Risk of herpes zoster after exposure to varicella to explore the exogenous boosting hypothesis: self controlled case series study using UK electronic healthcare data
BMJ 2020; 368 doi: https://doi.org/10.1136/bmj.l6987 (Published 22 January 2020) Cite this as: BMJ 2020;368:l6987Linked Opinion
What is the varicella zoster virus boosting effect?
All rapid responses
Rapid responses are electronic comments to the editor. They enable our users to debate issues raised in articles published on bmj.com. A rapid response is first posted online. If you need the URL (web address) of an individual response, simply click on the response headline and copy the URL from the browser window. A proportion of responses will, after editing, be published online and in the print journal as letters, which are indexed in PubMed. Rapid responses are not indexed in PubMed and they are not journal articles. The BMJ reserves the right to remove responses which are being wilfully misrepresented as published articles or when it is brought to our attention that a response spreads misinformation.
From March 2022, the word limit for rapid responses will be 600 words not including references and author details. We will no longer post responses that exceed this limit.
The word limit for letters selected from posted responses remains 300 words.
RE: Risk of herpes zoster following exposure to a varicella case: a self-controlled case series study using UK electronic healthcare data exploring the exogenous boosting hypothesis.
As a peer-reviewer of this manuscript, I commend the authors for addressing an important (if contentious) topic using powerful techniques. Unfortunately, I fear that my initial review was insufficiently clear, and the final publication still contains flaws that leave me questioning the validity of its results, quantitatively and qualitatively.
The strength of self-controlled methods is that cases serve as their own controls, so analysts need only control for time-varying characteristics. The authors carefully controlled for age, calendar-time and season. However, they overlooked a key time-varying covariate, parenthood status: “exposed time” (defined by varicella episodes occurring among children in the household) necessarily restricted participation to parents, whereas “baseline time” (defined by lack of varicella episodes) did not restrict by parenthood. Given that the median age of children experiencing varicella was 3.8 years, whereas the median duration of the pre-risk baseline period was greater, 5.0 years, it is likely that many cases were parents during exposed time, but non-parents during baseline time. Here is why this is important:
Detection of HZ in such studies is limited to medically-attended cases (unattended cases are unrecognized). While seniors experiencing HZ generally seek healthcare since the symptoms are so distressing, even among seniors, healthcare seeking varies by patient characteristics [1]. Among younger adults (the bulk of eligible subjects), the proportion of persons with HZ seeking care is likely much lower, since the symptoms are much less distressing. That lower proportion is also likely modified by many factors. Even in settings with free healthcare and paid sick leave, parenthood status is a critical factor that can affect HZ healthcare seeking (ie, ascertainment), irrespective of endogenous boosting effects, mediated by differences in available time, budgets, and/or childcare requirement for visiting the doctor: time, funds, and childcare are typically more problematic for parents (and the effect probably varies by household-childrens’ age). For instance, one study found that the incidence of 15 of 16 acute symptomatic conditions was lower in parents versus non-parents, with incidence ratios ranging from 0.60 to 0.95 [2], ie, a similar magnitude to that for HZ during exposed versus baseline time. In summary, the current study should have restricted to parents.
Ascertainment of varicella is poorer still: published UK and EU data [3,4] show that about 90% of children experience varicella by age 18 years, yet just a minority (perhaps 25-40%) come to medical attention in the very setting studied in this report [5]. The household factors affecting healthcare-seeking for varicella and HZ are likely linked, providing many opportunities for artifacts.
Few self-controlled studies include negative controls; this is a great study enhancement, but one pair of exposure and outcome controls is inadequate to identify bias and provide reassurance. Regarding gastroenteritis (the exposure control), virtually all young children, particularly infants, experience episodes, but only a minority are medically-attended (ie, ascertained) [6-8]; conversely, regarding hip fractures (the outcome control), they are so severe that virtually all episodes come to medical attention, irrespective of patient characteristics (furthermore, this outcome was underpowered; while spline methods removed most age effect, it is easy to imagine residual effects of a similar magnitude to the reported protective effects of varicella).
