Longitudinal head-neck positional discordance during nasopharyngeal carcinoma radiotherapy and early CBCT prediction of neck-dominant geometric instability

Background Regional head-neck positional discordance may develop during radiotherapy for nasopharyngeal carcinoma (NPC), but its longitudinal evolution, measurement reproducibility, and value for identifying subsequent cervical geometric instability remain incompletely characterized. Methods This single-center retrospective longitudinal cohort study included 822 patients with NPC and 5,436 evaluable CBCT fractions. Paired head- and neck-region six-dimensional registration-derived positional measurements were evaluated across early (weeks 1–2), middle (weeks 3–4), and late (week 5 onward) treatment phases. The primary outcome was study-defined patient-level neck-dominant setup instability (NDSI) during the middle-to-late phases. Longitudinal mixed-effects models, generalized estimating equations, multivariable logistic regression, sensitivity analyses, registration-reproducibility assessment, an exploratory focused dosimetric analysis, and an internally validated early prediction model were used. Results Neck translational vector error was greater than head translational vector error (2.91 ± 1.54 vs. 2.38 ± 1.26 mm) and increased from 2.69 ± 1.42 mm in the early phase to 3.23 ± 1.70 mm in the late phase (region-by-phase interaction, P < 0.001). Phase-specific NDSI occurred in 20.9%, 29.4%, and 36.7% of phase-evaluable patients in the early, middle, and late phases, respectively, while the primary middle-to-late patient-level NDSI outcome occurred in 254/822 patients (30.9%). N2–3 stage (adjusted OR, 1.44; 95% CI, 1.08–1.92), GTVnd per 10-cm 3 increase (adjusted OR, 1.09; 95% CI, 1.04–1.15), early head-neck translational discordance ≥1.5 mm (adjusted OR, 1.91; 95% CI, 1.42–2.57), and early neck Pitch error ≥1.0° (adjusted OR, 1.58; 95% CI, 1.16–2.15) were associated with subsequent NDSI. Inter-observer agreement was high for translational discordance (ICC, 0.94) and NDSI classification ( κ = 0.84). In the exploratory dosimetric subset, NDSI-positive scans showed a greater reduction in lower-neck PTV D95 than NDSI-negative scans (−2.8% vs. −1.1%; P < 0.001). The early prediction model achieved AUCs of 0.77 (95% CI, 0.73–0.81) and 0.74 (95% CI, 0.68–0.80) in the training and validation cohorts, respectively. Conclusion Neck-region positional discordance increased during NPC radiotherapy, and early regional CBCT measurements together with nodal burden were associated with subsequent study-defined geometric instability. Reproducibility and exploratory dosimetric findings provided preliminary support for the geometric construct, while the early model showed moderate internal predictive performance. Prospective and external validation are required before the endpoint or prediction model is used for clinical decision-making.