This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on http://www.jmir.org/, as well as this copyright and license information must be included.
The emergence of SARS-CoV-2, the virus that causes COVID-19, has led to a global pandemic. The United States has been severely affected, accounting for the most COVID-19 cases and deaths worldwide. Without a coordinated national public health plan informed by surveillance with actionable metrics, the United States has been ineffective at preventing and mitigating the escalating COVID-19 pandemic. Existing surveillance has incomplete ascertainment and is limited by the use of standard surveillance metrics. Although many COVID-19 data sources track infection rates, informing prevention requires capturing the relevant dynamics of the pandemic.
The aim of this study is to develop dynamic metrics for public health surveillance that can inform worldwide COVID-19 prevention efforts. Advanced surveillance techniques are essential to inform public health decision making and to identify where and when corrective action is required to prevent outbreaks.
Using a longitudinal trend analysis study design, we extracted COVID-19 data from global public health registries. We used an empirical difference equation to measure daily case numbers for our use case in 50 US states and the District of Colombia as a function of the prior number of cases, the level of testing, and weekly shift variables based on a dynamic panel model that was estimated using the generalized method of moments approach by implementing the Arellano-Bond estimator in R.
Examination of the United States and state data demonstrated that most US states are experiencing outbreaks as measured by these new metrics of speed, acceleration, jerk, and persistence. Larger US states have high COVID-19 caseloads as a function of population size, density, and deficits in adherence to public health guidelines early in the epidemic, and other states have alarming rates of speed, acceleration, jerk, and 7-day persistence in novel infections. North and South Dakota have had the highest rates of COVID-19 transmission combined with positive acceleration, jerk, and 7-day persistence. Wisconsin and Illinois also have alarming indicators and already lead the nation in daily new COVID-19 infections. As the United States enters its third wave of COVID-19, all 50 states and the District of Colombia have positive rates of speed between 7.58 (Hawaii) and 175.01 (North Dakota), and persistence, ranging from 4.44 (Vermont) to 195.35 (North Dakota) new infections per 100,000 people.
Standard surveillance techniques such as daily and cumulative infections and deaths are helpful but only provide a static view of what has already occurred in the pandemic and are less helpful in prevention. Public health policy that is informed by dynamic surveillance can shift the country from reacting to COVID-19 transmissions to being proactive and taking corrective action when indicators of speed, acceleration, jerk, and persistence remain positive week over week. Implicit within our dynamic surveillance is an early warning system that indicates when there is problematic growth in COVID-19 transmissions as well as signals when growth will become explosive without action. A public health approach that focuses on prevention can prevent major outbreaks in addition to endorsing effective public health policies. Moreover, subnational analyses on the dynamics of the pandemic allow us to zero in on where transmissions are increasing, meaning corrective action can be applied with precision in problematic areas. Dynamic public health surveillance can inform specific geographies where quarantines are necessary while preserving the economy in other US areas.
The emergence of SARS-CoV-2 in 2019 led to its associated disease, COVID-19, becoming one of the most severe and widespread disease outbreaks in modern history [
Timeline of COVID-19 in the United States. CDC: Centers for Disease Control and Prevention; FDA: Food and Drug Administration.
