HomeMy WebLinkAbout07-06-20 Public Comment - T. Fiddaman - COVID in our communityFrom:Tom Fiddaman
To:Agenda
Subject:[SENDER UNVERIFIED]Fwd: Re: [bozone-orgs] COVID in our community
Date:Monday, July 6, 2020 10:25:01 AM
Attachments:oicokblilmjblmod.pngigmnegijnomemfeg.png
ceaalfhefjhnjgip.pngeaifmklmojiefnef.pnglklehnmcmjdnmfdp.pnganhndjjkioebpgcc.pngbnabllchbnoccdnb.png
nkiiignfgfjnjnlb.pnggimdldiafbiplgdi.png
Dear Bozeman public health managers,
I've copied a letter below that I shared with the Bozeman community, examining the state ofCOVID in Montana, from a fairly sophisticated model I run for another state.
Details aside, I don't think you need a model to realize that the following are true.
Infection growth in June was over half the rate that prevailed in the early epidemic, with adoubling time of 7 to 10 days, implying a reproduction ratio (R) substantially greater than1.Due to incubation, accumulation and reporting lags, infections today are the consequenceof behavior and policy a few weeks ago.Behavior and policy today are looser than at the end of May, i.e. there are more and riskiercontacts.
Therefore, R will remain well above 1, driving continued growth in cases, and it is likely that thegrowth rate itself will accelerate. There is no plausible way for the course of the disease tostabilize on its own, until prevalence is orders of magnitude higher.
So my question to you is, where are we headed? Will we allow case loads to rise until hospitalcapacity is 10% utilized? 100% utilized? At present growth rates, these milestones could bereached in less than 30 and 60 days, respectively.
The White House reopening guideline indicates that we should "Monitor conditions andimmediately take steps to limit and mitigate any rebounds or outbreaks by restarting a phase orreturning to an earlier phase, depending on severity." We are clearly at (or well past) that point.Aggressive testing and contact tracing and exhortations for good behavior simply are notworking. It is time to act with closures, stricter distancing, a mask requirement, and othermeasures.
A month is not much time to remobilize the population, change behavior, and expand measuresto protect the vulnerable. But if we don't act now, I fear that we will soon find ourselves in thekind of crisis that is now unfolding in other regions. I realize that a reversal of reopening is adifficult decision to make, but better to do it now, than later when rising infections triggervicious cycles that lower the effectiveness of testing, contact tracing and critical care.
Having worked on the same problems in another area, I know that you are under great timepressure, with many conflicting constituencies to manage. I wish you the best in making thesedifficult decisions.
Respectfully,
Tom Fiddaman
Bozeman
On 6/26/2020 6:43 PM, Tom Fiddaman wrote:
Presumably most of you have seen today's news reports - big case numbers, and nomask order:
https://www.bozemandailychronicle.com/coronavirus/gallatin-county-adds-9-new-cases-of-coronavirus/article_d1ab82db-ea4e-5c9a-9ca2-3f38fd905662.html
https://www.bozemandailychronicle.com/coronavirus/gallatin-county-health-board-declines-to-impose-new-restrictions-as-virus-cases-increase/article_993d8e5c-7b8e-5293-9a64-6165c03508cf.html
I've been running a COVID19 simulation model for another state, so I thought I'dshare some of our findings as they relate to this discussion.
Some tech details for what follows:
The model is an extended SEIR model, with additional states to better represent thedelays in the infection process and pre-symptomatic spread. There's an explicitdistinction between people who will have severe or mild/unnoticed symptoms. All50 states are included, so we learn from other locales with very different populationdensities and infection numbers. There's a mild temperature effect (benefitting usnow) and cross-state exposure from travel and commerce. We distinguish amongtesting of symptomatics, contacts, and general population surveillance, so that wecan back out the underlying true infections from the reported test results.
As a caveat, what I'll show below is a "best guess" simulation, but the uncertaintyon some of these estimates - especially future behavior - is large. Unfortunately Idon't have time to delve into that right now, so what I show here should be takenwith a grain of salt. Also, you should heed the usual George Box admonition that"all models are wrong, some are useful," but bearing in mind that mental models arenot immune to this warning. In a situation this complex, formal models properlyregarded beat mental models; paraphrasing Jay Forrester, if you have a model you'llbe the only person in the room who can speak for 20 minutes without self-contradiction.
In spite of the caveats, there are some things that are robustly true across oursimulations, and most other models I've seen. First, it's not the flu. It's 5 to 10 timesmore fatal. But it's not the plague either; the fatality rate is a lot less than what youget by dividing reported deaths by reported cases (2-6% typically). There are a lot ofasymptomatics, maybe 70-90% of all infections, but the idea that 99% areasymptomatic and we all had it back in November is impossible to square with thedata.
First, where are we?
Here's the data from a couple days ago (red) and our model (blue).
There was a huge increase of testing in March, followed later by (roughly) 2 moredoublings in May and June. So it's reasonable to wonder how comparable today'snumbers are to the first peak. It turns out that test coverage is not all that important.
Prevalence today is about what it was at the first peak, and the infection rate mayactually be higher.
