Thoughts on the COVID-19 situation

This post is for our friends, family members, and acquaintances who may be struggling with finding reliable sources of information in today’s world of misinformation and sensationalism regarding the corona virus. We hope that this will provide a perspective that is largely based on data to inform how we view the risks and public policy decisions that have been made and should be made going forward.

Dr. Jose-Miguel Yamal is an Associate Professor of Biostatistics and Data Scientist at University of Texas School of Public Health. These are his own thoughts and not representing his institution. A special thanks to Dr. Luis Leon Novelo and Dr. Ashraf Yaseen for their help with this article.

1. What we know about COVID-19 so far

a. Risk factors for worse outcomes

b. Trajectories versus taking numbers at a specific timepoint

b. Case fatality rate estimates

c. Flattening the curve and health care system capacity

2. Public policy decisions and rationale

a. Public health impact

b. Economic impact

3. What we need and what we currently know about it

3a. More accurate estimate of CFR

3b. Vaccine

3c. Treatments

4. Conclusions (for now)

If you haven’t seen them already, these three articles by Tomas Pueyo (a data scientist of sorts) are VERY interesting reads and I highly recommend that you have a look at them. One, two, and three. There are some issues from a statistical perspective, but those issues do not change the overall conclusions and message. I have borrowed several graphs and arguments below from those articles as well as complemented it with information from other sources.

1. What we know about COVID-19 so far

a. Risk factors for worse outcomes

So far, we know that COVID-19 has manifested in various levels of severity. Some people are asymptomatic (don’t even know they are sick, maybe around 25%), some have very severe flu-like symptoms (majority of cases), some require hospitalization with ventilator support (maybe around 2%), and others die (maybe around 1%). The probabilities of each of these outcomes change based on a few things including age and co-morbidities. However, this does not only affect those that are old and have co-morbidities. This will be important to keep in mind when we talk about estimating the probability of dying from this, particularly since there are many demographic differences between cities and countries.

b. Trajectories versus taking numbers at a specific timepoint

When considering the severity of a novel pandemic disease, it’s extremely important to consider the trajectory and rate of growth instead of just the raw numbers at a specific time point. Many countries were questioned why they took drastic measures when they had so few cases. For example, take the US which declared a national emergency on March 13. On March 12, there were 1,645 cases known in the US with about 40 deaths.This didn’t seem like that big of a deal, but the wave that was sweeping the globe (128, 352 cases and 4720 deaths) was very clear and we have seen how it has affected the US since then. Today, April 9 2020, there are more than 431,000 confirmed cases with almost 15,000 deaths!

b. Case fatality rate estimates

There has been a lot of debate about the case fatality rate (CFR) estimates. This is the proportion that die out of those that have COVID-19 and is a hard number to estimate, but we will look at some of the evidence here. This is important because this is used to estimate how many deaths would occur in the population and thus help to justify public policy decisions. The estimate that Dr. Anthony Fauci has recently been using is 1%. 

In this Nature article , they estimated the overall symptomatic case fatality risk (only for the subset that develop symptoms) as 1.4% with a 95% confidence interval between 0.9-2.1%. It’s more important to pay attention to the confidence intervals than the single point estimate because that gives a lot more information about what the range of likely values are, given the data at the time. Now, we know that many people are asymptomatic (see for example this article), so that suggests that the case fatality rate (not just those that are symptomatic) will be lower than this. Some estimate that 25% are asymptomatic and therefore that would put this point estimate at 1.4*0.75=1.05%. Let’s say that we don’t believe the 25% asymptomatic and want to assume even higher percentage that are asymptomatic, we can estimate the case fatality rate. For example, if we assume 50% of those with coronavirus are asymptomatic, then the point estimate would be .7% (take 1.4*.5=0.7%). This is probably an underestimate, so the true case fatality rate is probably >0.7% and thus >7 times the death rates compared to influenza (which is .1%). We won’t know this exact number for a while.

