By now, everyone should have seen the Coronavirus graphs created by the IHME at UW showing projections of death rates and hospital resources needed over the next few months in the USA.
The graphs were trotted out by the White House yesterday, this time without any sharpies. The media is gushing over it, totally oblivious to the fact that such models and predictions were being made by experts over the past 3 months. trump was crowing about the claim that this will cost “only” 100k lives, thanks to his amazing discovery of this thingamajig called social distancing .
Let’s take a closer look at the model from IHME at UW and see what it is telling us with its nation-wide and state-by-state projections of hospital resources needed and death rates over the next few months.
Here is a quick summary of some of the important assumptions and caveats mentioned in the paper and in the FAQ section of the website, which do not get much media coverage —
- The model assumes Wuhan-like shutdown and lockdown measures for 2-3 months until early June (actually the authors use social distancing measures described in the New Zealand Government alert system Level 4, which pretty much shuts down everything).
- The model assumes a nationwide shutdown, all states, no exceptions.
- The plot is for the best case scenario with maximum mitigation and suppression measures in place. The range values shown in shade are not for best vs worst case scenarios, they are caused by errors and uncertainties in the model parameters.
- To calculate the death rate, the model assumes that adequate ICUs and hospital resources will be available/created to meet demand, even though the model calculates beds and ICU shortage values.
- The model recommends mass screening, contact tracing, and selective quarantine to be in place before we lift the mitigation measures. The model address does not address what happens after restrictions are lifted.
Any deviations from these assumptions (and there will be many) will change the results.
Here are some more details on the above points -
1. Shutdown Period. The model assumes that we will have New Zealand Government alert system Level 4 shutdown and lockdown measures, for the entire period of the graph — i.e., 2-3 months.
- The White House and even Dr. Birx mentioned a one month of social distancing period. That is a not what the model says.
- Wuhan had extreme lockdown measures — no one could leave home for over 2 months. We have not even closed restaurants or beaches in many of the red states yet.
2. Nationwide shutdown. The model assumes that lockdown measures will occur within a week (of whatever date the FAQ was written, probably March 28).
- Is that even realistic?
- As of March 30, 22 states had not instituted stay-at-home orders. That number dropped to 17 on March 31. PA and FL issued stay-at-home orders today.
- What are we waiting for?
3. Best case or range of cases. Almost everyone interpreted the shaded range across the plot lines to mean best and worst case results, depending on how well we implement social distancing measures.
- That is incorrect. The plot is for the best case scenario with maximum mitigation and suppression measures in place. The shaded range just captures various sources of inaccuracies and uncertainties in the model parameters. The model does not consider worst case and anything besides the best case scenario.
4. ICUs and Hospital resources. The model shows the number of hospital beds and ICU units required per day and indicates where existing capacity will be inadequate. But the death numbers assume that everyone who needs an ICU will get one, i.e., ICU capacity and hospital resources will be increased correspondingly.
- If that does not happen, death tolls will be higher.
- Also, if all ERs and resources are taken up for Coronavirus, what happens to the thousands of people who need hospital treatment every day for a variety of illnesses and accidents?
- China built 14 temporary hospitals in Wuhan in weeks, brought in the army, performed very aggressive testing. They even closed them down by mid March as infections subsided. We have barely started.
- Heck, we don’t even have sufficient face masks or PPEs at this point, let alone ICU beds or ventilators or ER doctors or nurses.
5. The day after. No one talks about what happens at the end of the lockdown period (early June). Do we open up schools and businesses?
- At that point, a substantial fraction of the population (97%) will be uninfected and susceptible.
- We will end up starting the death graph all over again.
- We cannot remain locked down for 18 months until the vaccine shows up.
5.1. Intermittent mitigation. One way to manage this after the end of the first wave was described in the Imperial College report. We lift some restrictions, not all and let the disease build up again to a certain level, then we lock down again. We repeat this every few months for 18 months.
5.2. Contact Tracing. The other way, which is what China, Singapore, S Korea and Taiwan have used, is to deploy mass screening, contact tracing, and selective quarantine procedures.
- This is something the IHME study mentions too, hidden in the FAQ section.
- Primarily, this means very fast contact tracing and quarantine. As soon as a person is detected as infected, you quickly, within hours, discover all the people the individual has been in close contact with in the past several days and test/quarantine/hospitalize them. You repeat the process with this group of suspected people and find all of their close contacts and test/quarantine/hospitalize them. You can do 3 or 4 levels deep. This way you quickly catch asymptotic infections and stop them from infecting others.
Check out diary How we can tame the Coronavirus for some more info on contact tracing and suggested medium and long term solutions.
6. Time Lag. Even if we perfectly deployed all mitigation measures today, it will take weeks for their effect to show up, because of the time lag between infection and symptoms. That is why the graphs peak in mid April. The peak will shift out if shutdowns get delayed or are less stringent than what is assumed in the model.
7. The Economy. The refrain we hear from republicans is that they don’t want to sink the economy, that the cure may be worse than the disease, blah blah. A recent study by eminent economists, which I described in diary “Will Coronavirus Public Health Interventions Depress the Economy? History tells us No.” clearly shows, based on data from the 1918 pandemic, that cities and states that were slow to implement mitigations and shutdowns could not grow their economy well after the pandemic was over, compared to those that did,
8. Model data. FYI, most of the data used to calibrate the model is from Wuhan. The graphs are updated every day to reflect the situation in the states.
Critiques
Prof. Carl T. Bergstrom has a twitter thread on this issue, discussing the model, its unusual methodology, and its assumptions and projections. Most of the info in this diary was culled from that thread and its responses.
Some Tech Info on the Model
Unlike other modeling efforts that use simulations based on the SEIR model, this model uses a simple Gaussian error function with 3 parameters. You can see the equation below. The 3 parameters (p, alpha and beta) were estimated using curve-fitting with data from Wuhan.
Epilogue
Note that this is not a knock on the study or its methodology, although we need to see more of how the parameters were estimated. The authors have provided many of the assumptions and caveats, although not too prominently, and most of the media and the administration does not talk about them.
Where does that leave us? Up the proverbial creek without a paddle. The White House is more interested in damage control, rewriting history, blaming others and showing trump as the great savior. They will fix the numbers and the narrative to suit themselves. Things are likely to be worse than projected and there is no long term exit strategy even if we manage to follow the curve for 2-3 months.
Meanwhile, let’s do all we can to suppress the Coronavirus —
- Stay home
- Wash hands
- Avoid touching face
- Wear a face mask
- Social distancing
- Stay healthy
- Stay informed
- Listen to scientists, don’t listen to stupid
- Spread the message, not the virus
Stay safe, stay strong.