COVID-19 DATA-DRIVEN MODELLING SERIES
Over the past several months, the COVID-19 pandemic has brought untold hardship and disruption to the world. Now more than ever, it behooves us all to do what we can to advance the greater good of our communities. As community leaders and the general public navigate unclear social and economic uncertainties, concise and unbiased information is paramount.
We at Hinman are fortunate to be positioned to provide valuable insights into the COVID-19 pandemic trajectory; this includes consequences associated with potential policy directions and the behavior of the general public. Because of our history in research and development, we can utilize machine learning, data-driven modeling, optimization, and uncertainty quantification to meaningfully assess COVID-19 data.
Hinman is inspired by the brave efforts of first responders and critical services workers throughout California. Our goal is to provide critical information to state leaders and communities as important decisions are required to be made in short order, such as reopening our society and economy. All of which, importantly, have lasting consequences on our health, economy, and way of life.
Looking forward – this is the first in a series of publicly released studies to be provided by Hinman. New data will be continuously collected and incorporated into our models. Additionally, future studies will be expanded to look at the degree of risk associated with reopening communities throughout the State of California at varying rates.
HINMAN RELEASE, 1.0
This initial baseline study has two primary objectives.
- First, to provide critical, clear, and unbiased information to the public regarding the current status and trajectory of the SARS-CoV-2 virus in the State of California.
- Second, to inform policymakers and the public alike of what the risk-based data driven modeling suggests is the most viable path forward until an effective COVID-19 vaccine is produced and is widely available.
This is an initial baseline study that will be expanded in the coming weeks to include additional findings. The model’s assumptions, parameters, and recommendations will be continuously updated to reflect new data as it emerges.
In light of Governor Newsom’s May 4, 2020 announcement that Phase II reopening will commence throughout the State starting May 8, 2020, this first release emphasizes the roles of social distancing easing and timing of policy changes on transmission, hospitalization, and morbidity in California.
METHODS & COVID-19 MODELLING
An initial baseline study has been performed on the transmission, hospitalization demands, and morbidity characteristics of the COVID-19 spread in California. The epidemiological model was generated through combination of swarm optimization with the Artificial Bee Colony (ABC) method and a modified compartmental SEIR model. ABC optimization is an efficient optimization method that seeks optimal solutions through mimicking the foraging behavior of honeybees. The optimized models were used to explore the relative risk and consequences of various social distancing strategies moving forward to protect against exponential-like growth in infections that may overwhelm California’s healthcare system. Open source data was obtained through the California Department of Public Health COVID-19 resources web page.
HUMAN BEHAVIORS – SOCIAL DISTANCING EFFECT ON PANDEMIC TRAJECTORY
A future wave of increasing infection rates is expected between late Fall 2020 and late Spring 2021, regardless of degree of social distancing and the duration to which it is relaxed. However, there is room for optimism. The severity of the future spikes in infections, as well as the duration of time between successive waves is greatly influenced by a combination of government policy and personal behavior.
A critical challenge for policymakers is to determine the optimal timing and magnitude of shelter-in-place and social distancing restriction rollbacks. It is critical to balance the risks associated with three factors – all of which carry either implicit or explicit social consequences.
- Opening the economy too late risks crippling the economy, leading to social hardship through lost income, lack of investment, and supply-chain disruption
- Opening the economy too early risks shorter duration between successive spikes in new infections that require reintroduction of strict shelter-in-place/social distancing measures. Reintroduction of strict measures only reinforces uncertainty sentiment and further degrades the State’s economic outlook through reduced investment and consumer confidence.
- Direct costs to human lives and well-being due to illness and morbidity.
This analysis focuses on the the role of timing of social distancing reductions and the degree to which social distancing is decreased on the COVID-19 trajectory. The metric of interest is the expected duration between easing social distancing and the need for reintroduction of strict social distancing measures to avoid exponential spread that would result in overtaxing the State’s healthcare system.
Figure 1 depicts the amount of time expected between rollbacks in social distancing and strict reintroduction for varying degrees of reduction and start time relative to May 8, 2020 – the first day of California’s Phase II reopening. The degree of social distancing, α, is the degree to which transmission is limited through lack of human to human contact. The value of α is ultimately a function of human behavior, though it is greatly influenced by government policy and restriction. The social distancing factor, α, can be thought of as a fractional measure of the number of human contacts that leads to infection compared to what would occur given no change in behavior – i.e. no social distancing or mitigating measures. For example, α = 0 would be indicative of an extreme scenario where no measures are taken to reduce transmission. α = 1 would represent the other extreme scenario where zero interactions may occur that lead to transmission, thus reducing the effective reproduction number to zero. α = 0.5 would correspond to a reduction in transmission-resulting interactions by 50%.
