Development of a Continuously Integrated Alert for Dengue Fever using multiple data sources in Brazil

Coelho, Flávio

Development of a Continuously Integrated Alert for Dengue Fever using multiple data sources in Brazil

Tuesday, 19 August 2014

Exhibit hall (Dena'ina Center)

Flávio C Coelho, PhD , Fundação Getulio Vargas, Rio de Janeiro, Brazil

Claudia T Codeço, PhD , Fundação Oswaldo Cruz, Rio de Janeiro, Brazil

Nildimar Honorio, PhD , Fundação Oswaldo Cruz, Rio de Janeiro, Brazil

Magda C Ribeiro, PhD , Universidade Federal do Paraná, Curitiba, Brazil

Jean Barrado, MS , Secretaria municipal de saúde de Belo Horizonte, Belo Horizonte, Brazil

Oswaldo d Cruz, PhD , Oswaldo Cruz Foundation, Rio de Janeiro, Brazil

Mauro Teixeira , Universidade Federal de Minas Gerais, Belo Horizonte, Brazil

Wagner Meira, PhD , Universidade Federal de Minas Gerais, Belo Horizonte, Brazil

Pedro A Martins, PhD , Rio de Janeiro City, Rio de Janeiro, Brazil

Crysttian A Paixão, PhD , Fundação Getúlio Vargas, Rio de Janeiro, Brazil

Sabrina Camargo, PhD , Fundação Getulio Vargas, Rio de Janeiro, Brazil

INTRODUCTION:

Dengue Fever, a viral mosquito-borne disease, causes high morbidity in a large portion of the tropical world. Its geographical range and epidemic dynamics is highly dependent on climate and urbanization. Dengue may appear in large and sudden outbreaks that challenge proper public health response and a timely alert is a useful and desired tool for informing decisions.

METHODS:

We have developed of an alert system that extends traditional approaches as it combines data from different sources: social media (Twitter), climate data (temperature, precipitation, humidity), mosquito abundance (trap data) and case data. To accommodate the heterogeneous data, the alert system provides two flags, codified as green, yellow, and red, for increased levels of risk. The first flag indicates “Dengue season”, which occur when weather conditions are proper for mosquito population growth. In a tropical climate with dry winter, temperature > 18ºC and humidity > 55% define the threshold condition for mosquito population growth. Other thresholds are required for different climates. The second flag measures the risk of dengue outbreak, observed by changes in the reference to “dengue” in social media, as collected by the Dengue Observatory project.

RESULTS:

Preliminary results reveal a strong linear relationship between dengue notification data and twitter data in many large cities(Fortaleza, Cases = −115.36 + 4946 ∗ tweets (R² = 0.93). Using a negative binomial regression, a threshold of 38 (29-50) was found for the tweets above which dengue incidence is compatible with an outbreak.

CONCLUSIONS:

The system innovates by proposing a continuous integration of many data streams and its modeling. Integrating streams of different time and spatial scales is the biggest challenge. In our system, the disease alerts feed directly into Rio de Janeiro's situation room, allowing for instant responses to outbreaks. Implementation of the system in other cities is being planned.