RAISE: IHBEM: Behavioral Heterogeneity and Uncertainty in Epidemiological Models

Cornell University, Ithaca NY

Investigators

Abstract

In this project, existing epidemiological models are developed further to address two fundamental challenges: heterogeneity and uncertainty. Epidemiological models are mathematical and computational tools that can help guide public-health decisions to mitigate morbidity, mortality, and economic impacts of communicable diseases. Their effectiveness could be further improved by addressing two issues. First, people differ from one another in their perceptions of risk, social network size, number of contacts per day, willingness to engage with public-health measures, and in their receptivity to mis- and dis-information. This heterogeneity is a major challenge in accurately characterizing human behavior that is central to ensuring prediction accuracy in epidemiological models. Second, uncertainty pervades all epidemiological modeling. All models rely on input parameters, such as the fraction of infections that are asymptomatic, that cannot be known with certainty, yet model predictions can vary dramatically as the unknown parameters vary over plausible ranges. Ignoring this intrinsic uncertainty in key input parameters can lead to overly confident predictions of models and, in turn, to poor decisions that do not consider an appropriate range of potential outcomes. This model development is carried out working with organizations of public health officials at state and federal levels to ensure relevance and practicality. This project is funded jointly by the Division of Mathematical Sciences (DMS) in the Directorate of Mathematical and Physical Sciences (MPS) and the Division of Social and Economic Sciences (SES) in the Directorate of Social, Behavioral, and Economic Sciences (SBE). In this project, a range of statistical and machine-learning methods are employed on publicly available data sets to fit parametric and non-parametric models of human behaviors relevant to epidemiological models. Advanced techniques are developed and used to elicit human behavior from contact-tracing data, adjusting for the missing data and selection bias that is inherent in such data. The resulting analyses provide input into optimization methods that are designed to optimize policy interventions while explicitly accounting for heterogeneity and uncertainty. Tailored optimization methods optimize over both short time scales (days), through optimization of graph models that depict social networks and through eigenvalue optimization problems on contact-structure models that minimize rates of spread, and over longer time scales (weeks or months), that take into account behavioral response to interventions, such as fatigue. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.