Site Editor
Gregory J. Riely, MD, PhD
Gregory J. Riely, MD, PhD
Posted: Thursday, November 13, 2025
Despite the success of immune checkpoint inhibitors (ICIs) such as pembrolizumab and nivolumab in treating advanced non–small cell lung cancer (NSCLC), only a subset of patients achieve durable benefit. Current biomarkers—including PD-L1 expression and tumor mutation burden—are limited in predictive accuracy and clinical reproducibility. Zhaojun Wang, of Nanfang Hospital, Southern Medical University in Guangdong, China, and colleagues, sought to improve the prediction of immunotherapy response in patients with NSCLC by developing the Lung Cancer Immunotherapy Response Assessment (LIRA) model, a machine learning–based integrative RNA-pathology model that leverages gene expression profiles and histopathologic features to forecast ICI outcomes more precisely. Their findings were published in a recent edition of Advanced Science.
Study Design and Methods
The study included 1127 participants with advanced NSCLC from three multicenter randomized clinical trials (OAK, n = 192; POPLAR, n = 699; and ORIENT-11, n = 171) and one in-house cohort from Nanfang Hospital (NFH, n = 65). Patients had received either ICIs (PD-1 or PD-L1 blockade), ICI combination therapy, or chemotherapy alone. The LIRA model, which was developed by integrating RNA sequencing transcriptomic data from pretreatment tumor samples, utilizes interaction analysis and a random forest algorithm to predict immunotherapy outcomes.
Validation of the model was performed across three patient cohorts—training, internal validation, and external validation. General clinical information and subtypes of driver mutation were well-matched between the immunotherapy and chemotherapy groups. The mean LIRA score within each cohort was used as a threshold to define patient subgroups. Patients above the threshold (i.e., those with a high LIRA score) were predicted to be at low risk, suggesting improved survival benefits with ICI treatment. Those with a low LIRA score were classified as high-risk patients who might not achieve the expected survival benefits.
Key Results
According to the authors, LIRA demonstrated superior predictive performance compared with established biomarkers such as PD-L1 expression and tumor mutation burden, particularly in identifying the risk of early progression during ICI monotherapy (hazard ratio = 0.15, 95% confidence interval = 0.11–0.20). Tumor profile analysis showed that LRP8 and HDAC4 are associated with immunotherapy outcomes. In addition, scRNA-seq (single-cell RNA sequencing) analysis of NSCLC tumors revealed a higher prevalence of T cells and fewer epithelial cells in samples with a high LIRA score.
The study findings demonstrate that LIRA can predict survival benefit in patients receiving PD-L1 blockade monotherapy, as well as identifying responders among those treated with PD-L1 blockade combined with chemotherapy. In addition, patients with a low LIRA score treated with ICI monotherapy were more likely to experience early disease progression than those receiving chemotherapy, which is consistent with previous findings.
Conclusions
The LIRA model was developed and validated to offer a superior method of predicting the survival benefit of ICI treatment in patients with NSCLC. “By incorporating gene-level feature engineering based on survival outcomes and treatment interactions, LIRA achieved enhanced performance in prognostic stratification of patients treated with immunotherapy,” the authors concluded.
Disclosure: For full disclosures of all study authors, visit advanced.onlinelibrary.wiley.com.
JNCCN Spotlights
