AACR Release Highlights Pathomics Model For NSCLC Immunotherapy Response

Key Takeaways

• Path-IO applied to routine pathology slides in metastatic NSCLC was reported to separate patients into higher- and lower-risk groups with different outcomes.
• Investigators observed stronger discrimination than PD-L1 in both discovery and test cohorts across the reported survival endpoints.
• Adding radiomics and clinical data further improved discrimination, while the analysis remained retrospective and prospective validation is still needed.

A biology-guided AI/pathomics framework called Pathology-driven Immunotherapy Optimization, or Path-IO, used routine pathology slides in metastatic NSCLC to estimate immunotherapy risk. In a UT MD Anderson cohort with added external validation, it separated higher- and lower-risk groups with significantly different outcomes. High-risk patients in that cohort had more than double the risk of death or disease progression versus low-risk patients, and investigators said the model outperformed PD-L1. An April 20, 2026 AACR news release on AACR Annual Meeting 2026 findings described the work as a biomarker-comparison and risk-stratification report that may help inform treatment decisions, not a treatment recommendation.

Because only a subset of patients benefit from immunotherapy, the analysis aimed to derive prognostic information from routine pathology material available in metastatic NSCLC. Path-IO was evaluated in 797 immune checkpoint inhibitor-treated NSCLC patients from UT MD Anderson. External validation included 280 patients from Mayo Clinic, Gustave Roussy, and the phase III Lung-MAP S1400I trial. That Lung-MAP group comprised immunotherapy-naive patients with lung squamous cell carcinoma who received immune checkpoint inhibitors, extending the assessment beyond the originating center. Investigators said the framework focused on tissue niches within the tumor microenvironment and grounded predictions in structures familiar to clinicians, fitting routine slide-based workflows.

Using C-index as the discrimination metric, Path-IO outperformed PD-L1 across discovery and test cohorts for both overall survival and progression-free survival. PD-L1 yielded C-indices of 0.58 for OS and 0.57 for PFS in discovery, then 0.50 for OS and 0.51 for PFS in test. Path-IO reached 0.69 for OS and 0.65 for PFS in discovery, followed by 0.63 for OS and 0.58 for PFS in test. These findings showed a discrimination advantage over PD-L1 in both development and independent testing sets.

When pathology-based predictions were combined with radiomics and clinical data, the C-index rose from 0.58 to 0.70 for PFS and from 0.63 to 0.75 for OS. Separately, the model's predictions correlated with immune profiling and multiplex imaging data. These associations were presented as biologic correlates rather than proof of mechanism. Investigators also highlighted an ongoing need to predict not only who may benefit from immunotherapy, but which immunotherapy type may fit best. Together, these layered data were associated with better discrimination than pathology alone.

The analysis was retrospective, limiting conclusions beyond the reported conference findings. Next steps included prospective validation and paired, more comprehensive molecular profiling. Investigators added that workflow integration could be feasible without major expense only if the tool is validated as predictive. Broader validation is still needed before the approach can be assessed more broadly.