Exploring the Seasonal Drivers of Varicella Zoster Virus Transmission and Reactivation

Kevin M. Bakker; Marisa C. Eisenberg; Robert Woods; Micaela E. Martinez


Am J Epidemiol. 2021;190(9):1814-1820. 

In This Article


Using Thailand's national-level disease data, we observed strong seasonal cycles in chickenpox outbreaks, which was expected based on observations from other countries.[6,16] Less expected, however, was the seasonal occurrence of shingles, which was less explosive than that of chickenpox but still prominent (Figures 1 and 2 and Web Figure 1). Chickenpox dynamics were characterized by outbreaks that began to take off each year during November and December, culminating with seasonal peaks in February or March. Peaks were followed by deep troughs that lasted from June to October. The observed national-level seasonal outbreak patterns also persisted at the regional (Web Figure 12) and provincial (Web Figure 13) levels, though amplitude decreased closer to the equator. The seasonality of shingles was more complicated. January consistently had a relatively high number of cases compared with other months; however, there were typically 2 annual troughs for shingles cases—a shallow trough and a deep trough. The shallow trough was not always present, but it was coincident with the peak in chickenpox (February–March). The deep trough was consistent and occurred in October–December. Shingles cases peaked in May–June, meaning that the shingles peak was delayed relative to the chickenpox peak. Importantly, a cross-wavelet analysis confirmed that both diseases displayed significant annual periodicity, with peak numbers of cases spaced 3 months apart (Web Figure 8).

The country-level shingles cases indicated a novel seasonal pattern; therefore, we investigated the possibility of spatial variation in shingles seasonality. We discovered a latitudinal gradient in shingles seasonality, since regions in higher latitudes had more pronounced seasonal cycles (Web Figures 1, 7, 9, 10, and 14), with diminished seasonal variation as populations approached the equator. The northern region, where the seasonal signal of both shingles and UV radiation was strongest, also had the highest shingles incidence in Thailand.

We observed a significant positive correlation between the monthly UV radiation index and monthly numbers of shingles cases in each region (Web Figures 9 and 10) and at the national level (Figures 1 and 2). At the national level, approximately 47% of the seasonal variation in shingles cases was explained by UV radiation exposure (Figure 2C). On a regional scale, the correlation between UV radiation and shingles was stronger in the north, northeastern, and central regions, relative to the southern region (Web Figures 9 and 10). This was at least partially due to the lack of seasonal cyclicity of shingles in the south (Web Figure 7). As a country, Thailand spans the latitudinal band from approximately 6°N to 21°N, with the northern region encompassing roughly 15°N–21°N and the southern region spanning 6°N–12°N. Throughout the study period, annual periodicity in both UV radiation levels and shingles was significant, with a 1/8-year (i.e., approximately 1.5-month) lag between the initial increase in UV radiation and the increased number of shingles cases (Web Figure 8).

We used the national-level data to test models representing different hypotheses regarding seasonal transmission, reactivation, and boosting of immunity. In total, we explored 14 mechanistic models and tested which of these models was most capable of capturing the observed data and seasonal patterns therein, using maximization by iterated particle filtering, a likelihood-based method for statistical inference of dynamical systems.[23,24] The best-fitting model (Web Tables 1 and 2, Web Figure 15, Web Appendix 5) had flexible seasonal components and fitted both chickenpox transmission and shingles reactivation (Figure 3). This model utilized separate B-splines for each chickenpox and shingles case, which allowed for a high amount of flexibility in fitting the shape of the seasonal processes. Although our models examining Thai school terms as a seasonal step function for chickenpox transmission and/or UV radiation as seasonal covariates for shingles reactivation were not the best-fitting, the parameters of the best-fitting model suggested that school terms and UV radiation are important in VZV transmission and reactivation, respectively. This is because the transmission spline estimated for the best-fitting model closely matched school terms[25] and the reactivation spline closely matched UV radiation data (Figures 3B and 3D). The splines afforded the model additional flexibility that may have allowed it to capture the nuance in school-term time forcing that a step function could not capture. Similarly, although there was a high correlation between numbers of shingles cases and UV radiation, the UV radiation data were unable to capture the entirety of the seasonal nuance of shingles reactivation. This could be due to a multitude of factors, including UV radiation not being the primary seasonal driver, UV radiation acting in combination with other seasonal drivers, or imperfect data—both in reporting and in aggregation at the national level. The B-splines estimated for our best-fitting model correlated with lagged UV radiation levels for shingles. We then used the best-fitting model to conduct 2,500 stochastic simulations to show how they compared with the chickenpox and shingles data (Figure 3). The model was able to capture the seasonal cycles of both chickenpox and shingles, but the shingles data had a high amount of interannual variation in its seasonal shape that the model was unable to capture with its fixed seasonal spline.

Figure 3.

Comparison of simulated cases of chickenpox and shingles from the best-fitting model with monthly log-detrended numbers of clinical cases in Thailand, 2003–2010. A) Reported log-detrended numbers of chickenpox cases (black line) with 2,500 fitted simulations (blue) through 2010. B) Maximum likelihood (B-spline) estimate of the monthly seasonal forcing for chickenpox from the best-fitting model (black line) and school terms for Thailand (green). C) Reported log-detrended numbers of shingles cases (black line) with 2,500 fitted simulations (red) through 2010. D) Maximum likelihood (B-spline) estimate of monthly seasonal forcing for shingles from the best-fitting model (black line) and mean monthly ultraviolet (UV) radiation values for Thailand (orange).