The Epidemiology of Herpes Zoster After Varicella Immunization Under Different Biological Hypotheses: Perspectives From Mathematical Modeling

Abstract

The impact of varicella vaccination on the epidemiology of herpes zoster (HZ) critically depends on the mechanism of immunological boosting, through which reexposures to varicella-zoster virus are thought to reduce the individual risk of HZ development. However, the qualitative and quantitative dynamics of this process are largely unknown. Consequently, mathematical models evaluating immunization strategies need to rely on theoretical assumptions. Available varicella-zoster virus models can be classified in 3 main families according to the postulated effect of exogenous boosting: 1) progressive accumulation of immunity following repeated reexposures; 2) partial protection that wanes over time; or 3) full but temporary immunity against HZ. In this work, we review and compare quantitative predictions from the 3 modeling approaches regarding the effect of varicella immunization on HZ. All models predict a qualitatively similar, but quantitatively heterogeneous, transient increase of HZ incidence. In particular, novel estimates from the progressive immunity model predict the largest increase in natural HZ and the largest incidence of HZ cases from reactivation of the vaccine strain, which in the long term will likely outnumber prevaccination numbers. Our results reinforce the idea that a better understanding of HZ pathogenesis is required before further mass varicella immunization programs are set out.

Herpes zoster (HZ) is a painful skin disease arising from the reactivation of varicella-zoster virus (VZV) in individuals previously exposed to varicella (1). Although recovery from varicella confers (in most cases) lifelong immunity to varicella itself (2), VZV is not cleared and remains latent in the dorsal root ganglia where it will reactivate at later ages, causing HZ in about 30% of those who experienced varicella (3). HZ pathogenesis is still poorly understood. Reactivation is thought to result from the decline of VZV-specific cell-mediated immunity (CMI) (4, 5), which represents the critical component of the host immune response against HZ (1, 6). In his seminal study, Hope-Simpson (4) formulated the main scientific hypotheses about the relationship between temporal trends of the host's immune response and VZV reactivation, postulating that CMI is boosted exogenously via reexposure to VZV through contacts with varicella-infected individuals. The exogenous boosting hypothesis has received increasing support by a range of epidemiologic (7, 8), immunological (9, 10), and modeling (11) studies. Although the issue is still controversial (12), a recent systematic review concluded that exogenous boosting exists, even if the magnitude of its effects is yet to be determined adequately (13). Hope-Simpson (4) further specified the mechanism of exogenous boosting by proposing that each boosting episode pushes the host's immunity to levels incrementally larger than those conferred by previous episodes (1), thereby progressively reducing the risk of VZV reactivation. To date, this hypothesis, which we will refer to as “progressive immunity” (14, 15), has not been tested further in immunological studies.

The complex relationship between varicella and HZ is mirrored by the complex epidemiologic patterns of HZ incidence. In a varicella serosurvey conducted in 8 European countries (16), a larger HZ incidence was associated with lower varicella serological profiles, that is, a slower VZV circulation. Broad intercountry differences also exist in the total HZ incidence. In addition, the age-specific HZ incidence has a common pattern across countries until about age 75: It remains about constant in younger adults (about 20–45 years of age) and grows steeply during middle age (>45–75 years). However, at older ages some countries exhibit a decline (e.g., Italy (17), Spain (18, 19), Finland (20), Belgium (21)); others show saturation (United Kingdom (22), France (23), Iceland (24)); and still others have continued growth (United States (25), the Netherlands (26)).

In relation to varicella immunization, the unknowns in HZ epidemiology and pathogenesis have been much debated in Europe in recent times. Varicella vaccination has been delayed in most European countries (27) because of (among other factors) the scare that HZ incidence will increase for several decades after vaccination as a consequence of the decline in the incidence of exogenous boosting episodes. Surveillance of HZ in contexts where vaccination programs are ongoing has shown ambiguous evidence, with some sites showing indication of increasing HZ incidence (28–31), others which do not (32), inconclusive studies (33), and sites where HZ incidence has increased in the absence of any immunization (34). These conflicting results reveal further complex patterns in HZ epidemiology even in the absence of vaccination.

