Dynamic Panel Surveillance of COVID-19 Transmission in the United States to Inform Health Policy: Observational Statistical Study

Abstract

Background The Great COVID-19 Shutdown aimed to eliminate or slow the spread of SARS-CoV-2, the virus that causes COVID-19. The United States has no national policy, leaving states to independently implement public health guidelines that are predicated on a sustained decline in COVID-19 cases. Operationalization of “sustained decline” varies by state and county. Existing models of COVID-19 transmission rely on parameters such as case estimates or R0 and are dependent on intensive data collection efforts. Static statistical models do not capture all of the relevant dynamics required to measure sustained declines. Moreover, existing COVID-19 models use data that are subject to significant measurement error and contamination. Objective This study will generate novel metrics of speed, acceleration, jerk, and 7-day lag in the speed of COVID-19 transmission using state government tallies of SARS-CoV-2 infections, including state-level dynamics of SARS-CoV-2 infections. This study provides the prototype for a global surveillance system to inform public health practice, including novel standardized metrics of COVID-19 transmission, for use in combination with traditional surveillance tools. Methods Dynamic panel data models were estimated with the Arellano-Bond estimator using the generalized method of moments. This statistical technique allows for the control of a variety of deficiencies in the existing data. Tests of the validity of the model and statistical techniques were applied. Results The statistical approach was validated based on the regression results, which determined recent changes in the pattern of infection. During the weeks of August 17-23 and August 24-30, 2020, there were substantial regional differences in the evolution of the US pandemic. Census regions 1 and 2 were relatively quiet with a small but significant persistence effect that remained relatively unchanged from the prior 2 weeks. Census region 3 was sensitive to the number of tests administered, with a high constant rate of cases. A weekly special analysis showed that these results were driven by states with a high number of positive test reports from universities. Census region 4 had a high constant number of cases and a significantly increased persistence effect during the week of August 24-30. This change represents an increase in the transmission model R value for that week and is consistent with a re-emergence of the pandemic. Conclusions Reopening the United States comes with three certainties: (1) the “social” end of the pandemic and reopening are going to occur before the “medical” end even while the pandemic is growing. We need improved standardized surveillance techniques to inform leaders when it is safe to open sections of the country; (2) varying public health policies and guidelines unnecessarily result in varying degrees of transmission and outbreaks; and (3) even those states most successful in containing the pandemic continue to see a small but constant stream of new cases daily.

SARS-CoV-2 Surveillance in the Middle East and North Africa: Longitudinal Trend Analysis

“…This study is part of a broader research program at Northwestern University Feinberg School of Medicine, The Global SARS-CoV-2 Surveillance Project: Policy, Persistence, & Transmission . This research program developed novel surveillance metrics to include rates of speed, acceleration, jerk, and 7-day persistence [ 3 , 36 ]. We have also derived surveillance metrics for all global regions.…”

“…An empirical difference equation was specified in which the number of new positive cases in each country on each day is a function of prior numbers of cases, level of testing, and weekly shift variables that measure whether the contagion was growing faster, at the same pace, or slower than the previous weeks. This resulted in a dynamic panel data model that was estimated using the generalized method of moments (GMM) approach by implementing the Arellano-Bond estimator in R (The R Foundation) [ 3 , 36 , 39 ].…”

“…The objective of our research is to provide additional surveillance metrics to add to the public health arsenal using dynamic panel modeling and method of moments to minimize sampling bias by measuring significant negative or positive weekly shifts in the increase, decrease, or plateaued transmission of SARS-CoV-2. We will apply novel indicators derived specifically to inform policy makers about the COVID-19 pandemic [ 3 , 36 ], including weekly shifts in the pandemic, speed, acceleration and deceleration, jerk, and 7-day persistence. These dynamic metrics will better inform public health policy when combined with standard surveillance measures.…”

Dynamic Panel Data Modeling and Surveillance of COVID-19 in Metropolitan Areas in the United States: Longitudinal Trend Analysis

“…This paper uses the dynamic panel data (DPD) modeling application and surveillance approach [ 50 , 51 ], which has been applied to sub-Saharan Africa, South Asia, and at the state level in the United States [ 52 - 54 ]. We apply these methods to the 25 most populous US metropolitan areas, as defined by the US Census Bureau [ 55 ].…”

“…Our surveillance method can quickly be applied to existing data to generate indicators of pandemic outbreak locations and where explosive growth is likely to occur. The DPD-based modeling and surveillance method was validated previously [ 50 , 51 ]; these papers contain more detailed explanations and additional references to other methods.…”