Surveillance and control of aedes-borne arboviruses in italy

The human surveillance of arboviral infections in Italy is coordinated by the Ministry of Health, implemented with the technical-scientific know-how of the Italian National Institute of Health (ISS) according to the National Arbovirus Response Plan 2020–2025³⁹.

In the Italian regionalized health system organization, Regional local health units are in charge of reporting the occurrence and most likely place of exposure of acute human arboviral infections to the National Surveillance System and coordinate activities in the event of health emergencies. Human surveillance of imported and locally acquired cases of DENV and CHIKV is active throughout the year, with case definitions aligned with the EU case definition⁴⁰. Any detection of Aedes-borne human infections in Italy (imported or local) is performed by clinicians (GPs or hospitals depending on case severity) who request laboratory confirmation (via PCR or serological tests). Any probable/suspected case as per the EU case definition⁴¹ triggers mandatory reporting within 12 h to the Public Health services and local response within 24 h including case investigation, vector capture, vector control and risk communication activities. In the presence of either confirmed or suspected human cases of arbovirus infection, whether imported or autochthonous, the competent health authority activates the vector control interventions within 24 h of notification. Control interventions are based on disinfestation of the affected area (~200 m radius around the place where the human case presumably was exposed) with insecticides, giving priority to adulticide interventions, both on public land and on private premises, and research and elimination of peri-domestic larval breeding sites, with “door-to-door” inspections of the homes included in the reported area.

If a locally acquired case is suspected, based on the patient interviews, local level active case investigation, proximity vector monitoring (including xenomonitoring) and control are further enhanced. This is why the presence of CHIKV and DENV also in local mosquito pools were confirmed during the observed outbreaks^9,18,42. Substance of Human Origin (SoHO) safety measures (CHIKV/DENV testing/suspension) are implemented at municipality level³⁸ upon confirmation of autochthonous transmission as well as a 28-day deferral from donation of people who traveled in the affected areas. Additional response measures include national and regional enhanced support for surveillance and active case finding (including voluntary screening campaigns⁹), national referral laboratory activities (including genomic epidemiology), medical entomology and risk communication. Activities conducted by local health authorities include individual case investigations that are not reported at national level. Similarly, control activities aimed at preventing mosquito population growth in the absence of suspected cases are not reported at the national level.

Epidemiological data

We retrieved data pertaining 1577 laboratory confirmed travel-related cases and 481 autochthonous cases of DENV and CHIKV, notified between 2006 and 2023 to the national database of human surveillance of arboviral infections in Italy hosted by the ISS. These records were obtained after excluding 64 travel-related cases and 10 autochthonous cases, due to data incompleteness or inconsistencies between the dates of symptom onset and notification.

Analyzed variables for each reported case included: pathogen laboratory confirmation, date of symptom onset, date of notification, age, sex, the list of clinical symptoms experienced by the case, classification of cases as travel-related or locally acquired, and geolocation of likely exposure for autochthonous cases and the country of likely exposure as identified during epidemiological interviews. Laboratory data were available only for cases diagnosed from 2013 onwards, encompassing a total of 1193 cases (371 autochthonous and 822 travel-related), with 67% confirmed via PCR and 33% through serological testing.

Statistical analysis of imported cases

We analyzed the time series of imported cases of both DENV and CHIKV in Italy using a generalized additive model. Generalized Additive Models (GAMs) were preferred over linear models because they accommodate nonlinear effects through smooth functions of covariates. This is particularly useful for modeling travel-related cases, which are expected to exhibit seasonal patterns rather than a simple, monotonic relationship with the months. Furthermore, GAMs mitigate overfitting by employing penalized regression splines, resulting in a more robust and interpretable model. Specifically, we considered the number of imported cases per month, aggregated by the date of symptom onset, as the response variable and assumed that it follows a Poisson distribution. A log link function was considered. The model incorporates two qualitative covariates: the disease (dengue or chikungunya), and whether the case importation occurred during travel restrictions put in place during the COVID-19 crisis (i.e., symptom onset between February 2020 and December 2020)⁴³. Regarding quantitative covariates, the year and month of symptom onset were modeled as penalized cubic splines and penalized cyclic cubic splines, respectively. To explore the relationships between each disease and both the month and year of importation, interactions between these temporal variables and the disease were considered, resulting in four smoothing functions. Finally, an autoregressive term of order two was included in the model to control for temporal autocorrelation. The analysis was carried out in a frequentist framework using the function gamm implemented in the R package mgcv (R Project for Statistical Computing, software version 4.3.2).

Following a similar approach, we investigated possible temporal changes in the delay associated with the notification of imported cases to the central health authority, defined as the time between the symptom onset of cases and their reporting to the National Surveillance System through the ISS surveillance platform. In this case, we employed a Generalized Additive Mixed Model (GAMM), by assuming that the notification delay follows a Negative Binomial distribution with log link. Whether the case was dengue or chikungunya was included as a qualitative covariate. The year and month of symptom onset were modeled as penalized cubic splines and penalized cyclic cubic splines, respectively. The interaction between these terms and the disease was accounted for as in the GAM. The region of importation was considered as a random effect.

