Herpesvirus infections and risk of Parkinson’s disease

Abstract

Introduction

Herpesviruses might play a role in the pathogenesis of neurodegenerative disorders. We sought to examine a possible association between alpha herpesvirus infections and Parkinson’s disease.

Methods

We conducted a population-based case-control study of incident Parkinson’s disease in 2009 Medicare beneficiaries age 66-90 (89,790 cases, 118,095 randomly-selected comparable controls). We classified beneficiaries with any diagnosis code for “herpes simplex” and/or “herpes zoster” in the previous five years as having had the respective alpha herpesvirus(es). In beneficiaries with Part D prescription coverage, we also identified those prescribed anti-herpetic medications. We calculated odds ratios (ORs) and 95% confidence intervals (CIs) between alpha herpesvirus diagnosis/treatment and Parkinson’s disease with logistic regression, with adjustment for age, sex, race/ethnicity, smoking, and use of medical care.

Results

Parkinson’s disease risk was inversely associated with herpes simplex (OR=0.79, 95% CI 0.74-0.84), herpes zoster (OR=0.88, 95% CI 0.85-0.91), and anti-herpetic medications (OR=0.87, 95% CI 0.80-0.96).

Discussion/Conclusion

Herpesvirus infection or treatment might reduce risk of Parkinson’s disease, but future studies will be required to explore whether this inverse association is causal.

Introduction

Human alpha herpesviruses are chronic neurotropic viruses that typically enter the peripheral nervous system of the human host. These viruses can reactivate, or cause recurrent disease, within neuronal sensory cells after a latency period [1]. The alpha herpesvirus subfamily includes herpes simplex virus (HSV) 1 and 2 and varicella-zoster virus (VZV), the cause of herpes zoster (HZ). Reactivation may occur due to immunosuppression, although reactivation can occur in immunocompetent hosts as well [1]. However, HZ more commonly affects older adults and immunocompromised individuals than HSV [2]. Several epidemiologic studies have explored whether these neurotropic viruses might play a role in the pathogenesis of Parkinson’s disease (PD), with conflicting results for both HSV and HZ [3–12]. Given that these two alpha herpesviruses establish latency in the peripheral nervous system, such as the dorsal root or trigeminal ganglia [13, 14], typically well before the onset of PD, these associations might have pathophysiologic implications that further our understanding about possible mechanisms involved in the occurrence of PD. Therefore, we investigated the relationship between prior infection with HSV and HZ, as well as anti-herpetic medications as indicators of infection, and risk of PD in a large, population-based study of United States Medicare beneficiaries.

Methods

Study design and eligibility criteria

We constructed a population-based case-control study of incident PD using comprehensive Medicare claims data including carrier (physician/supplier Part B), outpatient, inpatient, skilled nursing facility, home health care, and durable medical equipment claims data from 2004-2009 [15, 16]. All participants met the following criteria for 2009: Enrolled in Medicare Parts A/B without other coverage, residence in the United States, age-eligible for Medicare for ≥2 years (age ≥66 years, 11 months) at PD diagnosis/randomly-assigned control reference date, and age ≤90 years at diagnosis/reference. Nearly all Americans in this age group are enrolled in Medicare.

Identification of cases and controls

Newly diagnosed PD cases (N=89,790) had an International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) code 332 or 332.0 in 2009 and no prior year, and no ICD-9-CM code for Lewy body dementia (331.82) and/or atypical parkinsonism (333 or 333.0). Controls (N=118,095) were a random sample of remaining beneficiaries eligible for the study with none of the above codes. Because of our large sample size, matching was unnecessary [17]. Instead, we addressed confounding analytically, and avoided matching to obtain unbiased ORs for demographic risk factors of PD. This was essential for our previous development of a PD predictive model and confirmation that cases demonstrated all known demographic associations [15].

Ascertainment of herpesvirus diagnosis/treatment and covariates

We extracted ICD-9-CM diagnosis/procedure codes, and Healthcare Common Procedure Coding System codes, including Current Procedural Terminology codes, from all available claims up to the PD diagnosis/control reference date in 2009. We created a dichotomous variable for each code observed in 2004-2009 prior to this date. We classified beneficiaries as having HSV and/or HZ if they had ≥1 ICD-9-CM codes 054.xx and/or 053.xx, respectively [18]. We similarly used ICD-9-CM codes to identify ten other medical conditions considered as potential confounders elsewhere [6]. We used >600 diagnosis/procedure codes to calculate the probability of having ever smoked regularly [15, 19]. We also counted the number of unique diagnosis codes as an indicator of use of medical care [19].