Given the sparse empirical evidence, mathematical models designed to evaluate the impact of varicella immunization strategies are forced to rely on theoretical, rather than data driven, assumptions to incorporate exogenous boosting assumptions while balancing parsimony in model structure with biological plausibility. To date, the available formulations of the exogenous boosting hypothesis can be classified into 3 main groups:

1. models including Hope-Simpson's progressive immunity hypothesis, where reexposures confer a partial protection (i.e., a risk reduction of the HZ risk) that, although waning over time, can accumulate upon repeated episodes (14, 15);

2. models postulating that partial protection is restored at the same level after each new boosting episode (i.e., does not accumulate with successive reexposure) (14, 20); and

3. models assuming that boosting provides complete protection for a given (exponentially distributed) time, at the end of which individuals revert to full HZ susceptibility (11, 22, 35–40).

Despite these structural differences, modeling studies have consistently confirmed the likelihood of a detrimental effect of varicella vaccination on HZ incidence (11, 15, 20, 22, 35–40). However, no direct comparison has been performed yet among predictions from alternative models. In this paper, we aim to contribute to the debate by analyzing the impact of varicella vaccination on HZ epidemiology, as predicted by the 3 previous formulations of the exogenous boosting hypothesis, applied to 4 European countries under the same immunization scenarios.

METHODS

Models for the natural history of VZV and HZ

Available models used in the evaluation of varicella vaccination programs can be categorized into 3 main formulations. The first one considered here has been recently proposed under the name of “progressive immunity” (14, 15). According to this model, all newborn individuals are assumed to be susceptible to varicella infection (compartment denoted by S), which they acquire at an age-specific rate (compartment denoted by I) given by the force of infection, λ(a), where a denotes chronological age. Varicella-recovered individuals develop full immunity to varicella and, at the same time, become susceptible to HZ, entering stage HZS₁, which represents the epidemiologic class of HZ-susceptible individuals who have experienced 1 exposure to VZV. Individuals in this stage can either develop HZ at a stage-specific reactivation rate ρ₁(a, τ), where τ (τ ≤ a) represents the time spent in the current stage, or be exposed to VZV for a second time through contacts with infectious individuals, thereby receiving an exogenous boosting of CMI. Reexposures boost CMI at a rate (termed “force of boosting,” λ_B (a)) equal to the force of infection, λ(a). Reexposed individuals enter the successive HZ susceptibility stage HZS₂, where, again, they either develop HZ at rate ρ₂(a, τ) or are exogenously boosted (moving to HZS₃), and so on. Individuals who have experienced HZ are assumed to have become lifelong immune to HZ (compartment denoted by R). HZ development is assumed to depend on 3 concurring mechanisms: 1) senescence, by assuming that the reactivation rate is exponentially increasing along chronological age a after a certain threshold age a₀, set here at 45 years (14, 20); 2) CMI decline following the last exogenous boosting episode, by assuming that the reactivation rate is exponentially increasing in the time τ elapsed since the last exposure (i.e., each exogenous boosting episode resets τ to zero); and 3) progressive immunity, which reduces the HZ risk by a factor increasing in the number of experienced boosting episodes. Specifically, the VZV reactivation rate is assumed to have the form:

$${\rho }_i(a,\tau )={\rho }_0{e}^{{\theta }_a{(a-a_0)}_{+}}{e}^{{\theta }_{\tau }\tau }{q}^{{(i-1)}^2},$$

where (a − a₀)₊ = max(a − a₀, 0), ρ₀ is a scale factor, θ_a and θ_τ are rates respectively tuning the exponential increase along chronological age and time elapsed since last exposure, q (0 < q < 1) is the progressive immunity factor, and i is the number of exposures to VZV (i = 1 corresponds to primary varicella infection).

The second formulation has been proposed to describe the natural history of HZ and to project the effect of varicella vaccination on HZ incidence in Finland (20). It can be obtained as a special case of the progressive immunity model where q is set to 1: In this case, the reactivation risk is partially reduced by resetting τ = 0 at each reexposure; however, it becomes independent of the number of exposure episodes i, thereby dropping the progressive immunity effect. We will refer to this model as “partial immunity.”

The third model structure considered in this study represents both the oldest and the most common approach adopted in modeling studies (35–40), and it assumes that VZV reexposures provide full immunity to HZ for a given (exponentially distributed) time, at the end of which an individual reverts to full HZ susceptibility. For this reason, this model will be referred to as “temporary immunity.”

