Predictive modeling of immunotherapy efficacy in driver gene-negative non-small cell lung cancer

Immunotherapy combined with chemotherapy is highly beneficial for patients with driver gene-negative non-small cell lung cancer (NSCLC), the predominant form of lung cancer. However, the overall response in unscreened patients receiving immunotherapy is still modest, which is crucial for immune response prediction. Accurate prediction of single indicators [such as programmed death-ligand 1 (PD-L1) expression] has grown challenging, and multidimensional and multi-indicator combinations may produce more accurate results. Meanwhile, additional combination radiotherapy may increase survival benefits for individuals with poor immune responses. We aimed to predict the effectiveness of immunotherapy in patients by screening out individuals who have no durable benefit (NDB) using optimum models, and compare the survival of patients treated with and without radiotherapy based on the immunotherapy received in the first line.

The data were randomly divided into training and validation sets in a 7:3 ratio. Ten machine learning algorithms were used to build predictive models using a univariate and multivariate logistic regression approach to screen variables with an endpoint of whether or not there was durable benefit (no progression for more than 6 months). For the training and validation sets, compute area under receiver operating characteristic curve (ROC-AUC), area under precision-recall curve (PR-AUC), draft calibration curves and decision curve analysis (DCA) in order to determine the best model and display the representation.

A total of 161 patients were enrolled in the study, with 113 in the training set and 48 in the validation set, and tumor stage, tumor mutation burden (TMB) and Ki-67 were found to be independent predictors of prognosis following screening. Light gradient boosting machine (LightGBM) was chosen as the best model following a thorough comparison, with the ROC-AUC values of 0.858 and 0.852 for the training and validation sets, respectively. Fifty-two patients were selected for potential NDB based on model prediction, 27 were in the combination therapy group (radiotherapy plus immunotherapy), and 25 were in the control group (immunotherapy). The combination therapy group experienced a significantly lower incidence of NDB than the control group (29.6% vs. 64.0%, χ²=6.1706, P=0.03). The survival analysis revealed that the two groups' median progression-free survival (PFS) were 9.6 [95% confidence interval (CI): 7.882-11.318] and 5.6 (95% CI: 4.988-6.212) months, respectively, and that the patients in the combination therapy group had a better PFS than the control group (log-rank test P=0.04).

We have successfully developed a prediction model for the effectiveness of immunotherapy for NSCLC in this study. This model allows us to more precisely identify those individuals who have NDB from immunotherapy, and for these patients, additional radiotherapy combined with first-line treatment can produce better therapeutic effects.