Oncological Phenotype as an Independent Predictor of Aspiration in Oropharyngeal Dysphagia: A FEES-Based Multivariable Analysis of 328 Patients Across Seven Etiological Phenotypes

Oropharyngeal dysphagia (OD) affects diverse patient populations in tertiary referral centers, yet independent predictors of aspiration across the full etiological spectrum remain poorly defined. We conducted a retrospective, cross-sectional study (N=328) at a national tertiary referral center (April 2024–April 2025) to identify biomechanical and etiological predictors of aspiration in patients with OD confirmed by fiberoptic endoscopic evaluation of swallowing (FEES). Aspiration was defined as Penetration-Aspiration Scale (PAS) ≥6. Patients were classified into seven etiological phenotypes adapted from Roden and Altman. A block-entry binary logistic regression model was constructed with leave-one-out cross-validation (LOOCV), constrained to events-per-variable (EPV)=8.0 (80 aspiration events, 8 predictors). Neurological (30.5%) and systemic/metabolic (29.3%) phenotypes predominated. Global aspiration prevalence was 24.4% (95% CI: 19.7–29.6%). Contrary to the primary hypothesis, the oncological phenotype had the highest crude aspiration rate (38.9%; 95% CI: 23.1–56.5%) and was the strongest independent etiological predictor in the multivariable model (OR=2.82; 95% CI: 1.10–7.25; p=0.030), exceeding the neurological reference. Decreased lingual propulsion was the dominant biomechanical predictor (OR=4.50; 95% CI: 2.39–8.81; p<0.001); altered swallowing reflex also reached significance (OR=1.92; 95% CI: 1.03–3.56; p=0.037). The model had acceptable–good discrimination (AUC=0.777; 95% CI: 0.721–0.832; AUC-LOOCV=0.727; optimism=0.050) and satisfactory calibration (Hosmer-Lemeshow p=0.845). Lingual propulsion is the dominant biomechanical determinant of aspiration in OD. The oncological phenotype exceeds the neurological reference as an independent predictor—a counterintuitive finding that warrants prospective confirmation in dedicated oncological cohorts.