We refer the reader to Web Appendix 1 (available at http://aje.oxfordjournals.org/) for complete equations of the progressive, partial, and temporary immunity models.

The model for the epidemiology of HZ after varicella immunization

For all models, we consider the evolution of HZ incidence following a vaccination schedule where a fraction of individuals, represented by the program coverage, is routinely vaccinated at 1 year of age, starting at the beginning of year 2017. Among vaccinated individuals, a fraction depending on vaccine efficacy becomes fully and permanently protected from varicella, and the remainder become susceptible to breakthrough varicella and subject to the same force of infection as unvaccinated individuals. Given the milder symptoms, we assume that breakthrough varicella contributes to the force of infection with a relative infectiousness of 50% with respect to unvaccinated infectious individuals, as proposed in previous models (40).

Successfully vaccinated individuals are subject to a risk of developing HZ from the vaccine strain, with a reduced relative risk (41). We assumed that the risk of HZ in vaccinees is proportional to that of individuals recovered from natural varicella according to an attenuation coefficient χ (41), representing the relative HZ susceptibility in vaccinees with respect to unvaccinated individuals. Vaccinees at risk of HZ are boosted with the same force of boosting as unvaccinated ones, as in previous modeling studies (40).

We consider 3 immunization scenarios based on different combinations of the coverage and efficacy parameters: one that does not eliminate varicella in any of the considered countries (coverage, 70%; efficacy, 85%); one that eliminates varicella in some countries but not in others (coverage, 80%; efficacy, 95%); and one that is above the elimination threshold in all countries (coverage, 90%; efficacy, 95%). The 95% value for the efficacy parameter derives from a recent large-scale trial based on a 2-dose measles-mumps-rubella-varicella vaccine (42); however, for the first scenario, we made a pessimistic assumption and chose a value of 85%.

Furthermore, we consider an additional illustrative scenario where varicella is abruptly eliminated by routine immunization of all newborn individuals with 100% efficacy at times t ≥ t₀, supplemented by catch-up vaccination of all those still susceptible at t₀. Although this is not feasible in practice, this scenario is useful as it offers insight on the worst case possible for HZ.

Full details of model formulation and implementation are reported in Web Appendix 1.

Model parametrization

The considered models were fitted to HZ incidence data from 4 European countries (Finland, Italy, Spain, and the United Kingdom) following the same approach as in previously published works (14, 40). The chosen value for the attenuation coefficient, χ, is based on a published epidemiologic study (41), which estimates it to be between 8% and 25%; therefore, we take χ = 10% as a baseline and explore the sensitivity of model predictions with respect to values of χ. Populations are assumed stationary through country-specific yearly constant inflows of births (B) and age-dependent, time-invariant mortality rates μ(a) (43).

RESULTS

Figure 1 shows the predicted incidence of HZ from natural varicella, from vaccine strain, and total in the different vaccination scenarios over a time span of 100 years postvaccination (until 2117). We do not separately show the contribution of HZ from breakthrough varicella, because its contribution is negligible (<0.5 cases per 1,000 individuals) in all simulations. For the sake of brevity, we present only the results for Italy; similar considerations apply for the other considered countries, whose results are reported in Web Appendix 2 and Web Figures 1–12. In all models and vaccination scenarios considered, a transient increase of HZ is expected shortly after immunization, mainly because of reactivation of natural varicella infections. After a peak, natural HZ declines (up to extinction in scenarios predicting varicella elimination). Meanwhile, HZ cases from vaccine strain become dominant and plateau at an equilibrium incidence level. Although the qualitative dynamics are similar, predictions from different models show striking quantitative differences. In particular, the progressive immunity model predicts a much more important effect of vaccination on HZ incidence with respect to both the transient rise in natural HZ and the long-term level of HZ caused by reactivation of the vaccine strain. This can be interpreted by considering that, if the progressive immunity hypothesis were true, a large part of the population would be virtually immune to HZ because of repeated reexposures. Therefore, estimates of the HZ risk growth rates need to be higher compared with concurrent models in order to provide the same amount of prevaccination HZ incidence (14). These higher rates produce the large differences in predictions about the impact of immunization. For the same reasons, dramatic quantitative differences exist also for what concerns HZ incidence from reactivation of the vaccine strain. In particular, the progressive immunity model shows a plateau at levels that, in most scenarios, are surprisingly higher than the corresponding prevaccination HZ incidence. This effect depends on the rarity of reexposure episodes occurring in the postvaccination epoch, which is only partially mitigated by the lower risk assumed for the attenuated vaccine strain with respect to natural wild-type infection (41).

