Multifeature endoscopic swallowing patterns associated with pneumonia in hospitalized geriatric patients: a four-year longitudinal cohort study

BACKGROUND: Oropharyngeal dysphagia is common in geriatric patients and a major risk factor for pneumonia. Fiberoptic endoscopic evaluation of swallowing (FEES) can identify specific swallowing abnormalities; however, apart from aspiration, the relationship of these abnormalities to pneumonia risk is not well understood. This study aimed to identify FEES-based swallowing abnormalities associated with long-term pneumonia risk beyond airway invasion alone and to develop and internally evaluate a transparent multifeature risk score. METHODS: In this retrospective cohort study, 98 geriatric patients underwent FEES. Nine predefined FEES-derived swallowing features were analyzed using a clustering approach to identify a multifeature constellation associated with pneumonia. In addition, three methods were tested to develop a scoring system: clustering-based feature selection, penalized logistic regression, and a weighted ensemble of decision tree stumps, all assessed using fivefold cross-validation. RESULTS: The clustering approach revealed a swallowing pattern comprising prolonged oral phase (excluding bread), delayed swallow reflex, reduced whiteout intensity, repetitive swallowing (excluding bread), piriform sinus residue, and airway invasion at PAS ≥ 3. A simple scoring system assigning one point per feature yielded an area under the receiver operating characteristic curve of 0.73 (95% CI 0.61-0.82). Each additional deficit increased pneumonia risk (odds ratio 1.82, 95% CI 1.24-2.67). The optimal Youden-optimized cut-off was ≥ 4 deficits, yielding a sensitivity of 0.46 (95% CI 0.29-0.63) and specificity of 0.89 (95% CI 0.79-0.96). CONCLUSIONS AND IMPLICATIONS: Pneumonia risk in hospitalized geriatric patients with oropharyngeal dysphagia arises from the accumulation of functional swallowing impairments rather than isolated endoscopic findings. A transparent, exclusively FEES-based multifeature score provides a pragmatic framework for risk stratification.