Herpes Zoster Risk Reduction through Exposure to Chickenpox Patients: A Systematic Multidisciplinary Review

Varicella-zoster virus (VZV) causes chickenpox and may subsequently reactivate to cause herpes zoster later in life. The exogenous boosting hypothesis states that re-exposure to circulating VZV can inhibit VZV reactivation and consequently also herpes zoster in VZV-immune individuals. Using this hypothesis, mathematical models predicted widespread chickenpox vaccination to increase herpes zoster incidence over more than 30 years. Some countries have postponed universal chickenpox vaccination, at least partially based on this prediction. After a systematic search and selection procedure, we analyzed different types of exogenous boosting studies. We graded 13 observational studies on herpes zoster incidence after widespread chickenpox vaccination, 4 longitudinal studies on VZV immunity after re-exposure, 9 epidemiological risk factor studies, 7 mathematical modeling studies as well as 7 other studies. We conclude that exogenous boosting exists, although not for all persons, nor in all situations. Its magnitude is yet to be determined adequately in any study field.


Introduction
Primary infection by varicella-zoster virus (VZV) causes the clinical syndrome 'chickenpox' (CP), mainly in childhood. An effective commercial childhood CP vaccine has been available for nearly 20 years. It is recommended to be used in a two-dose schedule because experience with a single dose has led to frequent (milder) breakthrough infections [1,2]. After primary infection VZV remains latent in neural ganglia until reactivation. Herpes zoster (HZ), also called shingles, is caused by the symptomatic reactivation of VZV and this reactivation is assumed to be a consequence of a lower cellular immunity mainly in immunocompromised or older individuals [3][4][5]. Other risk factors for HZ have been identified and include gender, ethnicity, host susceptibility and depression (see review by Thomas and Hall [6]). Compared to CP, HZ is associated with relatively higher morbidity and costs (see for e.g. Bilcke et al [7]). A vaccine against HZ exists and is shown to be effective probably due to the long-term augmentation of VZV-specific cellular immunity [8]. The efficacy of this vaccine partially supports the exogenous boosting hypothesis, although one should take the different routes of exposure (i.e. subcutaneously vs. through mucosa) into account. Hope-Simpson first postulated that re-exposure to circulating VZV could inhibit reactivation of VZV [9]. A consequence of this so-called 'exogenous boosting' hypothesis would be a temporary increase in HZ cases following the reduced circulation of VZV, under the influence of a universal childhood CP vaccination program. This HZ increase is expected to be temporary because of the increasing proportion of CP vaccinated individuals who are generally assumed to be no longer at risk for HZ. Several population-based mathematical modeling papers (the first of which published by Schuette and Hethcote [10]) predicted substantial HZ incidence increases in over 30 years following the introduction of widespread CP vaccination and thus called into question the overall public health impact of CP vaccination. Although several countries across the world have already implemented universal childhood CP vaccination (USA, Australia, Germany, Japan, Taiwan, Greece) many other countries continue to wait for more conclusive data regarding the existence, duration, and thus the effect of exogenous boosting. This paper presents the first systematic and multidisciplinary in-depth assessment of the literature with respect to exogenous boosting.

Search Background
The multidisciplinary approach of our review is focused on two primary end points: (1) HZ incidence as a function of exposure to CP and (2) the longitudinal course of VZV-specific immunity elicited by contact with CP. Both end points will depend on the type and duration of contact with CP. Other factors potentially of influence include the age of the CP patient, the age of the exposed person, environmental factors (e.g. season), the type of contact, the elapsed time between current and previous exposures, and the preexposure immunity level.
The results from our review are presented following the PRISMA methodology, when applicable [11].

Search Strategy
Both PubMed and Web of Science (v5.8) were searched up to 27 th November 2012 without a restriction on the publication date. Our review included original research articles and letters, published at any time.

la] OR English Abstract[pt]) AND (hasabstract OR letter[pt]) NOT (review[pt] OR guideline[pt] OR editorial[pt]).
