A nurse-led bedside diagnostic model using cervical tracheal breath sounds to detect silent aspiration in dysphagia patients: A prospective diagnostic accuracy study

Abstract

Background

Silent aspiration poses significant clinical risks in dysphagia patients, contributing to recurrent respiratory infections and aspiration pneumonia that ultimately result in elevated hospitalization rates and increased mortality. While video fluoroscopic swallowing studies (VFSS) and fiberoptic endoscopic evaluation of swallowing (FEES) remain the diagnostic gold standards, their high costs create substantial barriers to efficient large-scale screening. Conventional bedside swallowing assessments demonstrate limited sensitivity for silent aspiration detection due to inherent subjectivity in evaluation process. This diagnostic dilemma presents nursing professionals with critical challenges in achieving timely and accurate identification of aspiration events.

Objective

To develop and validate a diagnostic model for detecting silent aspiration in dysphagia patients through acoustic analysis of breath sound parameters.

Design

A prospective diagnostic accuracy study.

Settings

Four hospitals in Nanjing, China.

Participants

A total of 212 patients with dysphagia were included in this study.

Methods

Breath sounds were recorded with an electronic stethoscope, and acoustic parameters were extracted via Fourier transform spectral analysis. The relative change in acoustic parameters was quantified by comparing post-swallow measurements against baseline values. Based on VFSS/FEES findings, breath sounds were categorized into no aspiration, silent aspiration, and overt aspiration. Through computer-generated randomization, samples were stratified into modeling (70 %) and validation (30 %) groups. A logistic regression diagnostic model was developed in the modeling cohort, with subsequent evaluation of its performance in the validation group using receiver operating characteristic (ROC) curve analysis.

Results

Silent aspiration breath sounds demonstrated comparable prevalence between cohorts, occurring in 41.6 % (226/543; modeling group) and 41.8 % (97/232; validation group). The derived logistic model comprised three acoustic parameters with the following equation: logit(P) = − 0.621 − (0.672 × change of aggregated entropy) + (0.279 × change of average power) + (0.632 × change of delta time). ROC analysis revealed comparable diagnostic performance between cohorts, with the modeling cohort demonstrating an AUC of 0.785 (95 % CI: 0.744–0.821) and the validation cohort achieving 0.746 (95 % CI: 0.679–0.806).

Conclusion

A novel acoustic-based diagnostic model for silent aspiration risk stratification in dysphagia patients demonstrated good discriminatory power across both modeling and validated cohorts. This decision-support tool enables nursing staff to proactively identify high-risk patients through continuous monitoring, facilitating early implementation of aspiration-prevention protocols.