The social distancing factor, α, can be influenced by any number of behaviors that reduce transmission, such as wearing masks in public and maintaining safe distance from others. Other actions such as increased testing, combined with contact tracing, can have a significant impact on the number of additional people that may be infected by a given individual, thus increasing α and lowering the effective reproduction number. The degree to which α is influenced through specific interventions and behaviors will be quantified in future studies.
The current social distancing intensity of α = 0.53 translates to a 47% reduction in transmissions per each infected persons that has been achieved due to adherence to California’s current strict statewide shelter-in-place order. In other words, the effective reproduction number has been decreased by 47% – a monumental feat. It is observed that two critical thresholds exist, a long-term threshold at α = 0.47 and short-term threshold that varies between α = 0.38 and α = 0.42, depending on the delay in initiating reduction in social distancing. Anything greater than an 11% reduction from current mean social distancing levels will dip below the long-term critical threshold and will result in a meaningful drop in time expected until strict measures must be re-instituted. The short-term threshold represents a second boundary which marks the severe transition to greatly reduced time between phases of strict social distancing.
The severity in decline is seen to consistently reduce with increased delay in the rollback of social distancing. Figure 1 also depicts the relative level of risk associated with each combination of social distancing reduction and delay of onset. The risk scoring balances the expected outcomes of the time until strict measures must be reintroduced and the variability of those predictions. As expected, combinations of greater social distancing and delays in reduction lead to lower risk. Note that economic components of risk are not currently included in the model but will be explored in future releases.
There are four distinct zones of behavior relative to the outcomes predicted in Figure 1. These four zones are depicted in Figure 2 and Figure 3 and represent regions of different degrees of stability and outcome success. Higher stability regions reflect lower variability where small changes in social distancing behavior will not have outsized effects on the outcome. The effects of the short and long-term critical thresholds are apparent in Figure 3 where significant changes in outcomes are observed for slight changes in social distancing magnitude and timing. Lower proximity to these critical thresholds results in higher potential variability. The long-term critical threshold provides the boundary between Zone I and Zones II/III and the short-term critical threshold provides the boundary between Zone II and Zone IV.
CALIFORNIA’S PHASE II REOPENING
In light of Governor Newsom’s announcement that California will move into Phase II of the State’s reopening plan, the outcomes predicted in Figure 1 are depicted further, with an emphasis on May 8, 2020 as the start date of social distancing reduction. A start date of May 8th opens the possibility that the State may pass through both the short and long-term critical social distancing thresholds, each with significant effect on the spread of the virus.
Three cases of increased reduction in social distancing behavior are evaluated – see Figure 4. In each case, it was assumed that strict social distancing measures will be reintroduced after the rate of new infections reaches what was observed on March 19, 2020, the day of the initial statewide shelter in place order. Figure 5 and Figure 6 depict the trajectories of the COVID-19 pandemic for Case A and Case B, as well as the two additional extreme cases – (1) current social distancing remaining in effect indefinitely and (2) all social distancing is reduced to pre-pandemic levels indefinitely.
- Case A describes only a 10% reduction in social distancing, placing it immediately above the long-term critical threshold. New infection rates would be expected to remain near current levels through the Summer and Fall of 2020. Late November would begin to see increasing rates of new infections; strict social distancing measures would need to be reinstated in mid to late December. This would afford California residents and businesses over seven and half months of slightly reduced social distancing. Peak hospitalization rates would increase marginally beyond current levels, peaking in mid-March, 2021. Approximately 26,775 additional deaths would result beyond what would be expected if current social distancing behavior were maintained indefinitely. If strict social distancing reductions are not reintroduced in mid to late December, a significant resurgence of infections would peak in early May, 2021 and would max out current ICU bed capacity. 509,961 deaths would be expected.
- Case B describes a 20% reduction in social distancing behavior, which lies between the short and long-term critical thresholds. A drop in duration between phases of strict social distancing occurs from Case A, with a little less than six months of time expected between phases. A mild increase in infection rates would be observed after nearly two weeks with approximately linear growth until October, 2020, when the rate of new infections would begin to rapidly increase. A return to strict social distancing would be required at the beginning of November to mitigate against transition to steep and uncontrolled growth. Peak hospitalization rates would increase beyond current levels by approximately 300% and would reach approximately 56% of ICU bed capacity. Additional capacity would likely be required to include all non COVID-19 related ICU cases, especially during peak flu season. An increase of 46,837 deaths would be expected beyond what would occur with no reduction in social distancing. If strict social distancing were not to be introduced upon seeing an alarming uptick in new infections in October, hospitalizations would peak in March, 2021 and current ICU capacity would be exceeded by approximately 90% by COVID-19 cases alone. Approximately 511,047 deaths would result.
