Avoiding Mayday

Today, May Day, is a rare triple entendre.

Mayday is the international distress signal[1]

May Day is an ancient festival of spring, still recognised in much of Europe.

May Day is International Workers Day[2], celebrating Labour, the key economic factor of production.

All three are appropriate today. Europe, and much of the world, including autumnal Australia, appears to be re-emerging from its coronavirus-induced hibernation. Labour has little cause for celebrating, with unemployment temporarily at least, at its highest since the Great Depression in many places, while avoiding a distress signal, is what this article focusses on.

In recent Linkedin posts, I have shared my “manifesto”[3] for coronavirus from here. One of the key pillars is developing an Early Warning System. I have moved my data focus away from what was a relatively accurate forecast of the path of deaths in the first wave, to trying to think about risk, and whether we see a second wave.

The purpose of tracking the data is an awareness of risk. What I can see from public data is still 10-days behind the true picture, and relies on testing outcomes missing the untested and asymptomatic.

But if transmission rates start to pick up, we must be alert to the frightening risks of exponential growth. The last few days have seen us move back in the wrong direction, so while it’s too early to tell for sure, the signs are already concerning; we absolutely must avoid complacency.

The maths behind it is simple, and while I look here at Australia, this is applicable anywhere and I am tracking this crudely on a weekly basis for several countries most affected at this point.

We all learned about exponential growth in the first wave, and Australia through good fortune dodged a bullet despite locking down late and having about a five-fold increase in infection between Scott Morrison’s footy comments, through the Ruby Princess disembarking and the crowded Bondi Beach that sparked my interest in this topic, before we finally locked down on 23 March. Self-attribution bias explains why people will take credit for what appears to have been good fortune (in the same way, we tend to blame bad luck when we make poor decisions).

As we open back up, we re-expose ourselves to risks of exponential growth that would cause the second wave. Opening borders later will expose us not only to our own community infection, but to that of the world, and poses a difficult political decision as it likely requires preferential treatment and strict quarantine for arrivals.

For now we are starting to go back into the community – it was already getting busier ahead of any government relaxation - with messages starting to be confused again with politicians trying to juggle votes and lives. The tracing app provides us with reassurance that if we come into contact with infected people, and spend 15 minutes within 2 metres of them we will be able to trace that, as long as they have done their national duty too.

Social distancing, self-isolation, and hygiene will become features of our daily lives. We need to be alert to avoid reigniting the spread. Identifying the vulnerable and protecting them is a project in co-morbidity that actuaries will be able to help with.

But, how do we assess the risks we face. While there are flaws in relying solely on confirmed, active cases[4] as the extent of community infection, we can use it to get a sense of the spread as we continually test more and more people.

We can model on a daily basis, from the public data, the number of new cases and compare them with the current active cases in the population. This gives us an effective rate of tranmission from the infected population. We can smooth this using five days of data. This is close to having a 3-day delay on the trend. Add to that the fact that it may be a week on average from contracting the virus to being tested, then we are still behind the actual spread, but we can quickly monitor if the trend is changing.

I would like to think there are additional adjustments we can make, eg collaborating with Google, Apple and others to track movements, and even searches – I suspect searches for symptoms will go up as outbreaks emerge. Google mobility data[5] shows a strong relationship between movements in the community and the rate of transmission, so real-time data here would be very helpful for governments. Using such additional data sources, calibrating between active cases and community infection, and projecting forward to offset the delay from infection to testing, would all enhance the output, but the trend picture can emerge nonetheless.

But for now, the crude exercise for the population is to measure and plot this 5-day rolling average transmission rates, as this chart shows for Australia.[6] To remind you of the phrase "flattening the curve" – the chart is reflects the steepness of the curve, so when this line is falling, the curve is flattening. 

What we notice here is that ahead of lockdown daily case transmission rate was over 20%. However, partly this was due to increased awareness and testing, and calibrating against deaths, which likely occur around two weeks after testing, the rate was more likely in the teens.[7]

For context, the trajectory of the graph coupled with mobility data and the assumption of testing coming a few days after infection, seems to indicate that our population made conscious personal choices, ahead of being told what to do. This is encouraging, and commendable, and hence perhaps shows an inherent understanding of the situation, despite the scenes of Bondi.

On the basis that the active population stays infected for 3 weeks[8], the critical rate of transmission is approximately 5% per day, or 1x over 20 days. This ties into the idea of the reproductive rate, R0, which leads to exponential growth when it is above 1.

As at the latest reading, based on data to 29 April, we are close to a transmission rate of 1%, (or R0 of about 0.2)however, we must not be complacent. This crude observation hides that it has steadily risen from 0.7% to 1.2% over the past 6 days, commensurate with a greater sense that we are going “back to work”.

To refine this analysis, one would incorporate data on the proportion of tests that are positive[9] and could look more closely at cities, or states -it is interesting to note at present that almost 75% of active cases in Australia are in NSW[10] but only 42% of new cases in the last 5 days[11] - ie it is growing slower here than in Victoria and Tasmania, which on the other hand have lower infected populations.  

One might also consider age groups – with the case fatality rate for the under 60 running at about 0.1% in Australia[12] - this might shape policy around the elderly, or make adjustments for the risk of clusters which could lead to pockets of rapid transmission and distort the population data.[13] At this stage, the model is simple and reductive and tells an adequate story of risk.

The more we are aware of the risks, the more we can temper our behaviour. The number of people carrying the virus in our community is low for now. But, if we see that this transmission rate is picking up, remembering the ten-day or so lag in this “Early Warning System”, and noting it has almost doubled in the last few days, we can start to focus more intensely on adjusting our behaviour.

Its easy to forget as we try to get back to normal, how conducive normal was for the spread of the coronavirus, and the implementation of medical advances is still some time away.

The more data we have, and the more we are aware, the less likely we are to have to call Mayday!

Stay safe, and spread data, and not the virus.

[1] Wikipedia notes that Mayday was derived from the French m’aider meaning “help me!” and is the distress call used by ships and airplanes, replacing the dot-dot-dot dash-dash-dash dot-dot-dot of SOS (Save our Souls)

[2] International Workers’ Day commemorates the Haymarket Affair, the aftermath of a bombing at a rally for the eight-hour working day in Chicago in 1886.

[3] To exit lockdown domestically, we need to Identify/Protect the Vulnerable, Identify/ Isolate the Sick and create an Early Warning System to refine restrictions. We then need to put health over politics in opening our borders. We then need to identify and fully understand the natural (distance, climate, time of year) and human advantages (eg demographics, density, medical system) that enabled Australia to get lucky (my friend Felicity Menzies has pointed me to some interesting work to explore on cultural factors but prima facie Australia looks similar to the less fortunate US so it may not be a factor), and prepare for either a second wave or another pandemic. Finally we need a reform agenda to compensate the young taxpayers for the budget blow-out.

[4] https://www.actuaries.digital/2020/04/15/covid-19-is-more-widespread-in-australia-than-the-headlines-suggest/

[5] https://www.google.com/covid19/mobility/

[6] Source data is the very useful worldometers.info

[7] On a crude assumption of a 3-week infectious period, this equates to an R0 of about 3x.

[8] This logic is derived from WHO work on the Chinese experience and backed up by the lag between new cases and deaths in the Australian experience.

[9] If there is a rapid change in the rate of daily testing, or any changes in testing criteria, this could also be an explanatory factor in changes in the number of active cases reported.

[10] Source : NSW Health

[11] To 30 April 2020, source: covid19data.com.au

[12] Source: Government Health Department

[13] Cruise ships and care homes have been widely publicized areas of risk.