One afternoon in mid-July, Juan gets up, gets ready, and goes out into the street. You have COVID-19 but you don’t know it (it’s one of those “asymptomatic”). He has a drink on a terrace with some friends and, since it is hot, he decides to remove his mask. At noon, he takes a bus and goes to his parents’ house to eat with them, as he has done so many times.
Three months later, without him knowing it, some of the people with whom he had contact became ill, infected others or even died. If Juan infected a single person and, over the next week, that person infected another and so on, after 3 months the outbreak that started will have spread to about 12 people. And if it had infected 2, which in turn infected another 2 each week, those affected would reach 4,000. Even worse: if the weekly rate of infection were 3 for each infected, it would be directly responsible for 500,000 infected.
Or maybe none of that happened. Perhaps Juan did not leave the house, did not meet his friends, or did not remove his mask at any time. Who could know?
The difficulties of crawling
This story shows three difficulties that we encountered in predicting the evolution of the pandemic. First, we don’t know for sure who is infected and who is not. Second, although “tracking” allows us to detect some asymptomatic patients early, we cannot always determine the network of contacts (some in public transport or in the supermarket). So we don’t know how many of those encounters have been contagious. Finally, and this is the central idea of this article, epidemics grow (and sometimes diminish) at an exponential rate.
Although exponential growth can overwhelm our healthcare system in a few weeks, humans are not psychologically equipped to understand all its implications (for example, did the reader know that if we could fold a sheet of paper about 103 times, it would be as wide as our universe?). In general, it is difficult for us to assimilate that on February 26 the first case of local contagion was notified and on March 14 the state of alarm was decreed.
However, the reality is that the dynamics of the epidemic do not just follow an uncontrolled exponential growth. That is why epidemiologists use mathematical models to capture all the subtleties of its spread. The most common models divide the population into groups (called “compartments”) that describe the groups of susceptible, asymptomatic, sick, hospitalized or other individuals. Different models include more or less details and give rise to a real alphabet soup.
It is common to think that the inability to predict the dynamics of the epidemic is due to the poor quality of the data and not to the quality of the models. So let’s fix the data, and we’ll have a “crystal ball” (which we call a mathematical model) that will tell us precisely when the second wave will begin or end. Even when the end of the epidemic will come. After all, if similar mathematical models have put man on the moon, why shouldn’t they work in this case?
Well, because it is not so simple. The difficulty is that, regardless of the details of each model, all contain the relentless exponential growth. This factor not only affects the growth of the epidemic, but also the rate at which our ignorance about it grows. As we mentioned in the short story that opens the article, all the little decisions one makes can be amplified and completely blur our ability to predict.
The bus we get on, or a meal at our parents’ house, can have unforeseeable consequences weeks later. And this epidemic of uncertainty does not depend solely on the quality of the mathematical models, but is intrinsic to the very dynamics of the spread of the infection. And let’s not forget that the real scenario is even more bleak, since we have incomplete, inconsistent data and, on many occasions, inaccessible to scientists and citizens.
Learning from meteorologists
The reader will wonder what we can do in this hopeless outlook. The solution could come from an issue in which our inability to look to the future is known and accepted: weather forecasting.
As in the dynamics of the pandemic, small changes in weather conditions (perhaps the flapping of a butterfly) could trigger a hurricane on the other side of the planet. A century ago, the weather forecast was not much better than that provided by individual experience: in winter it is cold and in summer it is hot. Over time, a great level of sophistication has been reached, as the three main sources of uncertainty in weather forecasting have been attacked.
The first one is the quality of the data. Millions of temperature, humidity, precipitation or wind speed readings are collected daily. There are specialized institutes in all countries that integrate this data and provide updated information every hour. In the case of the coronavirus, we hardly know the positive cases by PCR or antibody test and the data of hospitalized and deceased. Let’s compare these 4 or 5 data sources with the dozens of meteorological probes distributed in a single location.
Second, climate models have improved and are able to describe its dynamics in exquisite detail. Interactions in the atmosphere and with land and sea at the local level are described by very detailed equations. This has been possible thanks to the increase in the computing capacity of computers. In the past, equations of such sophistication could not be solved in time to make a prediction, and very simplified versions had to be used (and of course, much less precise).
The third problem is inevitable and nothing can be done about it: chaos. The equations of climate are chaotic. This means that the slightest error in a measurement will be amplified over time and will produce divergent results that, ultimately, will have nothing to do with the real situation.
How has this problem been “solved”? In two ways: one, by accepting that there is a reliable “prediction window” beyond which we can no longer “see.” That is why weather forecasts never exceed one week, and what happens over three days has a very relative reliability. In addition, the predictions are corrected on the fly.
The second “innovation” has been to incorporate probabilistic prediction. Since exact prediction is impossible, why not change it for a range of situations with different probabilities? That is why now in the weather news it is not stated that tomorrow it will rain, but that the probability of rain for tomorrow is 90%, for example. What it means is that, on average, one in 10 times that there is such a prediction will fail.
In the case of the coronavirus pandemic, we must take on the challenges that we cannot change (the exponential growth of our inability to predict) and work towards an efficient (and transparent!) Data collection system and a probabilistic approach to prediction. .
But for now, with the data and models we have, we can’t do better than good old Mariano Medina.
Susanna Manrubia, Researcher in Evolutionary Systems, National Center for Biotechnology (CNB – CSIC); José A. Cuesta, Chair professor, Carlos III University; Mario Castro Ponce, Professor and Researcher at the Higher Technical School of Engineering (ICAI), Universidad Pontificia Comillas and Saúl Ares, Associate research scientist, National Center for Biotechnology (CNB – CSIC)
This article was originally published on The Conversation. Read the original.
