Predictive Model Forecasts Advanced NSCLC Outcomes

A prediction model incorporating genomic and pathological data can be used to improve prognostic accuracy for patients with advanced non-small cell lung cancer (NSCLC), according to a new study.

Writing in NPJ Precision Oncology, a team of investigators developed a model that incorporated next-generation sequencing and pathological information in order to determine patients’ likelihood of progression-free survival (PFS) and overall survival (OS).¹ The model can be used to develop personalized treatment plans for patients, the authors said.

The investigators found that tumor mutational burden and PD-L1 expression, which have traditionally been used as indicators of immunotherapy efficacy, had little predictive value for survival in this cohort. | Image credit: Dragana Gordic - stock.adobe.com

Improvement in NSCLC diagnosis is important because most patients with the disease are not diagnosed until the cancer is in an advanced stage, the authors explained. They noted that National Cancer Institute data show the 5-year OS for patients with NSCLC is only approximately 10%.²

One reason for optimism, the authors said, was the development of newer therapeutic options, such as immunotherapy combined with chemotherapy.¹ Still, the factors influencing the efficacy of the therapy remain opaque. That has led the investigators, and others, to turn toward tumor sequencing as a means to better understand biomarkers and signatures associated with treatment response. Investigators have also looked to artificial intelligence (AI) to explore the tumor microenvironment (TME), since certain conditions within the TME have been linked with cancer occurrence, progression, and response to treatment.

The authors recruited 162 patients with advanced NSCLC undergoing first-line immunotherapy plus chemotherapy. Each of the patients in the cohort lacked EGFR/ALK driver alterations.

The patients underwent next-generation sequencing, and then clinical data like tumor site, tumor node metastasis, and histological type of treatment were collected. Most of the participants underwent assessment of PD-L1 expression using immunohistochemistry. All of the participants underwent tumor proportion scoring, which gave investigators a sense of the proportion of tumors expressing PD-L1.

The investigators found that tumor mutational burden and PD-L1 expression, which have traditionally been used as indicators of immunotherapy efficacy, had little predictive value for survival in this cohort. However, they found that by integrating genomic and cellular characteristics from stained pathological images into machine learning and deep learning models, they were able to establish a more reliable prognostic model. The model, which was called the Prognostic Multimodal Classifier for Progression (PMCP), provided a significant benefit in forecasting PFS and OS, they said.

Patients who were put into a subgroup called PMCP1 had a low risk of progression and a high proportion of epithelial cells, they said. Patients in PMCP2 had a high risk of progression and a high proportion of epithelial cells or a low risk of progression and a low proportion of epithelial cells. Patients in the PMCP3 subgroup had a high risk of progression and a low proportion of epithelial cells. The model’s area under the curve (AUC) for PFS was 0.807, the investigators said.

The authors said in clinical practice, patients in the PMCP1 cohort may benefit most from immunotherapy plus chemotherapy. “This is because of the favorable factors of simultaneously exhibiting both a low risk of genomic mutations and pathological images,” they wrote.

On the other hand, people in the PMCP3 cohort might not benefit significantly from the therapy.

“This is because their tumor immunogenicity is low, and chemotherapy may not effectively enhance the effects of immunotherapy,” they wrote. The investigators said their findings offer promising new biomarkers for patients with advanced NSCLC, but they acknowledged that a lack of external validation limits the generalizability of their results for now.

References

1. Han Y, Ma J, Liu Z, et al. Integrating genomic and pathological characteristics to enhance prognostic precision in advanced NSCLC. NPJ Precis Oncol. 2025;9(1):271. doi:10.1038/s41698-025-01056-8

2. National Cancer Institute. SEER*Explorer Application. National Cancer Institute website. Accessed August 6, 2025. https://seer.cancer.gov/statistics-network/explorer/application.html?site=1&data_type=1&graph_type=2&compareBy=sex&chk_sex_3=3&chk_sex_2=2&rate_type=2&race=1&age_range=1&hdn_stage=101&advopt_precision=1&advopt_show_ci=on&hdn_view=0&advopt_display=2#resultsRegion0.