Integrated histopathology-transcriptomic biomarker enhances survival prediction in HNSCC patients treated with...

Recurrent and metastatic head and neck squamous cell carcinoma (R/M HNSCC) remains a challenging disease with modest response to immune checkpoint inhibitors and a need for more robust predictive biomarkers. In a real-world (RW) cohort of patients treated with pembrolizumab alone or in combination with chemotherapy, we evaluated transcriptomic and histopathologic features associated with therapeutic benefit. PD-L1 expression, measured by combined positive score (CPS), was not significantly associated with progression-free survival (PFS). In contrast, immune-related gene signatures, particularly those linked to T cells and tertiary lymphoid structures (TLS), were predictive of improved outcomes. TLS presence identified from Hematoxylin and Eosin-stained (H&E) whole slide images correlated with favorable survival and showed association with RNA-derived TLS signatures. TLS-associated features demonstrated treatment-specific prognostic patterns, with stronger predictive power in pembrolizumab monotherapy versus combination therapy. We developed multimodal risk prediction models integrating molecular features with imaging data which better associated with RW outcomes. Evaluation using concordance index analysis revealed that traditional pathological markers and individual molecular signatures had modest predictive capability. Digital pathology features achieved better performance than clinical or molecular features alone, but the combination of imaging and molecular features yielded the highest predictive accuracy with concordance index values of 0.86 and 0.81 in pembrolizumab and combination therapy cohorts, respectively. Kaplan-Meier analysis confirmed that our multimodal risk signature achieved significant separation between high- and low-risk groups in both treatment arms, substantially outperforming molecular features alone. These findings highlight that integrating transcriptomic and histopathological data enables precise patient stratification for immunotherapy in R/M HNSCC.