The Web of Science search used the following search string: Topic = (zoster OR shingles OR varicella OR chickenpox) AND Topic = (exposure OR reinfection OR boost OR seroepidemiology) with further document refinement for document types ''article'' or ''letter'' and languages ''English''. This search was done with lemmatization ON (the search includes inflected forms of words in a Topic and/or Title search query, which allows for a broader scope of functionality and includes synonyms, plurals, and singulars).
All search results were aggregated in Endnote X5 for MAC OS and duplicates were discarded, such that 1090 unique references were retained at first. The reference lists from these publications were assessed by screening title and abstract of references not identified in our original search. Additionally, publications citing our selections were identified and screened using Web of Science, implying published comments on our selections were also considered. That is, publications being cited by and citing the papers we retained originally, were in turn considered for additional inclusion. Citations to and by the selected publications were again screened for inclusion (see Figure 1). This way, an additional 770 publications were screened.

Selection Criteria and Inclusion
The inclusion criteria were agreed upon by all authors and are presented in Table S1. BO first read all titles & abstracts of references from the combined PubMed and Web of Science search and organized them in 8 categories for additional assessment by PB and PVD to decide whether full text assessment was deemed suitable. All references in disagreement were further discussed until agreement was reached. The categories were: epidemiological and modeling papers included (PB) and excluded (PB), immunological or clinical papers included (PVD) and excluded (PVD), papers not related to or without a focus on VZV (PB), not original research papers (PVD) and papers without abstracts included (PB, PVD) and excluded (PB, PVD).
One hundred and twenty-eight papers were included for full text assessment and 27 papers were retained after full text assessment. The identification of 770 additional papers through citations led to the selection of 12 extra papers (see Figure 1). One non-English paper was selected for full text assessment, but could not be retrieved [12].
The methodology and results from these papers are summarized in Tables 1, 2, 3, and 4, using 5 categories: (1) 13 HZ incidence studies in countries with widespread CP vaccination, (2) 7 mathematical modeling studies, (3) 9 epidemiological risk factor studies, (4) 4 prospective longitudinal studies on VZV-immunity post-exposure and (5) 7 other studies (non-longitudinal immunological studies and an epidemiological study). We note that 40 studies are mentioned, whereas only 39 papers were included. This is caused by the Brisson et al paper [13] that had both a mathematical modeling aspect and an epidemiological risk factor aspect.
Due to large heterogeneities between different study designs, we chose to apply a two-stage grading system, based on pre-set criteria agreed upon between all co-authors. This way, the relevance of the study design to assess exogenous boosting (study design levels from D to A+, see Table S2) and the overall quality based on potential biases, given its specific study design (Low-Medium-High, see Table S3), were assessed separately.

Results
HZ Incidence in Countries with Widespread CP Vaccination Studies (see Table 1) Mullooly et al (quality: medium, relevance: A) found a difference when comparing pre-CP-vaccination HZ incidence in one USA region with the post-CP vaccination HZ incidence in another USA region [14]. HZ incidence was also noted to increase during the post-CP-vaccination years, but only in 10-17 year olds. Poisson multiple regression showed the observed HZ increase to be mediated mainly by an increase in oral steroid exposure. Using telephone survey data, Yih et al (quality: medium, relevance: A) observed that the age-standardized HZ incidence in Massachusetts (USA) increased by 90% in the 5 years after vaccination reduced CP incidence [15]. Unfortunately, Yih et al did not control for other potential causes for such an increase. These may include differences in reporting practices over time. For instance, gradually increased reporting of clinical diseases may occur during long term run-in periods following fundamental changes in surveillance, such as the implementation of electronic data collection/reporting. Furthermore, a temporal rise in HZ incidence could also be due to rising numbers of immunocompromised individuals. Jumaan et al (quality: medium, relevance: A) examined the medical records of a health maintenance organization (HMO) in Washington (USA). They observed no change in overall HZ incidence in the postvaccination era, but noted an increase in HZ in unvaccinated children [16]. Importantly, the overall CP incidence only started to decrease during the last four years of observations. Patel et al (quality: low, relevance: A) showed a marked increase in HZ hospitalizations in .65 year olds after the introduction of CP vaccination in the USA [17]. Rimland et al (quality: medium, relevance: A) observed an HZ increase in veterans aged older than 40 years during 8 years after the start of widespread CP vaccination [18]. Carville et al (quality: medium, relevance: A) reported an increase in after hours telephone HZ diagnosis after, but also prior to, introduction of CP vaccination in Victoria (Australia) [19]. Updates by Grant et al [20] and Carville et al [21] further supported this observation. Nelson et al (quality: low, relevance: A) found an increase in the proportion of GP consultations for HZ in Australia [22]. However, due to data limitations, they could not exclude a trend that may have started in Figure 1. Modified PRISMA flow diagram. PubMed and Web of Science (WoS) search results were combined and after controlling for duplicates, titles and abstracts from references were screened for further full text assessment. Citations from and to the selected references were also screened using title and abstract for further full text assessment. Citations from and to the additional selected references were again screened as discussed. The additional selected references thus found were added to the earlier found references in order to obtain the Selected Total. doi:10.1371/journal.pone.0066485.g001 Analysis includes pre-vaccination HZ data to compare with post-vaccination data Analysis takes changing demography in account Analysis takes a change in underlying diseases or immunocompromised states in account An age-standardized 98% HZ incidence increase was noted over the 13y period and an increase remained present when only focusing on immunocompetent individuals (factor 1.3-1.4 increase in age-standardized rate over 8y); however, HZ incidence increases were already detected in 1993-1996 (p,0.001) and age-specific HZ incidence was the same for adults living in states with high varicella vaccine coverage and those in low-coverage states (p = 0.3173), although it was difficult to assess the value of the observation since no specific information on the actual differences in CP vaccine uptake or CP incidence between high and low vaccine coverage states was shown Adults 20-50y with dependents aged ,12y initially had lower HZ incidence compared with adults without dependents (P,0.01); importantly, though the HZ incidence increased for both groups, it increased significantly more in those with dependent children, such that HZ incidence in both groups completely converged over time the pre-vaccination era. Also, as observed by Heywood et al [23], the lack of age-standardization prohibited a correct interpretation of Nelson et al's [22] observations. Jardine et al (quality: low, relevance: A) noted an increase in emergency room visits for HZ and an increase in antiviral use in Australia after introduction of CP vaccination [24]. Five years after universal CP vaccination was funded in Ontario (Canada), Tanuseputro et al (quality: medium, relevance: A) found no overall increase in HZ incidence, but observed an increase over the last year for those aged over 60 years [25]. Based on a medical insurance database Leung et al (quality: high, relevance: A) reported a gradual convergence of HZ incidence between adults, aged 20-50 years, with and without dependent children that started after the introduction of universal CP vaccination in the USA [26]. This convergence became complete in 2005, strongly suggesting that the association between living with children and a lower likelihood of HZ weakened due to the increasing uptake of childhood CP vaccination. However, they also found an overall increase in HZ incidence with time that started before the introduction of universal CP vaccination. A HZ incidence increase was also observed in immunocompetent individuals separately (factor 1.3-1.4 increase in age-standardized rate over 8 post-CP-vaccination years) and thus indicates that the HZ increase was not solely caused by an increase in immunosuppressed individuals. It was noted that the overall HZ incidence did not differ between US states with higher and lower CP vaccine uptake compared to the national median coverage, but no information was given in regard to the actual differences in CP incidence. Chao et al (quality: high, relevance: A) reported an increase in HZ coinciding with a decrease in CP after introduction of CP vaccination in Taiwan [27].
Overall, 8 studies supported the exogenous boosting hypothesis, whereas 3 studies did not and 2 remained inconclusive in regard to the exogenous boosting hypothesis.
Mathematical Modeling Studies (see Table 2) All mathematical models described in this section are based on a deterministic modeling approach in which individuals are grouped in a number of compartments describing the state of the individual in relation to VZV infection. This is a population-based approach in which individual characteristics are averaged within the compartments and age groups. The transition between compartments is mainly modeled by differential equations governed by rates describing the movement from one compartment to another. In the described models, transition rates are given by 1/(average time an individual stays in a compartment).
