The itchy-scratchy misery of a chickenpox was until recently a rite of passage for children around the world. The varicella-zoster virus causes chickenpox infections. This virus persists in small numbers in nerve cells for many years after infection, and can reactivate from these cells. Often this reactivation causes no symptoms, but sometimes it results in a painful skin condition called shingles (or herpes zoster), especially in older adults.

Some countries—including the United States, Australia, Taiwan and Greece—have virtually wiped out childhood cases of chickenpox by requiring that children be vaccinated against the varicella-zoster virus. But some countries have hesitated. One reason for this hesitation is that exposure to individuals with a chickenpox infection helps boost the immunity of individuals who have previously been infected. This may help reduce the likelihood of these people developing shingles later in life. So, some countries have worried that chickenpox vaccinations might inadvertently increase the number of shingles cases. To assess this risk, many scientists have created computer models, but the models have some limitations.

Now, Ogunjimi et al. report a new individual-based model to assess the effect of childhood varicella vaccination on shingles cases that factors in the immune responses to varicella infection. The model suggests that re-exposure to the varicella virus through contact with infected people would only provide extra protection for about two years; this is much shorter than previous predictions that suggested it might last 20 years. The model also predicts that implementing a varicella vaccination program for children would almost double the number of shingles cases 31 years later. But this increase would be temporary.

The predicted increase in shingles cases is likely to disproportionately occur among 31- to 40-year-olds. This is unexpected because most previous models predict that older age groups would bear the brunt of a rise in shingles, but this younger population would be less likely to develop lasting complications of shingles. Together, these findings may allay some fears about implementing childhood varicella vaccination programs by showing that the benefits of re-exposure are limited.

Varicella-zoster virus (VZV) causes the itching, erythematous vesicular disease called varicella or chickenpox, mainly during childhood, and remains latent in neural ganglia afterwards. Latency can then be interrupted by episodes of (primarily subclinical) reactivation of VZV as shown by the detection of VZV in saliva or blood from otherwise healthy individuals (Schünemann et al., 1998; Nagel et al., 2011). VZV reactivations may also cause herpes zoster (HZ), which presents clinically as a painful dermatomal rash. HZ occurs most frequently in individuals with a drastically declined cellular immune status (Dolin et al., 1978), but ageing itself is often assumed to substantially reduce resilience of the VZV-specific immune response (Miller, 1980; Berger et al., 1981; Levin et al., 2003, 2008). Recent research supports the hypothesis that waning of VZV cellular mediated immunity (CMI) by age is also influenced by cytomegalovirus (CMV) infection (Ogunjimi et al., 2014).

Effective and safe pediatric vaccines against varicella exist and have been universally implemented in some countries including the US, Australia, Greece, Germany, Japan and Taiwan (Marin et al., 2008). However, in many other countries policy makers have been hesitant to introduce childhood VZV vaccination due to the general population's perception of varicella as a relatively mild disease, as well as the so-called exogenous boosting hypothesis (Hope-Simpson, 1965; Ogunjimi et al., 2013). This hypothesis is based on the concept of the secondary immune response and assumes that boosting of VZV-CMI occurs upon re-exposure to varicella. Boosted VZV-specific cellular immunity would subsequently reduce the risk of VZV reactivation and thence HZ. The introduction of widespread childhood VZV vaccination would reduce opportunities for varicella re-exposure and could therefore increase HZ incidence, as shown in model-based projections (Schuette and Hethcote, 1999; Brisson et al., 2000; Bilcke et al., 2013). Although current data on HZ-incidence post introduction of childhood VZV vaccination has caused controversy, a systematic review rating the quality of the evidence, concluded that exogenous boosting exists, but that its population-wide effect after widespread childhood VZV vaccination remains highly uncertain (Ogunjimi et al., 2013). One of the main points of criticism on current predictions by deterministic VZV simulation models is the entanglement of exogenous boosting, waning of immunity, immunosenescence and reactivation undermining the possibility of estimating the magnitude and duration of exogenous boosting accurately. A concern is that these entangled parameters can be chosen to fit these models well to observed HZ incidence data, but that they are too artificial to allow real and verifiable biological interpretations. This leads to currently unverifiable and potentially poor predictions of VZV dynamics beyond the fitted equilibrium states. An additional, hitherto ignored uncertainty, is the potential occurrence of endogenous boosting by subclinical VZV reactivation. Indeed, some studies have shown VZV to reactivate subclinically both in healthy individuals, immunocompromised and stressed individuals (Schünemann et al., 1997; Mehta et al., 2003; Nagel et al., 2011). However, the effect on VZV-CMI has not yet been quantified.

In the current paper, we describe the first individual-based VZV model, explicitly combining within and between-host dynamics. This model is based on experimental viro-immunological data and allows an accurate estimation of the exogenous boosting characteristics and explicit insertion or validation of experimental data.

It is hypothesized that exogenous re-exposure to varicella increases VZV-specific cellular immunity (VZV-CMI). This natural consequence of the secondary immune response will reduce an individual's risk for HZ. When the probability of contact per unit-time between currently infectious and previously recovered varicella patients reduces, through a reduction in varicella incidence, it can be expected that HZ incidence temporarily increases due to a lack of exogenous boosting (Hope-Simpson, 1965; Ogunjimi et al., 2013).

To our knowledge, this study is the first to use an individual-based dynamic transmission model for VZV. This model allowed us to combine immunological and virological data to estimate key parameters in VZV population dynamics, such as the peak CMI response following re-exposure, duration of boosting and VZV subclinical reactivation characteristics. This means that in contrast to the deterministic models where abstract and ad hoc compartments are created to define the transition between different epidemiological states (for example the transition of a varicella recovered state to a zoster susceptible state), we can actually work with true biological, and verifiable, concepts such as VZV-CMI. Our best fitting parameter sets for Belgium suggest the effective duration of exogenous boosting to last only between 1 and 2 years. These predictions are significantly lower than those from the highly cited deterministic VZV model by Brisson et al., in which the average duration was predicted to be 20 years ([7–41 years] 95%CI) (Brisson et al., 2002). Our peak immunological response was estimated to be 2.8–4.0 times larger than the pre-re-exposure value. These predictions are consistent with those experimentally found in adults re-exposed to VZV by varicella contacts in the household (Arvin et al., 1983; Vossen et al., 2004) or by vaccination (Levin et al., 2008). A possible limitation of our study was that all close contacts with varicella patients were assumed to exert an equal ‘average’ boosting effect on the exposed individual. Future studies could assess how important it would be to incorporate variability in the impact of an exposure through direct contact with a varicella case, based on characteristics of both the exposed and the infectious person (e.g., age, comorbidity).

The VZV reactivation probability was estimated to be 5% per week. Observed data on VZV reactivation probability are rather sparse and highly divergent regarding study design, sampling site (saliva, blood, cerebrospinal fluid) and results (Schünemann et al., 1997; Mehta et al., 2003; Engelmann et al., 2008; Birlea et al., 2011; Nagel et al., 2011; van Velzen et al., 2013). The observed weekly VZV reactivation probability is a topic of discussion in the literature and varies between 0 and 71%. As such it is difficult to compare our estimates with the range of observed values. We also note that VZV reactivation in our model might only be detectable at the neural ganglia at not (always) necessarily in peripheral tissues. In order for HZ to occur, we assumed that VZV-CMI should be below a relative threshold (following a specific distribution) during VZV reactivation. None of our best fitting parameter sets predicted a significant effect of VZV reactivation on VZV-CMI, implying that endogenous boosting most likely does not have an impact on the occurrence of HZ. This finding is important since the existence of endogenous boosting has been proposed to have an effect in reducing the negative effects of varicella vaccination on HZ incidence. Future experimental studies should focus on confirming our predicted VZV reactivation probability and the lack of endogenous boosting.