There are a few other important unknowns in the CFR estimate that we need to consider. The numerator is the total number of people that died from COVID-19. Here is an article about how the CDC counts only deaths in which the presence of the coronavirus is confirmed in a laboratory test. Some who die at home or in overburdened nursing homes are not being tested. Given that US has a lack of testing, tests are being prioritized to living people. Also, there is the problem of differentiating between people who die with COVID-19 versus people who die of COVID-19. It’s true that this exists, but according to professor of epidemiology at Harvard Marc Lipsitch, this is a minor issue that is “swamped by the opposite problem: deaths that are caused by covid but never attributed, so the death count is underestimated.”. One way to estimate the covid-specific deaths is by comparing the number of people that would normally die in a certain period of time to the number that die during the time that there is a pandemic outbreak. We will have to wait for this data. 

The denominator is the total of people who have the virus. This is probably undercounted but it’s a bit more complex. First, there are many people who have not been tested (see graph below). Also, some people who have the virus test negative. This is why we have to rely more on other countries that have done a much better job of testing a lot of people and following up. Look at Chart 6 (from this post). This gives some hints to who is being tested in each country. The countries that have red bars and high percentages of positives of those that are being tested are countries that are probably reserving testing for those that are very sick. That’s been true in the US. The countries with green bars are those that are doing more widespread screening and thus are testing a lot of people who don’t actually have the virus.

It’s likely that Spain is missing many cases (i.e. cases >> confirmed cases) since so many people are testing positive. They are probably not testing many asymptomatic people. In contrast, Vietnam is probably testing a lot of people. The green bar countries are controlling the virus well. Their data is probably better for estimating CFR. Since the number of cases will always be higher that the number of confirmed cases, estimating the proportion of confirmed cases that die of covid19 most likely will overestimate the CFR. In South Korea, this estimate is currently 1.9% and they have very good testing and healthcare system. In comparison, Italy’s estimate is 10% but it’s pretty clear that they are undercounting the denominator (i.e., cases >> confirmed cases). 

We can compare this to the observed percentage of deaths from Johns Hopkins website:

Vietnam: [omitted since they only have 250 cases so too little data, 0 deaths]

Taiwan: 1.3%

Singapore: 0.4%

South Korea: 1.9% (they have very good testing and healthcare system)

Germany: 2.1%

Probably not accurate estimates of CFR (probably overestimated by undercounting the denominator)

Spain: 10%

Italy: 12.7%

US: 3.4%

Switzerland: 3.8%

France: 9.6%

Based on these, the 1% estimate falls within the range of the green bar countries’ estimates. It’s true that we have incomplete numbers, but given the data so far from countries around the world and from our own states, it’s clear that this is a serious situation. Given that there are uncertainties in the numbers, there is a philosophical question of whether one should err on the side of caution in making decisions. Unfortunately, in these types of situations we HAVE to make decisions before the picture is clear with more data because each day of inaction can result in hundreds or thousands more dead — even a 1 day delay! When a new virus comes out, we cannot collect the data about the true CFR at the beginning. All of this takes time. If we wait for all of that to occur, the pandemic can have disastrous consequences with no action. 

There has been some criticism of the CFR estimates and the public policy decisions. See below for this discussion. 

c. Flattening the curve and health care system capacity

One primary argument for taking aggressive action to lockdown cities is to “flatten the curve” — basically, just buy ourselves time so that we can work on a vaccine (trials have already started!), treatment (clinical trials are underway), and have enough resources in hospitals such that they are not overwhelmed and there is collateral damage where other people die due to not being able to get healthcare. This seems to be generally working so far (see more details of several countries’ response and their subsequent flattening of the curve about 2 weeks later), although this may change in the next few weeks.

When we don’t have visuals of how this is affecting people, the public often doubts its seriousness. See this article from a journalists’ perspective of trying to cover the news. People start questioning why we are taking extreme measures when the problem doesn’t seem too bad. Unfortunately, it’s hard to convince some people about this. Consider three scenarios:(1) the situation is very serious and the city shutdown, effectively making the situation much better than it would have been; (2) the situation didn’t turn out to be as bad as we thought it could be; (3) the situation is still dire but it’s worse effects are being seen in the emergency departments, where we cannot actually see it with our own eyes. It’s hard for the public to be able to decipher between those three scenarios and thus the justification for shutting down a city may be questioned. 

