Abstract
<title>Abstract</title> <p>Background Neoadjuvant immunochemotherapy can induce pathological regression in resectable locally advanced head and neck squamous cell carcinoma (LA-HNSCC), but whether pretreatment programmed death ligand 1 combined positive score (CPS) identifies responders remains uncertain. Methods This single-center retrospective cohort included 31 consecutive patients with treatment-naive, resectable stage III-IVA HNSCC. All received four 3-week cycles of pembrolizumab 200 mg, nab-paclitaxel 260 mg/m², and cisplatin 75 mg/m², followed by surgery after 3–6 weeks. Pretreatment biopsies underwent PD-L1 IHC 22C3 pharmDx testing. Two blinded pathologists independently assessed residual viable tumor (RVT) in primary and nodal specimens. Major pathological response (MPR) was RVT ≤ 10%, including pathological complete response (pCR). The primary endpoint was primary-tumor MPR. CPS 10 was prespecified; Fisher's exact and exploratory receiver operating characteristic analyses were used. Results Primary-tumor MPR occurred in 22/31 patients (71.0%) and pCR in 15/31 (48.4%). Patient-level overall MPR and pCR occurred in 18/31 (58.1%) and 9/31 (29.0%). Primary-tumor MPR was 66.7% with CPS < 10 and 72.7% with CPS ≥ 10 (odds ratio 1.33, 95% CI 0.25–7.11; P = 1.000). Continuous CPS showed limited discrimination (AUC 0.525, bootstrap 95% CI 0.277–0.767). Any-grade and grade ≥ 3 adverse events occurred in 71.0% and 12.9%; there were no treatment-related deaths. Conclusion Four-cycle pembrolizumab plus chemotherapy showed substantial pathological activity in this p16-negative cohort. CPS was not associated with response and should not be used alone for patient selection without validation.</p>