Here's the model's estimate for the fraction of all cases that are confirmed throughtesting. In spite of the increase in testing in May and June, this has actually fallenfrom its peak in April. That happens when infections are growing, because you have
more people early in the infection chain, who (a) don't have symptoms yet, (b)haven't been found yet by contact tracing, or (c) don't have enough viral load to testpositive. Differential behavior, with people likely to have low symptoms taking onmore risk, might also be a factor.
One consequence of this is that June's case growth cannot be attributed to increasedtesting.
The relative risk level of behavior that's driving the epidemic appears to be asfollows:
This indicates that, in June, Montana had reverted to behavior that was about 10%less risky than the pre-COVID situation. This is far short of what it would take tostop the spread of infection, which requires at least a 50% reduction in transmission.(Contact tracing and isolation is evidently picking up some of the slack.) Googlemobility data is shown for comparison (red), and it's even less optimistic, in that itshows greater mobility than pre-COVID. Unacast and other cellular data vendorsare similar ( https://www.unacast.com/covid19/social-distancing-scoreboard?view=county&fips=30031 ). Cellular mobility data doesn't tell you anything abouthygiene and PPE, so it's not strictly comparable to the model estimate.
Note that there are delays in disease incubation, development of serious symptomsand reporting that limit our ability to observe recent behavior and infections. Thecases we see today are a product of behavior a couple weeks ago.
Is it us or them?
The model includes an estimate of the relative contribution of contacts withinMontana and interstate contacts to total infections. Early on, this had to be high(that's how the disease got here). It fell as community spread took off inMarch/April. Then it rose, as we successfully halted infection in April/May. At thatpoint, Montana had few cases, but was subject to intense pressure from other stateswith much higher prevalence, in spite of limited travel. But today, communityspread again seems to dominate. Bottom line: we can't blame our problems onFlorida.
What will the future bring?
This is hard to say, because it's mostly contingent on what we do. If you holdcurrent behavior constant through next March, the best-guess simulation is a secondwave that dwarfs the first, with several thousand deaths statewide.
If you take 10% of this burden (roughly Gallatin County's share of MT population),that's a pretty big surge (a cozy 2 or 3 people per bed at Deaconess). The surgedynamics here also neglect three vicious cycles (they're in the model, butdeactivated here). First, when the number of cases rises, test coverage falls, makingit harder to find and isolate cases, leading to more infections. Second, as the contacttracing caseload rises, it's harder for public health capacity to keep up, also leadingto more infections. Third, as hospital loads rise, you lose staff to infection andexhaustion, increasing the burden on remaining staff; quality of care declines, andfatality rates increase.
What about masks?
In one version of our model, we estimate the effectiveness of a list of state policies,using implementation dates compiled by a group at BU. For a few weeks, theparameter on public mask orders has looked very strong:
The problem is that the confidence bounds on this parameter are wide, including 0.Presumably this is because there aren't too many mask orders, compliance is hard toassess, and they're confounded with many other overlapping policies and behaviors.So we can't make a definitive statement; it's just suggestive. Unfortunately much ofthe other mask evidence is similar, but there's starting to be a lot of "suggestive"around. It would be nice to do a better job of triangulating the different assessments.
My Takeaways
We may be under significant external infection pressure, but that's not the principalproblem. We have met the enemy and they are us. Certainly no one else is going tostep in to save us.
Current behavior is not stopping the growth of infection. It has lowered the growthrate to about half what it was in early March, but it's still growing. It is virtuallycertain that a "pause" in reopening (as the Texas Governor put it) won't pause theepidemic, any more than you can pause a wildfire. Without some positive action,growth will continue until enough people are infected to provide herd immunity -that is, if there is any long term immunity.
The do-nothing strategy involves a substantial chance of a massive surge withconsiderable death and suffering. If you prefer to think of it economically, the lossescould easily amount to a year of Montana GDP. The extent depends a lot onwhether we can keep the disease out of skilled nursing facilities and othervulnerable populations. We haven't had any catastrophes yet ... but can we keep thatup with current resources and higher prevalence?
The thing is, we're not that far away from a better outcome. In the simulationsabove, the reproduction ratio is presently about 1.4-1.5. So stopping growth couldtake as little as a 30% reduction in transmission, which we might get in a variety ofways: 10% from improved quarantine compliance, 10% from hygiene and PPE,10% from a small rollback of reopening, and 10% from added sarcasm. Maybe itwill take more than that - fall/winter weather might be more challenging, forexample.
Given the delays, vicious cycles and uncertainty in the system, I think it's better(and in the long run likely cheaper) to overreact and correct, than to move tooslowly and struggle to catch up. Also, diversity is key - we shouldn't put all our eggsin the mask basket, or any other, but neither should we ignore any option.
The thing that troubles me most is that I'm not hearing any clear message about ourstrategy. Are we going to let the surge happen and deal with it? (If so, what does"deal with it" really look like?) Are we going to let infections double, or quadruple,then lockdown for a while and start over? (There's actually some sense to an intervalpolicy.) Are we going to get even better at contact tracing and testing, which isalready doing the heavy lifting? (How?) Are we simply pretending this isn'thappening, until we can pretend no more?
I'd like to see every decision about COVID policy prefaced by a clear statementabout where we think we are, what proposed actions will do to the infectiontrajectory, and what are perceived to be the biggest contingencies. If behavior needsto change, people need to hear that in the clearest possible terms.