web analytics

Delta pandemic and uncertainty

In 1919-20, the second wave of influenza fanned out throughout the world. Although the exact reasons are disputed (mutation, troops coming back home after WW 1), the effect was devastating.

In 2020-21 we have so far experienced two large waves of COVID-19, the original one in February-June 2020 and the second one – fuelled by a combination of a Summer 2020 reopening and the replacement of the original strain by the UK strain, now called Alpha, as well as Brazilian and South African strains (Beta and Gamma).

We are now facing the third (or fourth) wave, cause again by a combination of behavioural changes and Delta strain (and other similar ones). Yet, there is a large uncertainty as to what is likely to happen in the next few weeks, as illustrated below by two consecutive predictions by Scottish Government modellers:

Why is the earlier prediction so dire and the current one a bit better? Why has one week of data made us change the prediction so drastically?

We are in a better situation now, compared to the world in 1919 as well as to the previous waves. But we are also in uncharted territory as we never have seen a massive epidemic growth in highly vaccinated populations (perhaps with exception of measles outbreaks after the mass vaccination drove the numbers down).

The UK is indeed a highly vaccinated country, although not with the highest proportion of fully vaccinated individuals and with a substantial proportion of those vaccinated with AZ which has lower efficacy against Delta variant. Currently, the UK has 50%, Israel 60% and Malta 77% population fully vaccinated, but all three now see an increase in numbers.

But the UK is also carrying out a unique experiment of removing all NPIs while the cases are going up exponentially. As there is no precedence for this strategy, it is difficult to capture all details with the models.

We simply do not know yet how Delta will spread in a highly vaccinated but also highly stratified population. What I mean by “stratified” is that age-limited vaccination created a pocket of susceptibility in school children who are by nature clustered in schools.

At (still) relatively small levels of disease, “stochastic” events like superspreading events are also important and affecting our way to predict. A large group of Scottish fans travelling to a match in London brought the virus back to Scotland. As they mixed together and with other fans, nearly 2,000 reported having caught COVID-19 while at the match.

Epidemiological models by nature are quite sensitive to changes in assumptions. Also, the reproductive number is currently around 1-1.5. This is a region where even very small changes in parameters or assumptions produce large changes in the dynamics. To illustrate this, think about R=1. If R=0.99, the epidemic dies out. If R=1.01, the epidemic grows exponentially; the difference in R is of 1%, but the outcome is very different!

Sensitivity of the epidemic growth in respect to small changes in R.

So to summarise, we are now in a similar situation to weather forecasters facing an unusual weather pattern: We have excellent modelling tools, but our predictions are quite variable. If we want to know the weather on Sunday and check the forecast on Wednesday, we can be predicting blistering heat and sunshine. But on Thursday, the prediction might already be a torrential rainfall.

This lack of predictability, while understandably annoying at times, is also why I love Scotland. But of course, the lack of predictability in COVID-19 can be a matter of life or death.

This is a companion post to an article by Mark McLaughlin from The Times in which he is quoting me. I am indebted to Mark for pointing out the uncertainty in the predictions – and for may stimulating questions.