Among Medicare beneficiaries with Part D outpatient prescription coverage with ≥1 fill of any medication in available years (2008-2009) (48,295 cases and 52,324 controls), we created a secondary herpesvirus exposure variable, ever having ≥1 fill of any anti-herpetic medication (acyclovir, valacyclovir, valganciclovir, famciclovir, ganciclovir, or penciclovir). We also identified beneficiaries with ≥1 fill of an immunosuppressant [20].

Statistical analysis

We performed all analyses in Stata [21]. We used logistic regression, with PD as our outcome variable, to determine the association between PD and each dichotomous herpesvirus exposure variable (HSV, HZ, and anti-herpetic medication). We reported the odds ratio (OR) and 95% confidence interval (CI) as an estimate of relative risk of PD. We adjusted a priori for age, sex, and race/ethnicity. In a fuller model, we also adjusted for smoking probability and number of unique diagnosis codes. Compared to comorbidity indices, the latter more completely adjusts for overall use of medical care [19], a potentially strong confounder of associations between any two medical conditions [19, 22]. However, we examined the effect of additionally adjusting for ten specific medical conditions considered elsewhere [6] (including Alzheimer disease/dementia) and immunosuppressant use [20] (among those with Part D data). There were no missing data except race/ethnicity in 0.1% of both cases and controls, which we categorized as “unknown” race/ethnicity in all models.

We performed three sensitivity analyses. First, we examined the effect of requiring more stringent PD diagnostic criteria (≥1 ICD-9-CM code for PD from a neurologist or ≥3 PD codes) [15, 16, 23]. Second, we applied 1-4 years of exposure lagging for alpha herpesvirus diagnoses, and one year for anti-herpetic medication. Exposure lagging ensures that only exposures prior to the lag period are counted, as a means to minimize any potential influence of prodromal period symptoms on ORs. Finally, among beneficiaries with ≥3 outpatient visits, we re-defined herpesvirus exposure as having ≥3 outpatient visits with a herpes diagnosis code, as a potential indicator of chronicity of infection and/or diagnostic accuracy [24].

We also performed three secondary analyses in which we estimated the PD-herpesvirus associations in other datasets or considered other herpesviruses. First, in our primary dataset we considered the association between PD and cytomegalovirus (CMV, ICD-9-CM 078.5 or 484.1) and Epstein-Barr virus (EBV, ICD-9-CM 075, i.e., infectious mononucleosis). These are members of other herpesvirus subfamilies that do not specifically target neuronal cells and were uncommon in our dataset. Second, among the controls in our primary dataset, i.e., a random sample without PD, we followed the 115,492 who were alive on January 1, 2010, forward for PD or death through December 31, 2014 [25]. For both HSV and HZ (but not CMV and EBV, which were too rare), we conducted a survival analysis using Cox proportional hazards regression while adjusting for the same variables as in our primary analysis, as of baseline (January 1, 2010). We excluded PD that occurred within one year from baseline [6]. We repeated this cohort design for an analysis of Alzheimer disease (AD, ICD-9-CM 331.0), for comparison to the results for PD. Finally, we conducted an analysis in a fully independent population-based case-control study of PD with lifetime medical history data from participant interviews [3]. Cases and controls were identified from a health maintenance organization in Washington State (482 newly diagnosed PD cases, 634 controls frequency matched by age, sex, and race/ethnicity). All cases were reviewed by three neurologists to verify the PD diagnosis. The questionnaire inquired about conditions related to two herpesviruses, VZV (chicken pox and shingles [HZ]), and EBV (mononucleosis). We used logistic regression to assess the association between these conditions and PD, while making similar adjustments as in our claims-based analyses (age, sex, race/ethnicity, and an indicator of use of medical care).

Data availability

The authors are not authorized to make data from Centers for Medicare and Medicaid Services available.

Standard protocol approvals, registrations, and patient consents

The study was approved by the Washington University Institutional Review Board, the Centers for Medicare and Medicaid Services, and the Group Health Cooperative Center for Health Studies and the University of Washington.

Results

PD cases demonstrated typical demographic associations, as well as greater use of medical care compared to controls (shown in Table 1). All PD-herpesvirus associations were modestly, sometimes significantly, above one when only adjusting for age, sex, and race/ethnicity (shown in Table 2). However, after accounting for smoking probability and number of unique diagnosis codes, PD was inversely associated with both HSV (OR=0.79, 95% CI 0.74-0.84) and HZ (OR=0.88, 95% CI 0.85-0.91). We observed the inverse association for HSV regardless of type of infection, and ORs were essentially identical in men and women (not shown). PD was also inversely associated with any anti-herpetic medication (OR=0.87, 95% CI 0.80-0.96) (shown in Table 2). Further adjustment for dementia and other medical conditions, or use of immunosuppressants, did not materially alter the PD ORs from the full models for HSV, HZ, or anti-herpetic medication (not shown).