2. Public policy decisions and rationale

a. Public health impact

The primary public health strategy is to identify all of the infected, isolate them, trace their contacts, and quarantine them. Additionally lockdown the cities. This has worked in East Asian countries. The US generally has a mitigation strategy, to have relaxed social distancing measures, but this is much weaker than a suppression strategy. Currently, this is coordinated at the state level instead of at the national level. Having each state fend for itself has created all kinds of logistical nightmares including them trying to outbid each other for a limited number of ventilators and heath care equipment. Also, if two adjacent states take very different measures, then they may need to seal their borders to each other. People from one state may look for healthcare in another state once his/her home state healthcare system is saturated. This is almost happening between TX and Louisiana due to the higher number of cases over there. Let’s look at the effect that these public policy decisions have on flattening the curve to see if this seems to be working or not.

Let’s see if the curve is actually being flattened or not. Go to this page and look at the graphs for the individual countries.

• Italy had the nationwide lockdown in effect on 3/9/2020. The number of new cases each day started to flatten and then drop about 2 weeks after this.

• Spain had a state of emergency declared on 3/14/2020. The number of new cases started to drop about 2 weeks after this.

• China locked down Wuhan 1/23/2020. Cases started going down 2 weeks later, then had another spike, but then went way down. They have now reopened a lot of their economy.

• Germany started their lockdown 3/13/2020 and further made a national curfew 3/22/2020. Cases started going down at the end of March and beginning of April.

• Netherlands started some measures 3/15/2020, but they weren’t very strong (they argued that they wanted to create population immunity). They then had to make them stronger since the number of cases kept increasing substantially on 3/23/2020. They may have reached a plateau now, but we will see.

• Belgium started social distancing measures 3/18/2020. The number of new cases started flattening about 2 weeks after this.

• US has not had a coordinated lockdown across the whole country. It’s been more piecemeal. However, the state of national emergency was declared on March 13. We see that the number of cases may be plateauing, but it’s too early to tell and models predict that it will continue to rise in the next few weeks. 

There’s a very strong pattern here that provides strong evidence that strong lockdown actually works. So why is there a 2-week delay? That’s because it takes about 2 weeks for people who get infected to then start showing signs and then get to the level where they go get tested and results come back. Therefore, the increase in new cases in those 2 weeks are people who were already sick but didn’t know it yet or didn’t have the test results back yet.

We can look at the US States response in the future. I think there still is too little data to come up with conclusions here. However, there are a lot of differences on policy decisions by state. For example:

· Florida, until very recently, had no ban for large gatherings, no business closures, and no stay-at-home orders beyond quarantines for travelers. Check out this interesting visual of the potential spread from Spring Breakers that went to the Florida beaches. The decision to not close the beaches probably caused a significant increase in cases. Florida cases are increasing at one of the highest rates.

· Georgia is similar to Florida except that bars and restaurants are still open with limited service and high-risk groups are asked to stay home. 

· Mississippi required a fever of 100.4 and severe cough or chest pain to receive testing and they reversed non-essential shutdowns applied by local jurisdictions. They finally took more measures at the beginning of April.

It’s hard to tease out all of these differences in policy since people travel within the US and our state borders are not closed.

b. Economic impact

What’s the benefit/risk of not responding too strongly (take some measures)?

- benefit: avoid economic shock early on

- cost: during the epidemic, people will avoid going to work or consuming for fear of getting infected. This can strain the economy for as long as people believe the epidemic is uncontrolled. Psychological toll of the fear of the virus or the loss of a job can depress consumers and their income, reducing spending. This can close businesses and financial system could collapse. People will probably spend less as well because of losses in their retirement accounts.

What’s the benefit/risk of responding too strongly (stop the economy for a few weeks or months)?

- benefit: can bring infections to nearly zero while giving time to do more testing, contact tracing, immunizations, not strain the health care system

- cost: hard in the short-term. Psychological toll of the fear of the virus or the loss of a job can depress consumers and their income, reducing spending. This can close businesses and financial system could collapse. People will probably spend less as well because of losses in their retirement accounts. Depression and poverty can also lead to deaths. How many is unknown.

There are few interesting analyses of the economic impact of public policies towards pandemics. In this article, the authors analyzed the economic impact of the 1918 Spanish Influenza. They found that the Spanish Influenza of 1918 reduced GDP per capita of the average country by 6% and consumption by 8% for a year. In their analysis, they concluded "At this point, the probability that COVID-19 reaches anything close to the Great Influenza Pandemic seems remote, given advances in public-health care and measures that are being taken to mitigate propagation. In any event, the large potential losses in lives and economic activity justify substantial expenditure of resources to attempt to limit the damage."

