Interobserver Agreement in Acute Toxicity Assessment During Head and Neck Radiation Therapy/Chemoradiation Therapy:...

Acute toxicity is a key endpoint in head and neck radiation therapy (RT) and chemoradiation therapy (CRT) trials and is most commonly reported as maximum toxicity per patient. However, the reproducibility of clinician-reported toxicity grading remains incompletely characterized. We evaluated interobserver agreement in acute toxicity assessment, focusing on maximum toxicity as the primary endpoint.

This predefined subgroup analysis was conducted in a prospective randomized controlled trial of patients with head and neck squamous cell carcinoma undergoing curative-intent RT/CRT. Acute toxicities were assessed weekly using the National Cancer Institute Common Terminology Criteria for Adverse Events version 5.0 by 2 independent, blinded radiation oncologists. Four domains were evaluated: dermatitis, oral mucositis, dysphagia, and xerostomia. The primary endpoint was patient-level maximum toxicity, dichotomized as grade ≥2 and grade ≥3. Interobserver agreement was assessed using Cohen's κ.

Seventy-four patients were included. For maximum grade ≥2 toxicity, the incidence (observer 1 vs observer 2) was 55.4% vs 45.9% for dermatitis, 91.9% vs 70.3% for mucositis, 95.9% vs 91.9% for dysphagia, and 55.4% vs 58.1% xerostomia, with κ values of 0.33, 0.35, 0.41, and 0.23, respectively (fair to moderate). For grade ≥3 toxicity, κ values were 0.79 for dermatitis, 0.41 for mucositis, 0.63 for dysphagia, and 0.25 for xerostomia. Pooled weekly agreement for grade ≥2 toxicity was higher (κ = 0.49-0.89), with peak concordance during weeks 5 to 6.

Interobserver agreement for maximum acute toxicity was fair to moderate across domains of head and neck RT/CRT, with the largest discordance for oral mucositis, even between observers at the same center. Although maximum toxicity remains a pragmatic and widely used endpoint, caution is warranted in its interpretation, and weekly assessment may offer a more reproducible complementary endpoint, particularly for cross-trial comparisons.