It is logical that the United States experienced technical difficulties at the beginning of the pandemic because we were dealing with a new virus, a pandemic unprecedented in modern times, and uncharted territory [
Without a national plan to control COVID-19, some mayors and governors implemented policies such as public masking mandates, prohibitions on gatherings, school closings, restrictions on commercial activities, and broad social distancing measures in an attempt to “flatten the curve” [
Research has explained why the Unites States consistently leads the world in COVID-19 cases and how we lost control of the outbreak; however, what has not been widely recognized is the need for ongoing systematic public health surveillance to gain control of the pandemic [
The objective of this paper is to use static and dynamic surveillance metrics to measure the caseload and dynamics of the COVID-19 pandemic at the national and state level [
This study relies on a longitudinal trend analysis study design to investigate the transmission of COVID-19 at the national and the state level over time. The COVID Tracking Project [
Arellano-Bond estimation of difference equations has several statistical advantages: it allows for statistical examination of the model’s predictive ability and the validity of the model specification, it corrects for autocorrelation and heteroscedasticity, it has good properties for data with a small number of time periods and large number of states, and it corrects for omitted variables issues and provides a statistical test of correction validity. With these advantages, the method is applicable to ascertaining and statistically validating changes in the evolution of the pandemic within a period of a week or less, such as changes in the reproduction rate. Oehmke et al [
To ascertain whether pandemic growth was explosive, we examined whether the acceleration and jerk were positive and if the persistence effect was positive as well as if it was larger than the speed. We examined these indicators week over week for 7 weeks along with the daily caseload. The persistence effect is an indicator for mathematically explosive growth (ie, it indicates if the difference equation has a solution that lies outside the unit circle), but large positive acceleration and jerk as well as persistence are indicative of explosive growth in a practical sense (ie, the COVID-19 caseload is expected to increase much more rapidly than in the current or recent week). However, this is a forward-looking statement and must be interpreted with due circumspection. To discuss potential explosive growth, we looked at five data points including speed, acceleration, jerk, persistence, and daily caseload each week for 7 weeks.
We present regression results for the panel comprising all 50 US states and the District of Columbia in
Arellano-Bond dynamic panel data model of COVID-19 dynamics at the state level.
Variable | Values | |
7-day lag, coefficient | 1.09 | <.001 |
Cumulative tests, coefficient | 0.01 | .48 |
7-day lag shift, coefficient | 0.108 | <.03 |
Wald statistic for regression, chi-square ( |
13,504 (8) | <.001 |
Sargan statistic for validity, chi-square ( |
48.3 (367) | >.99 |
The regression Wald statistic was statistically significant (χ28=13,504,
The coefficients on the 7-day lag were both positive and statistically significant (
The lagging indicators and shift parameters suggested recent change in disease transmission in the United States between September 28 and November 15, 2020. Specifically, the most recent 7-day lag this week was 10% faster than last week. The shift in the most recent 14 days, or 2 weeks, was positive and significant (
State pandemic dynamics
Date, State | Speed | State | Acceleration | State | Jerk | State | 7-day persistence effect | State | 7-day moving average | |
