Jonathan D. Schoenfeld

Purpose: To determine the maximum tolerated dose (MTD) of the pan–class I PI3K inhibitor buparlisib (B) in combination with weekly cisplatin and radiotherapy (RT) among patients with locally advanced (LA) squamous cell cancer of the head and neck (SCCHN) and tobacco history. Patients and Methods: Patients with stage III/IV LA-SCCHN [American Joint Committee on Cancer (AJCC), Seventh Edition] and >10 pack-year tobacco use history were enrolled. Patients received B during a 2-week run-in phase and during chemoradiotherapy (CRT) consisting of standard 70 Gy of RT plus weekly cisplatin. An exploratory analysis of genomic sequencing was performed on biopsy specimens. Results: Twenty-three patients were enrolled {n = 17 at the MTD [B 40 mg daily and cisplatin (CDDP) 30 mg/m2/week]}. Ninety-one percent (21/23) had stage IV disease. Human papillomavirus (HPV) was detected in 15 of 18 cases with oral/oropharyngeal disease. The median follow-up was 23 months. Recurrence was detected in five patients among whom four had evaluable tumor samples. Fifty-percent (2/4) of patients with evaluable tumor samples were characterized by activation of the PI3K pathway. In five patients whose disease responded during the 2-week run-in phase with B alone, 3 of 4 with sequencing data showed loss-of-function mutations in either tumor necrosis factor receptor–associated factor 3 (TRAF3) or cylindromatosis lysine deubiquitinase (CYLD). Preclinical studies with mutations in TRAF3 or CYLD modeled via CRISPR/Cas9 knockout in HPV+ SCCHN cells demonstrate that loss of TRAF3 or CYLD may sensitize SCCHN cell lines to PI3K through mechanisms other than blocking the NF-κB pathway. Conclusions: Low-dose B with CRT was feasible and active in this high-risk cohort, although escalation to the standard weekly cisplatin dose of 40 mg/m2 was not possible. Our data suggest that TRAF3/CYLD-mutant SCCHN may be susceptible to PI3K inhibition with B, whereas PI3K pathway activation seemed to be associated with poor outcomes in this limited dataset.

Immune-related adverse events (irAEs) affect up to 40% of patients receiving immune checkpoint inhibitors, yet identification requires laborious, inconsistent manual chart review. Here we show that agentic artificial intelligence coupled with human verification could enhance efficiency, performance and consistency for irAE assessment. We developed and evaluated an agentic large language model system to extract the presence, temporality, severity grade, attribution and certainty of six irAE types from clinical notes. Retrospectively, the system achieved a macro-averaged F ₁ of 0.92 for current-event detection and 0.66 for multi-class severity grading; self-consistency improved F ₁ by 0.14. The best-performing configuration cost approximately US$0.02 per note. In prospective silent deployment over 3 months, the binary F ₁ was 0.72–0.79. In a randomized crossover study of clinical trial staff, agentic assistance reduced annotation time by 2.71 min ( P < 0.001), increased complete-match accuracy (odds ratio 1.45; 95% confidence interval 1.01–2.08; P = 0.045) and improved interannotator agreement (Krippendorff’s α from 0.22–0.51 to 0.82–0.85).

Eligible and completed plasma TTMV tests at each study time point

Number of subjects eligible for and with completed TTMV testing at each study time point

Representativeness of study participants

Clinical evaluations performed after detectable surveillance TTMV until diagnosis of recurrence for early true-positive subjects

Test performance metrics for subjects with baseline detectable TTMV and HPV16

Test performance metrics for subjects with baseline TTMV>200 and HPV16