The annual decline of VZV-CMI was predicted to be between 1 and 1.5%. This result is lower than the 2.7–3.9% experimentally observed by Levin et al. for individuals older than 60 years (Levin et al., 2008). The twofold difference in waning rate can be explained by the explicit disentanglement of waning and boosting (with younger age groups having—on average—higher probabilities of being boosted recently thereby actually increasing the observed waning rate). A limitation in our study is the use of a fixed waning rate for all ages. Our results might be interpreted as an averaged result of lower waning rates for younger age groups and higher waning rates for older age groups (as documented by Levin et al. (2008)). A higher waning rate in older age groups could for example be caused by chronic CMV infection (Ogunjimi et al., 2014). Although different types of waning (both in model specification and age dependency) can be used in this kind of simulation models, we believe that further experimental data documenting VZV specific cellular memory as a function of age is needed so that new waning models can be appropriately formulated. One future avenue of research could be the fitting of our predicted VZV-CMI to observed VZV-CMI data, as this will be a mixture of waning, immunosenescence and boosting. Better VZV-CMI datasets than those currently available are needed to be able to do this. These datasets could and should contain immune responses against different VZV peptides and could differentiate between the different cellular immunity compartments (CD4 vs CD8 and central vs effector memory cells). The use of our VZV IBM could help us identify which VZV-CMI compartment is of importance in controlling HZ.

We used our best fitting models to analyze the effects of a 100% effective varicella vaccine implemented for all 1-year-old children. We predicted a net increase in HZ incidence during 50 years and a 1.75 peak fold increase 31 years after introduction of the vaccination program. This delay in the HZ peak incidence is caused by cohorts born close to the time varicella vaccination was introduced experiencing less repeated boosting instances during their childhood than previously born cohorts (a mechanism that is similar to the progressive immunity model proposed in the deterministic VZV model by Guzzetta et al. (2013)). Although increases in HZ incidences following universal childhood varicella vaccination have been noticed, some authors have attributed these increases to a background evolution that was already present prior to CP vaccination (see discussion in Ogunjimi et al. (2013)). However, our analysis shows that proper documentation of significant increases in HZ incidence might not be possible during the first 10 years, even if a 100% effective vaccine would be used. For instance, during the first 10 years of the US program, both uptake and efficacy with the initial single dose vaccination programme were far below 100%. Although our model predicts a much lower duration of boosting than used hitherto in compartmental models (Ogunjimi et al., 2013), some of our overall HZ projections are qualitatively similar. However, in contrast to earlier model estimates our VZV IBM predicts that 31–40 year olds contribute the most to the peak in HZ incidence following varicella vaccination. Some observational studies found no effect of varicella vaccination on HZ incidence for those aged 60 years and older (Hales et al., 2013). This could be compatible with an overall HZ incidence increase due to rising HZ incidence among 31–40 year olds. Importantly, younger adults have been shown to be less likely to develop post-herpetic neuralgia (Opstelten et al., 2002; Drolet et al., 2010). Thus although our aggregated predictions regarding the increase in HZ incidence following varicella vaccination may appear similar to those published using deterministic models, cost-effectiveness analyses using our VZV IBM would be more in favor of universal childhood varicella vaccination.

A major benefit of our modeling approach is the possibility to verify our best–fit parameter values via experimental studies, or vice versa, to adapt parameter values when empiricism delivers new insights. For example, if a new experimental study would prove the existence of endogenous boosting, this could be readily implemented in our VZV IBM so that parameter sets could be estimated, conditional on the existence of endogenous boosting. Although our IBM has given us valuable insights into between-host and within-host VZV dynamics, other influential factors could be introduced in future versions. These factors could relate to CMV-immunosenescence (Ogunjimi et al., 2014), maternal antibodies, reduced VZV-CMI induction if infected during the first year of life as this might improve the prediction of the teenage group (Terada et al., 1994), co-infection with other viruses (Ogunjimi et al., 2015), risk groups (for e.g., immunosuppressed individuals) and HLA types (Meysman et al., 2015). Although current deterministic compartmental models should be able to partially account for these effects, a VZV IBM seems better suited to directly model the influence of these immunity-perturbing factors. Future VZV IBM should also explore more realistic vaccination scenarios as well as inter-country variability (Poletti et al., 2013).

We conclude that our VZV individual-based model has explicitly estimated the duration of exogenous boosting to be limited to only 1 or 2 years and that there was no significant effect from endogenous boosting.

1. 1. Arvin AM
   2. Koropchak CM
   3. Wittek AE

   (1983) Immunological evidence of reinfection with varicella-zoster virus

   The Journal of Infectious Diseases 148:200–205.

   https://doi.org/10.1093/infdis/148.2.200

   • Google Scholar
2. 1. Berger R
   2. Florent G
   3. Just M

   (1981)

   Decrease of the lymphoproliferative response to varicella-zoster virus antigen in the aged

   Infection and Immunity 32:24–27.

   • Google Scholar
3. 1. Bernstein HH
   2. Rothstein EP
   3. Watson BM
   4. Reisinger KS
   5. Blatter MM
   6. Wellman CO
   7. Chartrand SA
   8. Cho I
   9. Ngai A
   10. White CJ

   (1993)

   Clinical survey of natural varicella compared with breakthrough varicella after immunization with live attenuated Oka/Merck varicella vaccine

   Pediatrics 92:833–837.

   • Google Scholar
4. 1. Bilcke J
   2. Jan van Hoek A
   3. Beutels P

   (2013) Childhood varicella-zoster virus vaccination in Belgium: cost-effective only in the long run or without exogenous boosting?

   Human Vaccines & Immunotherapeutics 9:812–822.

   https://doi.org/10.4161/hv.23334

   • Google Scholar
5. 1. Bilcke J
   2. Ogunjimi B
   3. Marais C
   4. de Smet F
   5. Callens M
   6. Callaert K
   7. van Kerschaver E
   8. Ramet J
   9. van Damme P
   10. Beutels P

   (2012) The health and economic burden of chickenpox and herpes zoster in Belgium

   Epidemiology and Infection 140:2096–2109.

   https://doi.org/10.1017/S0950268811002640

   • Google Scholar
6. 1. Birlea M
   2. Arendt G
   3. Orhan E
   4. Schmid DS
   5. Bellini WJ
   6. Schmidt C
   7. Gilden D
   8. Cohrs RJ

   (2011) Subclinical reactivation of varicella zoster virus in all stages of HIV infection

   Journal of the Neurological Sciences 304:22–24.

   https://doi.org/10.1016/j.jns.2011.02.030

   • Google Scholar
7. 1. Brisson M
   2. Edmunds WJ
   3. Gay NJ
   4. Law B
   5. De Serres G

   (2000) Modelling the impact of immunization on the epidemiology of varicella zoster virus

   Epidemiology and Infection 125:651–669.

   https://doi.org/10.1017/S0950268800004714

   • Google Scholar
8. 1. Brisson M
   2. Gay NJ
   3. Edmunds WJ
   4. Andrews NJ

   (2002) Exposure to varicella boosts immunity to herpes-zoster: implications for mass vaccination against chickenpox

   Vaccine 20:2500–2507.

   https://doi.org/10.1016/S0264-410X(02)00180-9

   • Google Scholar
9. 1. Castro Sanchez AY
   2. Aerts M
   3. Shkedy Z
   4. Vickerman P
   5. Faggiano F
   6. Salamina G
   7. Hens N

   (2013) A mathematical model for HIV and hepatitis C co-infection and its assessment from a statistical perspective

   Epidemics 5:56–66.

   https://doi.org/10.1016/j.epidem.2013.01.002

   • Google Scholar
10. 1. Dolin R
    2. Reichman RC
    3. Mazur MH
    4. Whitley RJ

    (1978) NIH conference. Herpes zoster-varicella infections in immunosuppressed patients

    Annals of Internal Medicine 89:375–388.

    https://doi.org/10.7326/0003-4819-89-3-375

    • Google Scholar
11. 1. Drolet M
    2. Brisson M
    3. Levin MJ
    4. Schmader KE
    5. Oxman MN
    6. Johnson RW
    7. Camden S
    8. Mansi JA

    (2010) A prospective study of the herpes zoster severity of illness

    The Clinical Journal of Pain 26:656–666.

    https://doi.org/10.1097/AJP.0b013e3181eef686

    • Google Scholar
12. 1. Engelmann I
    2. Petzold DR
    3. Kosinska A
    4. Hepkema BG
    5. Schulz TF
    6. Heim A