|
||||||||||
|
ND | 51.9 | SD | 4.6 | SD | 5.4 | ND | 55.7 | TX | 4083.1 |
|
SD | 49.1 | AK | 2.6 | AK | 1.6 | SD | 42.4 | CA | 3292.4 |
|
WI | 42.7 | ID | 1.7 | MO | 1.5 | WI | 37.7 | WI | 2489.0 |
|
UT | 30.9 | MT | 1.3 | ID | 1.2 | UT | 31.4 | FL | 2250.7 |
|
MT | 30.2 | WI | 1.3 | TN | 1.0 | OK | 31.2 | NC | 2102.9 |
|
AR | 27.3 | MO | 1.0 | NV | 0.8 | AR | 30.1 | IL | 2051.4 |
|
||||||||||
|
ND | 59.4 | ND | 2.9 | VA | 2.7 | ND | 56.5 | TX | 4184.7 |
|
SD | 52.9 | MT | 2.5 | ND | 2.0 | SD | 53.4 | CA | 3016.1 |
|
WI | 42.6 | VA | 2.3 | IA | 1.8 | WI | 46.6 | WI | 2477.7 |
|
MT | 42.5 | UT | 2.2 | UT | 1.7 | UT | 33.6 | FL | 2363.7 |
|
UT | 35.2 | SC | 1.9 | RI | 1.6 | MT | 32.9 | IL | 2215.7 |
|
ID | 29.6 | TN | 1.5 | NJ | 1.4 | AR | 29.8 | NC | 1784.0 |
|
||||||||||
|
ND | 80.0 | SD | 4.7 | SD | 7.2 | ND | 64.6 | TX | 4002.0 |
|
SD | 74.3 | MS | 3.6 | MS | 6.7 | SD | 57.6 | CA | 3371.6 |
|
MT | 56.9 | ND | 3.4 | LA | 4.6 | WI | 46.3 | WI | 3093.6 |
|
WI | 53.1 | NE | 2.1 | KY | 4.4 | MT | 46.2 | IL | 3031.3 |
|
UT | 37.9 | WI | 1.9 | AK | 3.1 | UT | 38.3 | FL | 2648.1 |
|
NE | 37.4 | AL | 1.8 | MT | 3.0 | ID | 32.3 | NC | 1934.6 |
|
||||||||||
|
ND | 101.2 | AL | 3.9 | SD | 7.5 | ND | 87.1 | TX | 5041.7 |
|
SD | 80.8 | ND | 3.4 | MO | 7.4 | SD | 80.9 | IL | 4155.4 |
|
MT | 62.9 | SD | 2.8 | ND | 7.2 | MT | 61.9 | WI | 3528.7 |
|
WI | 60.6 | MT | 2.7 | AL | 6.2 | WI | 57.9 | FL | 3231.6 |
|
ID | 46.0 | ID | 2.7 | MT | 2.5 | UT | 41.3 | CA | 3189.0 |
|
NE | 43.1 | RI | 2.7 | FL | 1.9 | NE | 40.8 | TN | 2154.3 |
|
||||||||||
|
ND | 113.9 | ND | 6.3 | WI | 4.3 | ND | 110.2 | TX | 5960.0 |
|
SD | 112.8 | WI | 3.7 | CT | 3.0 | SD | 88.0 | IL | 5203.6 |
|
WI | 75.8 | MN | 3.3 | WY | 1.8 | MT | 68.5 | CA | 4597.1 |
|
MT | 69.7 | CT | 3.3 | NV | 1.3 | WI | 66.0 | WI | 4411.7 |
|
WY | 58.9 | MO | 2.8 | MO | 1.2 | ID | 50.1 | FL | 3717.9 |
|
AK | 51.7 | MI | 2.7 | MN | 1.2 | NE | 47.0 | MI | 2852.4 |
|
||||||||||
|
ND | 163.2 | IA | 9.3 | SD | 11.2 | ND | 134.6 | IL | 7654.1 |
|
SD | 131.6 | ND | 5.9 | IA | 7.0 | SD | 132.9 | TX | 6882.0 |
|
WI | 90.5 | SD | 5.8 | MS | 4.2 | WI | 89.5 | WI | 5266.6 |
|
MT | 81.7 | NE | 4.9 | RI | 2.4 | MT | 82.1 | FL | 4577.9 |
|
WY | 71.6 | UT | 4.3 | IL | 2.4 | WY | 69.4 | CA | 4524.6 |
|
IA | 71.3 | IL | 4.0 | KY | 2.2 | AK | 61.2 | MI | 4031.4 |
|
||||||||||
|
ND | 175.0 | WY | 18.3 | WY | 25.5 | ND | 195.3 | IL | 11,827.4 |
|
SD | 154.5 | SD | 10.7 | LA | 12.8 | SD | 157.5 | TX | 8406.7 |
|
WY | 125.1 | LA | 9.7 | ND | 6.5 | WI | 108.3 | CA | 6719.0 |
|
WI | 112.9 | MN | 8.3 | MN | 5.6 | MT | 97.8 | WI | 6571.6 |
|
IA | 112.6 | UT | 5.0 | UT | 4.0 | WY | 85.7 | MI | 5845.7 |
|
NE | 103.8 | ND | 4.9 | SD | 3.8 | IA | 85.3 | OH | 5612.4 |
Although Texas and California ranked in the top six states for daily average of new infections for 7 consecutive weeks, they did not rank high for rates of speed, acceleration, jerk, or 7-day persistence. Florida ranked in the top for daily average of new infections for 6 of the 7 weeks, and in the middle of the study during the week of October 22, 2020, Florida had a positive jerk of 1.9 per 100,000 people, indicating that the pandemic was not just accelerating but accelerating at an increasing rate. North Carolina ranked in the top six states for novel infections for 3 of the 6 weeks but did not rank high for rates of speed, acceleration, jerk, or persistence. Illinois was ranked in the top six states for novel infections all 7 weeks; during the week of November 5, Illinois ranked high in acceleration at 4.0 additional new daily infections per 100,000 people per day, or 28 per week, and had a positive jerk of 2.4 infections per 100,000 people during the same week, indicating that acceleration increased from 1.6 in the prior week to 4.0 the week of November 5.
