Introduction

Chickenpox is a very contagious disease caused by varicella-zoster virus (VZV) that mostly affects children in endemic areas, although the age group affected during epidemics depends on the background prevalence of VZV in the area and the immunization coverage rate. The incidence and complications of varicella infections have decreased after the introduction of varicella vaccine, especially in countries starting vaccination in children 12 to 18 months of age [1]. There is a remarkable difference in the epidemiology of varicella infection between temperate and tropical regions. In temperate areas, varicella has a clear seasonality, with one or two peaks each year, usually in winter and spring [2–4], affecting preschool and school-aged children. However, in the tropics, it occurs all year round and affects adolescents and young adults, especially in rural populations [5–7]. Garnett et al. in 1993 proposed that the combination of high ambient temperatures and humidity in tropical regions might adversely affect VZV transmission [8], but reduction of social interaction and other factors may also intervene [7, 9]. In general, the factors involved in the seasonality and different epidemiology of varicella infections are not fully understood, but VZV genotypes, climatic factors, and population density have been considered to play a role in the dynamics of disease [7, 10, 11]. In addition, ultraviolet (UV) radiation has been hypothesized to be involved in the worldwide distribution of VZV genotypes [10], but there is no experimental or epidemiologic confirmation of this hypothesis [11].

To study the influence of climatic factors on the epidemiology of varicella infections, multivariate regression models considering all the plausible variables are applied, using time-series analysis when necessary. To date, there is only one report using these techniques to determine the association of hospitalization for varicella with temperature, humidity, and wind speed [12]. However, since meteorologic conditions are closely associated variables, the analysis of a small number of these factors cannot exclude false associations or establish the degree of association between the variables involved. Therefore, we need more studies that consider the most relevant meteorologic factors together. In the present study, we investigated the association between the number of cases of varicella in a population and the values of temperature, humidity, rainfall, water vapor pressure, atmospheric pressure, wind speed, and solar radiation.

Materials and methods

The study was conducted on the island of Mallorca (Spain), with a population of 876,147 inhabitants, and winter and summer mean temperatures of 10.2 and 24.3 °C, respectively. The number of cases of varicella infection in the period 1995 to 2012 was used to study correlation with meteorologic factors. Varicella cases were obtained from the Balearic Islands Regional Government Network of Epidemiologic Surveillance. In Spain, varicella infection is a disease of compulsory notification, but no data about gender and age are available, as the only data reported by primary care physicians are the total number of cases by week.

In our region, although the varicella vaccine was commercialized for private use in 2003 (with an estimated coverage of 20 %), it was not included in the National Immunization Program until 2005 (with 95 % coverage), for 11-year-old children who had not suffered the illness and had not been previously vaccinated. To evaluate trends of varicella infection and the impact of varicella vaccine, we studied the incidence rates of varicella from 1995 to 2012.

Meteorologic data

Data on temperature, humidity, rainfall, water vapor pressure, atmospheric pressure, wind speed, and solar radiation were obtained from the Spanish Agency of Meteorology. This agency collects daily information from a weather station located at latitude 39.55°N, longitude 2.73°E, and 4 m in height. Approximately 90 % of the island’s population lives within a radius of 30 km around the weather station.

Data analyses

The rates of varicella infection were calculated as cases per 100,000 population. The Wilcoxon test for trends was used to determine changes in varicella notification rates, and the Kruskal–Wallis test was used to assess pre- and post-vaccination differences. Regression and time-series techniques were used to model the relationship between the number of varicella cases and the meteorologic factors. Analysis was started with normal distribution analyses (Shapiro–Wilk test), simple and multivariate regression, partial regression calculation, and collinearity tolerance testing. Different week lags were tested for regression analyses considering the most common varicella incubation time (10–21 days). The multivariate regression model was finally tested for autocorrelation using Breusch–Godfrey tests. Autocorrelation involves the existence of correlation between a variable (e.g., number of varicella cases) and the same variable at particular lag periods. If autocorrelation was observed, an autoregressive moving average with exogenous variables (ARMAX) model was built [13]. For ARMAX building, correlograms, partial autocorrelation and autocorrelation graphs, were used to determine the autoregressive and moving average order. Variable selection was obtained by backward adjustment of the meteorologic factors. The final model was considered adequate if a portmanteau Q test for residuals resembled white noise, and residuals were normally distributed and showed no systematic trends. To analyze the influence of schooling on varicella epidemics, the variable “school holidays” was introduced as a dummy variable.