    (2008) Rapid quantitative PCR assays for the simultaneous detection of herpes simplex virus, varicella zoster virus, cytomegalovirus, Epstein-Barr virus, and human herpesvirus 6 DNA in blood and other clinical specimens

    Journal of Medical Virology 80:467–477.

    https://doi.org/10.1002/jmv.21095

    • Google Scholar
13. 1. Guzzetta G
    2. Poletti P
    3. Del Fava E
    4. Ajelli M
    5. Scalia Tomba GP
    6. Merler S
    7. Manfredi P

    (2013) Hope-Simpson's progressive immunity hypothesis as a possible explanation for herpes zoster incidence data

    American Journal of Epidemiology 177:1134–1142.

    https://doi.org/10.1093/aje/kws370

    • Google Scholar
14. 1. Hales CM
    2. Harpaz R
    3. Joesoef MR
    4. Bialek SR

    (2013) Examination of links between herpes zoster incidence and childhood varicella vaccination

    Annals of Internal Medicine 159:739–745.

    https://doi.org/10.7326/0003-4819-159-11-201312030-00006

    • Google Scholar
15. Book

    1. Hens N

    (2012)

    Modeling infectious disease parameters based on serological and social contact data : a modern statistical perspective

    New York: Springer.

    • Google Scholar
16. 1. Hope-Simpson RE

    (1965)

    The nature of herpes zoster: a long-term study and a new hypothesis

    Proceedings of the Royal Society of Medicine 58:9–20.

    • Google Scholar
17. 1. Levin MJ
    2. Oxman MN
    3. Zhang JH
    4. Johnson GR
    5. Stanley H
    6. Hayward AR
    7. Caulfield MJ
    8. Irwin MR
    9. Smith JG
    10. Clair J
    11. Chan IS
    12. Williams H
    13. Harbecke R
    14. Marchese R
    15. Straus SE
    16. Gershon A
    17. Weinberg A

    (2008) Varicella-zoster virus-specific immune responses in elderly recipients of a herpes zoster vaccine

    Journal of Infectious Diseases 197:825–835.

    https://doi.org/10.1086/528696

    • Google Scholar
18. 1. Levin MJ
    2. Smith JG
    3. Kaufhold RM
    4. Barber D
    5. Hayward AR
    6. Chan CY
    7. Chan IS
    8. Li DJ
    9. Wang W
    10. Keller PM
    11. Shaw A
    12. Silber JL
    13. Schlienger K
    14. Chalikonda I
    15. Vessey SJ
    16. Caulfield MJ

    (2003) Decline in varicella-zoster virus (VZV)-specific cell-mediated immunity with increasing age and boosting with a high-dose VZV vaccine

    The Journal of Infectious Diseases 188:1336–1344.

    https://doi.org/10.1086/379048

    • Google Scholar
19. 1. Malavige GN
    2. Rohanachandra LT
    3. Jones L
    4. Crack L
    5. Perera M
    6. Fernando N
    7. Guruge D
    8. Ogg GS

    (2010) IE63-specific T-cell responses associate with control of subclinical varicella zoster virus reactivation in individuals with malignancies

    British Journal of Cancer 102:727–730.

    https://doi.org/10.1038/sj.bjc.6605542

    • Google Scholar
20. 1. Marin M
    2. Meissner HC
    3. Seward JF

    (2008) Varicella prevention in the United States: a review of successes and challenges

    Pediatrics 122:e744–e751.

    https://doi.org/10.1542/peds.2008-0567

    • Google Scholar
21. 1. Mehta SK
    2. Cohrs RJ
    3. Forghani B
    4. Zerbe G
    5. Gilden DH
    6. Pierson DL

    (2003) Stress-induced subclinical reactivation of varicella zoster virus in astronauts

    Journal of Medical Virology 72:174–179.

    https://doi.org/10.1002/jmv.10555

    • Google Scholar
22. 1. Meysman P
    2. Ogunjimi B
    3. Naulaerts S
    4. Beutels P
    5. Van Tendeloo V
    6. Laukens K

    (2015) Varicella-zoster virus-derived major histocompatibility complex class i-restricted peptide affinity is a determining factor in the HLA risk profile for the development of postherpetic neuralgia

    Journal of Virology 89:962–969.

    https://doi.org/10.1128/JVI.02500-14

    • Google Scholar
23. 1. Miller AE

    (1980) Selective decline in cellular immune response to varicella-zoster in the elderly

    Neurology 30:582–587.

    https://doi.org/10.1212/WNL.30.6.582

    • Google Scholar
24. 1. Nagel MA
    2. Choe A
    3. Cohrs RJ
    4. Traktinskiy I
    5. Sorensen K
    6. Mehta SK
    7. Pierson DL
    8. Tyring SK
    9. Haitz K
    10. DiGiorgio C
    11. LaPolla W
    12. Gilden D

    (2011) Persistence of varicella zoster virus DNA in saliva after herpes zoster

    The Journal of Infectious Diseases 204:820–824.

    https://doi.org/10.1093/infdis/jir425

    • Google Scholar
25. 1. Ogunjimi B
    2. Buntinx F
    3. Bartholomeeusen S
    4. Terpstra I
    5. De Haes I
    6. Willem L
    7. Elli S
    8. Bilcke J
    9. Van Damme P
    10. Coenen S
    11. Beutels P

    (2015) Herpes zoster is associated with herpes simplex and other infections in under 60 year-olds

    The Journal of Infection 70:171–177.

    https://doi.org/10.1016/j.jinf.2014.08.016

    • Google Scholar
26. 1. Ogunjimi B
    2. Hens N
    3. Goeyvaerts N
    4. Aerts M
    5. Van Damme P
    6. Beutels P

    (2009) Using empirical social contact data to model person to person infectious disease transmission: an illustration for varicella

    Mathematical Biosciences 218:80–87.

    https://doi.org/10.1016/j.mbs.2008.12.009

    • Google Scholar
27. 1. Ogunjimi B
    2. Smits E
    3. Heynderickx S
    4. Van den Bergh J
    5. Bilcke J
    6. Jansens H
    7. Malfait R
    8. Ramet J
    9. Maecker HT
    10. Cools N
    11. Beutels P
    12. Van Damme P

    (2014) Influence of frequent infectious exposures on general and varicella-zoster virus-specific immune responses in pediatricians

    Clinical and Vaccine Immunology 21:417–426.

    https://doi.org/10.1128/CVI.00818-13

    • Google Scholar
28. 1. Ogunjimi B
    2. Theeten H
    3. Hens N
    4. Beutels P

    (2014) Serology indicates cytomegalovirus infection is associated with varicella-zoster virus reactivation

    Jornal of Medical Virology 86:812–819.

    https://doi.org/10.1002/jmv.23749

    • Google Scholar
29. 1. Ogunjimi B
    2. Van Damme P
    3. Beutels P

    (2013) Herpes zoster risk reduction through exposure to chickenpox patients: a systematic multidisciplinary review

    PLOS ONE 8:e66485.

    https://doi.org/10.1371/journal.pone.0066485

    • Google Scholar
30. 1. Opstelten W
    2. Mauritz JW
    3. de Wit NJ
    4. van Wijck AJ
    5. Stalman WA
    6. van Essen GA

    (2002) Herpes zoster and postherpetic neuralgia: incidence and risk indicators using a general practice research database

    Family Practice 19:471–475.

    https://doi.org/10.1093/fampra/19.5.471

    • Google Scholar
31. 1. Poletti P
    2. Melegaro A
    3. Ajelli M
    4. Del Fava E
    5. Guzzetta G
    6. Faustini L
    7. Scalia Tomba G
    8. Lopalco P
    9. Rizzo C
    10. Merler S
    11. Manfredi P

    (2013) Perspectives on the impact of varicella immunization on herpes zoster. A model-based evaluation from three European countries

    PLOS ONE 8:e60732.

    https://doi.org/10.1371/journal.pone.0060732

    • Google Scholar
32. 1. Schuette MC
    2. Hethcote HW

    (1999) Modeling the effects of varicella vaccination programs on the incidence of chickenpox and shingles