At first, individuals are assumed to be susceptible 'S' to infection (mostly after a period of protection through maternal antibodies). Next, individuals can be infected with VZV at a certain age and they will move from S to the CP recovered 'R' compartment. The force of infection l(a,t) describes the annual risk for a susceptible person of age a to be infected at time t with VZV. For example dS(a,t) dt~{ b(t)(a,t)~{l(t)(a,t) means that susceptibles leave their compartment at a rate proportional to the number of susceptibles. The force of infection l(t) is the product of the contact rate b and the number of infectious individuals at time t, I(t). Note that l(t) also represents the number of re-exposure episodes per year. The contact rates are described by the Who-Acquires-Infection-From-Whom 'WAIFW' matrices. Initially these matrices were estimated by using simplified and a priori assumptions on the way age groups interact with each other whereas in later work empirical social contact matrices were used [28]. Finally, modulated by the modeling approach, a reactivation Descriptions of vaccination uptake are as reported in the respective original papers.
*H = High: the quality of methods used in this paper permits the results, within the scope of the study design, to be interpreted with at the most a few remarks. M = Medium: the quality of methods used in this paper permits the results, within the scope of the study design, to be interpreted, but with some caution. L = Low: the quality of methods used in this paper urges the reader to interpret the results, even within the scope of the study design, with sufficient caution.
£See Simplified model: HZ(a,t) r 0 (a,t)~r(a,t) : (1{V(a,t,t)) r(a,t)~azC WAIFWmatrixfittedtoserologicallyobtainedl,dynamicallusingonlyCP Fitting data: weekly rates of CP and HZ incidence collected from the records of 106 GP   The HZ simulated incidence seemed to be higher than the observed data, however the authors noted that the latter could have been lower than in reality due to a registration run-in period  S susceptibles compartment; l force of infection; r reactivation rate; HZ herpes zoster; PDE partial differential equations; WAIFW who-acquires-infection-from-whom; CP chickenpox; R CP recovered compartment; S boost susceptible to boosting compartment; ODE ordinary differential equations.
*H = High: the quality of methods used in this paper permits the results, within the scope of the study design, to be interpreted with at the most a few remarks. M = Medium: the quality of methods used in this paper permits the results, within the scope of the study design, to be interpreted, but with some caution. L = Low: the quality of methods used in this paper urges the reader to interpret the results, even within the scope of the study design, with sufficient caution.
£See  rate r(a) describes the transition from R to HZ, also allowing a route for exogenous boosting. Garnett and Grenfell (quality: medium, relevance: C) were the first to incorporate exogenous boosting in a mathematical model. In their model, they added a 'time since initial infection' variable in addition to the ageing variable [29]. The reactivation rate was assumed to be a function of age. They formulated an exogenous boosting function that was applied to proportionally reduce the reactivation rate. The boosting function was constructed as an integration of all past exposures to CP but damped with time since exposure thereby assuming the most recent re-exposure to have the biggest effect. Also, the effect of exogenous boosting was assumed to decrease with age. An epidemiological observation revealed a downward shift in the mean age of CP coinciding with an increase in CP in 15-44 year olds and a small but significant increase in HZ in individuals aged 15-44 years. A simulation with an artificial increase in CP in  year olds predicted a decrease in HZ, which was qualitatively in accordance with the observations. Time series analysis showed CP incidence to be cyclical and HZ incidence not, and no correlation was found between the two on a weekly scale.