Last, we determined the meteorologic values of maximum and minimum varicella activity. To explore if the higher rate of varicella during certain meteorologic values was not only dependent on the frequency of this factor, we used the ratio between the number of weekly cases per meteorologic value and the number of weeks with the same meteorologic value. All the statistical analyses were performed using STATA 12.1 (StataCorp, College Station, TX, USA). Statistical significance was set at p < 0.05.

Results

Epidemiology and trends of varicella infections

From January 1, 1995, to December 31, 2012, a total of 68,379 cases of varicella infections were notified, equivalent to 46.3 % of the birth cohort of that period (147,574 live newborns). The incidence rates of varicella infections decreased during the 18 years of surveillance (z = −2.09, p <0.05). A decrease in the incidence rate (H = 6.2, p < 0.05) was observed from the pre-National Immunization Program period (1995–2004; 617/100,000) to the vaccine introduction in the National Immunization Program (2005–2012; 398/100,000), as can be seen in Fig. 1. Varicella infections were reported all year round, with a peak incidence in June (week 24) and the nadir in September (weeks 36 to 38), as can be seen in Fig. 2.

Fig. 1
figure 1

Incidence of varicella cases on the island of Mallorca from 1995 to 2012. Trend lines for the pre-immunization period (1995–2004) and the vaccine immunization period (2005–2012)

Full size image
Fig. 2
figure 2

Seasonality of varicella infections on the island of Mallorca from 1995 to 2012. Solar radiation and water vapor pressure are plotted for comparatives

Full size image

Bivariate and multivariate analyses

The linear correlation between varicella cases and meteorologic factors were studied for different lag periods. In Table 1, linear correlations are shown for the most probable transmission time, which corresponded to 3 and 4 weeks lag from diagnosis. Correlation with the number of varicella cases was positive for solar radiation and wind speed, but was inversely related to water vapor pressure and mean temperature. As expected, analyses of collinearity showed a very strong association between mean temperature and water vapor pressure (r = 0.987, p < 0.001) and was confirmed by tolerance testing, which suggested the drop of the mean temperature variable.

Table 1 Bivariate linear correlations between varicella cases and meteorologic factors
Full size table

Linear multivariate analyses revealed an independent association between the mean number of varicella weekly cases and meteorologic factors. The best model fitted with a 3-week lag period, showing a positive correlation with solar radiation, rainfall, and relative humidity, and an inverse correlation with water vapor pressure and atmospheric pressure (see Table 2). However, the contribution of rainfall (2 %), atmospheric pressure (1 %), and relative humidity (3 %) was scarce, as shown in the partial regression analyses (see Table 3). Breusch–Godfrey tests showed autocorrelation (Chi2 174, p < 0.001) in the multivariate regression model, involving the need for further time-series analyses.

Table 2 Multivariate regression analyses of the mean weekly number of varicella cases and meteorologic factors at 3 weeks lag
Full size table
Table 3 Partial regression and semipartial square correlation of the mean weekly number of varicella cases and meteorologic factors at 3 weeks lag
Full size table

Time-series analyses

The autocorrelation and partial autocorrelation graphs used for model fitting are shown in Fig. 3. The resulting ARMAX model showed that 73 % of varicella weekly cases could be forecasted (see Fig. 4) with two meteorologic variables; that is, solar radiation and water vapor pressure (see Table 4). However, analysis of the model’s residuals revealed the existence of non-random noise or disturbances (portmanteau Q statistic = 333, p < 0.001) and did not show a normal distribution (Shapiro V index = 175, p < 0.001). To improve the residuals’ performance, we added the “school holidays” variable to the analyses. School holidays increased the prediction rate of varicella cases (75 %), but the model’s residuals remained inadequate. To diminish the effects of possible confounders, data were transformed into the average of the different weekly means of the variables throughout the 52 weeks of the different years, and were analyzed again using ARMAX analyses. The best fitting model was obtained using a second-order moving average with 3 weeks lag for water vapor pressure, 4 weeks lag for solar radiation, and 3 weeks lag for school holidays (Table 5). This final model predicted 95 % of varicella cases (see Fig. 5), and the residuals showed no non-random noise, were normally distributed, and had no systematic trends.