    Bulletin of Mathematical Biology 61:1031–1064.

    https://doi.org/10.1006/bulm.1999.0126

    • Google Scholar
33. 1. Schünemann S
    2. Mainka C
    3. Wolff MH

    (1998) Subclinical reactivation of varicella-zoster virus in immunocompromised and immunocompetent individuals

    Intervirology 41:98–102.

    https://doi.org/10.1159/000024920

    • Google Scholar
34. 1. Terada K
    2. Kawano S
    3. Yoshihiro K
    4. Morita T

    (1994) Varicella-zoster virus (VZV) reactivation is related to the low response of VZV-specific immunity after chickenpox in infancy

    The Journal of Infectious Diseases 169:650–652.

    https://doi.org/10.1093/infdis/169.3.650

    • Google Scholar
35. 1. van Velzen M
    2. Ouwendijk WJ
    3. Selke S
    4. Pas SD
    5. van Loenen FB
    6. Osterhaus AD
    7. Wald A
    8. Verjans GM

    (2013) Longitudinal study on oral shedding of herpes simplex virus 1 and varicella-zoster virus in individuals infected with HIV

    Journal of Medical Virology 85:1669–1677.

    https://doi.org/10.1002/jmv.23634

    • Google Scholar
36. 1. Vossen MT
    2. Gent MR
    3. Weel JF
    4. de Jong MD
    5. van Lier RA
    6. Kuijpers TW

    (2004) Development of virus-specific CD4+ T cells on reexposure to Varicella-Zoster virus

    The Journal of Infectious Diseases 190:72–82.

    https://doi.org/10.1086/421277

    • Google Scholar
37. 1. Weinberg A
    2. Zhang JH
    3. Oxman MN
    4. Johnson GR
    5. Hayward AR
    6. Caulfield MJ
    7. Irwin MR
    8. Clair J
    9. Smith JG
    10. Stanley H
    11. Marchese RD
    12. Harbecke R
    13. Williams HM
    14. Chan IS
    15. Arbeit RD
    16. Gershon AA
    17. Schodel F
    18. Morrison VA
    19. Kauffman CA
    20. Straus SE
    21. Schmader KE
    22. Davis LE
    23. Levin MJ

    (2009) Varicella-zoster virus-specific immune responses to herpes zoster in elderly participants in a trial of a clinically effective zoster vaccine

    The Journal of Infectious Diseases 200:1068–1077.

    https://doi.org/10.1086/605611

    • Google Scholar
38. 1. Westera L
    2. Drylewicz J
    3. den Braber I
    4. Mugwagwa T
    5. van der Maas I
    6. Kwast L
    7. Volman T
    8. van de Weg-Schrijver EH
    9. Bartha I
    10. Spierenburg G
    11. Gaiser K
    12. Ackermans MT
    13. Asquith B
    14. de Boer RJ
    15. Tesselaar K
    16. Borghans JAM

    (2013) Closing the gap between T-cell life span estimates from stable isotope-labeling studies in mice and humans

    Blood 122:2205–2212.

    https://doi.org/10.1182/blood-2013-03-488411

    • Google Scholar
39. 1. Wood SN

    (2011) Fast stable restricted maximum likelihood and marginal likelihood estimation of semiparametric generalized linear models

    Journal of the Royal Statistical Society Series B-Statistical Methodology 73:3–36.

    https://doi.org/10.1111/j.1467-9868.2010.00749.x

    • Google Scholar
40. 1. Yawn BP
    2. Wollan PC
    3. Kurland MJ
    4. St Sauver JL
    5. Saddier P

    (2011) Herpes zoster recurrences more frequent than previously reported

    Mayo Clinic Proceedings 86:88–93.

    https://doi.org/10.4065/mcp.2010.0618

    • Google Scholar

[Editors’ note: this article was originally rejected after discussions between the reviewers, but the submission was accepted after an appeal against the decision.]

Thank you for choosing to send your work entitled “Integrating between-host transmission and within-host immunity to analyze the impact of varicella vaccination on zoster” for consideration at eLife. Your full submission has been evaluated by Prabhat Jha (Senior editor) and three peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the decision was reached after discussions between the reviewers. Based on our discussions and the individual reviews below, we regret to inform you that your work will not be considered further for publication in eLife.

The following individuals responsible for the peer review of your submission have agreed to reveal their identity: Mark Jit (Reviewing editor); Alessia Melegaro (reviewer #3). Another reviewer chose to remain anonymous.

All of us agreed that it is innovative and addresses an important issue which is being explicitly modelled for the first time, i.e. the role of the dynamics of cell-mediated infection and endogenous boosting in VZV reactivation. Given that this concern has had a large influence on vaccine decision making in Europe and elsewhere, the topic is of considerable public health importance. However, we had a number of major concerns with the paper which would require substantial changes to the methodology to address.

Below are the most substantive concerns; however we would encourage the authors to look at the more detailed comments from each reviewer as well.

1) The model is itself is hard to understand and therefore to assess what it adds to the existing literature. The Methods section plunges straight into a discussion of the estimation of the various parameters. It is not clear whether the follows the traditional Brisson/Edmunds/van Hoek approach, the Guzzetta progressive immunity one (AJE 2013) or a different approach entirely. It would be helpful to start with an overview of the model structure (with a written description and/or compartmental flow diagram), as well as the main ways it differs from previous models.

2) It is not clear what the conclusions add to existing modelling work on the same topic. The manuscript criticises existing deterministic models used in current research in the field. However, the main conclusions of the paper (e.g. with regards to vaccine impact) are in line with those of available deterministic models. Although the rise in herpes zoster is predicted to occur earlier in life, it is unclear whether the magnitude of the HZ rise would be smaller, the same or larger than in previous models, and hence the overall significance of this age shift. The fact that a perfect vaccine is used makes the comparison particularly difficult, although it is understandable why this was done in order to show effects more clearly. Hence it would be important to explain clearly what the gains achieved by an individual-based framework taking into account cell-mediated infection explicitly are.

3) In addition, the manuscript points out that a variety of models are able to fit age-specific incidence or prevalence data so conclusions may be a function of model structure rather than biological fact. However, the model as it stands has at least 6 free parameters (i.e. more than in currently used deterministic models), which appear to be fitted to a single output, a zoster age-specific incidence profile. This should worse over-fitting problems, thereby increasing rather than reducing the lack of identifiability. There needs to be some way to address this problem, or at least more extensive sensitivity analyses to show that the conclusions are robust to the range of possible parameter values.

4) Furthermore, like previous models, the conclusions are not completely independent of subjective model choices, since the form and sometimes the parameter values of functions representing cell-mediated immunity and boosting have to be pre-determined (albeit with reference to empirical work). In particular, in the three exogenous boosting scenarios, a number of parameters are simply assumed and no explanations on their values are provided. For example the force of boosting (subsection “Modeling exogenous re-exposure to VZV”) is assumed to occur at the same rate as the force of infection. This assumption (which contrasts with the earlier remark that the actual magnitude of the boosting effects is unknown) should be emphasized, as it might be a main responsible of the fact that seven of the best models “include exogenous boosting” (see Results). Better justification for all the assumptions and assumed parameter values is required; at present it is not clear that this is the “final answer” to the exogenous boosting question.

Reviewer #1:

This paper deals with an important issue, i.e. the extent to which childhood varicella vaccination may cause an increase in HZ in adults due to reduction in exogenous boosting. The observational and ecological evidence about boosting are hard to interpret precisely. As the authors point out, a variety of mathematical models are able to fit age-specific incidence or prevalence data so conclusions may be a function of model structure rather than biological fact. This paper offers a number of improvements that may enhance the believability of conclusions, particularly an individual-based structure which allows tracking of within-host immune dynamics, allowing the duration of exogenous boosting to be fitted rather than largely assumed.

The main difficulty is that the model itself is hard to understand and therefore to assess what it adds to the existing literature. The Methods section plunges straight into a discussion of the estimation of the various parameters. It would be helpful to start with an overview of the model (with a written description and/or compartmental flow diagram), as well as the main ways it differs from previous models.

