NEW estimates by the Institute of Health Metrics and Evaluation (IHME) of the University of Washington, according to which the number of deaths from coronavirus in Croatia could reach 1,500 by the beginning of October, resonated on social networks and alarmed the Croatian public.
Some of the comments raised doubts about the reliability of such dramatic predictions.
Outbreaks of new hotspots
As we have already written on the Index, IHME presents interactive graphs showing the current number of deaths from coronavirus and estimates of how the number of deaths could move in the future, in the next few months.
On Wednesday, we published estimates for Croatia that reached August and were based on data on the number of infections, restrictive measures and other factors before the recent increase in the number of cases, ie before the emergence of new outbreaks in Zagreb, Đakovo and Zadar.
At that time, the IHME predicted an increase in the total number of deaths from 107 to 117 for the beginning of August. At that time, no significant increase was predicted, and the curve was fairly flat.
Soon about 1,500 deaths from covid-19
But IHME then released new ones on Thursday, updated data for Croatiain which they go back to October and show a much steeper curve and an estimate of 1498 dead by October 1 in the blackest scenario. According to the new projection, the curve is significantly steeper, and predicts 225 dead for the beginning of August (chart above).
In a more optimistic scenario, in which a majority of about 95% of people would wear masks and adhere to certain basic hygiene and distance measures, only 270 people would die by October 1st.
The chart shows that the daily death toll in early August could jump from one to seven people a day. At the end of September, we could have as many as 25 covid-19 deaths a day, and we could need about 240 respirators.
Interestingly, the IHME for most other countries does not predict such rapid growth. For example, for neighboring BiH, where the epidemiological situation is currently less favorable, it predicts an increase in the number of deaths from today’s 178 to 188 by October 1 (chart below).
For Sweden, it predicts an increase in the total number of deaths from the current 5,280 to about 5,753, and for Germany from today’s 9,012 to 11,624.
Three American scenarios
Since these figures and curves for Croatia may seem like an exaggeration, we asked physicist Dr. sc. Davor Horvatić from the Zagreb Faculty of Science (pictured below) and his colleague, mathematician dr. sc. Petra Posedel Šimović, who teaches at Agronomy and FER (pictured below), for their expert interpretation of the model that gave such predictions.
“These models are based on a number of assumptions,” says Davor Horvatic.
“The graph of the total number of deaths represents three scenarios. The first, ‘current projection’, which could be translated as a projection according to the current state, is marked in gray. The second, ‘mandates easing’, which implies permanently relaxed measures, as we now have in Croatia, is marked in red. It should be borne in mind that in some countries the measures have not yet been relaxed. The third, the ‘universal masks’, which means that about 95% of people wear masks, but also the introduction of stricter measures if the death toll rises above an acceptable threshold, is marked in green. It is interesting to note that in the case of Croatia, the first two scenarios are identical, which can be seen by the fact that the gray and red curves completely coincide. What does that mean? This implies that anti-epidemic measures in Croatia are expected to remain relaxed even if we reach the daily number of 8 dead per million inhabitants, or 32 dead per day. This number of 8 per million authors of the model did not come by chance. Namely, he assumes that the states, when they reach that number, should introduce stricter measures every six weeks. This is a limit that no one would want to cross, and at the same time no one would like to reintroduce strict measures for the sake of the economy. This is the idea of a scenario called current projection. Croatia has never had 32 deaths from covid-19 in a day, nor is it expected to reach that number in the coming months. This chart for the case of Croatia essentially shows what could happen to us in the best and worst case. The authors write in the summary that based on the experiences of various countries, they reached that maximum acceptable number of deaths. The predictions are based on a number of parameters collected for all the countries represented, ”explains our physicist.
How reliable is the presented numbers?
Some commentators, skeptical of the fact that the chart for Croatia has a large 95% reliability range, concluded that this should mean that the projection of 1,500 dead by October 1st is very unreliable. But such an interpretation does not stand.
“Since the prediction was made for a fairly large period, it is logical that the 95% confidence range in which the model predictions can move is quite wide,” says Posedel Simovic.