Table 1.

Characteristics of Incident Parkinson’s Disease Cases and Randomly Sampled Controls, United States, Medicare, 2009

	PD Cases N = 89,790 (%)	Controls N = 118,095 (%)	Unadjusted OR (95% CI)a	p-value
Age, Years, Mean (SD)	78.7 (6.1)	76.0 (6.2)	1.074 (1.072, 1.075)	<0.01
Female	50.2	57.0	0.76 (0.75, 0.77)	<0.01
Race/Ethnicity
White	88.8	86.4	1.00 (Ref)
Black	6.0	7.5	0.78 (0.75, 0.80)	<0.01
Hispanic	2.1	1.9	1.11 (1.04, 1.18)	<0.01
Asian	1.7	2.2	0.78 (0.74, 0.83)	<0.01
Pacific Islander/Other	1.0	1.5	0.63 (0.58, 0.68)	<0.01
Native American	0.3	0.4	0.75 (0.65, 0.87)	<0.01
Unknown	0.09	0.09	0.95 (0.71, 1.27)	0.74
Smoking Index ≥ Medianb	42.9	55.4	0.60 (0.59, 0.61)	<0.01
Unique Diagnosis Codes, Mean (SD)	100.9 (54.4)	65.6 (47.1)	1.0140 (1.0138, 1.0141)	<0.01

^aAbbreviations: OR - Odds Ratio; CI - Confidence Interval

^bSmoking probability divided by the person’s total number of unique diagnosis codes (or one for 292 cases and 6,227 controls without any diagnosis codes).

Table 2.

Risk of Parkinson’s Disease in Relation to Herpesviruses and Treatment, United States, Medicare, 2009

	Cases N (%)	Controls N (%)	Basic model OR (95% CI)a,b	p-value	Full model OR (95% CI)a,b	p-value
All beneficiaries	89,790	118,095
HSVa 1 & 2	2,273 (2.5)	2,545 (2.2)	1.18 (1.11, 1.25)	<0.01	0.79 (0.74, 0.84)	<0.01
HZa	6,936 (7.7)	7,572 (6.4)	1.14 (1.10, 1.18)	<0.01	0.88 (0.85, 0.91)	<0.01
Restricted to Part D Coverage	Cases N (%)	Controls N (%)	Basic model OR (95% CI)a,b	p-value	Full model OR (95% CI)a,b	p-value
Beneficiaries	48,295	52,324
HSVa 1 & 2	1,259 (2.6)	1,275 (2.4)	1.09 (1.00, 1.18)	0.04	0.79 (0.73, 0.86)	<0.01
HZa	3,763 (7.8)	3,738 (7.1)	1.06 (1.01, 1.11)	0.02	0.87 (0.83, 0.92)	<0.01
Anti-herpetic Medicationc	955 (2.0)	1,025 (2.0)	1.06 (0.97, 1.16)	0.22	0.87 (0.80, 0.96)	0.01

^aAbbreviations: OR - Odds Ratio; CI - Confidence Interval; HSV - Herpes Simplex Virus; HZ - Herpes Zoster.

^bBasic Model adjusted for age (as two linear splines with a knot at age 85), sex, and race/ethnicity (7 categories). Full model adjusted for age, sex, race/ethnicity, smoking probability, and overall use of medical care (number of unique diagnosis codes), as continuous variables.

^c≥1 Part D fill for any of the following anti-herpetic medications: acyclovir, valacyclovir, valganciclovir, famciclovir, ganciclovir, or penciclovir.

The inverse association between PD and HSV was unchanged in all three sensitivity analyses. This was also true for PD and HZ, except that the association tended to weaken with more lagging. After the maximum lag of four years, the association between PD and HZ remained inverse, but was very modest (OR=0.94, 95% CI 0.88-1.004). The median time between first HZ and first PD diagnosis codes was 2.62 years (interquartile range 1.28-4.12 years), precluding >4 years of lagging for either HZ or HSV (median 2.74 years, interquartile range 1.35-4.16 years). None of the sensitivity analyses materially weakened the inverse association between PD and anti-herpetic medication.