The comparison with previous models (which have been influential to vaccine policy in much of Europe) needs to be a bit sharper. In particular, it is not clear whether the magnitude of the HZ rise is smaller, the same or larger than previous models (e.g. Brisson). The fact that a perfect vaccine is used makes the comparison particularly difficult, although it is understandable why this was done in order to show effects more clearly. Hence even though the rise in HZ occurs earlier in life, it is not clear what the impact on cost-effectiveness will be.

Also it should be noted that the conclusions are not completely independent of subjective model choices, since the form and sometimes the parameter values of functions representing cell-mediated immunity and boosting have to be pre-determined (albeit with reference to empirical work), as is inevitable. Hence it is not clear that this is the “final answer” to the exogenous boosting question.

However, the model clearly adds a great deal of additional insight to the existing debate, including the use of data that was not previously used in this fashion.

Reviewer #2:

The paper has some innovative aspects, as it represents a first attempt (to my knowledge) to explicitly include into a single model for VZV transmission dynamics and reactivation (and vaccination), the trend in cell mediated immunity (CMI) and endogenous boosting. As trends in CMI and the role of boosting are thought to be the primary determinants of VZV reactivation, explicit consideration of such effects (instead than e.g. implicitly as in available models) is worthwhile.

I have however some major points that should be carefully tackled. Authors criticize deterministic models used in current research in the field (Introduction, “One of the main points of criticism ... equilibrium states.”) Since however some of the main conclusions of this paper are in line with those of available deterministic models (e.g. in regards of the impact of vaccination) authors should explain clearly which are the possible gains allowed by IBMs, particularly in what respects the present approach should be superior to the deterministic models used in the literature. My main point here deals with identifiability of model parameters. The current model has at least 6 free parameters (i.e. more than in the currently used deterministic models), which as I understand are fitted to a single output, a zoster age-specific incidence profile. This should worse over-fitting problems, thereby emphasizing rather than reducing, the aforementioned criticism.

I am not entirely clear about the three hypotheses about CMI decrease after a boosting event (subsection “Modeling exogenous re-exposure to VZV”). In subsection “Modeling waning of VZV-CMI” a hypothesis has been made about CMI decline after primary CP infection. I believe one should take this – for reasons of both modeling parsimony and simplicity – as a baseline holding also for CMI decline after a boosting event, especially given that the data by Levin et al. used by the authors to parametrize the model did not seem to refer to CMI decline after primary chickenpox only. Other hypotheses could then naturally be grounded against this baseline. In particular I am not clear about the meaning of HP1, in subsection “Modeling exogenous re-exposure to VZV”, “Exponential decrease from peak to pre-boosting value based”. Why do you want to prevent that decline continues beyond the pre-boosting situation? The Hp3 formula is incomprehensible and needs careful explanation.

The force of boosting (subsection “Modeling exogenous re-exposure to VZV”) is assumed to occur at the same rate as the FOI. This assumption (which contrasts the authors' remark that the actual magnitude of the boosting effects is unknown) should be emphasized, as it might be a main responsible of the fact that seven of the best models “include exogenous boosting” (see Results).

About the modeling of the Force of reactivation (subsection “Modeling VZV endogenous reactivation”), recent papers have modeled it as an increasing function of age (=senescence), and of time elapsed since last boosting episode. In the paper this is modeled as a purely random factor “The FoR deviates per time step and individual by means of a gamma function.” Am I understanding correctly? In my opinion that should be taken as a (useful) baseline against which, however, to ground alternative, possibly more realistic hypotheses. Therefore this point should be motivated more carefully. How was this gamma (density?) parametrized? This is important because one of the main results on the impact of vaccination i.e. the young age of zoster cases in the post-vaccination period (“Noteworthy is the relative dominance of the 31–40 year...”) might strongly follow from this hypothesis.

Reviewer #3:

This work aims at answering some relevant questions about VZV infection and in particular estimating parameters that describe the exogenous and endogenous boosting effect, a key aspect when considering the introduction of varicella vaccination. For this reason I believe that the paper, in principle, is very relevant and of interest for the research community that is involved in VZV modeling and policy decision making.

However, also considering the standard of the eLife journal, I believe the paper cannot be accepted in its current form. Some major improvements and changes are required to make it fit for publication. In particular, specific modeling assumptions and fitting procedures need to be carefully re-addressed and properly clarified before consideration for publication. See the list below for the required amendments:

1) The analysis is based on an Agent-Based Modeling framework. These models require a very thorough validation procedure, especially when considering a realistic demographic population. The authors do not show any model output that convinces us that the demographic component is suitable to model VZV in Belgium. I believe that these validations outputs should be presented.

2) Details of the model structure used are missing. Is the structure following the traditional Brisson/Edmunds/van Hoek approach or the Guzzetta progressive immunity one (AJE 2013)? Details on this should be available to the reader.

3) The three exogenous boosting scenarios: a number of parameters are simply assumed and no explanations on their values are provided. In particular, why six weeks to reach the peak after exogenous boosting? Also, in scenario 2, 60% of pre-boosting value (why?), steady state for x years (?) and why >= 3 years? A better explanation on these parameter values is required. Also what is the biological explanation for the return to pre-boosting values after x years? Finally, in scenario 3 there's a little bit of confusion between delta t and x. The authors should check the formula of Y(t).

4) The lines “HZ individuals are assumed to be infectious for 1 week and receive a VZV-CMI reset to 1 +/− random factor (normally distributed, cf. primo infection) and “HZ only occurs in about 5% of individuals [Yawn 2011] and that VZV-CMI is higher in recovered HZ patients than in age-matched controls” seem contradicting. What is the post HZ value of VZV CMI? As after primary infection or higher?

5) The sentence “Although HZ infectiousness was needed to maintain circulation of VZV in our model, the relative effects on overall VZV transmission are – compared to CP – marginal”: it is not clear here why you need HZ infectiousness to maintain circulation of VZV infection. If it is post-vaccination we don't care to maintain VZV circulation. If it is pre-vaccination, then there is something wrong if the model need HZ to keep VZV infection.

6) The approximation of HZ infectiousness to varicella breakthrough is a strong assumption. at the least we need to check the sensitivity of model results to variations of this parameter value.

7) The fitting procedure needs to be rethought. It seems that the authors explore in a two-step approach a grid of values and parameter sets. First of all, where is the decision to run the simulation three times coming from? And how is the grid of parametric space explored? And also how are the values in the grid selected? What is the criteria? From Table 2 I notice some very weird combinations of parameters going from step 1 to step 2. It does not seem we are considering the whole parametric space in step 1 and then narrowing the ranges as a function of the results obtained in step 1. A much more comprehensive sensitivity analysis should be performed to evaluate more thoroughly the entire parametric space.

8) The best parameter sets are chosen based on a fit of the HZ incidence data, considered to be binomially distributed. Wouldn't it be better to consider a Poisson distribution for the HZ incidence data (Guzzetta et al., 2013), considering that incidence refers to count of HZ episodes per age group?

9) The authors built an IBM that accounts for both between and within host dynamics, namely, taking into consideration both exogenous and endogenous boosting. The best parameter sets to inform the mathematical model are chosen based on repeated fitting of the HZ age-specific incidence estimates from the model to observed HZ incidence data from Belgium. As regards the fit of the best sets of parameters to the observed HZ data, as shown in Figure 1 and 2, the fit is very poor for the younger age groups, even for the two best parameter sets (9 and 13), as it is duly noted by the authors. Is there any way to improve the fit in this age group, considering that in the deterministic model by Guzzetta et al. (2013) it can be observed a better fit of HZ in young age groups? Maybe the authors should aggregate the incidence into age classes as otherwise the stochastic nature of the data might prevail over the goodness of fit.

10) It seems, from parameters sets 8, 9, and 13, that a higher peak fold increase following exogenous boosting is associated with a better fit of HZ incidence data in the first ten years of age. What about the age group between 10 and 20 years of age, the one suffering the poorest fit?

