COVID-19 epidemiology as emergent behavior on a dynamic transmission forest

TL;DR

This paper develops a stochastic compartmental model for COVID-19 transmission that captures dynamic behaviors and can generate transmission trees, aligning well with observed data and contact tracing insights.

Contribution

It introduces a novel stochastic model with dynamic mean and variance, and reframes it as a branching process to analyze transmission structures.

Findings

Model fits Washington COVID-19 data from 2020-2021.

Hidden states like prevalence agree with surveys.

Transmission trees are consistent with contact tracing.

Abstract

In this paper we create a compartmental, stochastic process model of SARS-CoV-2 transmission, where the process's mean and variance have distinct dynamics. The model is fit to time series data from Washington from January 2020 to March 2021 using a deterministic, biologically-motivated signal processing approach, and we show that the model's hidden states, like population prevalence, agree with survey and other estimates. Then, in the paper's second half, we demonstrate that the same model can be reframed as a branching process with a dynamic degree distribution. This perspective allows us to generate approximate transmission trees and estimate some higher order statistics, like the clustering of cases as outbreaks, which we find to be consistent with related observations from contact tracing and phylogenetics.