- Case C describes a 25% decrease in social distancing from current levels, placing it below the short-term critical threshold. As expected, a sharp decrease in effectiveness is seen when the short-term critical threshold is crossed. California’s current flat trajectory of new infections would continue for approximately two weeks before infection rates would noticeably increase. At two and a half weeks, a sharp increase would occur that would require almost immediate action to dampen the trajectory of the virus’s spread. Little change would be observed in peak hospitalization rates or morbidity due to the rapid need for re-instituting current levels of social distancing. Without return of strict social distancing measures, new infection rates would peak in mid-February, 2021 and ICU bed capacity would be exceeded by approximately 140%, and 510,432 deaths would be expected.
KEY OBSERVATIONS AND TAKEAWAYS
A number of insights into the likely trajectories of the pandemic and the origins of its spread in California are yielded from this initial study. Specific cases were depicted in the previous section to reinforce the key observations that are presented.
- Prior to widespread vaccination, rapidly increasing rates of COVID-19 will not be controlled without consistent social distancing.
- The current level of social distancing in California appears to be above the minimum required to maintain a controlled rate of new infections until a viable COVID-19 vaccine can be produced and distributed, but not by much.
- To avoid future uncontrollable spikes in new infections, only relatively modest reductions in social distancing levels may be introduced until sufficient vaccination is available. Otherwise, uncontrollable spread is expected to return with the need for reintroducing strict social distancing.
- California’s planned Phase 2 reopening measures appear to be appropriate due to the small change in interactions that have the potential to lead to transmission. It does not appear that the drop in social distancing intensity will fall below the critical thresholds depicted in Figure 1.
- Elimination of all social distancing measures (i.e. back to normal) is not economically or ethically viable. Over 500,000 deaths within 18 months would be expected in California.
- Results of this study are consistent with the conclusions from recent studies that the virus arrived in California and began to rapidly spread prior to the first reported infection. Data analysis suggests that SARS-CoV-2 began propagating within California’s population in December, 2019.
- Data analysis suggests that less than 10% of SARS-CoV-2 infections were reported prior to widespread public awareness in early March. An increase in reporting percentage of new cases is apparent as public perception of risk increased.
- California’s statewide shelter-in-place order was optimally timed, as depicted in Figure 7. Only a shallow decrease in infection rates would have resulted from earlier issuance of the order. However, significant increases in infection rates would have resulted from even modest delays in issuing the March 19th shelter-in-place-order.
Difficult decisions – The State of California faces a difficult task over the next year and beyond. Forging a path toward safe public health while moving toward economic recovery is the extraordinary challenge, made difficult by the uncertainties associated with community spread and of future human behavior. The timing and policies associated with California’s planned Phase 2 reopening is a positive step toward achieving these goals.
Focus on the goal – Strategies for easing restrictions must balance positive and negative effects on the health and financial well-being of California’s workforce and greater population. Neither the current level of economic shutdown, nor total elimination of social distancing measures would balance these objectives.
The data modeling performed thus far informs us that eliminating resurgence of the virus is not practical until an effective COVID-19 vaccine is widely distributed. The degree to which social distancing is maintained and the timing of potential reductions in strict social distancing are the principally influential factors in affecting the pandemic’s trajectory. Regardless of strategy, a second wave of infections is expected in early to mid-2021. However, the severity of secondary and other future waves is greatly influenced by human behavior – i.e. reduced human contact, which can be encouraged or mandated by public policy.
Data analysis suggests that California’s strict social distancing measures have successfully leveled the transmission of the virus, but a new normal may be expected until widespread vaccination is achieved. Such a new normal would include a combination of reduced (but still present) social distancing that would be accompanied by sustained progression deaths, albeit at a controlled non-exponential rate. Increases in testing, combined with a robust contact tracing program, will provide an opportunity to further reduce the effective reproductive number to keep California on a stable trajectory, even with a commensurate decrease in social distancing.
