Dynamic panel estimate–based health surveillance of SARS-CoV-2 infection rates to inform public health policy: Model development and validation

James Francis Oehmke, Theresa B. Oehmke, Lauren Nadya Singh, Lori Ann Post*

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

3 Scopus citations

Abstract

Background: SARS-CoV-2, the novel coronavirus that causes COVID-19, is a global pandemic with higher mortality and morbidity than any other virus in the last 100 years. Without public health surveillance, policy makers cannot know where and how the disease is accelerating, decelerating, and shifting. Unfortunately, existing models of COVID-19 contagion rely on parameters such as the basic reproduction number and use static statistical methods that do not capture all the relevant dynamics needed for surveillance. Existing surveillance methods use data that are subject to significant measurement error and other contaminants. Objective: The aim of this study is to provide a proof of concept of the creation of surveillance metrics that correct for measurement error and data contamination to determine when it is safe to ease pandemic restrictions. We applied state-of-the-art statistical modeling to existing internet data to derive the best available estimates of the state-level dynamics of COVID-19 infection in the United States. Methods: Dynamic panel data (DPD) models were estimated with the Arellano-Bond estimator using the generalized method of moments. This statistical technique enables control of various deficiencies in a data set. The validity of the model and statistical technique was tested. Results: A Wald chi-square test of the explanatory power of the statistical approach indicated that it is valid (χ210=1489.84, P<.001), and a Sargan chi-square test indicated that the model identification is valid (χ2946=935.52, P=.59). The 7-day persistence rate for the week of June 27 to July 3 was 0.5188 (P<.001), meaning that every 10,000 new cases in the prior week were associated with 5188 cases 7 days later. For the week of July 4 to 10, the 7-day persistence rate increased by 0.2691 (P=.003), indicating that every 10,000 new cases in the prior week were associated with 7879 new cases 7 days later. Applied to the reported number of cases, these results indicate an increase of almost 100 additional new cases per day per state for the week of July 4-10. This signifies an increase in the reproduction parameter in the contagion models and corroborates the hypothesis that economic reopening without applying best public health practices is associated with a resurgence of the pandemic. Conclusions: DPD models successfully correct for measurement error and data contamination and are useful to derive surveillance metrics. The opening of America involves two certainties: the country will be COVID-19–free only when there is an effective vaccine, and the “social” end of the pandemic will occur before the “medical” end. Therefore, improved surveillance metrics are needed to inform leaders of how to open sections of the United States more safely. DPD models can inform this reopening in combination with the extraction of COVID-19 data from existing websites.

Original languageEnglish (US)
Article numbere20924
JournalJournal of medical Internet research
Volume22
Issue number9
DOIs
StatePublished - Sep 2020

Keywords

  • COVID-19
  • COVID-19 guidelines
  • COVID-19 health policy
  • COVID-19 surveillance system
  • Dynamic panel data
  • Infectious disease modeling
  • Models
  • Reopening America
  • Surveillance

ASJC Scopus subject areas

  • Health Informatics

Fingerprint Dive into the research topics of 'Dynamic panel estimate–based health surveillance of SARS-CoV-2 infection rates to inform public health policy: Model development and validation'. Together they form a unique fingerprint.

Cite this