Dengue impacts travelers and local populations in areas of ongoing transmission. This dashboard uses current and historical dengue and travel data to estimate the number of cases reported in travelers arriving in the United States and Europe and the relative risk to travelers visiting endemic areas.
ESTIMATE: –
This is the number of cases expected to be reported in the Centers for Disease Control and Prevention ArboNET national surveillance system. This is likely an underestimate of symptomatic cases as many cases may be mild and not recognized by clinicians. It is also an underestimate of infections, as many dengue infections result in no apparent disease at all. Thus the actual number of sick travelers with dengue and the number of infected travelers are higher than these estimates or the reported data.
Several states are combined in groups due to shared airports and traveler data tied to airports: Connecticut, New Jersey and New York; Delaware and Pennsylvania; and Washington DC, Maryland, and Virginia. Additional local movement between states (e.g., New Hampshire travelers flying from Boston) is not explicitly modeled but state-specific adjustments are made to account for some additional movement.
Risk to travelers is generally proportional to the prevalence of infection in the local population but is not the same for all travelers. Risk to a traveler depends on where they spend time, how long they stay in a country, and how they protect themselves from mosquitoes. For example, a business traveler spending a few days in an air-conditioned hotel generally has lower risk than a traveler spending a month staying with friends or family. The risk here represents risk for an average traveler from a non-dengue endemic region. Similar to the travel-associated cases, this risk is only estimated for reported cases and is therefore an underestimate of the true number of cases and infections in travelers.
All the estimates include substantial uncertainty. We provide a downloadable mean, median, 50% prediction interval, and 90% prediction interval for each estimate. Many of the estimates have high and asymmetrical uncertainty, which can produce mean estimates that are much higher than the median. The median represents the midpoint of all estimates and is therefore often a more reliable single indicator than the mean when the estimates are skewed, which is generally the case here.
The 50% prediction interval is the 25% and 75% quantile, indicating a 50% chance that the reported cases are within that range, a 25% chance that cases are lower and a 25% chance that they are higher. The 90% interval is similar but wider, with only a 5% chance that reported cases are lower and a 5% chance that they are higher. Uncertainty for a given location may be related to variability in local surveillance data, uncertainty in the relative risk to travelers, and low numbers of local cases or travelers.
Estimates are made in a 4-step process.
We compiled historical surveillance data from the World Health Organization, OpenDengue, and multiple national data sources for the French Antilles, Mexico, Nicaragua, and Puerto Rico. We collected travel-associated dengue case data from previous research on dengue in travelers (Hitchings et al., PLOS NTD 2025), the European Centre for Disease Prevention and Control, France, Germany, and the US Centers for Disease Control and Prevention. We compiled traveler data from OAG.
Traveler risk is proportional to local risk but not the same. Specifically, risk can vary depending on the particular exposure of each traveler (e.g., staying in a hotel vs. a home or staying for a week vs. a month). In any given location, individual traveler exposures vary but they also can vary systematically between locations. Moreover, local surveillance systems are also varied, so different infection prevalence across countries may not be reflected in incidence rates calculated from each country's surveillance data. Detection of cases in travelers may also be varied.
We therefore implemented a generalized linear modeling framework to estimate the number of infected travelers moving between one source location and one destination location. In the model, we included monthly traveler volume as an offset, local prevalence (reported incidence divided by the population size) with both a linear and a squared term, a term for estimated prevalence when no surveillance data are available, a source location-specific effect, a destination location-specific effect, a source-destination effect for select location pairs, and a year effect. The year effect allows for additional unexplained variability between years. We used a negative binomial model with a log link fitted in Stan to fit this model to the historical data and forward-fold testing to validate its performance.
We update global dengue data on a weekly basis; however, there are many delays in reporting which vary across countries and particularly impact the most recent data. We first assess the completeness of data for recent months and then estimate up to 6 months of incidence using a simple, country-specific forecasting model. The country-specific forecasting models consider historical seasonality and the most recently reported data from that country. Estimates are always made for the current month (nowcast) and the next month (forecast). If no data are available for the last 6 months, we do not make a nowcast or forecast for that country. If no data are available for the last 5 months, we make a nowcast but no forecast for that country.
Two countries are treated differently: Cuba and Haiti. There is very limited surveillance data for these countries, so relative prevalence is estimated by fitting a model to the travel-associated case risk. The time series of relative prevalence is then used in the same way as the surveillance data to forecast recent trends.
We use the traveler risk model described above to estimate recent, current, and future risk to travelers using the most up-to-date dengue data available, short-term forecasts for incidence, and travel data from the same month of the previous year. Travel follows strong seasonal trends and long-term growth except for major disruptions such as the COVID-19 pandemic. While we do not have access to real-time travel data, testing on recent data suggests a 1–2% error.
Supported by cooperative agreement CDC-RFA-FT-23-0069 from the CDC's Center for Forecasting and Outbreak Analytics.