Fig. 3
figure 3

Autocorrelation (a) and partial autocorrelation (b) for varicella infections

Full size image
Fig. 4
figure 4

Time-series line graph comparing the reported mean weekly varicella infections with the predicted values [autoregressive moving average with exogenous variables (ARMAX) model]

Full size image
Table 4 Autoregressive moving average with exogenous variables (ARMAX) model for the mean weekly varicella infections
Full size table
Table 5 ARMAX model for varicella infections obtained from the analyses of the average weekly means of the variables
Full size table
Fig. 5
figure 5

Time-series line graph with polynomial smoothing comparing the reported varicella infections using the average of the different weekly means of the variables and the predicted values

Full size image

Varicella activity according to meteorologic values

Figure 6 shows the varicella activity according to the meteorologic values. The final model of meteorologic factors (vapor pressure and solar radiation) was analyzed, as well as temperature; the latter because of its significant collinearity with vapor pressure. These analyses were performed with 3 weeks lag for vapor pressure and temperature, and 4 weeks lag for solar radiation. The best trend line was a curvilinear association.

Fig. 6
figure 6

Varicella activity according to the meteorologic values. Varicella activity is expressed as the ratio between the number of weekly varicella cases per meteorologic value and the number of weeks with the same meteorologic value

Full size image

Discussion

VZV is a fairly labile lipid-enveloped DNA virus that causes varicella (chickenpox) and zoster (shingles). In varicella, the transmission of VZV is primarily by direct contact and secondarily by airborne transmission. As occurs with other lipid-enveloped viruses like influenza, VZV survives longer at lower relative humidity, and sunlight (UV radiation) acts as its principal natural virucide in the environment [14]. Varicella is a highly contagious disease, especially in households, where 80 to 90 % of susceptible individuals exposed go on to develop clinical infection. The diagnosis of this infection is usually clinical, but for viral detection, polymerase chain reaction (PCR) testing of vesicles or scabs samples during the early stages of varicella is more sensitive than culture or immunofluorescence assays.

On the island of Mallorca, the incidence of varicella infections from 1995 to 2012 showed a seasonal pattern, with a peak in June and the nadir of incidence in September. Peak incidences from May to July and low incidences towards October have been reported in other areas of Spain [15]. Therefore, varicella seasonality in Spain seems to be slightly different to the winter–spring pattern usually reported from other temperate regions, although the pattern of infection affecting preschool and school-aged children remains identical [16]. In tropical areas, the winter–spring pattern is also frequent, but peaks in August, for instance, have also been reported in tropical areas of Mexico [11]. These seasonal diversities increase the interest in determining the role of meteorologic factors in chickenpox epidemics.

Previous reports evaluating the seasonality of varicella and its association with meteorologic factors have analyzed either simple correlations or only a few meteorologic factors, which could lead to inaccurate conclusions [12, 17]. In our study, when simple correlations were obtained, four out of the seven meteorologic factors studied had a positive correlation with varicella epidemics. In contrast, when multiple regression or ARMAX analyses were performed, the contribution of some of the factors was found to be irrelevant, as seen with partial regression, or was dropped from the model. ARMAX is a statistical model based on autoregressive integrated moving average (ARIMA) analyses, which uses one or more predictive independent variables and allows the analysis of data considering the time factor [13]. These techniques are usually employed in econometrics to forecast markets and have been recently used in epidemiologic studies [18, 19].

In a previous report from Greece [12], hospitalizations for varicella correlated negatively with temperature, positively with wind speed, and showed no correlation with relative humidity. Unfortunately, the authors did not analyze other important meteorologic variables. In our study, we observed a strong correlation between temperature and water vapor pressure. Indeed, the replacement of temperature for water vapor pressure scarcely modified the resulting ARMAX model, making the decision of which variable to remove from the model particularly difficult. The existence of reports of the peak incidence of varicella during the summer (high temperature) and the better stability of our model using water vapor pressure made us choose to eliminate the mean temperature variable. However, our study does not exclude the contribution of mean temperature in the peaks of varicella infections.

Our predicting model using two meteorologic factors, namely, water vapor pressure and solar radiation, forecasted 95 % of varicella cases. Interestingly, in our model, the association between varicella cases and solar radiation was stronger at 4 weeks lag, suggesting that its main target would be the host. In contrast, for water vapor pressure and school holidays, the best association occurred at 3 weeks lag, and would probably influence VZV transmission.

