Modeling COVID-19 latent prevalence to assess a public health intervention at a state and regional scale

Philip J. Turk; Shih-Hsiung Chou; Marc A. Kowalkowski; Pooja P. Palmer; Jennifer S. Priem; Melanie D. Spencer; Yhenneko J. Taylor; Andrew D. McWilliams

The lack of widespread COVID-19 testing, both for symptomatic and asymptomatic individuals, presents a major limitation of unknown scale and implications to forecasting models [32, 33] . Data sources are known to undercount cases, only including asymptomatic illness by chance, and to define cases inconsistently, based on variable testing criteria, between and within geographies. Collectively, these contribute to imperfect detection. As a result, high-level models may not comprehend the full extent of the outbreak, creating challenges in producing accurate forecasts. Our decision to base our modeling strategy on estimated latent prevalence addresses this inconsistency by adjusting observed prevalence counts. Modeling only the observed prevalence has the effect of shifting the SIR curves ahead in time by several days or more. While our estimate of the detection probability (0.14) is heuristically motivated, a thorough search of the literature supports our use of this estimate as reasonable. Future work will focus on refining this estimate as new research appears and to allow it to vary as a function of time.