Abstract
The prevalence of asymptomatic COVID-19 infections is largely unknown and may determine the course of future pandemic waves and the effectiveness of interventions. Using an epidemiological model fit to COVID-19 hospitalization counts from New York City, New York and Austin, Texas, we found that the undocumented attack rate in the first pandemic wave depends on the proportion of asymptomatic infections but not on the infectiousness of such individuals. Based on a recent report that 22.7% of New Yorkers are seropositive for SARS-CoV-2, we estimate that 56% (95% CI: 53-59%) of COVID-19 infections are asymptomatic. Given uncertainty in the case hospitalization rate, however, the asymptomatic proportion could be as low as 20% or as high as 80%. We find that at most 1.26% of the Austin population was infected by April 27, 2020 and conclude that immunity from undetected infections is unlikely to slow future pandemic spread in most US cities in the summer of 2020.
Introduction
During the early stages of the COVID-19 pandemic, a critical unknown was the "asymptomatic proportion"—the percentage of individuals who contract the virus but never develop symptoms. This knowledge gap created an "iceberg" problem for public health officials: if the asymptomatic pool was large, herd immunity might be closer than case counts suggested; if small, the vast majority of the population remained susceptible. This study seeks to resolve this uncertainty by triangulating between reported hospitalizations and serological data to determine the true extent of the virus's reach in two distinct urban environments: New York City (NYC) and Austin, Texas.
Methodology
The researchers employed a stochastic Susceptible-Exposed-Symptomatic/Asymptomatic-Recovered (SEIR) compartmental model. Unlike confirmed case counts, which were heavily biased by testing shortages in early 2020, hospitalization data provided a more reliable anchor for the actual number of symptomatic infections.
The model was fitted using the mif2 algorithm, a stochastic optimization procedure that performs maximum likelihood estimation. Key inputs included daily hospitalization data from the NYC Department of Health and Mental Hygiene and the Austin Metropolitan Statistical Area. The team varied three key parameters to test sensitivity: the asymptomatic proportion (0–99%), the relative infectiousness of asymptomatic individuals (29–142% of symptomatic cases), and the symptomatic hospitalization rate, adjusted for local demographics using 2015 US Census data.
Results
The model successfully captured the dynamics of hospital admissions in both cities (Figure 1A). By aligning the NYC model with a reported seroprevalence of 22.7%, the researchers derived a most likely asymptomatic rate of 56% (95% CI: 53–59%).
A key finding was the "Infectiousness Paradox": the total attack rate (percentage of the population infected) depends heavily on the proportion of asymptomatic cases but is remarkably insensitive to how infectious those individuals actually are. In Austin, the model estimated an infection detection rate of only 9.7%, meaning roughly 9 out of 10 cases went uncounted. Despite this "silent spread," Austin's cumulative infection rate was estimated at only 0.94% (95% CI: 0.72–1.26%) by late April 2020, reflecting the success of early social distancing interventions.
Discussion and Conclusion
The study highlights a significant vulnerability for cities that successfully "flattened the curve" early. Because Austin prevented a massive first wave, it possessed almost zero natural immunity, leaving it susceptible to future outbreaks. Furthermore, even with a high asymptomatic rate, the cumulative incidence in both cities remained far below the estimated 60% threshold required for herd immunity (Figure 1B).
These findings suggest that immunity from undetected infections was unlikely to slow pandemic spread in most US cities during the summer of 2020. Limitations of the study include sensitivity to the assumed symptomatic hospitalization rate and the assumption that asymptomatic infections provide lasting immunity. Future work should focus on refining these parameters as more comprehensive retrospective data becomes available.
Summary updated