Brisson et al (quality: medium, relevance: C) modeled the transition from CP to HZ in a three part compartmental model where individuals went from R to HZ via a transitory compartment named 'susceptible to boosting 'S boost '' [30]. Thus, instead of one rate describing the reactivation process, two rates were assumed to exist. The rate s from R to S boost was imputed independently of age and represents continuous waning of immunity with time since primary infection. Next, an age dependent 'reactivation' rate, estimated by fitting to HZ incidence data, defined the transition from S boost to HZ. Exogenous boosting was assumed to move individuals from S boost back to R by a rate, which was assumed to be proportional to the force of infection. This way re-exposure was assumed to have only effect when immunity was sufficiently low. Without specifying whether the average time in R was 2 or 20 years the authors noted the predicted HZ incidence to be qualitatively similar to the observed HZ incidence data. There was no assessment made whether the inclusion of exogenous boosting created a better fit to the data. In a later paper Brisson et al (quality: medium, relevance: C) estimated s and the reactivation rate by fitting them simultaneously to HZ incidence data using sub-compartments to distinguish people living with and without children [13]. Brisson et al thus estimated that re-exposure to CP would boost immunity to HZ for an average of 20 years with a 95% CI of 7-41 years and they considered the fit to the data to be good. Several partially modulated papers using the Brisson et al methodology were published in the past ten years. Bonmarin et al (quality: low, relevance: C) adjusted the Brisson et al [30] methodology to French data under assumption of 20 years duration of boosting, however with limited details, and they qualitatively noted a good agreement between simulated and observed data [31]. Brisson et al (quality: medium, relevance: C) [32] applied the European empirical social contact matrices on Canadian data and allowed an age-dependent effect of boosting which was inspired by agespecific HZ vaccine efficacy results of the Shingles Prevention Studies [33]. They remodeled the England & Wales dataset from reference [13] and re-estimated s to be 1/(24.4) years. Next, they imputed these values for Canada, thereby estimating the reactivation rate. They observed that a simulation of the USA early post-CP-vaccination years only partially (qualitatively) agreed with the surveillance data on HZ incidence. Additional analyses showed important sensitivities to model components such as the contact matrix and vaccine efficacy estimates. Van Hoek et al (quality: medium, relevance: C) applied the empirical social contact data for England [34]. By using a s of 20 years they found a good fit between the predicted and observed HZ incidence data.
Karhunen et al (quality: medium, relevance: C) presented another approach to modeling exogenous boosting [35]. They constructed an innovative Bayesian method to estimate the reactivation rate, which they formulated as a log-linear function of age plus time since last exposure. Karhunen et al thus explicitly assumed exogenous boosting to be on the same level as ageing. Also, they assumed that the ageing effect could be delayed until a certain age. Fitting to HZ incidence data and allowing waning by age to start at 45 years, they estimated that each year since last exposure would increase the HZ risk with 3.3%. When delaying the ageing until 65 years, the HZ risk increased 8.4% with each year since last exposure. Model selection criteria preferred the model with an ageing threshold of 45 years. By having waning VZV varicella-zoster virus; HZ herpes zoster; CP chickenpox; HH household; OR odds ratio; aOR adjusted odds ratio; NS not significant; PY person-years. *H = High: the quality of methods used in this paper permits the results, within the scope of the study design, to be interpreted with at the most a few remarks. M = Medium: the quality of methods used in this paper permits the results, within the scope of the study design, to be interpreted, but with some caution. L = Low: the quality of methods used in this paper urges the reader to interpret the results, even within the scope of the study design, with sufficient caution. £See Table S2. **The 'B' statement expresses whether the study supported the existence of exogenous boosting ('+) or not ('2'). doi:10.1371/journal.pone.0066485.t003 steep contraction up to week 6 and slow decrease thereafter; 1y later VZV-specific CD4+ cells were detectable in all these RE at 0.08% (qualitatively constant from +/215 weeks); TT-specific CD4+ cells remained constant at all times 5/16 RE had no VZV-specific IFN-gamma