Water vapor pressure correlated inversely with varicella cases, and, as suggested for airborne viruses such as influenza, it is likely to influence VZV transmission [20]. VZV is transmitted by airborne bioaerosols from respiratory secretions or contact with skin vesicles [21–24]. As previously suggested for other viruses [25], our hypothesis is that VZV bioaerosols or bioparticles from exhaled air or ruptured vesicles remain airborne and start moving towards the surface with an acceleration and speed depending mostly on gravitational and Stokes’ forces. As the droplet travels, it starts to evaporate and is reduced until the water vapor pressure exerted by the droplet surface is equal to the water vapor pressure in the air. The lower the water vapor pressure in the air, the faster the evaporation, increasing the chances of airborne transmission. The residue of the droplet after evaporation (droplet nucleus), containing VZV, could remain airborne for prolonged periods of time [23].

The impact of UV radiation on differences in the epidemiology of varicella and the evolution and global distribution of VZV genotypes has been previously suggested but not proven [10]. Animal and human studies have demonstrated the immunosuppression caused by UV radiation and its effects in the reactivation of herpes simplex virus (HSV), human immunodeficiency virus (HIV) replication, or human papillomavirus (HPV)-triggered skin cancer [26]. In herpes zoster disease, although most of the cases show no seasonality, there are some reports of a summer predominance [17, 27], suggesting an effect of UV radiation on the reactivation of VZV and local and systemic cell-mediated immunosuppression [26]. One of the most remarkable findings in our study was a positive correlation between solar radiation and the number of cases of varicella infection. To our knowledge, this is the first evidence of the influence of solar radiation on varicella epidemics. Our results suggest that the effects of solar radiation, directly on the host immune system and/or indirectly through vitamin D metabolism, may play a role in varicella epidemics.

The cutaneous immunosuppressive mechanisms of UV-B radiation and vitamin D have been recently studied and have important therapeutic implications in immunodermatological disorders [28]. Both UV-B and vitamin D seem to stimulate Fox-p3+ Tregs cells, producing IL-10, a cytokine with immunosuppressive activity in the skin [29]. In addition, vitamin D has been shown to inhibit human natural killer (NK) cells activity [30]. NK cells are lymphocytes that play a role in the innate defense against viral infections. This seems to be especially relevant in infections caused by the herpes virus family, where NK deficiency has been associated to severe infections, mainly those by VZV [31]. For these reasons, our findings could raise questions concerning the safety of vitamin D supplements during the course of varicella infections.

This study has the usual limitations of ecological studies where factors like human behavior are not considered. Our study tried to predict the occurrence of varicella in a Mediterranean region using its meteorologic conditions. However, we do not ignore or downplay the overwhelming influence of human behavior in the dynamics of varicella infections, and the fact that indoors transmission of VZV is far more efficient than outdoors transmission. We assessed the impact of schooling in the analyses, and although its contribution to the prediction of varicella cases was scarce, it was statistically significant and reduced the non-random noise. Other studies have shown clearly that reduction in contact between children during the school summer holiday leads to a decrease in the transmission of VZV [32]. Among other possible drawbacks of our study would be the reliability of the number of varicella cases, the lack of data about sex and age of the cases, and the influence of varicella vaccine. Notification of varicella cases is compulsory in Spain and is considered a consistent and adequate tool to monitor disease trends [33]. However, improvement of this reporting by adding age and gender is necessary in order to better identify the demographics of the disease.

As for varicella vaccine, its introduction in the National Immunization Program decreased the number of varicella reports. From the beginning, we were aware that changes in varicella notifications or the effects of varicella vaccine would influence the number of varicella cases during the study period. As a matter of fact, the initial ARMAX model (see Table 4) was not considered suitable due to the existence of non-random noise and a non-normal distribution of residuals. In other words, the differences between the cases predicted by our model and the number of reported cases seemed to be influenced by other significant factors (probably the aforementioned). To solve this problem, we used the average of the different weekly means of the variables for analyses. This approach not only diminished the distortions caused by these possible confounders, but resulted in a model with a very high predicting power. In addition, we built ARMAX models using values of the variables before and after the inclusion of varicella vaccine in the National Immunization Program (data not shown), but they revealed no differences to the final model built considering the whole 1995–2012 period (Table 5).

In conclusion, in the present study, we have observed that solar radiation and water vapor pressure predicted up to 95 % of cases of varicella over a period of 18 years in a temperate region. These climatic factors may participate in the transmission of VZV or changes in the host that makes them susceptible. Although we cannot establish the role of UV exposure on varicella infections, our results show, for the first time, the epidemiologic importance of solar radiation in the dynamics of the disease. However, we need more studies from different geographic areas in order to determine the role of meteorologic factors on the development of varicella epidemics.