Figure 1.

Predicted incidence of herpes zoster (HZ) after varicella immunization per 1,000 individuals in Italy over time between 2017 and 2117. Temporal trends are shown for the total incidence due to natural HZ (dashed line), HZ from vaccine strain (dotted line), and total (solid line), as predicted by the 3 models discussed and in the different immunization scenarios. A–D) Progressive immunity model; E–H) partial immunity model; and I–L) temporary immunity model. A, E, and I) Scenario with 70% coverage and 85% efficacy; B, F, and J) scenario with 80% coverage and 95% efficacy; C, G, and K) scenario with 90% coverage and 95% efficacy; D, H, and L) scenario with 100% coverage, 100% efficacy, and complete catch-up of the susceptible population.

Predictions are robust with respect to the considered scenario. In all models, the transient increase in natural HZ is partially mitigated in scenarios that do not result in varicella elimination, whereas vaccination programs above the elimination threshold are essentially equivalent to the illustrative scenario of sudden varicella elimination. This depends on the fact that residual varicella transmission after initiation of a highly effective program is too low and persists too shortly to provide noticeable protective effects at the population level. Programs above the elimination threshold differ mainly in the incidence values reached at equilibrium by HZ from vaccine strain: Lower coverages imply a smaller population exposed to the risk of reactivation and therefore lower incidence levels. The marginal importance of the immunization scenario suggests that differences in predictions across models are primarily ascribable to the different interpretation of prevaccination dynamics, driven by different parameter estimates and resulting boundary conditions at the beginning of the immunization era (Web Tables 1–3).

The incidence level of HZ caused by vaccine strain at equilibrium also depends significantly on the assumed value for the attenuation parameter, χ. Figure 2 shows the total equilibrium incidence in the different models and scenarios for a broad range of values for the attenuation factor. Results suggest that, in order to obtain an equilibrium HZ that is below the prevaccination level (horizontal line in Figure 2), χ needs to be less than 5% in elimination scenarios under the progressive immunity model and less than at most 30%–40% in concurrent models.

Figure 2.

Model sensitivity with respect to the value of the attenuation coefficient, χ, in the predicted herpes zoster (HZ) risk for vaccinated versus unvaccinated individuals. The predicted total incidence of HZ from vaccine strain at postvaccination equilibrium is shown for Italy in the 3 models discussed and in the different immunization scenarios. Progressive immunity model (solid line); partial immunity model (dashed line); temporary immunity model (dotted line); prevaccination incidence (horizontal solid line); plausible range of χ according to a published epidemiologic study (shaded area) (41). A) Scenario with 70% coverage and 85% efficacy; B) 80% coverage, 95% efficacy; C) 90% coverage, 95% efficacy; and D) 100% coverage, 100% efficacy, and complete catch-up of the susceptible population.

In Web Figure 13, we report the temporal trend in the predicted age-specific incidence of HZ from natural varicella. As immunization programs reach an increasing number of cohorts, the age-specific incidence of natural HZ is gradually forced to shrink rightward from its prevaccination profile. In particular, incidence becomes close to zero for cohorts born after the start of immunization and rises in virtually all other age classes, concentrating at increasingly older ages over time. Once again, this common qualitative pattern shows quantitative differences across models: In this case, a much milder effect is predicted in the temporary immunity model, while proportions between the partial and progressive immunity ones are more similar.

The predicted long-term equilibrium for the baseline case (χ = 10%) of age-specific incidence for HZ from the vaccine strain (Figure 3) is substantially overlapping in the temporary and partial immunity models, with a monotonic exponential increase that reflects the growth of the HZ risk with age. In the case of progressive immunity, however, the risk of HZ from vaccine grows so rapidly that the majority of vaccinated individuals will have already experienced HZ by age 80, where the incidence curve shows a peak followed by a brisk decline. The peak is lower and occurs slightly later in scenarios that allow for sustained varicella circulation at equilibrium, because of the mitigation of the overall HZ risk granted by residual boosting episodes. Intercountry differences in the 3 models, which depend directly on intercountry differences in the estimated risk of HZ, are shown in Web Appendix 2.

