Can the Best Climate Models in the World Predict Cases of Dengue?

Two ISGlobal teams were among the 16 teams chosen to meet a worldwide challenge launched by the United States: to predict the incidence of dengue in Iquitos (Peru) and San Juan (Puerto Rico). Xavier Rodó, ICREA Research Professor and Head of ISGlobal’s Climate & Health programme, explains the process and the results.

ISGlobal has taken part in a new initiative that has, for the first time, analysed the current capability worldwide for predicting outbreaks of dengue infection. Dengue is a major public health concern in many parts of the world and the problem is increasing because Aedes aegypti—the principal vector responsible for transmitting this infectious disease—is a mosquito that thrives in an urban environment.

Dengue is a mosquito-borne viral infection found most commonly in the Caribbean, the Americas and South East Asia. While the course of the disease is generally mild, the infection can also cause severe flu-like symptoms and even prove fatal. According to World Health Organisation (WHO) estimates, as many as 100 million dengue infections occur worldwide every year in over 100 countries, putting nearly half of the world's population at risk.

The research challenge launched by the U.S. Government’s Pandemic Prediction and Forecasting Science and Technology Working Group chose the best computational models currently or previously in use somewhere in the world to generate forecasts of the incidence of dengue fever. They then gave those sixteen teams from all over the world the same challenge: to predict outbreaks of dengue fever in Iquitos, Peru, and San Juan, Puerto Rico, based on disease incidence and climate data.

The research challenge launched by the U.S. Government chose the best computational models currently or previously in use somewhere in the world to generate forecasts of the incidence of dengue fever

Two of the 16 models chosen were developed by researchers from ISGlobal, the only teams from Spain who took part in the challenge. The first of these was a Bayesian model developed by a team led by Rachel Lowe, a visiting scholar at ISGlobal. The second, which I myself developed, is a complex dynamic model.

One part of the challenge was to generate the best operational forecast of dengue in the city of San Juan, Puerto Rico—the gateway for many of the cases of dengue that are imported into the United States and a location where all four serotypes of the disease are present. The dynamics of the disease are very different in Iquitos, the second location studied.

Dengue and climate data for Iquitos, Peru and San Juan, Puerto Rico. Fig. 1. PNAS. doi.org/10.1073/pnas.1909865116. 

This type of initiative is very interesting because it requires all the participants to generate forecasts under the same conditions, making it possible to test the different models. Predicting epidemics of climate-modulated infectious diseases has become a priority in many of the countries most affected by climate change, although appropriate tools are still scarce.

This type of initiative is very interesting because it requires all the participants to generate forecasts under the same conditions, making it possible to test the different models

The evaluation of the computational models based on climate and population data, published recently in the scientific journal PNAS, showed that our model and some other tools could improve early season forecasts as well as predict the week when the epidemic would peak and how many cases would occur during the year. Other models, in contrast, generated satisfactory forecasts of maximum incidence at the peak of the epidemic. However, none of the models evaluated performed as well as the ensemble multimodel developed by bringing together the forecasts generated by all the models in the project, which demonstrated predictive capacity for all the objectives.

Forecast skill by team, forecast week, and target in the testing seasons (2009/2010 to 2012/2013). Fig 3. PNAS. doi.org/10.1073/pnas.1909865116. 

As in the case of climate forecasts, the multimodel appears to be the best choice for forecasting arboviral infections. This is probably due to the many key aspects of epidemiology that are still poorly understood and the way the different models that make up the multimodel complement each other.

None of the models evaluated performed as well as the ensemble multimodel developed by bringing together the forecasts generated by all the models in the project, which demonstrated predictive capacity for all the objectives

This project has made it possible, for the first time, to conduct a joint exercise similar in intent—though not in magnitude—to the one carried out by the Intergovernmental Panel on Climate Change (IPCC) to predict the climate of the future. The exercise has allowed us to identify the limitations of current computational models and the areas in which further research is needed. The United States is already promoting the continuation of the challenge and the expansion of the concept to similar initiatives for other diseases, such as influenza. Similar international efforts are also needed, given the unexpected growth in the prevalence of these kinds of diseases worldwide in recent years.