Instrumental screening for dysphagia by combining high-resolution cervical auscultation with advanced data analysis tools to identify silent dysphagia and silent aspiration

点击翻译按钮获取中文摘要

ABSTRACT Dysphagia (disordered swallowing) causes nearly 150,000 annual hospitalizations and over 220,000 additional hospital days, and prolongs hospital lengths of stay by 40%. Dysphagia risk is typically identified through subjective screening methods and those identified through screening undergo gold standard imaging testing such as videofluoroscopy (VF). However, screening methods over- or underestimate risk, and completely fail to identify patients with silent dysphagia (e.g., silent aspiration) that can cause pneumonia and other adverse events. Pre-emptive detection of silent or near-silent aspiration is essential. The long term goal is to develop an instrumental dysphagia screening approach based on high-resolution cervical auscultation (HRCA) in order to early predict dysphagia-related adverse events, and initiate intervention measures to mitigate them. The overall objective here is to develop accurate, advanced data analysis approaches to translate HRCA signals to swallowing events observed in VF images. Our strong preliminary data has led us to our central hypothesis: advanced data analytics tools are suitable approaches for the analysis of HRCA in order to automate dysphagia screening. The rationale is that a reliable, robust early-warning instrumental dysphagia screening approach will reduce adverse events in patients with silent aspiration/dysphagia, shorten length of stay and improve overall clinical outcomes. Guided by strong preliminary data, we will pursue the following three specific aims: (1) develop machine learning algorithms to differentiate HRCA signals produced by swallowing physiologic events from similar, non-swallow related signals produced during swallowing; (2) translate HRCA swallowing-signal signatures to actual swallow physiologic events to detect abnormal swallowing physiology; and (3) discriminate normal from abnormal airway protection and swallow physiology via machine-learning analysis of HRCA signals with similar accuracy as VF. Under the first aim, a machine learning approach will be used to detect pharyngeal swallowing events and differentiate them from speech, cough and other non- swallow events, with 90% accuracy, when compared to a human expert’s interpretation of our VF data sets. Under the second aim, objective swallowing physiology observations from VF will be matched to swallowing events observed with HRCA in order to show that abnormal swallow physiology and airway protection will produce distinctive HRCA signal signatures that predict the same events identified with VF. Under the third aim, analytical algorithms will be used to detect signs of disordered airway protection in HRCA signal signatures with 90% accuracy when compared to a human expert’s airway protection ratings from VF images. The approach is innovative, as it will produce analysis tools that will infer about dysphagia and aspiration based on the analysis of HRCA with unprecedented accuracy, before patients are placed in harm’s way. Our work is significant, because it will translate to an early-warning HRCA screening tool that predicts dysphagia- related adverse events in asymptomatic patients reducing medical adverse events, and length of stay.