Xihong Lin will give a presentation at the virtual 2021 Joint Statistical Meeting (JSM) on a JASA Applications and Case Study Discussion Paper entitled:
Regression Models for Understanding COVID-19 Epidemic Dynamics with Incomplete Data
Wednesday, August 11
The session has three discussants. This is a joint work with Xihong’s postdoctoral fellows Corbin Quick and Rounak Dey.
This work develops a regression framework to jointly estimate the effective reproductive number Rt, ascertainment rates, incidence, and prevalence over time in one or multiple regions, and quantifies the effects of public health interventions on Rt. This Rt regression framework allows for incorporating geographic and time-varying covariates. To account for under-ascertainment, the team (a) modeled the ascertainment probability over time as a function of testing metrics, e.g., the number of PCR tests, and (b) jointly models data on confirmed cases and population-based serological surveys. To account for delays between infection, onset, and reporting, the team modeled stochastic lag times as missing data, and developed an EM algorithm to estimate the model parameters. The team analyzed data of US 52 states on COVID-19 daily case counts, PCR testing data, and serological survey data, giving an overall COVID-19 prevalence of 12.5% (ranging from 2.4% in Maine to 20.2% in New York) and cumulative ascertainment rate of 45.5% (ranging from 22.5% in New York to 81.3% in Rhode Island) in the US from March to December 2020. The team found substantial differences in Rt associated with several state containment policies, such as face-covering mandates and gathering restrictions.