“In that context, the authors present the explanatory variables on which they base their predictions. These are, for example, human mobility, testing rates, use of masks, mortality rates from pneumonia, air pollution, percentage of the population living by the sea, smoking, population density, opening of educational institutions such as kindergartens, schools and colleges, etc. The average estimate for Croatia is that around 1,500 people could die by October 1st. But the upper limit of the 95 percent confidence interval is 16,000. The range is expected to be huge, from 1,000 to 16,000, because we predict the number of those infected in the long run, so the uncertainty is greater. It should be borne in mind that the confidence interval is a random variable. This means that if we insert today’s data into the model, we will get a range that goes from about 1000 to 16,000. When new data is entered in a few days, it is possible that this range of reliability will narrow to 1000 to 14,000. But it can also increase to go over 16,000. The probability that 1500 is found in this interval is not 95%, but it is the most probable value. To clarify better. The number of infected in such models is a random variable that depends, as we have stated, on a large number of parameters. Imagine that we simulated 100 scenarios to predict the number of infected on a given day in the future; 95 numbers would fall in that interval, and 5 would not. Countries that have smaller confidence ranges than Croatia are the ones that probably have smaller jumps in the number of infected, so their projections are more in line with existing trends, “says Posedel Simovic.
Horvatić points out that when predicting the model, we should keep in mind that uncertainty will generally grow as far as possible in the future, we want to predict things.
“The parallel we can draw is meteorology. When the situation over an area is relatively stable, it is not a problem to give a more reliable forecast for a longer period of time. In case of unstable atmosphere, some sudden significant weather changes, the forecast will change constantly. their models with a better understanding of physical models and all factors influencing forecasting, smaller volumes observed in the atmosphere to increase numerical accuracy.Also, a team of scientists predicting covid-19 emphasizes that the uncertainty interval decreases when source data are widely available and consistent, and growing if the data is limited or in conflict with the dynamics observed so far.There are no perfect models.Models can be better only if they are constantly tested and refined in accordance with the observed, as in this situation where a lot of factors affect prediction for a certain number of days in the future “, explains our physicist ..
Saving tourism, ie the economy
Horvatić points out that the worst-case scenario for Croatia, which is a curve called ‘mandate easing’, shows what could happen if the relaxed measures of social distance, which were introduced in Croatia a month ago, continue, or if new restrictions are not introduced.
“I guess that, due to the protection of the economy and saving the season, it will be like that, if we don’t have a big increase in mortality. The third scenario, called the ‘universal masks’, predicts what would happen if about 95% of the country’s population wore masks in public. This percentage is also not randomly selected. It is actually the highest percentage ever recorded during the covida-19 pandemic, most likely in some Far Eastern countries like China or Singapore. In this scenario, as we have already said, it envisages the introduction of some stricter measures every six weeks in the event that the number of deaths exceeds the threshold of eight victims per day per million people. This is then considered the best possible scenario. Since Croatia is in the middle of the tourist season, the idea is that restrictions should be avoided, while people should follow the instructions on social distance, hygiene and wearing masks “, says Horvatić.
We need to figure out what we can do to keep the script from being the blackest
Posedel Šimović believes that the graph and the blackest projection should not be seen as a threat, but as one of the possible scenarios that tells us how we should behave in order for the best scenario to happen.
“I would interpret that as a message: ‘Caution, it’s possible that this scenario will happen to us.’ It can happen, but no concrete probability should be attributed to it, neither 95% nor 30%. I need to take this as information that tells me that in my own domain, according to my capabilities, I have to be extremely careful. I should also have confidence that the authorities will make sure that appropriate measures are taken to prevent the collapse of our health care system. Finally, we now know more about the virus than at the very beginning of the epidemic. It is up to the authorities to act on the basis of experience and new scientific research, respecting all new insights that point to caution in order to protect each individual and all those who belong to the risk group especially sensitive to covid-19 such as the elderly and the sick, “she said. mathematician.