Epidemics Under the Microscope: How Brazil Scans Dengue and Chikungunya Outbreaks ✛

Original: "Large-scale Epidemiological modeling: Scanning for Mosquito-Borne Diseases Spatio-temporal Patterns in Brazil"
arXiv:2407.21286 · 2024-07-31 · CC BY 4.0 · 3 min
Over 14 years, data were collected on 18,567 dengue epidemics and 2,199 chikungunya epidemics across all 5,570 municipalities in Brazil.

A fisherman casts his net into murky water, and an hour later the catch is thrashing on the deck. That's roughly how Episcanner works: it sifts through 14 years of disease records and pulls out every fever outbreak in every one of Brazil's 5,570 municipalities. From 2010 to 2023, the net caught 18,567 dengue epidemics and 2,199 chikungunya epidemics. Previously, such data were collected piecemeal, and comparing north with south was like comparing catches from different rivers. Now all the fish are in one boat.

Each epidemic in the net gets a passport: the basic reproduction number R0 (how many people one patient typically infects), the peak week, duration, and size. For dengue, the average R0 was 1.97; for chikungunya, it was 2.12. This means that one chikungunya patient transmits the infection to slightly more people than a dengue patient does. The average duration of a dengue epidemic is 26.6 weeks, which is more than half a year. Chikungunya is a bit shorter at 25.5 weeks. But these are national averages; in reality, the picture is as varied as fish scales.

In northern Brazil, dengue epidemics last longer than in the south. In some northern states, an outbreak can drag on for nearly a year, while in the south it may fizzle out within a couple of months. The peak of dengue occurs earlier than that of chikungunya, and earlier in the west than in the east.

Why is that? As the water warms, the fish behave differently. The epidemics are spread by Aedes mosquitoes, whose activity depends on temperature and rainfall. The researchers fed climate data into a machine learning model to predict the peak week. It turned out that January temperature and the El Niño index from the last quarter of the previous year are important for the forecast. When El Niño is positive, epidemics arrive earlier; when La Niña, later. A warm ocean seems to spur the mosquitoes on.

Before this work, there was no unified picture of how dengue and chikungunya epidemics differ in time and space across the entire country. Now there is an open dataset, and health authorities can plan where and when to set mosquito traps or stockpile medications. The fisherman isn't just showing off his catch; he's pointing out where in the river to cast the net next time.

The net has not been cast for the last time. The data will be updated weekly, and every new epidemic will join the overall catch. Eventually, the net will also include information on dengue virus serotypes and the population's immune status, to better understand why in some places the fish run in schools and in others they come one by one. Then the forecast will become even more accurate, and the fisherman will be able to warn of storms in advance.

The average duration of a dengue epidemic in Brazil is about 27 weeks, which is more than half a year. In northern states, epidemics can last nearly a year, while in the south they may last only a couple of months.

J(t) = L - L[1 + \alpha e^{b(t-t_j)}]^{-1/\alpha}
Describes the growth of cases during an epidemic; L is the final size, t_j is the peak week, alpha and b are shape parameters.
R_0 = \frac{1}{1-\alpha}
Shows how R0, the average number of infections caused by one infected person, is derived from the shape of the epidemic curve.
Concepts
epidemic model epidemic Multivariate ENSO Index Niño 4 spring predictability barrier ENSO IRI El Niño index interannual variability