Background:Traditional longitudinal regression models only estimate the average impact of time-varying covariates on outcome. Such impact may be different by time. Time-varying effect model(TVEM) could explore the way associations between variables of interest change over time. It has been applied into behavioral health research and intensive longitudinal data. However, there are few studies applying TVEM to investigate the transmission of infectious disease and its dynamic association with time-varying risk factor. Leveraging surveillance data of Coronavirus Disease 2019(COVID-19) and Twitter-based population mobility in South Carolina(SC), we applied TEVM to explore their dynamic relationship.

Methods:Cumulative COVID-19 cases through December 31, 2020 across the 46 counties in SC were collected from The New York Times database. Population mobility was assessed using the number of Twitter users with moving distance larger than 0.5 mile per day by county. County-level demographic characteristics were extracted from American Community Survey. TVEM was used to test the relationship between mobility and COVID-19 incidence adjusting for demographic characteristics. P-spline penalty with 10 knots was employed.

Results:The impacts of mobility on COVID-19 incidence were different over time, with significant and positive effects from June 20 to July 31. The coefficients ranged from 9.97(95%CI:1.70~18.25) to 22.16(95%CI:6.24~38.09). This timeframe was closed to the second phase of reopening in SC, which aimed to release the restrictions on attraction facilities and occupancy of retail establishment.

Conclusions:Intervention efforts are needed to maintain social distancing during economic recovery. TVEM could be applied into surveillance data and investigate the dynamic impact of risk factor on disease transmission.