11) The authors claim that their model can easily accommodate new parameter value coming from experimental studies. However, when it comes to consider the effect of an hypothetic vaccination program, they choose a very simplistic and unrealistic one. Wouldn't it be better if they considered more realistic scenarios, such as those taken into consideration by Poletti et al. (2013) and other works, or some actual vaccination program, as the one in the US or in other countries with an effective varicella vaccination program? Also, the amount and timing of zoster reactivation may be country-specific as shown in Poletti (2013). This should be mentioned in the Discussion, especially considering the current debate on whether or not to introduce universal VZV vaccination in several countries.

12) Figures 4 and 5 should be redone considering maybe an aggregation over time, to avoid all the stochasticity of the model output. In Figure 4 the authors plot the pre-steady state and pre-vaccination incidence which is not of interest. And there is also a part of the plot without any line (right hand side). The interesting bit is the post vaccination part to which the authors only dedicate a small portion of the graph. Also, why are there different colors?

We believe that the individual comments by the reviewers do not point to major shortcomings of our methodology but mainly request clarifications of methods and results. Moreover, we believe some of the reviewers' comments are based on misconceptions, probably due to the paradigm shifting nature of our paper. We hope that eLife is willing to reconsider its position.

Reviewer #1:

The main difficulty is that the model itself is hard to understand and therefore to assess what it adds to the existing literature. The Methods section plunges straight into a discussion of the estimation of the various parameters. It would be helpful to start with an overview of the model (with a written description and/or compartmental flow diagram), as well as the main ways it differs from previous models.

We understand the demand for a better structuring and more elaboration. We have added a new figure detailing an important part of the model, we have majorly restructured the Methods section and we have added a new paragraph called “Model overview”.

The comparison with previous models (which have been influential to vaccine policy in much of Europe) needs to be a bit sharper. In particular, it is not clear whether the magnitude of the HZ rise is smaller, the same or larger than previous models (e.g. Brisson). The fact that a perfect vaccine is used makes the comparison particularly difficult, although it is understandable why this was done in order to show effects more clearly. Hence even though the rise in HZ occurs earlier in life, it is not clear what the impact on cost-effectiveness will be.

We would like to emphasize that our paper is not intended to be a new cost-effectiveness analysis, but instead to present a whole new framework allowing new cost-effectiveness studies to be performed. This is indeed the reason why we discussed the scenario of a perfect vaccine and why we have minimized the direct comparison of results with those resulting from using the other framework. However, we do clearly state that the total peak increase in HZ incidence is similar compared to the results from the other framework, but that a shift to younger age groups could have an important effect on morbidity.

Also it should be noted that the conclusions are not completely independent of subjective model choices, since the form and sometimes the parameter values of functions representing cell-mediated immunity and boosting have to be pre-determined (albeit with reference to empirical work), as is inevitable. Hence it is not clear that this is the “final answer” to the exogenous boosting question.

However, the model clearly adds a great deal of additional insight to the existing debate, including the use of data that was not previously used in this fashion.

We agree with the reviewer that subjective model choices could be the reason of bias. However, the fact that our model is based on experimentally observed data, in contrast to the ad hoc compartmentalization of the deterministic models, helps us in model selection.

Reviewer #2:

[…] I have however some major points that should be carefully tackled. Authors criticize deterministic models used in current research in the field (Introduction, “One of the main points of criticism ... equilibrium states.”) Since however some of the main conclusions of this paper are in line with those of available deterministic models (e.g. in regards of the impact of vaccination) authors should explain clearly which are the possible gains allowed by IBMs, particularly in what respects the present approach should be superior to the deterministic models used in the literature. My main point here deals with identifiability of model parameters. The current model has at least 6 free parameters (i.e. more than in the currently used deterministic models), which as I understand are fitted to a single output, a zoster age-specific incidence profile. This should worse over-fitting problems, thereby emphasizing rather than reducing, the aforementioned criticism.

We would like to note that the 2002 Brisson paper, that lies at the base of all Brisson-alike papers, clearly states in the Appendix that 11 parameters were estimated. However, we do acknowledge that Guzzetta and colleagues only use four parameters to estimate their reactivation rate. An important distinction however is the fact that they do not explicitly account for endogenous boosting for which we have reserved two parameters.

HZ incidence is not a scalar variable but a vectorized variable meaning that – in our case – 80 data points are taken into account for the goodness-of-fit (sufficient to estimate the seven parameters as also shown by our analyses). We do agree that future research should try and account for other fit possibilities and we suggest in our Discussion that new studies should try and obtain better VZV-CMI data as a function of age.

We have added the following text to the Discussion: “This means that in contrast to the deterministic models where abstract and ad hoc compartments are created to define the transition between different epidemiological states (for example the transition of a varicella recovered state to a zoster susceptible state), we can actually work with true biological, and verifiable, concepts such as VZV-CMI.” We also note that the following text was already present in the Discussion: “A major benefit of our modeling approach is the possibility to verify our best–fit parameter values via experimental studies, or vice versa, to adapt parameter values when empiricism delivers new insights. For example, if a new experimental study would prove the existence of endogenous boosting, this could be readily implemented in our VZV IBM so that parameter sets could be estimated, conditional on the existence of endogenous boosting. Although our IBM has given us valuable insights into between-host and within-host VZV dynamics, other influential factors could be introduced in future versions.”

I am not entirely clear about the three hypotheses about CMI decrease after a boosting event (subsection “Modeling exogenous re-exposure to VZV”). In subsection “Modeling waning of VZV-CMI” a hypothesis has been made about CMI decline after primary CP infection. I believe one should take this – for reasons of both modeling parsimony and simplicity – as a baseline holding also for CMI decline after a boosting event, especially given that the data by Levin et al. used by the authors to parametrize the model did not seem to refer to CMI decline after primary chickenpox only. Other hypotheses could then naturally be grounded against this baseline.

Levin et al. presented two types of data: (1) the decline of baseline (thus prior to vaccination) VZV-CMI as a function of age and (2) the decay of VZV-CMI after vaccination. The first is used as a starting value for the waning of VZV-CMI, whereas the second is used as a starting value for the decay of VZV-CMI after vaccination. Our model is thus based on these experimental findings. In order to avoid confusion we have added the notion in the Methods section that the waning of VZV-CMI is informed by the baseline evolution of VZV-CMI as a function of age.

In particular I am not clear about the meaning of HP1, in subsection “Modeling exogenous re-exposure to VZV”, “Exponential decrease from peak to pre-boosting value based”. Why do you want to prevent that decline continues beyond the pre-boosting situation?

We assume that boosting causes a temporary increase of VZV-CMI, so that when VZV-CMI has reached the “pre-boosting” value we can assume that the system has returned to its equilibrium state meaning that waning of VZV-CMI can continue. Figure 6 illustrates this concept. This is indeed an assumption and we hope that future VZV-CMI data will provide disclosure.

The Hp3 formula is incomprehensible and needs careful explanation.

We have changed delta_t to x and we have added “This function is constructed by the boundary conditions that Y(t = x) = Y₀ and Y(t = 0) = P* Y₀” to the explanation. This clarifies the equation (given the prior explanation that the formula represents an exponential decay).

The force of boosting (subsection “Modeling exogenous re-exposure to VZV”) is assumed to occur at the same rate as the FOI. This assumption (which contrasts the authors' remark that the actual magnitude of the boosting effects is unknown) should be emphasized, as it might be a main responsible of the fact that seven of the best models “include exogenous boosting” (see Results).

The statement in the subsection “Modeling exogenous re-exposure to VZV” refers to a probability and not to a rate. We intentionally did not use the wording “force of infection” as this is a typical deterministic phrasing that is not applicable in this individual-based modeling setting. We followed the example of the progressive immunity model of Guzzetta et al. and assumed that all re-exposures to varicella resulted in effective boosting. Furthermore, the parameter values allow situations in which infectious contact occurs, without being boosted. These models, however, did not fit well to the data as none of these models were retained after our selection procedure.