It is impossible to know how the complex, intertwining factors noted above affected study outcomes, but the affects could easily be large enough to fully explain the observed results (or to diminish them), and I receive no reassurance by the fact that the results seem counterintuitive, as follows:
First, a varicella exposure protected for 20 years. If so, a large proportion of adults would be protected (particularly given that just a fraction of exposures were ascertained). There is no evidence for such widespread, durable protection in real-world data from Australia or the US [9]. As an aside, I question that the incidence ratios (0.67 [95% CI 0.62-0.73], 0.69 [0.63-0.76], 0.69 [0.61-0.77], and 0.73 [0.62-0.87] at 0-1, 2-4, 5-9, and 10-20 years post-varicella exposure, respectively) reflect a trend with the reported significance of P=0.031.
Second, there was no incremental protection from varicella exposures subsequent to first exposures. While not impossible, it is counterintuitive. Given the context of immunosenescence, the lack of effect modification by age at varicella exposure is counterintuitive as well.
Self-controlled methods can be powerful. The authors creatively applied these methods to an important topic, but used an extraordinarily long observation time and assumed that control of time alone would take care of important residual confounding. Given the inadequate control of parental status and the poor ascertainment of both exposures and outcomes, I am skeptical of the results, and their counter-intuitiveness only increases my skepticism. I would have been more convinced of the results if participation had been restricted to parents, and if there were additional control groups. In the end, healthcare seeking influences are challenging to mitigate; this remains an important topic for better analytic approaches, not only for self-controlled studies but for all medical-sector based studies. Absent such new approaches, readers should always bear healthcare seeking influences in mind as they consider all observational studies.
References:
1) Hales CM, Harpaz R, Bialek SR. Self-reported herpes zoster, pain, and health care seeking in the Health and Retirement Study: implications for interpretation of health care-based studies. Annals of epidemiology. 2016;26(6):441-6.e3.
2) Leung J, Harpaz R. Interpretation of data derived from healthcare settings: Cautionary findings related to parenthood status. Abstract 2053.0, APHA2018, Nov 10-14, San Diego, CA. https://apha.confex.com/apha/2018/meetingapp.cgi/Paper/416612.
3) Vyse AJ, Gay NJ, Hesketh LM, Morgan-Capner P, Miller E. Seroprevalence of antibody to varicella zoster virus in England and Wales in children and young adults. Epidemiol Infect. 2004 Dec;132(6):1129-34.
4) Bollaerts K, Riera-Montes M, Heininger U, Hens N, Souverain A, Verstraeten T, Hartwig S. A systematic review of varicella seroprevalence in European countries before universal childhood immunization: deriving incidence from seroprevalence data. Epidemiol Infect. 2017 Oct; 145(13): 2666–2677.
5) Walker JL, Andrews NJ, Mathur R, Smeeth L, Thomas SL. Trends in the burden of varicella in UK general practice Epidemiol Infect. 2017 Oct; 145(13): 2678–2682.
6) Jones TF, McMillian MB, Scallan E, Frenzen PD, Cronquist AB, Thomas S, Angulo FJ. A population-based estimate of the substantial burden of diarrhoeal disease in the United States; FoodNet, 1996-2003. Epidemiol Infect. 2007 Feb;135(2):293-301
7) Hall AJ, Rosenthal M, Gregoricus N, Greene SA, Ferguson J, Henao OL, Vinjé J, Lopman BA, Parashar UD, Widdowson MA. Incidence of acute gastroenteritis and role of norovirus, Georgia, USA, 2004-2005. Emerg Infect Dis. 2011 Aug;17(8):1381-8. doi: 10.3201/eid1708.101533.