A Harvard Business Review article looked at the possible economic impact of COVID-19 and compared it to prior epidemics. They found that usually, after a pandemic, the economy goes back to normal, probably in a few years.

In a recent article that was published, authors found that Pandemics depress the economy but public health interventions do not (based on analyses of various states’ reactions to the 1918 flu). They found that cities that intervened earlier and more aggressively did not perform worse (economically) and, if anything, grow faster after the pandemic is over.

Note that there are many differences between the response to the 1918 flu and this virus, so we cannot assume the exact same results in today’s economy. However, this gives us some data in support of suppression (stronger lockdown) instead of just mitigation (flexible social distancing).

Here is how the China stock market reacted to shutting down Hubei and also to the West’s mitigation strategies (from this post):

In general, pandemic suppression looks to have better economic outcomes than mitigation.

3. What we need and what we currently know about it

3a. More accurate estimate of CFR

Everyone is seeking more accurate estimates of CFR. To get this, we need: (1) testing of those that die; (2) screening tests for people who currently are infected; (3) antibody tests to see who had it and recovered; and (4) epidemiological analyses of those that died during the epidemic compared to those that were expected to die anyway in that time period (due to all other causes). In recent interviews, Dr. Bhattacharya has been advocating for needing studies that estimate the true CFR by doing complex surveys that are representative of the US that include antibody tests so that we can estimate the total number of people who had coronavirus but then recovered. This addressed the 3rd component. If the antibody test is accurate and once we have more data about the virus (for example, can people be infected again), this might be able to be used to determine when people should start going back to work. Note that if people can get re-infected, an antibody test may challenges. Dr. Bhattacharya (YouTube video) argues that we need to have an estimate of the CFR to justify our public policy decisions. The number of people who die (based on the CFR estimate of .1% or 1%) could be between 50,000-100,000 to 2-4 million deaths, respectively. He agreed that it’s true that the Novo coronavirus would kill millions without shelter in place orders and quarantines, then the extraordinary measures are surely justified. However, this needs to be balanced with deaths that could arise from a shutdown that goes on for too long and causes a global economic collapse (due to depression, opioid overdoses, shorter lifespan). His primary argument is that we need to do the study to estimate CFR.

I agree that we need to study the CFR. He has some criticism for public policies that have been put in place without the data to back it up. He noted that if the situation is that the CFR is about 1%, then extraordinary measures are justified. I believe (1) the current data suggests that there is a good chance that the CFR could be around 1% (see above); (2) more data is needed but public policy decisions needed to be made early on and NOT wait until all of the data is in; (3) better be safe than sorry for a pandemic that has the potential for such catastrophic loss; (4) we will get more data and we will have to use that to adjust public policy and we should not be scared of changing policies once more information comes in. A recent news article suggests that one of the antibody tests that have been developed has failed to detect some subjects who recovered.

3b. Vaccine

This could be a game changer. Vaccine trials have already started, including one in Houston (Houstonvaccine). This will take between almost a year to a few years, probably. Flattening the curve can delay major effects to give time for vaccines to be developed.

3c. Treatments

There are many potential treatments being investigated. Telling people that there is a treatment that works at this point is irresponsible (e.g.  hydroxycloriquine). Flattening the curve can delay major effects to give time for treatments to be developed.

4. Conclusion (for now)

Shutting down an economy is very hard for businesses, independent contractors, and others in the workforce. It’s hard on parents that are now trying to homeschool. People spend a lot of their lives building something up, only to be torn down. On the flip side, not doing enough and having millions die would be disastrous to families and the economy as well. The studies above suggest that the pandemic is what really causes these economic hardships and the public policy aspect of shutting cities down does not actually make it worse and may make the economic situation better in the long-run (perhaps within a few years). Economies are pretty robust and we will probably bounce back from this in a manner of a year or two. Inaction is irresponsible and will cause significant economic and public health catastrophes. Thankfully, the world has generally responded to this by taking action. The US action needs to be centralized to be more effective. This isn’t about fear-mongering. We will get through this. Yes, there will be some deaths and economic hardships. However, we are resilient. Most people are being smart about being cautious. We don’t need to live in fear right now. We just need to live smart and take appropriate precautions and not take unnecessary risks. The risk may be small, but the consequence is not worth it.

#statistics #publichealth