Development and validation of a respiratory-responsive vocal biomarker-based tool for generalizable detection of respiratory impairment: independent case-control studies in multiple respiratory conditions including asthma, chronic obstructive pulmonary disease, and COVID-19.

Abstract

Vocal biomarker-based machine learning approaches have shown promising results in detecting various health conditions, including respiratory diseases such as asthma. In this study, we aim to validate a respiratory-responsive vocal-biomarker (RRVB) platform initially trained on an asthma and healthy volunteer dataset for its ability to differentiate, without modification, active COVID-19 infection vs. healthy volunteers in patients presenting to hospitals in the US and India. The objective of this study was to determine whether the RRVB model can differentiate patients with active COVID-19 infection vs. asymptomatic healthy volunteers by assessing its sensitivity, specificity, and odds ratio. Another objective was to evaluate whether the RRVB model outputs correlate with symptom severity in COVID-19. A logistic regression model using a weighted sum of voice acoustic features was previously trained and validated on a dataset of about 1,700 patients with a confirmed asthma diagnosis vs. a similar number of healthy controls. The same model has shown generalizability to patients with chronic obstructive pulmonary disease (COPD), interstitial lung disease (ILD), and cough. In the present study, a total of 497 participants (46% male, 54% female; 94% < 65 years, 6% >= 65 years; 51% Marathi, 45% English, 5% Spanish speakers) were enrolled across four clinical sites in US and India and provided voice samples and symptom reports on their personal smartphones. The participants included symptomatic COVID-19 positive and negative patients as well as asymptomatic healthy volunteers. The RRVB model performance was assessed by comparison with clinical diagnosis of COVID-19 confirmed by RT-PCR. The RRVB model's ability to differentiate patients with respiratory conditions vs. healthy controls was previously demonstrated on validation data in asthma, COPD, ILD and cough with odds ratios of 4.3, 9.1, 3.1, and 3.9 respectively. The same RRVB model in the present study in COVID-19 performed with a sensitivity of 73.2%, specificity of 62.9%, and odds ratio of 4.64 (p<0.0001). Patients experiencing respiratory symptoms were detected more frequently than those not experiencing respiratory symptoms and completely asymptomatic patients (78.4% vs. 67.4% vs. 68.0%). The RRVB model has shown good generalizability across respiratory conditions, geographies, and language. Results in COVID-19 demonstrate its meaningful potential to serve as a pre-screening tool for identifying subjects at risk for COVID-19 infection in combination with temperature and symptom reports. Although not a COVID-19 test, these results suggest that the RRVB model could encourage targeted testing. Moreover, the generalizability of this model for detecting respiratory symptoms across different linguistic and geographic contexts suggests a potential path to development and validation of voice-based tools for broader disease surveillance and monitoring applications in the future. ClinicalTrials.gov (NCT04582331.