Analysis

We first performed a descriptive analysis of case characteristics, focusing on place of residence, demographics, month of onset and country and region of infection. Considering the short incubation period of dengue (< 10 days), we considered that the month of onset was the same as the month of infection. As proxy for the risk of infection, we calculated the travellers’ infection rates per 100,000 travellers (TIR) and the 95% confidence intervals (CI) around the TIR estimates based on a Poisson distribution. The TIR was calculated following Formula 1:

We identified seasonal patterns and trends in the number of travel-related cases and TIR by using centred moving averages: 3- and 6-month moving average to describe the seasonality (bi-annual and annual peaks, respectively) and 12-month moving average for the overall trend.

For the trend analysis, we used a harmonic regression model including Fourier terms for capturing seasonality. In this model, we adjusted for seasonality using three pairs of sine and cosine with 12, 6 and 3 months as length of the periods to capture both the two yearly peaks and to allow to capture the 'wavy' pattern in the data. This analysis was performed by regions.

We analysed the association between TIR for a given year and a given country of infection and disease incidence rate in the local population using a linear regression. We selected the four countries of infection (Cuba, French Polynesia, Réunion and Thailand) with the highest number of cases in their region, Americas, Oceania, Africa and Asia, respectively. For each selected country, the disease incidence rate in the local population was obtained by dividing the number of cases in the local population by the population estimate (cases/100,000 population).

To define the risk of autochthonous transmission in Europe, we assessed the association between the number of travel-related cases in receptive areas and the number of autochthonous outbreaks that occurred in Europe from 2015 through 2019 [9].

A receptive area was defined as a NUTS-3 region where Ae. albopictus was established and at a time when vectorial capacity was assumed sufficient to facilitate local transmission, estimated to be between 1 July and 31 October. The period of sufficient vectorial capacity was defined based on the recorded occurrence of autochthonous vector-borne transmission of dengue virus in Europe since 2010 (all outbreaks occurred during the period July to October) [9]. Per year and per European country, we selected the number of travel-related cases with date of onset between 1 July and 31 October and notified or residing in a NUTS-3 region where Ae. albopictus was established.

Alternatively, when the place of notification and place of residence were not available or not available in the right format (e.g. NUTS-2), we assumed that the geographical distribution of the travel-related cases was following the geographical population distribution of the country. Consequently, to estimate the number of cases in receptive areas (µ), we used the following Formula 2:

where N is the number and pop the population.

We used Stata software release 14 (StataCorp. LP, College Station, United States) for all data management and statistical analyses.