"Step up, speak up" could be the next mantra for COVID-19 screening at airports, universities and large employers if Vocalis Health Inc.'s screening tool takes off – and its recent CE mark approval makes that more likely. Vocalis reported that a large study found the tool, which uses AI-powered vocal biomarkers, had greater than 80% sensitivity and specificity, even in asymptomatic individuals.
Screening tools will likely be in high demand throughout most of 2021 in the U.S. and for several years in many other parts of the world. While vaccinations are steadily increasing, the majority of adults may not have access to immunization until this summer and no vaccine has yet received U.S. FDA emergency use authorization for children under age 16.
Better screening tools remain urgently needed. Many places that permit gatherings of large crowds still use elevated temperature as a screening tool, despite studies that show up to 50% of people hospitalized with COVID-19 did not have a fever on admission. Other organizations use online or paper questionnaires based on the U.S. CDC’s list of symptoms, though 40% of individuals who test positive for the disease are asymptomatic at diagnosis.
Colleges, long-term care facilities, and employers have turned to serial testing to quickly identify cases, but both the rapid antigen tests and gold-standard RT-PCR tests present problems. Generally the tests have low rates of false positives, but substantially higher rates of false negatives, particularly in the first several days after infection. While tests vary substantially, several studies have found that antigen tests return negative results in about 20% of confirmed, symptomatic COVID-19 cases and miss the majority of asymptomatic cases in real world settings. A pooled analysis of seven studies of PCR tests determined that they produced false negatives in infected individuals 100% of the time on day one of infection, declining to 67% on day 4, 38% on the first day of symptoms and 20% three days after symptoms appear.
Still, many organizations have embraced testing as a way to resume operations as safely as possible. That’s a costly decision, especially for those using PCR tests.
“PCR testing is being used to screen for COVID-19, which is an extremely expensive, resource-intensive and time-consuming approach,” said Shady Hassan, co-founder, chief operating and medical officer of Vocalis. “Instead of misusing PCR testing for screening, utilizing a highly scalable screening tool like Vocalischeck can fill a significant gap in the current approach to COVID screening, with the ability to effectively funnel those with high risk of infection to the appropriate diagnostic test.”
A recent study supports Hassan’s assertion. In collaboration with the Municipal Corporation of Greater Mumbai (MCGM), Newton, Mass.-based Vocalis evaluated Vocalischeck as a screening tool for COVID-19 at the NESCO COVID-19 Center. The unpublished study included more than 2,000 participants who spoke multiple languages including English, Hindi, Marathi and Gujarati.
Results from an unblinded validation set of 288 participants demonstrated an area under the curve (AUC) of 0.88 or an overall accuracy of 81.2%. That broke down to 80.3% sensitivity and 81.4% specificity. In comparison, a symptom checking survey identified about two-thirds of the individuals with COVID-19 infections and, of course, none of those who were asymptomatic.
“Our tool, unlike symptom checking-based screening, can also identify people who are asymptomatic. In the unblinded validation group, a third were asymptomatic COVID-19 positive, and Vocalischeck accurately detected over 80% of them as COVID-19 positive despite being asymptomatic,” Hassan told BioWorld.
For the study, participants used a mobile tablet to record a short voice sample in the Vocalischeck app. In this study, adding a symptom checker did not improve the accuracy of the vocal biomarker, though previous studies have shown some improvement with combined use, Hassan noted.
The other participants in the study comprised the training and blind validation groups. The voice recordings from the training group were used to optimize the vocal biomarker and so were excluded from analysis. Analysis of the blinded validation group is in process, “but initial results are very promising,” Hassan said.
Vocalischeck is available as an app or API. Any connected device can upload the recordings for analysis by the company’s artificial intelligence system. Results are returned in under a minute.
Vocalischeck is designed for use as a screening tool, not a diagnostic test. “Adding this layer of accurate screening can enable health care systems to allocate diagnostic resources more efficiently to those with the highest risk of being COVID-19 positive,” said Hassan.