The COVID-19 pandemic has affected wide swaths of the global economy, mostly in a negative manner, but it has spurred some types of innovation at a rate that would be unimaginable in ordinary times. That seems to be the take-away for an emergency use authorization (EUA) granted to Miami-based Tiger Tech Solutions Inc. for its COVID Plus monitor, which uses plethysmography and a machine learning algorithm to provide a screening mechanism at mass gatherings, thus bringing the world one step closer to a state of normalcy.
The FDA announcement regarding the Tiger Tech EUA was the second time in two days the agency announced a first in the world of products granted market access for the COVID-19 pandemic. The FDA announced March 17 that it had granted a de novo petition to Biofire Diagnostics LLC, of Salt Lake City, for the Biofire 2.1 mulit-analyte assay for the SARS-CoV-2 virus and several other pathogens, including a number of coronavirus and influenza pathogens.
The Biofire de novo was announced simultaneously with the withdrawal of the EUA for the test, thus offering some insight into the question of conversion of EUAs to conventional premarket review mechanisms. Acting FDA commissioner Janet Woodcock said that while the Biofire de novo was the first such conversion, “we do not expect this to be the last, and look forward to working with developers of medical products to move their products through our traditional review pathways.”
Device picks up indicators of hypercoagulation
The FDA’s press release for the Tiger Tech product said that this first machine learning-based screening device identifies characteristics that may be indicative of hypercoagulation or hyperinflammatory states, although the results might not be specific to COVID-19. Hypercoagulation is known to accompany sepsis and cancer, while hyperinflammation may be seen in severe allergic reactions.
The EUA allows for screening of asymptomatic individuals aged at least 6 years, and is to be used when temperature checks are being conducted in mass settings, such as airports. The individual whose temperature is lower than would be detected for someone suffering from fever would be referred to the COVID Plus, but the agency went to some lengths to emphasize that the COVID Plus is not a substitute for a diagnostic test, and that it is not intended for use with symptomatic individuals. For the purposes of this EUA, a fever is defined as anyone with a body temperature of at least 100.4° F (38° C)
The COVID Plus system employs an armband wrapped around the upper arm for up to five minutes, using two light sensors providing pulsatile signals via plethysmography. Those data are fed into the ML algorithm, which generates either a green light indicating that the individual does not exhibit symptoms suggestive of COVID-19, or a red light, which indicates possible hypercoagulation. The algorithm can also generate a blue light, indicating an indeterminate outcome, and any follow-up should be handled with PCR testing. The entire system is powered by a nine-volt battery.
The system comes with documentation guiding the test administrators as to whether to admit the individual onto the plane or into a large congregate setting. The clinical performance study requirements for the COVID Plus were considerable, starting with a hospital-based validation study of nearly 470 asymptomatic individuals. Of this group, 69 tested positive, and those data indicated a positive percent agreement with PCR of 98.6% and a negative percent agreement of 94.5%. The FDA said these numbers are similar to those seen in a confirmatory study conducted in a K-12 school study, but the company will have to conduct a postmarket study of 1,100 individuals at sites where staff have been trained by Tiger Tech. Unlike the pre-EUA study, the sites conducting the follow-up study will not have to offer diagnostic testing on site.
The company did not respond to contact for comment.
Other screening algorithms in development
Tiger Tech might not have the AI/ML space for COVID-19 to itself for much longer if recent developments in the literature are any indication. One example of this is a study appearing in the February 2021 issue of The Lancet Digital Health, which describes two machine learning algorithms that analyze data collected at the emergency department (ED), such as lab test results and vital signs. The algorithms were developed from electronic health records of patients seen in emergency departments in four hospitals in the U.K., while the validation set was based in part on controls that included both pre-pandemic and COVID-19 negative patients admitted after Dec. 1, 2019.
More than 155,000 patients were included in the study, and the results were a sensitivity of 77.4% and a specificity of 95.7% for the ED model while the hospital admission model achieved a sensitivity of 77.4% and a specificity of 94.8%. The authors said their models had performed effectively as a screening test, and had excluded COVID-19 negative patients within an hour of presentation at the ED. They also said this approach can be quickly scaled within the existing lab infrastructure and standard care capacity at most hospitals in high- and middle-income nations.