Figure 3.

Predicted age-specific incidence of herpes zoster (HZ) from vaccine strain at postvaccination equilibrium per 1,000 individuals for Italy in the 3 models discussed and in the different immunization scenarios. Progressive immunity (solid lines); partial immunity (dashed lines); temporary immunity (dotted lines). Lines for the partial and temporary immunity models are virtually superimposed for Italy (but not for other countries; refer to Web Figures 10–12). A) Scenario with 70% coverage and 85% efficacy; B) 80% coverage, 95% efficacy; C) 90% coverage, 95% efficacy; D) 100% coverage, 100% efficacy, and complete catch-up of the susceptible population.

DISCUSSION

The still limited understanding of the critical role of exogenous boosting in the relationship between varicella and HZ has led to widely different modeling hypotheses, resulting in a high uncertainty in the predicted outcome of varicella vaccination programs. In this work, we evaluate the evolution of HZ epidemiology in 4 European countries following highly effective varicella immunization programs, as predicted by mathematical models based on different biological hypotheses on exogenous boosting (14, 20, 40). The models considered are representative of the 3 main formulations proposed to date, which assume, respectively: 1) a partial protection that can progressively accumulate with further reexposures; 2) a fixed level of partial protection that gradually wanes over time and is restored at each reexposure; and 3) a full but temporary protection against HZ conferred by each reexposure episode.

We show that the different models result in qualitatively similar postvaccination patterns of HZ incidence, with an increase of natural HZ in the medium-term and a long-term emergence of vaccine-related HZ. However, predictions show dramatic quantitative differences. In particular, the progressive immunity model magnifies the perverse effects of immunization on natural HZ and predicts that the long-term incidence of vaccine-related HZ will exceed prevaccination levels in almost all scenarios, even considering a strongly attenuated risk with respect to natural infection. The emergence of a large HZ incidence from the vaccine is especially worrying because it is predicted to occur rapidly after staying silent for several decades. The reason why, somewhat counterintuitively, the progressive immunity model exacerbates the burden of postimmunization HZ lies in the higher reactivation risk, compared with the other hypotheses, estimated for individuals who are never boosted (which become the rule after effective immunization programs).

Results are robust with respect to the overall effectiveness of the vaccination program, with significant differences arising only when the vaccine uptake falls below the elimination threshold. In this case, the impact of varicella vaccination on HZ is partially mitigated by the sustained circulation of varicella and boosting events in the postimmunization era. However, this is not a desirable scenario, as suboptimal vaccination also increases the burden of severe disease (44).

Models used in this study are based on the exogenous boosting hypothesis, whose validity is still debated (12). In addition to the multidisciplinary evidence reported (13), we propose here 2 additional arguments supporting the existence of exogenous boosting. First, it provides a clue for interpreting the inconsistencies of multicountry observations on varicella and HZ (16). Countries with faster varicella circulation should have a larger population at risk of HZ. If the HZ risk were purely age dependent and not dramatically variable across countries (which can be expected under similar socioeconomic and demographic conditions), one would correspondingly expect a higher HZ incidence. However, observations contradict this conclusion by detecting instead less HZ in countries with faster varicella circulation (16), thereby suggesting that the HZ hazard is determined by some intermediate process inversely related to the intensity of varicella circulation.

The second argument stems from the incompatibility between the shape of age-specific HZ incidence profiles and the observed 30% lifetime hazard of HZ (3) under standard senescence mechanisms for HZ (i.e., exponential increase of the HZ risk with age) (1). Indeed, the saturation of HZ incidence at older ages actually observed in the majority of countries can only be justified by a depletion of susceptible individuals. Exogenous boosting keeps under control the size of the population actually susceptible to HZ via immunity conferred by boosting episodes, providing the needed depletion mechanism. In the absence of exogenous boosting, the HZ curve can be expected to decline with age only if the lifetime hazard of HZ is allowed to approach 100%. To date, no alternative explanations for mechanisms of depletion of susceptible individuals have been suggested.

