Thread by @mamund

"Georgia had an early surge of the virus, and now cases are spiking again. Brian Kemp has refused to learn a thing." here comes a thread on #COVID proxies, prediction, and information mismanagement #duringApandemic.
theatlantic.com/health/archive…

Georgia's problems are a microcosm of those experienced across the US. the intersection of policy, politics, information, and public health has -- for the most part -- gone badly.
theatlantic.com/health/archive…

The reasons for this failure are many but a few themes stand out including poor information management, confusing metrics, and a flawed attempt to predicting the virus' impact on the country.

let's first talk about information management. most states are unable to handle the volume and depth of data needed to clearly monitor the pandemic.
beckershospitalreview.com/data-analytics… (via @LauraMiller19)

some states, like California, are simply bungling the reporting of the data they have on hand. capradio.org/articles/2020/…

while others, like Florida, actively supressed COVID tracking data in an attempt to "control" the impact of the virus on the public perception.
jacksonville.com/story/news/cor…

errors of omission and comission have a simlar result, a poor quality of data that hinders decision-makers' efforts to protect and serve the public. *everyone* involved is complicit in speeding the spread of this deadly disease.

but clearing up the data, while important, begs the question. "What are we trying to do with this data? What is our gameplan? What is the expected outcome?" and this is where we need to talk about the idea of "Data Proxies"
en.wikipedia.org/wiki/Proxy_(st…

In statistics, proxy variables are "stand-ins" for variables in a study that cannot be easily measured. for example, GDP might be used as a proxy for measuing stanard of living. And, when it comes to COVID-19, data proxies are super important.

Imagine everyone who was infected had an orange dot on their forehead. And everyone who was immune from infection had a green dot on their forehead. It would then be easy to identify those who need to be isoalted and treated, and those who could safely interact with others.

The dots are proxies for the virus we cannot see. But people don't have dots on thier foreheads. so we need to look for other proxies to help us identify and treat COVID infections. and that's where the data comes in.

over the last five months, we've seen lots of attempts are focusing on a proxy varible to help us track the virus. things like "R-naught" and "positivity rates" have become a part of our daily vocabulary. but are these proxies helpful infighting infection?

"R-Naught" tells us about the rate of reproduction of the virus.
en.wikipedia.org/wiki/Basic_rep…

"Test Positivity Rates" tell us about the rate of _testing_ for the virus.
medicalxpress.com/news/2020-07-p…

neither one of those tell us much about the actual infection rate in a population.

other, more easily understood, proxies are the number of new cases, the number of active cases, and the number of virus-related deaths. most states report thse numbers every day. and, while accessible, they have limited value for managing COVID spread.

all these proxies share a feature which diminishes their value: time lapse. they report some aspect of the infection *from the past*. we can know how many people *were* infected, how many *have been* tested, how many *already died* from COVID-19.

and the value of each proxy is affected by the time-lapse involved. COVID deaths are a poor proxy for reducing infection since they are a long-lagging indicator. new cases are a better proxy but they still indicate who *already* has the disease.

if you want to _prevent_ the spread of COVID-19, you need to get ahead of it, block it from infecting others. and that's where the third element in our story comes up: prediction.

predicting the spread of COVID-19 has proved very tough. this is due partly to the fact that this virus has a long incubation (and infection) period -- the time before symptoms appear. Symptoms are a proxy. but COVID sufferers don't exhibit this proxy for up to two weeks.

without disease proxies we've tried to come up w/ other ways to predict the spread. the common focus is on the environment in which the virus thrives: indoor, poorly-ventilated rooms, and extended close exposure.
asiatimes.com/2020/07/covid-…

following these guidelines, we can "predict" that you're more likely to get infected in a crowded bar than an open-air market.
khn.org/news/packed-ba…

we can also predict that spending just a few minutes w/ someone outside is safer than working in a close-quarters office environment.
weforum.org/agenda/2020/05…

but these are not predictions of the virus' spread. they are proxies about where it can thrive; where is is likely to exist. and that's where we finally get to the reason proxies and prediction for COVID is so challenging.

the real thing we need to track and predict are _people_. the virus lives in people, it is spread by people's behavior, and it can be defeated by their behavior, too. *we* are the ones infecting each other.

so, predicting the spread of coronavirus is the work of predicting people's behavior. and predicting behavior is both complicated and complex.

think of the classic "murmuration of starlings" example.
theatlantic.com/photo/2019/01/…

starling flocks move in corellated but unpredicatable ways. this is also true for groups of people. fashion, fads, "ice bucket challenges", and even panic can spread through a community in unpredictable ways. so does COVID-19.

those of us engaged in activities w/ large groups for extended periods of time in close quarters, are "feeding" the epidemic.
livescience.com/covid-19-super…

those moving around (for vacations, commuting to work, etc.) are transporting the virus and growing the epidemic.
whdh.com/news/americans…

these are better proxies for the virus than statistics about where it has already been. and the people who do well at this kind of prediction are those who study "virality" of content online.
visme.co/blog/psycholog…

yes, "virality" is the realm of instagram, twitter, facebook, and others. they have their fingers on the pulse of communities; what goads them into action, what can start a trend -- or a panic.
sysomos.com/2016/12/22/imp…

i don't hear enough from these companies -- and others like them -- researching and publishing papers on how to use the virality of poeple's behavior as both a prediction and prevention for COVID-19.

If we want to help defeat this virus, we need to focus on prediction -- and influencing -- human behavior. we should focus on developing proxies that center around poeple's actions, not the past data of the infection's effect on us.

we need to track and report "prevention proxies". mask wearing is a prevent proxy, empty pubs is a prevent proxy, increases in carry out is a prevention proxy. and there are many more we can come up with.
ucsf.edu/news/2020/06/4…

we need to collect and manage data proxies that are _leading_ indicators of disease spread. we need to actively promote these prevention activities. i want to see nightly reports on these prevention proxies to help us track our success in defeating the virus.

in order to get ahead of this pandemic, all of us need to stop focusing on the past and face the future.
brainyquote.com/quotes/soren_k…

yo, @threadreaderapp, unroll, pro favor. #Thanks