Modeling Tool Informs Local Epidemiological Projections

May 17, 2020

By Elvin Geng, MD, MPH, professor, Division of Infectious Diseases, School of Medicine; director, Center for Dissemination and Implementation, Institute for Public Health

The end of shelter-in-place policies in Missouri has led to a new wave of uncertainty about the future: Will the epidemic get worse? If so, how quickly? And if it gets worse, how quickly will we know and how bad will it get?

To help understand the future, particularly at a local or regional level, teams of investigators from UC Berkeley, UCSF and Washington University created the Local Epidemiological Modeling for Management and Action (LEMMA) model of the COVID-19 epidemic.

Read Dr. Geng's full perspective of the COVID modeling tool here.

LEMMA is a simple compartmental model using an Approximately Bayesian Computation approach to the COVID-19 epidemic.  Unlike other prominent and highly sophisticated models, LEMMA is able to accept locally relevant information about the population, interventions and other information —  and thereby give projections tailored to a particular region or area.  Such local modeling is useful when epidemic conditions differ markedly from the larger average.

The Center for Dissemination and Implementation is dedicated to advancing the use of evidence-based interventions in public health and clinical practice.  Creating tools to enable adaptations and tailoring to context is an important approach to making scientific findings relevant and useful.

Over the last month, the LEMMA model has been used by both local health systems and public health authorities to understand the epidemic trajectory to date in the St. Louis region.  We’ve learned several things.

First, these analyses suggest that the St. Louis region has undoubtedly flattened the curve.  Early in March, the reproductive number was around three (3), if the pace of had been carried forward without change, by May 1, this reproductive number would have yielded a total of hundreds of thousands of active cases, with tens of thousands in the hospital.  Instead, on May 1, we had approximately 600 cases in the hospital in the region.

The numbers 600 and 10,000 seem inconceivably far apart until we consider that epidemics grow on an exponential scale.  For 600 to grow to 10,000 is less than four doubling cycles (which with COVID19 in early March would scarcely have taken three-weeks). While the tragedy of the last weeks should not be understated (including unacceptable obvious racial disparities), delayed or absent social distancing policies would have led to far worse outcomes.

Even though we can be applauded as a community for flattening the curve, the present reproductive number is still precariously near the epidemic threshold of one – and this has consequences.  If after lifting of social distancing, the number rises to greater than one, and each infected individual gives rise to more than one additional infected person, the epidemic will expand, and possibly do so rapidly because growth happens on an exponential scale.

The reproductive number is a function of three things: the probability of infection per contact, the average number of contacts per time, and the duration of infectivity.  A 10% to 20% increase in contacts due to relaxing of social distancing – an amount that seems entirely plausible – would lead to a ~ 10% to 20% rise in the reproductive number without concomitant changes in the other factors. If on May 15 the reproductive number rises to 1.1, the hospital census could reach 1,000 by late summer, and possibly higher (Figure).  Higher rises in contact and reproductive numbers to 1.2 or 1.3 would lead to far worse epidemic conditions if further tightening did not occur.

These projections suggest a number of implications.  First, if we are to avoid another wave, we must aggressively seek to offset the anticipated rise in contacts with successful reduction of infections per contact (through masking) and the duration of infectivity (isolation, quarantine, contact tracing).  Backing off shelter-in-place means a full court press for other strategies.

Second, even if we do what we can, we still need to anticipate that these measures will most likely not keep the reproductive number under one.  We should anticipate a second wave.  The Fall will be particularly complex as the prevalence of other respiratory viruses that could be mistaken for COVID-19 rises dramatically. Even if we make it through the summer, we may well be intensifying social distancing practices again by then.  Planning ahead can diminish the costs of doing so when we do have to intensify.  Third, planning future wave measures mean that we need to prepare and equip citizens, especially the most vulnerable, to withstand both the risk of infection and the socio-economic fallout of distancing policies.  When certain individuals do not adhere to social distancing policies, many assume the reason is motivation, but more often than not, it is a problem of capability.  The fact that this epidemic is not over must motivate urgent efforts to protect those in vulnerable social contexts. Finally, we must invest in a multi-pronged monitoring architecture to detect changes in the epidemic course.  This means rapidly scaling up testing, monitoring of cases and new hospitalizations by Zip Code, enhancement of syndromic surveillance, and exploration of mobility data and novel approaches such as monitoring of sewage systems.  No single source of information will be perfect, and we need to put up as many antennae as we can, to detect where and when COVID is beginning to spread.

The COVID-19 pandemic has introduced vast uncertainty in the lives of nearly every person on earth.  Despite our desire for a crystal ball, modeling does not necessarily help us see the future. But by helping to make the consequences of our choices now clear, they should help us prepare.  In the words of Maya Angelou, while we hope for the best, we must prepare for the worst, and not be surprised by anything in between.

Figure: Projections from LEMMA model under three scenarios for the St. Louis metropolitan area (population ~ 2.8 million). Panel A: Continue current social distancing practices as they stood on at the end of April. The projections suggest a slow decline over the summer. Panel B: On May 15, relax social distancing by ~ 10%-20% for an effective reproductive number of approximately 1.1. This scenario yield 1000 hospitalizations by late summer. Panel C: On May 15, relax social distancing by approximately 30% so that effective reproductive number is between 1.2 and 1.3. A 30% rise in reproductive number would quickly lead to unacceptable levels of hospitalizations. The blue line is the median trajectory of all combinations of supplied inputs and constrained by observed hospital case series (triangles) from the region. Hospitalization numbers are entered as a range given uncertainties on any given day from persons under investigation.

















Panel A:

Panel B:











Panel C: