Casual, Statistical and Mathematical Modeling with Serologic Data

We will develop methods to enhance the design and analysis of serologic studies of populations with respect to COVID-19, including methods that may be generalized in the future to address challenges raised by other seasonal diseases (such as influenza) and newly emerging diseases. In addition, we will use serologic data in innovative ways to underpin mathematical models that can project population-level trends. Early serosurveys using convenience samples of the population and serologic assays with variable and often uncertain sensitivity and specificity were heavily criticized, for unrepresentativeness and inadequate accounting for test characteristics, resulting in bias and overconfidence (unduly narrow confidence bounds). Aim 1 will develop methods for valid inference of seroprevalence, specifically by (a) accounting for biased sampling, (b) accounting for imperfect tests, and (c) developing and testing a novel approach to snowball sampling employing serologic tests to enhance outbreak detection and contact tracing. Valid comparisons that assess seroprotection-whether, how much, and how long an individual is protected by an immune response to a COVID-19 infection (specifically, by antibodies) against reinfection-rely on adequate control for confounding, an issue that arises in multiple ways specific to seroprotection studies. Likewise, waning of seroprotection may be inferred in error if studies are not carefully designed and analyzed. The unprecedented efforts to develop detailed serologic and systems serologic data sets provide new forms of data that can be leveraged to better inform these inferences. Aim 2 will develop a suite of methods to enhance causal inference in seroprotection studies, including (a) sample size and power calculations; and (b) improved exploitation of serological data to reduce biases due to confounding and risk compensation. Aim 3 will develop new mathematical modeling approaches and apply them to quantify the likely reduction in the herd immunity threshold for COVID-19 due to various forms of risk heterogeneity and assortativeness in mixing. Aim 4 will develop models of COVID-19 transmission that accommodate emerging evidence about the duration and nature of immunity to infection, shedding, and symptoms, to obtain estimates of how illness attack rates will differ under varying assumptions about the progress of immunity. Aim 5 will develop transmission models to assess optimal cohorting arrangements in congregate facilities (eg prisons and nursing homes), with special attention to the nature of immunity required for these arrangements to be beneficial. Finally, vaccine supplies may be initially limited, necessitating efficient use of them. Aim 6 will investigate the use of serologic data in combination with other types of data to optimize allocation of scarce vaccines.