In
South Dakota has 24 positive upward trending data points for speed, acceleration, jerk, and persistence week over week for 7 weeks, indicating explosive growth. South Dakota had the second highest rate of novel infections and persistence in the United States; for 5 of the 7 weeks, North and South Dakota’s acceleration and jerk were positive, indicating that the pandemic was strengthening at an increasing rate. Utah has 12 data points in
Arizona is more promising. Although Arizona ranked sixth in the rate of new infections during the week of October 1, 2020, and had a leading persistence rate for the weeks of October 1 and October 8, Arizona does not appear in the leading states in
Wisconsin had a high caseload, and it was in the top six states for speed, acceleration, and persistence. Wisconsin remained a leader in speed across all 7 weeks and cumulatively had the highest number of new infections per 100,000 people over those weeks. Wisconsin’s pandemic accelerated in 3 of the 7 weeks. During the week of October 1, 2020, Wisconsin had a speed of 42.7, and by the week of November 12, 2020, its speed had reached 112.9. Wisconsin ranked as a leading state all 7 weeks for persistence, indicating there was some underlying condition that persisted and echoed forward.
Wyoming showed the largest acceleration over the 7 weeks, with speed accelerating from 19.2 new daily cases per 100,000 people during the week of October 1, 2020, to 125.1 during the week of November 12, 2020.
Most populous US states.
State | Population as of 2020 |
California | 39,937,500 |
Texas | 29,472,300 |
Florida | 21,993,000 |
New York | 19,440,500 |
Pennsylvania | 12,820,900 |
Illinois | 12,700,381 |
US pandemic dynamics.
Date | Speed | Acceleration | Jerk | 7-day persistence effect |
Oct 1, 2020 | 12.91 | 0.12 | –0.12 | 13.99 |
Oct 8, 2020 | 13.85 | 0.41 | 0.13 | 14.05 |
Oct 15, 2020 | 16.03 | 0.34 | 0.11 | 15.08 |
Oct 22, 2020 | 18.39 | 0.45 | 0.24 | 17.45 |
Oct 29, 2020 | 22.97 | 0.63 | –0.16 | 20.03 |
Nov 5, 2020 | 28.09 | 1.24 | 0.18 | 27.11 |
Nov 12, 2020 | 39.38 | 1.45 | –0.34 | 33.62 |
Weekly US state statistics.
The United States has had an uncoordinated, decentralized response to COVID-19. Although traditional public health surveillance provides a static view of the pandemic, the data are limited by secondary bias from undercounting, reporting delays, testing issues, and other forms of contamination. The novel measures presented in this study take steps toward resolving these shortcomings and measure the dynamics of the pandemic. They also provide greater insight into the evolution of a pandemic, such as where COVID-19 is transmitted and whether rates of transmission are increasing. Measures like 7-day persistence control for incomplete case ascertainment and look retrospectively to understand current infection rates, and speed, acceleration, and jerk provide a dynamic perspective on future cases.
The data presented in Tables S5-S18 in
Overall, the state of the pandemic in the United States is concerning based on static surveillance and dynamic metrics. All 50 states and the District of Colombia have positive rates of new infections. Moreover, the rates of infection are higher and faster than at any other time since the onset of the pandemic.
COVID-19 has been politicized and polarized, flamed by social media and misinformation [
The US COVID-19 epidemic has rebounded and is accelerating rapidly with multiple outbreaks in most US states, indicating shortfalls in preparedness and a dearth of nonpharmaceutical control measures. Although standard surveillance techniques such as daily and cumulative infections and deaths are helpful, they have incomplete case ascertainment, err on the side of the most severe cases, and provide a static view of what has already occurred during the pandemic, which is less helpful in prevention. Public health policy that is informed by dynamic surveillance can shift the country from reacting to COVID-19 transmissions to being proactive and taking corrective action when indicators of speed, acceleration, jerk, and persistence remain positive week over week. Implicit within our surveillance is an early warning system that indicates when there is problematic growth in COVID-19 transmissions as well as to signal when growth will become explosive without action.