Viable strategies moving forward are rooted in acceptance that additional wave(s) of COVID-19 cases will emerge and that swift reintroduction of shelter-in-place orders will be needed if infection rates begin consistently increasing beyond those seen at the time of the original California shelter in place order on March 19, 2020. Neither complete elimination of social distancing measures, nor dependence on pre-vaccination herd immunity are viable strategies, as healthcare resources would be quickly overburdened and unacceptably high death tolls and economic disruption would follow. Restrictiveness of shelter-in-place orders may vary county-by-county but the most effective strategies to maximize time between spikes (i.e. maximize time without the most restrictive interventions) is for largely consistent shelter-in-place orders across geographic regions – especially those who’s inhabitants may readily travel between.
Future releases will incorporate new data and will include expanded areas of research. Some topics to be explored include the following:
- Correlate social distancing metric, α, with specific government interventions to inform specific policy selection.
- Interconnect county models to further explore the optimum combination of timing and degree of social distancing measures and government interventions. These combinations will be modelled as an interconnected set of open systems while considering county/regional-specific policies to balance economic and social risks.
- Apply machine learning and swarm optimization techniques to find most efficient combination of government initiatives and easing/tightening of restrictions for separate geographic regions of the State of California.
- Include economic factors to provide recommendations on risk-based policy decision making.
- Explore the effects of seasonal-based policies on social distancing.
The amount of time between return of strict social distancing measures will depend strongly on the vigilance of the general public, regardless of policy. Responsible behavior now to limit the virus’s spread is far preferable to the more severe restrictions and poor outcomes that would result from the alternative. Until an effective COVID-19 vaccination is widely distributed and administered, any point can be looked at as a potential inflection point in the virus’s transmission.
We applaud the State of California’s ongoing vigilance and public policy. The actions of the State Government and general public have saved hundreds of thousands of lives and have positioned us in a hopeful trajectory. That is, a condition with controlled infection rates that empower the general public to make responsible choices moving forward to have a direct impact on the health and economic future of their state.
Several assumptions are included to facilitate the numerical modeling that underpins this study. Current assumptions include the following:
- Vaccine research, trials, and production will be fast tracked to be widely available in 18-months. Development and production of an effective vaccine within 18 months is unprecedented but is considered nonetheless due to the significant resources being put to the task.
- Hospitalization demands that are listed are COVID-19 related only. Additional demands from non COVID-19 sources such as the flu would need to be added separately.
- Models currently compares hospital demands to capacity based on sum totals of hospital beds across all counties and additional capacity developed over time is not considered.
- Current simulations assume that social distancing measures revert to current levels upon reaching a new infection growth rate equal to the rate observed on March 19, 2020. The reintroduced social distancing intensity is maintained indefinitely for the purposes of this study.
Caig Y, Gaod D, Hee D, Lina Q, Loue Y, Musae S, Wangg W, Wangb M, Yangf S, Yangh L, Zhaob S. A conceptual model for the coronavirus disease 2019 (COVID-19) outbreak in Wuhan, China with individual reaction and governmental action. International Journal of Infectious Diseases. Volume 93, P211-216, April 01, 2020
“COVID-19.” COVID-19, California Department of Public Health, www.cdph.ca.gov/Programs/CID/DCDC/Pages/Immunization/ncov2019.aspx.
“COVID-19 Public Dashboard.” Tableau Software, CA.gov, public.tableau.com/views/COVID-19PublicDashboard/Covid-19Hospitals?:embed=y&:display_count=no&:showVizHome=no.
Lauer S, Grantz K, Bi Q, Jones F, Zheng Q, Meredith H, Azman A, Reich N, Lessler J. The Incubation Period of Coronavirus Disease 2019 (COVID-19) From Publicly Reported Confirmed Cases: Estimation and Application. Annals of Internal Medicine, 2020; DOI: 10.7326/M20-0504
Kissler S, Tedijanto C, Goldstein E, Grad Y, Lipsitch M. Projecting the transmission dynamics of SARS-CoV-2 Through the Postpandemic period. Science, doi:10.1126/science.abb5793, April 12, 2020
Rihan F, Anwar M. Qualitative Analysis of Delayed SIR Epidemic Model with a Saturated Incidence Rate. International Journal of Differential Equations. Volume 2012, Article ID 408637, doi:10.1155/2012/408637, December 6, 2012
Hinman is a data-driven company and a world leader in risk mitigation, committed to addressing emerging challenges and providing adaptable solutions in an ever-changing landscape. We have been trusted by federal, state, and regional agencies as well as private corporations for projects around the world. Whether the project pertains to a single facility, a complex operation, or a portfolio of properties, we help clients understand the level of investment required to adequately protect against known and unknown risk.
For more information on Hinman’s latest COVID-19 modelling efforts, contact the author, Kyle Haas, at firstname.lastname@example.org.