About the modeling of the Force of reactivation (subsection “Modeling VZV endogenous reactivation”), recent papers have modeled it as an increasing function of age (=senescence), and of time elapsed since last boosting episode. In the paper this is modeled as a purely random factor “The FoR deviates per time step and individual by means of a gamma function.” Am I understanding correctly? In my opinion that should be taken as a (useful) baseline against which, however, to ground alternative, possibly more realistic hypotheses. Therefore this point should be motivated more carefully. How was this gamma (density?) parametrized? This is important because one of the main results on the impact of vaccination i.e. the young age of zoster cases in the post-vaccination period (“Noteworthy is the relative dominance of the 31–40 year...”) might strongly follow from this hypothesis.

First of all it is important to emphasize that the deterministic models use an epidemiological compartmentalization in which the force of reactivation represents the transition rate between two groups. In case of the deterministic models, this force of reactivation integrates the concepts of waning of immunity, VZV reactivation and endogenous boosting. And depending of the type of model, exogenous boosting might also be found in the equation for the reactivation rate (Brisson: No; Guzzetta: Yes). We use the force of reactivation as an explicit and independent characteristic of the behavior of VZV. It simply represents VZV reactivation. Whether it will evolve into zoster or endogenous boosting will depend on the value of VZV-CMI. This kind of modeling is not possible when using a deterministic framework. We have added “The reader should imagine the FoR to represent independent reactivation behavior of VZV and whether this reactivation will lead to HZ or endogenous boosting will depend on the value of VZV-CMI.” and “The FoR deviates per time step and individual by means of a gamma probability density function. This gamma function is chosen as it represents the summation of unknown biological phenomena that are assumed to have an exponential distribution.” to the paragraph concerning VZV reactivation. Furthermore, we have changed the title from “Modeling VZV endogenous reactivation” to “Modeling VZV reactivation”. In our VZV-IBM we have tested four different gamma distributions and these were included into the parameter set selection procedures. We reckon that our choice for a gamma distribution might be considered to be arbitrary, but we would like to emphasize that even in deterministic models arbitrary choices are made given the fact the exponential distribution is implicitly assumed for all these models (although often not explicitly mentioned).

Reviewer #3:

[…] However, also considering the standard of the eLife journal, I believe the paper cannot be accepted in its current form. Some major improvements and changes are required to make it fit for publication. In particular, specific modeling assumptions and fitting procedures need to be carefully re-addressed and properly clarified before consideration for publication. See the list below for the required amendments:

We believe we have made the necessary changes that allow our innovative paper to be suitable for publication in eLife.

1) The analysis is based on an Agent-Based Modeling framework. These models require a very thorough validation procedure, especially when considering a realistic demographic population. The authors do not show any model output that convinces us that the demographic component is suitable to model VZV in Belgium. I believe that these validations outputs should be presented.

We have completely reformatted our description of demography to “The dynamics of the synthetic population of the individual-based model are based on Belgian population and mortality data normalized to a fixed total population of 998,400 individuals [Eurostat, 2012]. The chosen population size is the result of a trade-off between ensuring sufficient heterogeneity and a manageable computational burden. Natural deaths are selected based on the age-dependent mortality rate. Per time step, the number of newborns equals the number of deaths to obtain a constant population size. To conduct predictions over many years, a stationary demographic structure is required throughout the simulated period. The Belgian population from 2012 has an overrepresentation of people from age 40 to 60 years (see Figure 7). Applying a fixed age-dependent mortality rate in combination with the replacement by newborns would results in an oscillating population distribution over time with respect to age. Therefore, the initial age distribution is not based on the Belgian population count from 2012 but on the survivor function estimated from data on the number of deaths per capita. The survivor function by age is estimated from data on the number of deaths per capita by $m\left(a\right)=\text{exp}(-{{\displaystyle \int }}_0^a\mu \left(t\right)dt)$ with age-dependent mortality rate μ using a thin plate regression spline model with the gam-function in the R-package mgvc (Hens 2012, Wood 2011). Figure 7 illustrates the survivor function by age together with the Belgian population and mortality rates in 2012.

The model runs in time steps of one week and people with week-age 53 move to the next age class. To obtain homogeneous age transitions throughout the simulated period, initial week-ages are randomly assigned between 1 and 52. People from the last age class with 53 weeks are removed from the population so that the demographic structure remains stable throughout the simulated period.”

2) Details of the model structure used are missing. Is the structure following the traditional Brisson/Edmunds/van Hoek approach or the Guzzetta progressive immunity one (AJE 2013)? Details on this should be available to the reader.

We understand the demand for a better structuring and more elaboration. We have added a new figure detailing an important part of the model, we have majorly restructured the Methods section and we have added a new paragraph called “Model overview”.

3) The three exogenous boosting scenarios: a number of parameters are simply assumed and no explanations on their values are provided. In particular, why six weeks to reach the peak after exogenous boosting?

Our model is informed by experimentally observed data and most of the information is based on the 2008 Levin et al paper where the time point of six weeks was used as the time point to assess VZV-CMI. We agree that the time point of boosting could also be placed at four weeks after re-exposure, but on a larger scale this will have no effect on our results. We have added “The six week duration between the boosting event and the peak has been influenced by the Levin et al. data (Levin 2008)” to the “Modeling exogenous re-exposure to VZV” paragraph.

Also, in scenario 2, 60% of pre-boosting value (why?), steady state for x years (?) and why >= 3 years? A better explanation on these parameter values is required.

Again, these numbers come from the Levin et al. paper. In order to further clarify this we have expanded our citation to that reference to “(cf. Figure 3 in the Levin et al. paper)”.

Also what is the biological explanation for the return to pre-boosting values after x years?

All vaccine studies show the tendency for cellular immunity after boosting to decay (in many studies after some time) to the base line values. Speculations in the experimental studies where this has been observed and in the biomedical literature in general, to try and explain this goes beyond VZV and certainly beyond the purpose of our paper. From a biological perspective this means that the original infection creates an equilibrium value (set-point) for VZV-CMI. From a clinical perspective this could also be illustrated by the decline of Zostavax vaccine efficacy with time after vaccination. However, we note that our parameter sets allow the “x” to be very large so that the duration of boosting could even go up to 20 years.

Finally, in scenario 3 there's a little bit of confusion between delta t and x. The authors should check the formula of Y(t).

The reviewer is correct. We have changed all delta t’s to x.

4) The lines “HZ individuals are assumed to be infectious for 1 week and receive a VZV-CMI reset to 1 +/− random factor (normally distributed, cf. primo infection) and “HZ only occurs in about 5% of individuals [Yawn 2011] and that VZV-CMI is higher in recovered HZ patients than in age-matched controls” seem contradicting. What is the post HZ value of VZV CMI? As after primary infection or higher?

In our model successful VZV reactivation is assumed to cause an equal increase of VZV-CMI as compared to chickenpox. However, due to waning of VZV-CMI in individuals not having HZ the actual VZV-CMI could be higher in those after HZ.

5) The sentence “Although HZ infectiousness was needed to maintain circulation of VZV in our model, the relative effects on overall VZV transmission are – compared to CP – marginal”: it is not clear here why you need HZ infectiousness to maintain circulation of VZV infection. If it is post-vaccination we don't care to maintain VZV circulation. If it is pre-vaccination, then there is something wrong if the model need HZ to keep VZV infection.

Deterministic models do not need HZ infectiousness to maintain VZV transmission, but IBM do need HZ infectiousness (or multiple re-introductions from outside our populations, but that is beyond the scope of this paper) to reach endemic equilibrium (we realize that our model might be the only model so far to endemically model a viral infection by means of a IBM).

6) The approximation of HZ infectiousness to varicella breakthrough is a strong assumption. at the least we need to check the sensitivity of model results to variations of this parameter value.

We understand the hesitance of the reviewer regarding this approximation. We note that Brisson et al. (2000) used the same approximation. We added a sensitivity analysis to verify the robustness for our 13 best fitting parameter sets. This is shown as Figure 2–figure supplement 2. The reader can verify that for a range of parameter values of the HZ infectiousness (from 0.03 to 0.45) the predicted HZ incidence does not vary that much. We have also added “Figure 2–figure supplement 2 shows the results of a sensitivity analysis in which we varied HZ infectiousness from 0.03 to 0.45 for the thirteen best fitting parameter. As can be seen in Figure 2–figure supplement 2 our results are quite robust” to the respective Methods paragraph.

