Thread by @Farzad_MD

1/ Is there is any evidence that stay-home orders put into place in the past week or two have had a measurable impact on slowing the outbreak?

Where would we see it first? (We need better eyes!)

We *may* have some early signals in public data that infections in NYC slowed 🙏

2/ There are three sources of timely surveillance data we might look to, to understand the current dynamics of the outbreak, and all three are flawed right now. WE NEED TO FIX THESE FLAWS URGENTLY.

a) Case growth
b) Positivity rate
c) ED visits

3/ Case growth.

As you can see, it looks like the slope of the exponential line has slackened just a bit. That's actually a huge deal, if doubling time goes from 2 days to 5 days.

But to build a proper epi curve you need to plot cases by date of onset. cdc.gov/training/quick…

4/ if you do a lot more tests, old infections could show up as recently reported cases

The curve could look much different if you could reconstruct it properly by onset date- without the lag in testing and reporting

WE DON'T HAVE DATE OF SYMPTOM ONSET FOR REPORTED CASES
#FixIt

4/ At a minimum, reporters, please

Don't ask for how many new cases reported today, ask for the curve of cases *plotted by date of diagnosis* this is available.

Then we can see if there is a flattening of the daily growth rate.

5/ How can we get date of symptom onset, you ask?

We need to set up a systematic data collection system that goes along with lab testing in some controllable site

Dedicated testing sites, sentinel ILI surveillance, selected volunteer EDs collect 6 key Qs on every patient tested

6/ What are the 6 questions? (the CDC case report form has > 100 questions, basically not being used anymore)

1) Any comorbidities
2) Any Symptoms (cough, fever, shortness breath, taste)
3) Date of onset
4) Any exposure to known case
5) Any recent travel
6) High risk occupation?

7/ And you would ideally collect this from everyone being tested, so we can compare negatives to positives.

someone will build the app, I'm sure. @aneeshchopra will make it FHIR-enabled

We need smart coronavirus testing, not just more testing
statnews.com/2020/03/24/we-…

8/ The next type of data we would look for is to see if the rate of tests returning positive has changed. This can help account for testing intensity

Here's New York State in recent days. The % positive rate's come down from peak

(earlier low volume days are too noisy)

9/ The problem with this data is that we don't know who's being tested, and how that's changing over time.

If the test was only offered to hospitalized patients with pneumonia, then will have very high positive rate. If to asymptomatic people, then expect lower positive rate

10/ Again, what we need isn't just more TESTS, TESTS, TESTS

We need a consistent channel of testing that we can follow over time.

Again, I would suggest that sentinel Emergency Departments would be ideally suited to be able to serve this role, collect necessary info on severity

sorry- reconnecting threads

https://twitter.com/Farzad_MD/status/1243933012779315201?s=20

11/ So far, we see some suggestive evidence that NY outbreak *may* be abating after stay-home order- a flattening of exponential case growth, a drop in percent test positivity (despite flaws in both data sources)

here's the 3rd system-Syndromic Surveillance of ED ILI/resp visits

12/ For the first time since I started following the spike in COVID-related syndromic surveillance, there's been an apparent drop in the number of visits (not due to weekend effect). Same thing for the respiratory syndrome.

13/ The drop seems to be relatively consistent across age groups.

...and we all owe a huge debt of gratitude to @nycHealthy for maintaining transparency amidst uncertainty. this graph is theirs

I would love to see that more broadly, including w federal surveillance data

14/ The flaw in this system is that it's non-specific right now

We need to improve the signal to noise by fine-tuning it to COVID symptoms, removing noise due to worried well and/or other viruses, using age distribution to sharpening focus

And we need to be able to investigate

15/ In NYC context right now, I'd be cautious
a) It's just one day
b) lots of people have left NYC
c) health-seeking behavior could be shifting

But to me, ED visits not going up inexorably is a huge relief-If infections were up 30% 5 days ago, would expect to see it in ED visits

16/ Finally, the timing is about what you would expect to see a drop in symptomatic people getting sick enough to go to the ED- about 6 days after the stay-home order was put into place.

Interestingly, the drop in ED-ILI for 5-17 yo kids came about 6 days after schools closed

17/ We should be looking hard to see if NYC ED trend continues for Friday's data, and then again for Monday (weekend can be misleading).

I would expect new hospitalizations to drop soon.

New ICU admissions would begin to go down next week, though cumulative burden would build

18/ It can be hard to appreciate just how much misery is still to come, coiled up in infections already brewing, sicknesses intensifying, ARDS building.

In my toy model of infection dynamics, fully half of hospitalizations and 2/3 of ICU stays are yet to come- under best case

19/ I'd be curious to see what real modelers like Nathaniel Hupert are projecting, and how that might change if inflection point was reached.

It also elides the truth that the outbreak may continue to burn in pockets/ ethnic communities that aren't or can't protect themselves

20/ My toy model estimate is that we may have about 500,000 symptomatic infections already in NYC to produce this dynamic of ED visits and hospitalizations- maybe a million people already infected

If stay-home order had been placed 3 days sooner, it might have been half that?

21/ If this is validated, I hope the lesson for policymakers is not that we can lift stay-home orders if there is no syndromic spike

But if a region is seeing a spike in resp/ILI ED visits, there should be NO DELAY in instituting stay-home measures

How many regions is that now?