit prediction in patients with NSCLC and provides novel insights for combined machine learning in PD-1/PD-L1 blockade.

Artificial Intelligence and Machine Learning in Predicting the Response to Immunotherapy in Non-small Cell Lung Carcinoma: A Systematic Review

“…The studies utilized diverse AI/ML techniques, including deep learning (DL), artificial neural networks (ANNs), support vector machines (SVMs), random forests (RFs), and gradient boosting methods (e.g., XGBoost). These algorithms were applied to various data modalities, such as medical imaging (computed tomography (CT), positron emission tomography (PET)), genomic data (TMB, gene expression), clinical variables (performance status, blood counts), and immunohistochemical markers (PD-L1, TILs) [8][9][10][11][12][13][14][15][16][17][18][19][20][21][22][23][24][25][26]. The studies employed various performance metrics to evaluate the predictive accuracy of their AI/ML models.…”

Determining the prognosis of Lung cancer from mutated genes using a deep learning survival model: a large multi-center study

“…A total of 349 patients from a clinical trial and retrospective analysis (NCT01454102, NCT01295827) who received anti-PD-1/PD-L1 monotherapy or combinatorial treatment with anti-CTLA4 were included [ 20 ]. These patients constituted another validation cohort.…”

Advances in efficacy prediction and monitoring of neoadjuvant immunotherapy for non-small cell lung cancer

“…Several types of omics data, such as RNA expression levels, immune-related gene panels, and immune-related biomarkers, from peripheral blood samples or tumor tissues of NSCLC patients treated with ICIs, could be combined with bioinformatics and ML techniques to improve the predictive performance of the model ( 103 , 104 ). ML has also been applied in some clinical trials.…”