Validity of at-home rapid antigen lateral flow assay and artificial intelligence read to detect SARS-CoV-2

Background: RT-PCR—the current reference standard to detect SARS-CoV-2—is costly and
slow, requiring a sophisticated laboratory with an average test turnaround time of one to two days. Rapid at-home and point of care (POC) testing is urgently needed to prevent COVID-19 transmission throughout the world, particularly in low and middle-income countries.

Objective: We assessed the diagnostic accuracy of an at-home rapid lateral flow antigen assay (Cellex) with artificial intelligence interpretation by smartphone (Gauss) to detect SARS-CoV-2 antigen (Lateral flow with AI read; LFAIR). We also serially assessed the diagnostic accuracy of the LFAIR in SARS-CoV-2 positive individuals over several days and compared the LFAIR results with polymerase chain-reaction cycle threshold (CT) value.

Methods: We recruited 33 individuals from two clinical treatment trials of COVID-19 and 88 participants from a population-based surveillance study. Individuals recruited from the clinical trials were asked to complete the LFAIR daily for five days. Participants from the population-based study completed one LFAIR. All participants were provided kits with no instructions from study personnel and were asked to complete the tests at home. The LFAIR results were compared to RT-PCR, the current reference standard, as either dichotomous outcome or as continuous CT value. We also assessed LFAIR results by days since symptom onset.

Results: Using the first test completed by each subject, the LFAIR was 86.2% sensitive (95% CI 73.6% - 98.8) and 94.3% specific (95% CI 88.8% - 99.7%). With a GEE model that adjusted for days since symptom onset and repeat testing, the sensitivity was predicted to be 97.78% (95% CI 89.9% - 99.5%) on the first day with a 0.89% (95% CI 0.30 - 1.61%) decrease in sensitivity with each additional day after symptom onset. The LFAIR test line intensity decreased inversely with CT value, a proxy for viral load (McFadden pseudo R2 = 0.49; Nagelkerke pseudo R2 = 0.99)

Conclusions: An at-home LFAIR shows high levels of diagnostic accuracy. Intensity of the at-home antigen test line is inversely related with an individuals’ SARS-CoV-2 viral load. The test has potential for broad-scale implementation to minimize the spread of SARS-CoV-2.