Multilevel Regression and Poststratification Interface: An Application to Track Community-level COVID-19 Viral Transmission

TL;DR

This paper introduces a Bayesian multilevel regression and poststratification (MRP) workflow with new extensions and open-source tools, applied to estimate community COVID-19 transmission using hospital outpatient testing data in Michigan.

Contribution

It develops a novel MRP-based method with extensions for time-varying data and granular geography, along with open-source tools, to estimate community viral transmission from non-random testing data.

Findings

Successfully estimated community-level COVID-19 incidence in Michigan.

Demonstrated the effectiveness of MRP in adjusting for sample nonrepresentativeness.

Provided open-source tools for broad research adoption.

Abstract

We present a novel Bayesian workflow for multilevel regression and poststratification (MRP), introducing extensions to time-varying data and granular geography and publicly available open-source computation tools, facilitating broad research adoption and reproducibility. In the absence of comprehensive or random testing throughout the COVID-19 pandemic, we have developed a proxy method for synthetic random sampling to estimate community-level viral incidence, based on viral RNA testing of asymptomatic patients who present for elective procedures within a hospital system. The approach collects routine testing data on SARS-CoV-2 exposure among outpatients and performs statistical adjustments of sample representation using MRP, a procedure that adjusts for nonrepresentativeness of the sample and yields stable small group estimates. We illustrate the MRP interface with an application to track community-level COVID-19 viral transmission in the state of Michigan.