I will start with the disclaimer that I am not an epidemiologist. Of course, this means you take the conclusions with a little bit of salt on the side.
This is mostly an attempt to understand it myself. If you ask me to predict anything, my answer is a simple – I dont know. All I am doing here is tracking the data and trending it to find reasonable boundaries and necessary conditions for things to come under control.
The key driver of a pandemic – R0
New York Times has made an epidemiological model available (towards the bottom of the page) where you can play with the model assumptions. The model is very sensitive to this R0 input (Ro – Basic reproduction number) moving it between 2.3 – 3.0 has drastically different results, ranging from no problem to a big mess. Currently, the estimates for R0 range from between 2 and 5.
Recently, Neil Ferguson, the author of the very well cited Imperial College Report on the potential spread of the virus, revised up his estimate for the transmissability of the virus from 2.5ish to above 3.0. So, if anything, this virus is more contagious (and given we dont have a cure, dangerous). There was some confusion as to his remarks, and I would refer you to an FT article and his tweet below. To be clear, his revision means a lockdown is even more necessary.
In short, beyond the case fatality rate (~4%), the biggest driver of the pandemic remains its contagiousness. As such, the best tool to fight it remains slowing that spread through lockdowns and social distancing. That could create an exponential decay in its exponential growth.
Lockdowns – Hubei and South Korea vs Italy, Spain, UK, New York
Ofcourse, I am taking the number of infection data at face value and, ofcourse, it is not exact nor comparable from one jurisdiction to another. Yet, I think there is value in comparing some very simple metrics across differerent regions
- Daily new cases
- Growth rate of total cases, day over day
- Growth decay factor, the day over day decline in the above growth rate
It is the last one that I am most interested in comparing across different regions. And while this is not a very sophisticated analysis, it does compare in a simple manner how effectively the spread is being controlled. Again, I appreciate this is lagging and incomplete data because of the time delay between getting infected, showing symptoms, and the authorities recording a confirmed case. However, this is shared across all the regions and we need to remain aware of the lag when drawing conclusions. Indeed, the confirmed case data is only a proxy (with a lag) for how fast the virus might be spreading (given asymptomatic and untested cases). The only way to shorten that lag is to test, test, test (as in South Korea).
Using these three, I have made trended the trajectories of the current pandemic going forward. I dare not call them predictions, because of the multiple assumptions made and the dynamic nature of the inputs.
A few observations
- Lockdowns were enacted later vs sooner in western countries, despite lessons from the East. Hubei locked down when they reached 100 daily new cases, UK/Spain/Italy did so around ~1000-1500 daily new cases, NY did so when they reached 4000 daily new cases.
- The effectiveness of these lockdowns has varied hugely across regions. This is the biggest swing factor in how much we can “flatten the curve”. To fight a pandemic, it needs to be throttled down by a somewhat extreme lockdown. Hubei brought daily total case growth from 20% to 1% in 23 days, S. Korea in 16 days. Both of these used different techniques (Extreme quarantine vs testing + contact tracing). Their resulting growth decay factor were 13%-18%. In most of the west, this growth decay factor ranges from 6-9%. This is partly cultural (people not following advise or taking the threat seriously) and partly the mixed media and governmental messages (opening lockdown by Easter because “it was a beautiful time, a beautiful timeline, it’s a great day”).
- The peak of new daily infections is still ahead of us, perhaps another 1-2 weeks out, for the cities / regions below. However, Italy appears to be at its peak already. However, this peak is contingent on continuing with social distancing measures.
- If we relax a lockdown by mid-April, the daily cases will again start growing. To see a curve flatten like in the charts below, the lockdown needs to continue for 60 days or more, unless something else is done (for ex, South Korean/HK/Singapore response). Hubei, the province where it all started, has stayed in lockdown for over 60 days and is only now starting to very slowly relax the lockdown.
How widespread is the virus already?
A report from Oxford University stirred a debate when it claimed over half of the UK (and Italy) might already be infected (see the same FT article). Perhaps the intent of the researcher was to highlight an urgency for a serological antibody test for Covid-19 infection to get an idea of how widespread the virus is in the population already. Doing such a random test will help determine a) regions where herd immunity may be developing and thus can be slowly relaxed, b) regions where the virus is still very thin and strict quarantine, contact tracing, and testing can help isolate the few cases.
The data below shows the more you test, the more you find the virus. Beyond that, it also gives a rough idea of what % of population might be infected already.
Eyeballing the relationship below, you can estimate that roughly 10% of those tested have the infection. But the 10% tested are the ones with multiple and serious symptoms, so this is (for now) an overestimate and needs to be adjusted downwards. At the same time, we know there are plenty of asymptomatic cases and people who showed mild symptoms and recovered. So then you need to adjust this upward.