A public health approach that focuses on prevention can prevent major outbreaks in addition to endorsing effective public health policies and may resolve conflict between various government entities. Moreover, subnational analyses on the dynamics of the pandemic allows us to zero in on where transmissions are increasing, meaning corrective action can be applied with precision in problematic areas. Dynamic public health surveillance can inform specific geographies where quarantines are necessary, preserving the economy in other US areas.
Without a unified national policy to address COVID-19, individual cities and states have launched efforts to control the spread of the pandemic. Similar studies have found that states that imposed strict guidelines saw drastic reductions in COVID-19 spread and avoided significant increased case counts and fatalities [
Discord between New York City Mayor Bill De Blasio and New York Governor Andrew Cuomo has resulted in delays in shutting down new virus hot spots in neighborhoods across New York City [
An important change in the dynamic of COVID-19 transmission occurred starting mid-August when universities and schools around the country started to open for their new academic year [
States such as Wisconsin and Indiana have reported pandemic highs in daily COVID-19 new infection counts [
The variation in speed, acceleration or deceleration, and jerk between and within states is consistent with varying degrees of compliance with public health guidelines to combat COVID-19 in terms of social distancing, masks, hand hygiene, and crowd control. Some states have overall higher caseloads because they have higher populations, such as California, New York, Illinois, Texas, and Florida, and others have alarming speed, acceleration, and jerk. Nationally, the US COVID-19 epidemic has rebounded and is accelerating rapidly in multiple states, indicating shortfalls in preparedness and a dearth of nonpharmaceutical control measures. Although standard surveillance techniques such as daily and cumulative infections and deaths are helpful, they also have incomplete case ascertainment, err on the side of the most severe cases, and provide a static view of what has already occurred in the pandemic, which is less helpful for future planning. Public health policy that is informed by dynamic surveillance can shift the country from reacting to COVID-19 infections and deaths to being proactive and taking corrective action when indicators of speed, acceleration, jerk, and persistence remain positive week over week.
Implicit within our surveillance is an early warning system that indicates when there is problematic growth in COVID-19 transmissions as well as signals when growth is likely. For example, at the US level, the novel metrics indicated a need for strengthening public health measures as early as the week of October 22, 2020, when the highest numbers of new daily cases were 82,000-84,000 per day. In the absence of these novel metrics, little action was taken until the number of new cases had more than doubled to over 161,934 new cases on November 17, 2020 [
A public health approach that focuses on prevention can prevent major outbreaks in addition to confirming when public health guidelines are effective and controlling the pandemic. Moreover, subnational analyses on the dynamics of the pandemic allows us to zero in on where transmissions are increasing, meaning corrective action can be applied with precision on problematic areas. However, this approach requires subnational, dynamic public health surveillance that can inform specific geographies where lockdowns or other measures are necessary. This paper provides novel surveillance measures that can help fill that exact need.
Our data are limited by state-level granularity and differences in testing and reporting within states. State-level granularity, although superior to national reporting, provides less detail than county- or city-level reporting. This limitation is particularly pronounced in states with large urban centers governed by powerful mayors. Testing and reporting varies across and within states for many reasons, including the decentralization of US health care. Both insurers and providers have inconsistent policies in place and resources deployed for COVID-19 testing. To address this need for small area surveillance, we have generated static and dynamic surveillance for larger metropolitan areas in the United States in an additional publication.
This study is part of a broader research program at Northwestern 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 [
Supplementary Tables S5-S18.
Explosive growth potential.
US weekly SARS-CoV-2 trends.
Weekly US 7-day persistence map.
Weekly US acceleration and jerk.
This study and the resulting surveillance system were funded in part by the Davee Innovations Research Endowment for North America. Any opinions, findings, conclusions, or recommendations expressed here are those of the authors alone and do not necessarily reflect the opinions of funders or authors' employers.
None declared.