Blood Test May Predict Immunotherapy Response in Head and Neck Cancer

Why It Matters

Accurate early prediction of immunotherapy benefit could reduce unnecessary toxicity and health‑care costs while improving survival outcomes for head and neck cancer patients.

Key Takeaways

• cfDNA fragmentation patterns predict pembrolizumab response in HNSCC
• Test correctly identified responders in 68‑patient phase‑II trial
• Machine‑learning model outperformed current biomarkers for immunotherapy selection
• Validation in larger trials needed before clinical adoption

Pulse Analysis

Head and neck squamous cell carcinoma (HNSCC) accounts for hundreds of thousands of new cases worldwide each year, and despite the rise of immune checkpoint inhibitors such as pembrolizumab, only about 20 % of patients experience durable tumor regression. The lack of reliable predictive tools forces clinicians to administer costly infusions that can trigger severe immune‑related adverse events, while many patients endure months of therapy without benefit. Consequently, oncologists and payers have been searching for a minimally invasive biomarker that can separate likely responders from non‑responders early in the treatment course.

The Northwestern Medicine team turned to cell‑free DNA (cfDNA), short fragments shed by dying cells into the bloodstream, and examined genome‑wide fragmentation patterns rather than single‑gene mutations. In a multi‑institutional phase‑II trial, 185 serial blood draws from 68 HNSCC patients receiving pembrolizumab were analyzed. A proprietary scoring algorithm captured entropy differences between tumor‑derived and immune‑cell‑derived cfDNA, and when fed into a machine‑learning classifier, it achieved predictive accuracy that surpassed established markers such as PD‑L1 expression and tumor mutational burden. Responders identified by the test showed significantly longer disease‑free survival.

If larger, prospective studies confirm these findings, the cfDNA‑based assay could reshape treatment pathways by allowing physicians to triage patients toward immunotherapy only when the likelihood of response is high, thereby sparing the majority from unnecessary toxicity and reducing overall oncology spending. The approach also hints at a broader platform: similar fragmentation signatures might forecast therapeutic outcomes in melanoma, lung cancer, or even autoimmune conditions where tissue‑immune crosstalk is pivotal. Investors are likely to watch the upcoming validation trials, as a successful commercial test would create a new revenue stream at the intersection of liquid biopsy and AI‑driven oncology.