Using Machine Learning to Improve Screening for Oropharyngeal Dysphagia in Hospitalized Versus Primary Care Adult Patients With COVID-19 Disease: Retrospective Observational Study.

Abstract

Background: Oropharyngeal dysphagia (OD) commonly occurs in patients with COVID-19 disease, posing diagnostic challenges due to isolation protocols.

Objective: This study aimed at evaluating Artificial Intelligence Massive Screening for Oropharyngeal Dysphagia (AIMS-OD), a machine learning software for real-time OD screening, comparing OD prevalence and clinical outcomes using OD ICD-10 (International Statistical Classification of Diseases, Tenth Revision) R13 codes (R13-OD) and high-risk AIMS-OD (H-AIMS-OD) scores (>0.5), in hospital and primary care patients with COVID-19 disease. It explored clinical characteristics, OD risk factors, and clinical outcomes.

Methods: This retrospective, observational study analyzed patients with SARS-CoV-2 aged 18 years and older in Catalonia from January 1 to August 31, 2020, including hospital and primary care data on clinical information, International Classification of Diseases, Tenth Revision (ICD-10) codes, hospital stay, discharge destination, and mortality. AIMS-OD assessed OD risk, stratifying patients by age (aged 18-69 years and 70 years and older).

Results: Among 257,541 patients with COVID-19 disease, 59.3% (152,721/257,541) were aged 18-69 years and 40.7% (104,820/257,541) were aged 70 years and older. Hospital and primary care R13-OD prevalence was 3.5% and 4.3%, respectively; AIMS-OD showed 34.8% and 15.4%, with True prevalence at 16.7% and 7.4%. Patients aged 70 years and older had worse clinical outcomes and worse prognosis. Patients in R13-OD experienced significantly worse clinical outcomes than patients with H-AIMS-OD, who in turn fared worse than those with no R13-OD and with low AIMS-OD risk. Risk factors for patients with COVID-19 R13-OD included age, neuroleptic use, stroke, dementia, and delirium.

Conclusions: AIMS-OD screening revealed high prevalence and significant underdiagnosis in patients with COVID-19 disease across settings. Early detection and risk stratification using AIMS-OD could improve clinical decision-making, diagnosis, and management, particularly in older patients with comorbidities.