increase RE cells were mainly CD45RA-and showed a strong correlation between peak-level VZV-CD4+ percentages and CD27-negativity; this correlation returned to baseline at 1y VZV-specific IFN-gamma, TNF-alpha and IL-2 had similar kinetics in all RE; CD8+ and NK cell kinetics were similar to CD4+ kinetics in RE; VZV was not detected in plasma from RE There was no overall correlation between peak IgG and VZV-CD4+%, but a correlation was more common in those with boosted cellular immunity Epidemiological Risk Factor Studies (see Table 3) Solomon et al (quality: low, relevance: B) found, in a study with a low response rate, the cumulative HZ incidence in psychiatrists to be almost twice of that of pediatricians and comparable to that of dermatologists [36]. A case-control study by Thomas et al (quality: high, relevance: B) found that an increasing number of exposures to CP, social contacts with children or occupational contacts with ill children (not necessarily CP) reduced the risk of HZ [37]. Their multivariate analysis showed the OR for HZ to decrease from 0.9 to 0.29 when 1 to $5 known contacts with CP patients occurred over the last 10 years. Examining a national survey, Brisson et al (quality: high, relevance: C) showed adults The authors state (data not shown) that CP epidemiology from Japan did not change between 1999-2008 HZ incidence increased during the study period in females older than 60y The authors showed visually that after averaging over the study period HZ incidence in Miyazaki mirrored the national CP incidence at a seasonal level with a HZ peak in the summer L A + RE re-exposed; CP chickenpox; CO controls; VZV varicella-zoster virus; RIA radioimmunoassay; PBMC peripheral blood mononuclear cells; TT tetanus toxine; PHA phytohemagglutin; IFN interferon; Cpm counts per minute; FAMA fluorescent antibody to membrane antigen; FCM flow cytometry; ICS intracellular cytokine staining; ELISPOT enzyme-linked immunosorbent spot; GMR geometric mean response; ELISA enzyme-linked immunosorbent assay; RCF responder cell frequency; HCW healthcare workers. *H = High: the quality of methods used in this paper permits the results, within the scope of the study design, to be interpreted with at the most a few remarks. M = Medium: the quality of methods used in this paper permits the results, within the scope of the study design, to be interpreted, but with some caution. L = Low: the quality of methods used in this paper urges the reader to interpret the results, even within the scope of the study design, with sufficient caution. £See Table S2. **The 'B' statement expresses whether the study supported the existence of exogenous boosting ('+) or not ('2'). doi:10.1371/journal.pone.0066485.t004 currently living with children in the household to have a lower HZ incidence than same-aged adults who do not live with children (incidence ratio of 0.75) [13].  [41]. Gaillat et al (quality: medium, relevance: C) performed an innovative study in which HZ incidence was compared between members of monastic orders, who were selected to have had fewer exposures to CP, and the general population [42]. They found no difference in HZ incidence between the two groups. However several methodological issues, such as gender bias, limit the interpretation of this result. These issues have been discussed elsewhere [43,44]. Finally, after multivariate analysis in a prospective case-control design, Lasserre et al (quality: high, relevance: C) found living together with children not to be protective against HZ [45].
Overall, 4 studies supported the exogenous boosting hypothesis, whereas 3 studies did not and 2 remained inconclusive in regard to the exogenous boosting hypothesis.
Prospective Longitudinal Studies on VZV-immunity Postexposure (see Table 4) Arvin et al (quality: low, relevance: A) compared VZV-specific lymphocyte proliferation data and antibody titers at less than 4 days and at 3 to 4 weeks since re-exposure to CP in women [46]. An increase in cellular immunity was seen in 61% and an IgG response in 64% of re-exposed. Interestingly, the IgG response was influenced by the initial value with low initial values increasing and high initial values decreasing after re-exposure. This could be due to either a faster immune response or a longer time period between re-exposure and the first sample (causing a peak response on the first time point, followed by a decrease at the second time point). VZV-specific serum IgA positivity was frequently and IgM positivity was rarely noted after re-exposure. Gershon et al (quality: low, relevance: D) observed in a similar study design, solely focusing on VZV-specific antibodies, that 32% of re-exposed parents were immunologically boosted within 40 days after reexposure [47].
