Predicting the Flu using Mathematical Models
It was a debilitating winter this year in the Northern Capital: between November and February, St. Petersburg Hospitals treated more than 6,500 people with the flu, according to media reports. All pharmacies were not even adequately equipped with the necessary medication and medical masks. This year’s epidemic broke the records of previous years which was a surprise for many. In connection with this, at the end of the last century, a mathematical model was developed in Russia for predicting the spread of the flu, which worked successfully and which ITMO University specialists are currently trying to rework for our current reality. Vasiliy Leonenko, an engineer at the Research Institute of High-Tech Computer Technologies, and a participant in ITMO University’s recent Science Slam, expands on this further.
Is your job at the Research Institute of High-Tech Computer Technologies also associated with medicine like your colleague, Ilya Syomov, whom we recently interviewed?
Yes, the only difference is that Ilya is engaged in personal medicine, and I'm focused more on public health. We are concerned not with the diseases of specific individuals, but with the population as a whole. In particular, we analyze long-term data on seasonal flu epidemics, and try to learn to predict the dynamics of epidemic outbreaks, their magnitude and duration, and then on the basis of this information we can plan preventive measures such as the purchase of vaccines, the required number of beds in hospitals and so on.
How do you do this and what is this data?
The Research Institute of Influenza situated in St. Petersburg provided the data. The Institute also functions as the WHO National Influenza Centre of Russia. They monitor statistical data and if certain epidemiological thresholds are exceeded, they report it to the WHO and regional organizations. For example, the Federal Service for Surveillance on Consumer Rights Protection and Human Wellbeing compiles reports on the basis of the data. They used to have a very powerful Department of Cybernetics, in the 80s, I’ve been told, 5 graduates of the math department at St. Petersburg State University worked there and they were engaged in forecasting and modelling. They worked on implementing the unique Baroyan-Rvachev model, which was developed in the early 70s in Moscow and showed the spread of influenza epidemics amongst USSR cities.
The flu, unlike the majority of conventional acute respiratory infections, was brought into the country from the outside. Usually in South-East Asia there is a new strain of it, and air travel brings it to Western Europe. From Europe it usually came to the USSR through Moscow or Leningrad and after that spread across the whole Union. Leonid Rvachev was a graduate of the math department at MSU, and Oganes Baroyan, a Doctor of Medical Sciences, he worked in the Scientific Research Institute of Epidemiology and Microbiology in Moscow. Based on the analysis of traffic flows they created a model that gave the opportunity to get a very accurate prediction of the spread of the flu, it was epoch-making. To make predictions of the behavior of live systems is very difficult as there are too many random factors.
However, with the help of the model, in 5-7 days after the arrival of an epidemic to the USSR, there was a fairly accurate forecast of the number of people expected to contract the virus daily, for all the cities of the country for the entire period until the end of the epidemic. Imagine you are working in a regional office for the Ministry of Health and are told that in a few weeks an epidemic of the flu will start and all the relevant information is provided. For you this is priceless data – you will have time to plan, to calculate the date to announce quarantine in schools and to prepare for the possibility of associated diseases.
Unfortunately, this model is largely outdated due to the fact that with the collapse of the USSR, all previous traffic flows changed the movement of populations and the strains of the flu changed. What we want to do is to build a similar model which takes into account new circumstances and even if it fails to show results as good as before, at least the people of our country will know that such a model existed. In the West it is well known but in Russia very few people have heard that our scientists have created such a breakthrough model.
A couple of months ago, I spoke with a doctor at Botkin Hospital. He said that this winter the city didn’t have the necessary medication, because they expected the epidemic to come next year. If it works out to create your model, can we hope that epidemic preventions will be arranged in time?
There are different kinds of doctors, there are clinical doctors who work with patients and there are those who work with science. The degree of their awareness is different, because doctors at the clinic level don’t always see the big picture. I spoke about this with a Doctor of Medical Sciences, Yuri Grigorievich Ivannikov, one of the co-authors of the old model. Unfortunately Baroyan and Rvachev died a long time ago, but Yuri is alive and continues to work in St. Petersburg. He said that this year, the hysteria around the flu was largely contrived. I believe that it’s not appropriate to make big deal as to whether it is the “swine” flu or the “bird” flu for example, since all flus are of an animal nature. The media often uses these names to escalate the panic almost like a horror film trailer: “no one expected it and then the deadly swine flu came and now we all gonna die.” The situation was actually quite possible to predict.
But these days in Russia doctors are only predicting the starts of epidemics using statistical methods. Roughly speaking, current data is compared with an earlier period of time and if it exceeds a certain threshold, an epidemic is declared. Using statistics, you can try predict anything but you can hardly understand and explain it. When there is a good model, you can use it even when additional factors are added. For example, you can add a few more indicators, and the model can tell you when the epidemic will start. Not because last year was like that but because this was the most likely developmental path of the live system.
The problem with St. Petersburg is that the flu usually hits Russia through Europe and this is one of the first cities which “takes the hit”. During Soviet times data on incidences of the flu in Europe were not available, so they had to wait for the start of the epidemic in Leningrad, then they could measure the desired parameters and make predictions for other cities. Thus, for other cities the picture was clear in advance, and Leningrad and Moscow were in a less advantageous situation. These days, with the world becoming more connected it is possible to calculate migration flows in cooperation with Western colleagues and to make forecasts on a global scale, in particular to predict the course of epidemics in all cities of the Russian Federation in advance based on data from Western Europe.
What you are working on, how it will look in the end – like an app or some kind of server software?
Most likely it’ll be some kind of program, that can be rendered into a web interface. If you draw the level of incidences of ARI on a chart, you can see clearly that the number of registered incidences are small in summer while in September the level of the curve starts to rise and in the winter period an epidemic starts, and then the curve returns back to its previous level. You have to understand that the model does not issue a specific date for the start of the epidemic. It will not provide an exact date but rather an interval of time during which the number of incidences will be either good or bad.
Will the completed model need to “run” for some time before its' actual use?
It won’t be necessary to test it for a few years. There are so-called “retrospective” approaches, when we have all the data, but we pretend it doesn’t exist, and we try to predict it. We take incomplete data from the previous period, run the model and see if the results match the real data. Right now we have preliminary agreements with the specialists at the Research Institute of Influenza, if we can correctly build a model and demonstrate its first positive predictions then they will be ready to work with us.
Do you have a medical background?
I’m an applied mathematician, I studied in Omsk. I’ve been working in this field for some time, in my third year I wrote a paper on it. In general, in addition to mathematics, I dipped my feet into other branches, related to biology, sociology, I find them more interesting than physics. I know physics also has many interesting things, but I enjoy explaining natural processes with the help of Mathematics.
How did you end up at ITMO university?
I became involved with the Research Institute of High-Tech Computer Technologies through the scientific school workshop for young scientists. I was in my first year of post-graduate studies at the time and my supervisor gave me the information about the event. I already had experience participating in conferences before but this one stood out to me. They gave you time to think. It was also evident that the speakers took into account the different target audiences when preparing their presentations, and wanted to make sure that all students understood what they were saying. This is not so common in most Russian conferences.
Through consecutive experiences I got to know Alexander Boukhanovsky, the Director of the Institute, and when I was looking for a place to do my postdoc, I asked him if he could be my referee, but then he invited me to St. Petersburg. I had considered going overseas for some time and returning to Russia. But then I decided to work here, since I am more tied to Russia than any other country, and there is a demand for my work here.