We finally investigated whether the notification delay differed between autochthonous and imported cases by means of Generalized Linear Mixed Models (GLMM), assuming a Negative Binomial distribution with log link. To do this, we included autochthonous cases in our analysis and focused solely on regions and years where autochthonous transmission was documented (chikungunya: 2007 in Emilia-Romagna, 2017 in Lazio and Calabria; dengue: 2020 in Veneto, 2023 in Lazio and Lombardy). Two categories of autochthonous cases were considered: those with symptom onset preceding the outbreak detection and those occurring afterward. Considered covariates included the disease and case classification (i.e., imported, autochthonous preceding the outbreak detection, autochthonous following the outbreak detection); the outbreak was considered as a random effect. The analysis was carried out in a frequentist framework using the function glmer.nb implemented in the R package lme4 (R Project for Statistical Computing, software version 4.3.2). More details are provided in the Supplementary Information.

Transmissibility from surveillance of human cases

We estimated the net reproduction number (R_t) associated with each arboviral outbreak occurred in Italy between 2006 and 2023, by applying a consolidated Bayesian approach^44,45. Estimates of R_t were obtained using a Markov Chain Monte Carlo (MCMC) sampling method and applying the renewal equation to the time series of symptomatic confirmed cases by date of symptom onset. This probabilistic approach accounts for uncertainty in R_t estimates, making it particularly well-suited for real-time epidemic analysis^8,9. We assumed a Gamma-distributed generation time with a mean of 12.4 days and a variance of 18.5 for CHIKV¹⁴, and with a mean of 18 days and a variance of 66 for DENV⁸. The analysis was performed using codes widely adopted for estimating the net reproduction number available at https://github.com/majelli/Rt. More details are provided in the Supplementary Information.

Assessment of onward transmission risks from entomological and climate data

We assessed the potential risk of onward transmission for chikungunya and dengue at the national level during the outbreak years: 2007, 2017, 2020, and 2023. To this aim, we leveraged a model recently developed to estimate the spatiotemporal abundance of Aedes spp. mosquitoes and the consequent risk of autochthonous arboviral transmission at detailed spatiotemporal scales²⁶. The model was extended to incorporate year-specific daily temperature records, account for more refined data on human density (at a resolution of 100 m × 100 m), and provide for each year of interest (1) daily estimates of the CHIKV and DENV reproduction number R₀ (i.e., the average number of secondary human infections arising from a primary human infector in a fully susceptible population), (2) the likelihood of onward transmission, defined as the probability of experiencing local transmission chains after case importation, and (3) the duration of epidemic risks (defined as the number of consecutive days in which R₀ exceeds the epidemic threshold of 1). Specifically, the absolute density of adult mosquitoes per hectare on a given day was modeled as a logistic function of the mean temperature observed in the preceding days, where the maximum is defined as proportional to a location-specific climatic suitability index (σ_i)²⁶. This approach assumes that seasonal variations in mosquito abundance are primarily driven by the persistence of favorable temperature conditions throughout the mosquito lifecycle and that higher abundances are expected in locations associated with greater climatic suitability, as described by the following equation for any day d and location i:

where \(\widetilde{T}\left(d,i,w\right)\) represents the average temperature recorded at location \(i\) in the \(w\) days that precede \(d\), \({T}_{0}\) and \(k\) are the midpoint and steepness parameters of the logistic function, and \(\alpha\) is a scaling factor that accounts for variations in trap efficiency across mosquito capture data used for model calibration, as well as consequent rescaling factors to convert estimated captures into per-hectare abundance. The climate suitability index (\({\sigma }_{i}\)) was assumed to be driven by temperature and precipitation observed in location i as in Zardini et al.²⁶. Overall, the model was informed by georeferenced presence-absence records of Ae. albopictus across 4372 locations in Europe and 300 time-series of female adult mosquitoes collected between 2007 and 2018. The resulting estimates of vector abundance were combined with human density retrieved from the WorldPop data⁴⁶ to assess arbovirus transmission potential. Specifically, for each location and year of interest, we calculated the basic reproduction number (\({R}_{0}\)) and the likelihood of onward transmission using a Susceptible, Exposed, Infectious, Recovered (SEIR) host and Susceptible, Exposed, Infectious (SEI) vector model. Both measures were computed following standard approaches as proportional to the vector-to-host ratio (\({N}_{v}\left(i,d\right)/{N}_{h}\left(i\right)\)), while incorporating disease- and vector-specific parameters obtained from the literature. More details are provided in the Supplementary Information.

We compared estimates of R₀ resulting by assuming Ae. albopictus as the only vector for arboviral transmission in Italy with R_t values estimated from human cases identified during the different outbreaks. To do this, we combined R₀ estimates associated with geographical patches characterized by a population density greater than 10 residents per hectare and falling within a square of 900 m × 900 m centered on the site of likely exposure of each autochthonous case. The square dimension was determined based on the analysis of past chikungunya and dengue epidemics, suggesting that most transmission events for both diseases typically occur within a 900 m radius^8,14,45. Daily temperature records were retrieved at a spatial resolution of 0.1° × 0.1° from the E-OBS dataset, which is a collection of high-resolution gridded climate data covering Europe and part of the Copernicus Climate Change Service⁴⁷. Human density data were obtained at a spatial resolution of 100 m × 100 m from the WorldPop data⁴⁶. The analysis was performed using codes developed ad-hoc by our team in the programming language C, available at https://zenodo.org/records/10203374. Maps were created using QGIS software version 3.30.2. Administrative boundaries of the Italian regions were retrieved from the Italian Institute of Statistics⁴⁸. More details are provided in the Supplementary Information.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.