Dengue continues without respite in America. “Between epidemiological weeks 1 and 44 of 2024, a total of 12,479,437 suspected cases were reported (cumulative incidence of 1,307 cases per 100,000 inhabitants). This figure represents an increase of 204% compared to the same period in 2023, and 381% compared to the average of the last five years,” indicates the Report. Epidemiological issued by the Ministry of Health of Argentina on December 9.
Until the end of November, 13,647 suspected cases of dengue were reported in Argentina, of which 195 were confirmed (180 without travel history), according to the Bulletin. And he points out a no small fact: “From week 42 to 47, a progressive increase is observed in the detection of confirmed cases of dengue.”
All specialists expect that there will be a lot of dengue, because in addition to the previous history, there is also it in neighboring countries. Towards the end of October, ANMAT approved the first dengue detection kit developed by CONICET scientists. Now, an interdisciplinary team of experts is working on predicting outbreaks using an artificial intelligence tool. Mapping will make it possible to take targeted measures and anticipate the proliferation of cases where the mathematical probabilities point to an outbreak. The measures will be implemented in the province of Buenos Aires based on an initiative of the Buenos Aires government and the National University of San Martín (UNSAM). The physicist and researcher Ezequiel Alvarezof the International Center for Advanced Studies (ICAS) and the UNSAM School of Science and Technology, He is the one in charge of the project, based on advanced Bayesian AI techniques.
News: What are the characteristics of this type of AI?
Ezequiel Alvarez: Bayesian Artificial Intelligence is a particular kind of AI that is very useful for analysis when one has a lot of data. This type of AI allows you to see the internal structure of the data and how its correlation gives you information about what is really happening in the variables that you do not see. Something that affects mosquitoes is the socioeconomic level of the apples, which is observable; One can tell how many sick mosquitoes there are in a region from the number of sick people reported. One can see the weather, the number of sick people reporting, the socioeconomic level, how many people went to the hospital, how many people called the citizen assistance number 148 in each place. But we also have a number of variables that we don’t see, such as the number of mosquitoes in general and the number of sick mosquitoes in an apple. One cannot have an exact number, but it is feasible to obtain a probability distribution on that variable, which is called a latent variable.
News: And can that Bayesian AI learn which variables are of interest?
Alvarez: Absolutely. When you start to observe things like the weather, the number of sick people, the socioeconomic level and how that evolves, the system learns how the other variables that you don’t see take on a probability distribution to better explain the observable variables. So that relationship, that interrelation between the data means that automatic learning, machine learning, can learn the variables that interest us. In this case of an epidemic, what matters to us is how many sick mosquitoes there are per block or where the greatest number of sick mosquitoes are. Having these probabilities allows us to make decisions well in advance because we will be able to know where an outbreak is going to arise because our information, our artificial intelligence system, begins to show where it is most likely that there will be sick mosquitoes that are going to bite people who have not yet bitten. , because the mosquito needs to develop the disease in its saliva.
News: What data is useful to the system?
Alvarez: We have divided the Metropolitan Region of Buenos Aires (AMBA) into hexagons of three blocks. And we observe the weather and its forecast ten days in advance, the history of the weather, the temperature, the number of calls to 148. The geolocation of each of the reported cases, because then based on the spatial and temporal distribution From these cases, we can distinguish the probability of correlation between cases that are close in space and time, which is more likely to be a dengue outbreak. That is why we need to know what place they are from, the density of inhabitants of that place, the number of people who have reported symptoms and who are suspects and who are in the National Health Surveillance System, the socioeconomic level of each block, the history of that place before. A hexagon that is covered with people who have mosquito nets and air conditioners, or who live in buildings between which there are almost no green spaces or pools or flower pots where water can fall, is not the same as a place mostly populated by houses where maybe they have buckets outside. With all this, the AI system makes a distribution map of where the most fertile places are for the disease to develop.
News: The tool is fundamentally preventive…
Alvarez: Yes, the goal of this is to get dengue outbreaks before they occur. This is a project that is half research and half application, with which it is being developed. We started it in September, we have used the information from last year and now we are using the information that comes on stream, that which is arriving. There has not yet been any outbreak in AMBA, but we are beginning to have signs of outbreaks. The idea is that it can be anticipated in various ways. One of them, not ready yet because it is part of the development, is when we have the distribution of infected mosquitoes in the AMBA. And another way, which is what we are using now, is a system based on differentiating people who are truly in a dengue outbreak, from what we call background, which are those who have symptoms but who are likely not to have them. dengue.
News: How is that?
Alvarez: Due to the geographical distribution and with the information that I mentioned before, we have a method that allows us to distinguish the probability that people who report suspicious symptoms are from an outbreak and people who are from a background. And essentially, one of the most powerful ways is purely mathematical, because you know that the outbreaks are located in space. Because there is a correlation of suspected cases in places close in space and also close in time. They are techniques to distinguish background signal, the same ones that physicists use in the work we do to search for elementary particles at the Large Hadron Collider. In this case, what we achieve is to make a map of the probabilistic distribution of the outbreaks.
News: There is a widespread misconception that artificial intelligence can do anything.
Alvarez: Artificial intelligence does not do magic, it is a probabilistic tool, both in the GPT chat and in this case of data analysis, and what we do is give a probabilistic estimate of the evolution of the pandemic, both of the outbreaks and of the general evolution or the transfer of cases from one block to another. This means that we can give an anticipation of what is expected 10 days from now, but from a probabilistic perspective. It would never be possible to make an anticipation that has to do with a family from Lanús that goes to eat a barbecue in San Martín, everyone gets infected and takes dengue to Lanús, it is impossible for us to anticipate it. What we do is a probabilistic version, but it is detailed enough so that the government of the province of Buenos Aires can take public policy actions, which they are already beginning to take, based on the results we obtain. So in that sense it is working and it will work better and better because we are incorporating new tools and we are developing the mathematics and artificial intelligence part better, because the model itself is learning, but we are also learning, because we are humans implementing algorithms of artificial intelligence.
For the tool to work, data is needed. “Due to the volume of data needed in the province of Buenos Aires, it is the State that can collect it, because it has primary care, the health coordination network, and the human and informational resources to be able to respond to an epidemic.” says Pablo Palmaz, Undersecretary of Interinstitutional Relations of UNSAM. “We need a lot of articulation of public data. Data is required from different parts of the State and that has to be super well-oiled for it to work well. It is in real time, every day.”
Álvarez’s team will provide the information. Then, the political decision to respond will depend on the Ministry of Health. “The most important thing in dengue is the mess,” Palmaz concludes. It is the way to end the epidemic. House-to-house actions will be carried out. If you understand where it is happening, you can carry out a crackdown operation in a certain area. “It’s not just turning a cover over with water: it’s taking care of the grates, cleaning, among other things.”
Image gallery