Cautions relative to exogenous boosting hold even more so for progressive immunity, given that no experimental study has been performed yet to test this hypothesis. However, the model including progressive immunity fitted the age patterns of HZ incidence in multiple countries with parsimonious hypotheses, performing better than the partial immunity model (14). In addition, estimates for the HZ risk parameters under progressive immunity were stable with respect to geographical variation, without the need of forcing them to be equal, as was done for the temporary immunity model (40). Furthermore, the high intercountry variability associated with nonbiological parameters in the temporary model may reflect parameter identifiability issues (40) (Web Table 1) that do not appear in the progressive immunity model. For these reasons, despite the lack of direct empirical support, progressive immunity provides a robust interpretation of HZ dynamics.

HZ is a complex phenomenon with data shortcomings and heterogeneous epidemiologic trends, resulting in conflicting evidence that complicates quantitative assessments (45). Mathematical models can help to identify hidden factors that drive some of the counterintuitive trends in HZ. For example, a recent study has shown that the demographic changes that occurred in industrialized countries over the last century have likely contributed to the remarkable increase of HZ incidence observed in several countries in the absence of varicella vaccination (15). However, discrimination between competing models is still an ambitious objective with the current state of knowledge. Indeed, models based on exogenous boosting primarily attempt to explain the observed HZ data by embodying 2 different critical processes of the relationship between varicella and HZ: 1) the immunological dynamics of VZV-specific CMI in response to exogenous reexposures; and 2) the pathogenetic mechanism that allows VZV to overcome the residual CMI and reactivate into HZ. The latter process represents the link between the micro scale of immunological dynamics and the macro scale of clinical manifestation. The current interpretation of the 2 processes is based on the set of qualitative hypotheses developed by Hope-Simpson half a century ago (4), for which available empirical evidence is still sparse. Previous studies have provided valuable information on the existence of boosting by investigating the accumulated effect of reexposures on the clinical occurrence of HZ (8). Other studies have investigated the response of CMI levels to VZV reexposures, independently of HZ manifestation (9, 10). However, a full empirical evaluation of Hope-Simpson's hypotheses requires unpacking the 2 above-mentioned processes, by 1) a better characterization of the dynamic response of CMI to repeated reexposures and 2) the identification of a quantitative relationship between the clinical outcome and the CMI level. This might be obtained by longitudinal measures of CMI on adequately large samples, repeated at relatively short intervals and for a sufficiently long follow-up period, jointly with a diary of presumed reexposure events. These data would allow testing and quantifying the hypothesis of a progressive accumulation of CMI upon repeated reexposures (4), as well as provide information on the timing of CMI waning. In addition, the postulated existence of a protective threshold level for CMI (4), above which HZ does not develop, could be tested by comparing pre-HZ CMI levels in subjects who develop symptoms (so that the acute CMI response that follows HZ has not yet developed) with those of age-matched controls. This ideal design currently presents remarkable economical and practical limitations. However, proof of principle of feasibility of the proposed approach has already been demonstrated in smaller-scale studies (9, 10). The parallel development of mathematical models of the within-host VZV-specific immune response could greatly assist in the interpretation of data, following an approach successfully used for other chronic infections (46, 47). Only after more evidence is made available on the immunological dynamics and HZ pathogenesis will we be able to discriminate between available models or develop more biologically plausible ones.

An alternative indirect approach to providing evidence on Hope-Simpson's progressive immunity hypothesis may come from improved epidemiologic studies on selected subpopulations. For example, one could compare the average HZ incidence in frequently reexposed individuals (such as retired pediatricians, pediatric nurses, and teachers), segregated individuals without a history of boosting episodes (such as secluded monks and nuns and long-term prison inmates), and the general population. The additional collection of cross-sectional VZV-CMI levels may help in the interpretation of data in light of possible heterogeneities in HZ risk factors (e.g., ethnicity composition (48), psychological stress (48), or co-infection incidence (49)), especially in segregated populations.

Overall, this work demonstrates how the present lack of knowledge on exogenous boosting results in large uncertainty on the quantitative impact of varicella vaccination on HZ epidemiology. In particular, its expected negative effect is predicted to be much more dramatic under the progressive immunity formulation. These findings reinforce the need for further research to elucidate the immune and pathogenetic mechanisms of VZV and the opportunity of applying the precautionary principle before further mass varicella immunization programs are set out.