8) Schmidt MA, Salas SB, Yamshchikov V, et al. The CAGE Study: Prevalence of Acute Gastroenteritis and Enteric Virus Infection in the Community. Abstract 694, IDWEEK2018, Oct 3-7 2018, San Francisco, CA. https://idsa.confex.com/idsa/2018/webprogram/Paper69787.html.
9) Harpaz R. Do varicella vaccination programs change the epidemiology of herpes zoster? A comprehensive review, with focus on the United States. Expert Rev Vaccines. 2019;18(8):793-811.
Competing interests: No competing interests
Re: Risk of herpes zoster after exposure to varicella to explore the exogenous boosting hypothesis: self controlled case series study using UK electronic healthcare data
Dear Editor
We thank Dr Harpaz for his response to our article. He raises an interesting possibility: that parents may seek healthcare less than non-parents, so there may be lower ascertainment of zoster in the exposed (post-varicella) than unexposed (pre-varicella) period. If this were true, we may have overestimated the protective effects of exposure of adults to varicella on zoster risk.
Nevertheless, this possibility is not supported by an additional post-hoc analysis that we performed, restricted to parents only. The date adults become parents was derived using the date when the first child in the household was born. Among the 9,604 participants in the main analysis, 6,867 participants’ (72%) start of follow-up remained the same (i.e. they were already parents at the original start of follow-up date). After amending the start of follow-up in the remaining 28%, we dropped 778 participants, as their zoster diagnosis occurred prior to their new start of follow-up. This left a total of 8,826 participants for the analysis. The median length of the baseline (pre-risk period) changed from 5.0 years (IQR: 2.3-8.8) in the main analysis to 3.7 years (IQR: 1.8-6.8) in this post-hoc analysis. The results were unchanged when restricting follow-up time to time when the participants were parents (or grandparents) in this way (full results available on request).
Concerning the suggestion that decisions to seek healthcare for zoster and varicella could be linked, we acknowledged this in the paper. As discussed, there may be lower attendance for zoster during the child’s acute varicella illness, due to caring responsibilities. This would result in slightly lower zoster ascertainment in the week after varicella diagnosis, which may have moved the association between varicella exposure and zoster away from the null for the first risk period (0-2 years post exposure).
Dr Harpaz also highlights the challenge of choosing suitable exposure and outcome controls. However, our negative controls were determined a-priori, with a clear rationale and considerable thought given to controls that would be the most appropriate. Both controls gave the results we were anticipating a-priori, therefore we are very reluctant to add additional controls post-hoc. The negative exposure, rotavirus, was chosen as it would not plausibly be expected to influence the future risk of zoster. As Dr Harpaz points out, it is likely to be under-ascertained, so some exposed time will be included in baseline time. This would tend to bias effect estimates toward the null. However, to incur serious bias it would need a large effect size and a large amount of the baseline time to be wrongly classified, which is unlikely. The negative outcome, hip fracture, was chosen as it is strongly associated with increasing age and if the age-adjustment in our primary analysis was sufficient, we should observe a null association between varicella exposure and hip fracture. That we have seen no association between varicella and hip fracture risk, after adjusting fully for age, suggests this is not biasing our study findings. As Dr Harpaz says, hip fractures are severe enough almost all will come to medical attention, which will mean there is little or no reduction in power in this analysis.
Regarding the trend, when excluding the baseline and pre-exposure groups there was some evidence, albeit weak, of a linear trend (P=0.031, comparing a model with no exposure group and a linear exposure). We have checked the test for trend and would be happy to share our statistical code and outputs if helpful.
As Dr Harpaz states, there was no evidence to indicate that protection against varicella zoster virus reactivation accumulated upon repeated exposure (the “progressive immunity” hypothesis). However, as we acknowledged in the paper, our study may have been underpowered to detect such an effect. Future studies using a similar self-controlled case series study design with a larger study sample size may be able to address this issue with more precision.
Competing interests: No competing interests