Abstract
Aim
To develop a three-item short form of the EAT-10 by identifying the smallest subset of items that preserves the screening performance of the full questionnaire in hospitalized older adults.
Design
Multicentre cross-sectional observational study.
Methods
The study was conducted in nine public hospitals in Spain within the NUTRIFAG project. Hospitalized patients aged ≥65 years with a length of stay ≥48 h were included. Patients receiving enteral or parenteral nutrition at admission, those in terminal condition, or with a primary diagnosis of cancer were excluded. Suspected dysphagia was assessed using the Eating Assessment Tool (EAT-10), with a score ≥3 considered positive. Multivariate logistic regression models with forward item selection were used to identify the most informative symptoms indicators. The sample was randomly divided into a training subset (60%) and an internal assessment subset (40%). Model performance was evaluated using Nagelkerke’s pseudo
R
2
, the Hosmer–Lemeshow test, sensitivity, specificity, and the area under the receiver operating characteristic curve (AUC).
Results
The sample included 2,519 participants. Three symptom indicators emerged as the most informative for identifying patients with suspected dysphagia: requiring extra effort to swallow liquids, requiring extra effort to swallow solids, and coughing while eating. Together, these indicators showed high agreement with the classification obtained using the full EAT-10 in the training phase (
R
2
= 0.861; sensitivity = 0.863; specificity = 0.979; AUC = 0.982), which similar classification performance in the internal assessment subset (sensitivity = 0.881; specificity = 0.984).
Conclusion
The proposed three-item subset represents a shortened version of the EAT-10 that preserved much of the screening performance of the full questionnaire in hospitalized older adults. Although it may facilitate dysphagia screening by reproducing the classification obtained with the full EAT-10 in routine nursing assessment, independent external validation against clinical reference standards is required before its diagnostic use can be recommended.