Thread by @joshuasweitz

One more COVID-19 related modeling update for the week; this time more conceptual in scope, but something that has been on my mind for weeks: peaks - whether they are ahead of us, behind us, or whether they here at all.

To start -- a perusal of any number of sites suggests that fatalities have gone up rapidly in many places, but then have lingered, via plateaus and long shoulders, here is a subset of country-curves from @FT

ft.com/coronavirus-la…

The y-axis is key insofar that large impacts of #COVID19 with over 200,000 reported global fatalities still means that the vast majority of individuals remain susceptible.

Which also means that we are nowhere near 'herd immunity', nor (as @CT_Bergstrom and @nataliexdean wrote today) should we aspire to be:

nytimes.com/2020/05/01/opi…

But this also means that a 'peak' in cases does not imply an absence of risk.

If interventions work, then the population still remains susceptible to infection and a resurgence of cases.

This is precisely the scenario that @neil_ferguson and colleagues at @imperialcollege warned in their seminal report... of a second wave to follow widespread public health interventions.

imperial.ac.uk/media/imperial…

Yet, for a disease that is already the documented cause of far too many deaths, we ask: what is the anticipated shape of an epidemic if individuals modify their behavior in direct response to the impact of a disease at the population level?

Extending a SEIR framework, we find that dynamics with awareness of fatalities can lead to significant reduction in transmission and fatalities, but also to peaks that linger or lead to persistent shoulders.

This model assumes that significant reductions in awareness-driven transmission initiates closer to a critical level of fatalities/day controlled by a sharpness exponent.

But irrespective of details: awareness matters after (exponentially) fast take-offs.

This type of awareness driven change builds on work by @sbfnk and colleagues from over a decade ago, as well as more recent work w/@ceyhuneksin.

We also find that the duration of awareness matters.

Long-term awareness drives rapid declines after reaching a peak, while short-term awareness can lead to plateaus, shoulders, and oscillations.

We will soon learn more about how we collectively respond to the opportunity to go back to 'business as usual', and may find out that many of us do not, precisely because we are aware of the risk that remains even when the 'peak' has not even arrived...

Also, quite relevant to our modeling efforts is a thoughtful thread by @trvrb from yesterday on local/individual responses questioning the notion of peaks when we often see plateaus:

https://twitter.com/trvrb/status/1255976675252158465?s=20

Although the models we developed are simple, we contend that asymmetric post-peak dynamics of COVID 19, including slow declines and plateau-like behavior, may be an emergent property of awareness-driven epidemiological dynamics.

Many thanks to collaborators @sang_woo_park, @ceyhuneksin, and @jd_mathbio:

Write-up:
github.com/jsweitz/covid1…

Code:
github.com/jsweitz/covid1…

(and spelling 'whether they're here at all' correctly would have been helpful!)