7) The fitting procedure needs to be rethought. It seems that the authors explore in a two-step approach a grid of values and parameter sets. First of all, where is the decision to run the simulation three times coming from? And how is the grid of parametric space explored? And also how are the values in the grid selected? What is the criteria? From Table 2 I notice some very weird combinations of parameters going from step 1 to step 2. It does not seem we are considering the whole parametric space in step 1 and then narrowing the ranges as a function of the results obtained in step 1. A much more comprehensive sensitivity analysis should be performed to evaluate more thoroughly the entire parametric space.

Our results showed that the stochasticity of the simulations did not have a major influence on the predicted HZ incidences per parameter set as illustrated by the fact that the median relative variance of the deviance of the top 13 parameter sets for the three runs is only 1.9% (also see Figure 2–figure supplement 2). As such, we decided to use three runs per parameter set to account for a reasonable amount of stochasticity.

All parameter values are combined in the first parameter sets meaning that the entire parameter grid (informed by existing experimental data) is explored in the first step. In the second step we explore in more detail the most significant regions (as detailed in the Methods section). It is also important to realize that 1 simulation (thus one run for one parameter set) costs about 6-7 hours of computing time for one core meaning that it is not feasible to explore the entire parameter space with equal detail. Our presented results include data from more the 24,000 runs thus representing approximately 150,000 CPU hours (not including preparatory work).

8) The best parameter sets are chosen based on a fit of the HZ incidence data, considered to be binomially distributed. Wouldn't it be better to consider a Poisson distribution for the HZ incidence data (Guzzetta et al. 2013), considering that incidence refers to count of HZ episodes per age group?

Published data by Yawn et al. showed that HZ recurrence rate within two years is 2% (even not significantly different from 1 when only looking at immunocompetent individuals). As such we assumed that individuals had either HZ or not during a year and this is best modeled by a binomial distribution.

9) The authors built an IBM that accounts for both between and within host dynamics, namely, taking into consideration both exogenous and endogenous boosting. The best parameter sets to inform the mathematical model are chosen based on repeated fitting of the HZ age-specific incidence estimates from the model to observed HZ incidence data from Belgium. As regards the fit of the best sets of parameters to the observed HZ data, as shown in Figure 1 and 2, the fit is very poor for the younger age groups, even for the two best parameter sets (9 and 13), as it is duly noted by the authors. Is there any way to improve the fit in this age group, considering that in the deterministic model by Guzzetta et al. (2013) it can be observed a better fit of HZ in young age groups? Maybe the authors should aggregate the incidence into age classes as otherwise the stochastic nature of the data might prevail over the goodness of fit.

We disagree with the strong statement that our model fits “very poor” for younger age groups. We also note that Figure 3 in Guzzetta et al. shows a poorer fit for the teenagers (although the size and scale of Figure 3 in Guzetta et al. hampers full appreciation of the fit). We do agree that our fit for the younger age groups is not as good as our excellent fits for the older – and more relevant – age groups and this finding might be due to a lack of relevant biological data for these age groups. We believe that future models and experimental studies should look into the effect of having chickenpox as an infant as it has been observed (as mentioned in our Discussion) that these individuals have lower VZV-CMI compared to others. We have expanded our sentence in the Discussion to “reduced VZV-CMI induction if infected during the first year of life as this might improve the prediction of the teenage group (Terada 1994)”.

10) It seems, from parameters sets 8, 9, and 13, that a higher peak fold increase following exogenous boosting is associated with a better fit of HZ incidence data in the first ten years of age. What about the age group between 10 and 20 years of age, the one suffering the poorest fit?

Parameter sets 8, 9 and 13 also have other different parameters (including a higher VZV reactivation rate). As stated in the previous response, we believe that it is rather the observed HZ incidence in the teenage group that is higher than would be expected and we think that future versions of our model (in the same way that the Brisson models improved over the years) should try to account for this.

11) The authors claim that their model can easily accommodate new parameter value coming from experimental studies. However, when it comes to consider the effect of an hypothetic vaccination program, they choose a very simplistic and unrealistic one. Wouldn't it be better if they considered more realistic scenarios, such as those taken into consideration by Poletti et al. (2013) and other works, or some actual vaccination program, as the one in the US or in other countries with an effective varicella vaccination program? Also, the amount and timing of zoster reactivation may be country-specific as shown in Poletti (2013). This should be mentioned in the Discussion, especially considering the current debate on whether or not to introduce universal VZV vaccination in several countries.

Our paper presents the first VZV IBM to date and presents a whole new method for dealing with the VZV “problem”, and estimates hitherto poorly documented immunological parameters. We agree with the comment by the reviewer that future more applied analyses should account for a broader (and more realistic) range of vaccination scenarios than the illustrative example from this conceptual paradigm shifting paper. We have added “If future VZV IBM are applied to project the impact of interventions, they should of course explore more realistic vaccination scenarios and assess inter-country variability in these explorations (Poletti 2013)” to the Discussion with reference to the Poletti et al. paper.

12) Figures 4 and 5 should be redone considering maybe an aggregation over time, to avoid all the stochasticity of the model output. In Figure 4 the authors plot the pre-steady state and pre-vaccination incidence which is not of interest. And there is also a part of the plot without any line (right hand side). The interesting bit is the post vaccination part to which the authors only dedicate a small portion of the graph. Also, why are there different colors?

Aggregation can be understood as presmoothing and it has been shown in the statistical literature that presmoothing is often not appropriate to handle stochasticity (see e.g. [Aerts 2010]). In order to present our results as transparent as possible we prefer to present the results “as they are”, meaning with the inclusion of full stochasticity. Otherwise the reader might be confused with the traditional output from deterministic models. We strongly disagree that the “interesting bit is post vaccination”. We have fitted this model to data pre-vaccination, and verify simultaneously some poorly understood biomedical parameters using a highly novel individual-based approach. The concept of the model is the primary interest in this paper, not the projections that can be made with it. We thus prefer to show all data including the pre-steady and pre-vaccination data. We only analyzed the post-vaccination data up to 80 years after vaccination.

Author details

1. Benson Ogunjimi

   1. Centre for Health Economics Research and Modeling Infectious Diseases, Vaccine and Infectious Disease Institute, University of Antwerp, Antwerp, Belgium
   2. Interuniversity Institute for Biostatistics and Statistical Bioinformatics, Hasselt University, Hasselt, Belgium

   Contribution

   BO, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   For correspondence

   benson.ogunjimi@uantwerp.be

   Competing interests

   The authors declare that no competing interests exist.

   "This ORCID iD identifies the author of this article:" 0000-0002-0831-2063

2. Lander Willem

   1. Centre for Health Economics Research and Modeling Infectious Diseases, Vaccine and Infectious Disease Institute, University of Antwerp, Antwerp, Belgium
   2. Interuniversity Institute for Biostatistics and Statistical Bioinformatics, Hasselt University, Hasselt, Belgium
   3. Department of Mathematics and Computer Science, University of Antwerp, Antwerp, Belgium

   Contribution

   LW, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

3. Philippe Beutels

   1. Centre for Health Economics Research and Modeling Infectious Diseases, Vaccine and Infectious Disease Institute, University of Antwerp, Antwerp, Belgium
   2. Interuniversity Institute for Biostatistics and Statistical Bioinformatics, Hasselt University, Hasselt, Belgium
   3. School of Public Health and Community Medicine, University of New South Wales, Sydney, Australia

   Contribution

   PB, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

4. Niel Hens

   1. Centre for Health Economics Research and Modeling Infectious Diseases, Vaccine and Infectious Disease Institute, University of Antwerp, Antwerp, Belgium
   2. Interuniversity Institute for Biostatistics and Statistical Bioinformatics, Hasselt University, Hasselt, Belgium

   Contribution

   NH, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

• 4,182

  Page views

• 391

  Downloads

• 28

  Citations

Article citation count generated by polling the highest count across the following sources: Scopus, Crossref, PubMed Central.