Adjusting Mechanistic Epidemiological Models to Account for Urban Infrastructure Factors

Abstract Accurate and timely estimation of modeling parameters in a novel disease outbreak determines the proposed effectiveness of public health interventions. Parameters such as the incidence rate, recovery rate, and basic reproduction number are estimated from the first locations with cases when an outbreak begins to become a pandemic. However, these values, often based on case counts, include population characteristics that are specific to the index population. In this study, we propose a novel framework for parameter estimation for models that use inputs that come from index populations which differ in relevant urban crowding features. To prevent generalizability issues, this study introduces a modification to the SIR calibration method that takes into account population-level factors with accessible data. A set of calibrated parameters are used to train a neural network consisting of input features describing the risk factors for infectious diseases like COVID-19 including crowding on public transit, household crowding, and socioeconomic factors. The results show that when these factors are taken into account from even a small set of source locations, consistent improvements can be made to estimation for locations with no cases to use for calibrating their own models.