Vossen et al (quality: medium, relevance: A) emphasized that boosting of immunity was not seen in all re-exposed (11/16 showed an increase in cellular immunity) and that some individuals had an increase in antibody titers whereas others had a decrease [48]. Intracellular cytokine staining showed qualitatively an initial up-rise in VZV-specific IFN-gamma CD4+ cells (with similar kinetics for CD8+ and NK cells) at 4-6 weeks since re-exposure, followed by a decrease until stabilization (probably still higher than compared to a control group) around 15 weeks. Half of the re-exposed showed either an increase or a decrease in antibody titer. No VZV DNA was detected in the blood.
Ogunjimi et al (quality: medium, relevance: A) found a factor 1.6 VZV-specific IFN-gamma ELISPOT increase from 1 week to 1 year post-exposure, but could not demonstrate a higher cellular immunity in exposed versus control individuals (even the reverse was true at 1 month after re-exposure) [49]. The latter could be explained by cryopreservation, leading to cell apoptosis and reduced immune responses [50], and inter-assay biases [51]. They also observed no formal significant longitudinal effect by reexposure on antibody titers, although there was a tendency for overall higher antibody titers in the re-exposed group as compared to the control group.
Overall, all 4 studies supported the exogenous boosting hypothesis. However, only 2 studies had data up to one year post re-exposure.
Other Studies (see Table 4) Gershon et al (quality: medium, relevance: D) found VZVspecific IgM antibodies in 4 out of 6 household re-exposures in comparison to 11 out of 49 control individuals [52]. Terada et al (quality: medium, relevance: C) noted that pediatricians had a higher VZV-specific responder cell frequency compared to nonmatched healthy adults thereby suggesting a cellular effect of boosting [53]. In a later study, Terada et al (quality: low, relevance: D) also showed that health care workers frequently exposed to VZV had increased serum IgA levels [54]. A seroepidemiological study by Yavuz et al (quality: low, relevance: D) described higher VZV-specific IgG in health-care workers compared to office workers [55]. Saadatian-Elahi et al (quality: low, relevance: D) studied pregnant women and found no effect of the number of children in the household on VZV IgG titers [56]. Valdarchi et al (quality: low, relevance: D) noted during a prison CP outbreak study that some asymptomatic inmates developed VZV IgM, suggesting re-exposure, although only two had a known CP contact [57]. Toyama et al (quality: low, relevance: A) suggested a short term influence of CP on HZ incidence through visual inspection of plots, showing an increase in HZ during the summer period when CP circulation is at its lowest [58]. However, if this observation would prove to be based on a statistically significant association, it could easily be explained by other causal factors than CP (for e.g. solar exposure). Furthermore, a recent review did not find sufficient epidemiological support for the seasonality of HZ incidence [6].
Overall, 6 studies supported the exogenous boosting hypothesis, whereas 1 study did not.

Discussion
We presented an in-depth systematic review of the literature on exogenous boosting for VZV.
The majority (27/40) of HZ incidence studies post widespread CP vaccination showed the existence of exogenous boosting. However, the clear extent of this effect is not easy to interpret due to a substantial population and environmental heterogeneity (e.g. CP vaccination coverage, variations in surveillance practices) and the lack of pre-and post-CP vaccination data for a sufficiently long time. Indeed, in all of the studies in countries with a universal CP vaccination program, the time period since the occurrence of large reductions in CP incidence is perhaps too short and the role of less infectious breakthrough infections might be too important to allow differentiating between the different boosting hypotheses. Also, some studies noted an increase in HZ incidence preceding initiation of CP vaccination. This could be explained by an ageing population, increasing immunosuppression (by disease and/or medication), but also by more accurate surveillance and the implementation of electronic data collection/reporting. However, as much is yet to be learned about risk factors for HZ in otherwise healthy individuals, it remains challenging to control for changing risk factors over time. With the exception of one study [16], these studies did not correct for the decreasing presence of naturally infected children in the denominators. So, although exogenous boosting seems to exist, it is not clear from these studies to which extent it affects HZ incidence in the community when CP incidence is reduced.
