Researchers have developed a machine learning-based approach that significantly improves outcome prediction for patients with advanced renal cell carcinoma (RCC) by integrating multiple biomarkers, according to findings presented at the 2025 ASCO Annual Meeting. The study represents a shift from traditional single-biomarker analyses toward comprehensive predictive modeling.
David A. Braun, MD, PhD, from the Center of Molecular and Cellular Oncology at Yale Cancer Center, led the integrative post hoc analysis of the phase 3 CheckMate 9ER trial, which evaluated nivolumab plus cabozantinib versus sunitinib in 651 patients with advanced RCC.
"While most biomarker studies have typically focused on 1 biomarker at a time, integrating both tumor and circling factors is key," Braun explained. He noted that individual features like genomic alterations, tumor microenvironment characteristics, and circulating immune factors "provide really important biological insights, they are only modestly associated with differences in response and resistance."
Novel Biomarker Integration Approach
The research team analyzed biospecimens from 150 patients in the CheckMate 9ER trial, examining both tumor intrinsic and peripheral blood biomarkers. Their comprehensive approach included conventional assessments such as PD-L1 immunohistochemistry alongside novel AI-powered analysis of hematoxylin and eosin staining for single-cell resolution mapping of the tumor microenvironment.
A key innovation was the assessment of circulating extracellular matrix (ECM) proteins using competitive ELISA. High ECM levels were associated with improved progression-free survival in both the nivolumab-plus-cabozantinib arm (HR, 0.57; 95% CI, 0.39-0.83; P = .00358) and the sunitinib arm (HR, 0.52; 95% CI, 0.36-0.75; P = .000438).
"These ECM biomarkers were increased in patients with kidney cancer compared to healthy volunteers," Braun said. "In patients with higher-risk disease, whether that was IMDC poor risk disease or sarcomatoid histology, we saw that these ECM biomarkers were increased when we associated it with clinical outcomes."
Machine Learning Model Performance
The researchers organized their dataset of more than 4,000 measurements across 150 patients, identifying the top 16 most variable features that accounted for 85% of variability. These included PD-L1 staining, 4 ECM markers, 4 peripheral blood mononuclear cell markers, and 7 H&E human interpretable features.
Univariate machine learning analysis identified 6 features predictive of nivolumab plus cabozantinib response: PD-L1 level, VICM, Pro-C6, proportion of plasma cells in ESI, proportion of cancer cells in ESI, and FOXP3-positive CD4-positive T cells. However, Braun noted that "the predictive ability is incredibly modest."
The multivariate machine learning approach yielded dramatically different results, identifying 10 predictive features including PD-L1 level, VICM, proportion of plasma cells in ESI, proportion of cancer cells in ESI, LAG3-positive CD8-positive T cells, density ratio of macrophages to fibroblasts in tumor, PD-1-positive CD8-positive T cells, TUM, FOXP3-positive CD4-positive T cells, and PRO-C6.
"We see now substantially improved predictive capability," Braun said, emphasizing that the model incorporated "all 4 buckets of our data [ECM, H&E HIF, PD-L1, and PBMC]."
Clinical Trial Context and Outcomes
The original CheckMate 9ER trial demonstrated significant efficacy advantages for the combination therapy. Median progression-free survival was 16.4 months (95% CI, 12.5-19.3) with nivolumab plus cabozantinib compared to 8.3 months (95% CI, 7.0-9.7) with sunitinib (HR, 0.58; 95% CI, 0.49-0.70). Overall survival was also superior at 46.5 months (95% CI, 40.6-53.8) versus 35.5 months (95% CI, 29.2-42.8), reducing the risk of death by 21% (HR, 0.79; 95% CI, 0.65-0.96).
Future Applications and Limitations
The machine learning model is currently trained on short-term outcomes data to elucidate mechanisms of response and resistance. Braun indicated that future development will incorporate the model into clinical trial forecasting algorithms using tumor growth kinetics, population-level survival information, and model-based response probabilities.
"Using this forecasting approach, just based on baseline values, we see a fairly good forecast using this model approach for both PFS and for overall survival," Braun added.
The researchers acknowledged several limitations, including analysis of a discovery dataset only with limited features. They plan to integrate additional features such as T-cell phenotype, genetics, and molecular subtype, and evaluate the model's prognostic and predictive potential in additional clinical trials.
Implications for Kidney Cancer Treatment
Braun emphasized the unique challenges in RCC biomarker development: "We know that kidney cancer is immunobiologically distinct from other solid tumors, so we cannot just extrapolate biomarkers from other tumors like [tumor mutation burden] or PD-L1. Despite many biomarker studies, I would say that the determinants of response and resistance to new checkpoint inhibitors and to TKIs still remains largely unknown within kidney cancer."
The findings suggest that combining tumor and circulating biomarkers provides superior predictability compared to individual biomarkers alone, indicating that "the state of the tumor microenvironment and the circulating factors together influence outcome, and ultimately this provides a good foundation for future integrative biomarker discovery."