Prospective longitudinal immunological studies after re-exposure circumvent the difficulties associated with epidemiological studies. All such studies clearly showed exogenous boosting to exist, but clearly not for all re-exposure episodes. It is likely that the proportion of boosted individuals and the duration of boosting depend on several variables such as the intensity and duration of the contact [28,59], age-specific characteristics [59], the viral load [60] and the pre-re-exposure immunity levels at the mucosa and in the blood circulation [3]. The mere existence of exogenous boosting is not surprising as it should be interpreted as a secondary immune response. We found only two studies that analyzed samples up to one year post-re-exposure [48,49]. However, shortcomings in study design hampered estimating the duration of immunological boosting. Nonetheless, the combined immunological data suggest the effect of immunological boosting would not extend beyond two years. We underline however that both the threshold level, if it exists, and the cell type(s) of the immunological correlate of protection against HZ are yet to be defined with certainty. Also, endogenous boosting after asymptomatic reactivation could add to the complexity of the VZV immune response [61].
Weighting by study design and quality of epidemiological studies that assessed the effect of re-exposure to CP, both directly and indirectly, supported the existence of exogenous boosting although the magnitude and time frames for re-exposure varied between studies. It is important to note that some of these studies used contact with children as a proxy for VZV re-exposure. Overall, one can assume that women have more contacts with children than men. However, women tend to have higher HZ incidence rates than men. Possibly, just household re-exposures, which would be more equal between men and women, can be sufficient for exogenous boosting. Alternatively, gender susceptibility to HZ could have a greater effect on HZ occurrence than exogenous boosting.
The mathematical models that contributed to many countries' hesitation to start universal CP vaccination, involved an important and long term effect of exogenous boosting on HZ incidence. However, none of the models allowed for an explicit comparison of the goodness-of-fit between scenarios with and without boosting. These modeling papers predicted HZ incidence ratios around 1.15-1.4 10 years after introduction of CP vaccination for a 20 year boosting scenario and still increasing modestly for some years thereafter before a decrease would occur. We note that the modeled HZ incidence ratios do not differ substantially between high vaccine uptake scenarios with short (e.g. 2 years) or long (e.g. 20 years) durations of boosting over time horizons up to 10 years. This implies that current observational HZ post-CP-vaccination data can not yet be used to estimate the duration of the effect of boosting. The predicted incidence ratios are however relatively low when compared to some of the present HZ incidence data (up to 10 years since introduction of CP vaccination). Possibly, the observed HZ data could be caused by a combination of lack of exogenous boosting and a background increase in HZ incidence registration (for e.g. due to gradually improving surveillance). The pre-CP-vaccination HZ increases found in some studies support the latter statement. If however the observed HZ incidence ratios are solely caused by the lack of boosting, they could be explained through a variety of underlying mechanisms. Indeed, the observed HZ incidence is a consequence of the balance between the physiological reactivation rate, which is a driving force towards HZ, and the duration of boosting, which defines the time period of protection against HZ. Thus, for HZ incidence to remain constant in a pre-CP-vaccination mathematical model, the reactivation rate should increase when the duration of boosting increases and vice versa. Thus a higher observed post-CP-vaccination HZ incidence could mean that the duration of boosting was higher than predicted by the models (the models would seem to be insufficiently predictive in this instance). Consequently, the number of years with a net increase in HZ incidence post-CPvaccination would be higher as well. However, even a lower duration of boosting combined with a higher exogenous boosting frequency could lead to a higher initial HZ incidence, but within a shorter time frame as compared to the scenario with a longer duration of boosting. The effect of different contact rates on HZ incidence post-CP-vaccination was illustrated by Brisson et al [32].
Our review was limited by the exclusion of non-peer reviewed publications and of abstracts-only publications.
Our review has found sufficient support to conclude that the mechanism of exogenous boosting exists, although not for all persons, nor in all situations. Besides continuing post-CPvaccination surveillance studies over time periods with expected higher HZ incidence, we reckon that the duration of boosting would still be the most important and the most challenging parameter to estimate correctly. Both improved mathematical models and post-re-exposure immunological studies, combined with a better identification of the immunologic correlates of protection through HZ vaccination trials, should allow a more accurate estimation of the duration of boosting.