In the secondary analysis in which we followed random controls forward to incident PD (N=1,949 occurring >1 year after baseline), there was no association with HSV (hazard ratio=1.11, 95% CI 0.85-1.44) but a marginal inverse association for HZ (hazard ratio=0.86, 95% CI 0.72-1.03) (Supplemental Table 1). When we instead followed controls forward to incident AD, we observed marginal inverse associations for both HSV and HZ (Supplemental Table 1). As in our primary analysis for PD, there was no evidence of inverse associations between herpesviruses and AD, until adjusting for use of medical care adequately. Failure to adjust for use of medical care resulted in point estimates above null for both PD and AD (Supplemental Table 1).

In the secondary analysis in the fully independent dataset, 60 cases (12.5%) and 107 controls (16.9%) reported having had shingles prior to their PD diagnosis or control reference date (unadjusted OR=0.70, 95% CI 0.50-0.98; adjusted OR=0.77, 95% CI 0.54-1.10). This inverse association was not materially altered with one, five, or ten years of lagging, with ORs ranging from 0.76 to 0.78. With twenty years of lagging the inverse association between shingles and PD also remained (OR=0.66, 95% CI: 0.36, 1.19). In contrast, chicken pox was not associated with PD (OR=1.10, 95% CI: 0.68, 1.80), and no shingles-PD ORs materially changed if we restricted to participants with a history of chicken pox.

A diagnosis code for either CMV or EBV was uncommon in our primary dataset (Supplemental Table 2). Nonetheless, there was a significant inverse association between CMV and PD (OR=0.65, 95% CI 0.44-0.97). There was no association between EBV and PD (OR=0.92, 95% CI 0.66-1.27). Similarly, in the secondary analyses, which used a fully independent population-based case control study, lifetime occurrence of mononucleosis prior to PD diagnosis or control reference date was not associated with PD (adjusted OR=1.01, 95% CI 0.60-1.72).

Discussion/Conclusion

In this large population-based study of older adults, we observed an inverse association between PD and the alpha herpesviruses HSV and HZ, as well as anti-herpetic medications. Our findings were similar in sensitivity analyses that used more stringent definitions of both alpha herpesvirus and PD, to ensure greater certainty of both diagnoses. The inverse associations also fully (HSV) or partially (HZ) remained when we applied exposure lagging so that we did not include herpesvirus infections that occurred in the late prodromal period of PD. In addition, we confirmed the inverse association for HZ in two secondary analyses, including in a longitudinal analysis in our primary dataset. Notably, we could only observe the inverse HZ-PD associations when we adequately adjusted for smoking and use of medical care. As expected, given that confounding by use of medical care biases PD risk estimates upward [19, 22], the HZ-PD association was positive without this essential adjustment. Without this adjustment, our OR was quite similar to the most comparable risk estimate obtained within another population [6, 7]. Among individuals age ≥65 the incidence rate ratio was 1.12 (95% CI 1.05-1.20) when excluding PD diagnosed within one year of HZ diagnosis and adjusting for selected medical conditions but not overall use of medical care [6]. This study, like our primary analysis, was based on claims data.

Confounding by medical care may occur when patients visiting a physician for one condition, are more likely to be diagnosed with another condition, simply because they are accessing the healthcare system. Beneficiaries who receive a diagnosis code for one condition are also probably more likely to receive a diagnosis code for another condition. This phenomenon likely is not specific to PD and herpesviruses, or even PD [22]. In fact, we also observed strong confounding by use of medical care on the association between AD and herpesviruses. Confounding by use of medical care is a plausible explanation for some differences in results across the broader literature for herpesviruses and neurodegenerative diseases, perhaps especially in studies based on claims data. In the secondary analysis in which we used an independent population-based case-control study with lifetime medical histories rather than claims-based data, HZ infection was inversely associated with PD, regardless of adjustment for the number of other medical conditions. Moreover, this inverse association between PD and HZ remained with up to 20 years of lagging, in close parallel to another population-based case-control study of incident PD with lifetime medical histories [4]. Taken together with our primary results in our larger sample, these complementary results strongly suggest that there is an inverse association between PD and at least some herpesviruses, and that this association is not due to restriction to more recent medical histories in our study, the prodromal period, reverse causation, or confounding.

There are several potential mechanisms that could underlie the inverse association we observed between herpesviruses and PD. It is possible that alpha herpesviruses are neuroprotective against PD, given that they remain dormant in cells of the peripheral nervous system, which may have direct access to the central nervous system making it potentially relevant to PD. However, any truly protective mechanism in this regard remains to be identified, and furthermore, we also observed a significant inverse association between PD and CMV. Unlike the alpha herpesviruses, CMV, a beta herpesvirus, does not appear to remain dormant in neuronal cells of the peripheral or central nervous system. Although CMV can also affect the central nervous system, through other mechanisms, this beta herpesvirus appears to cause severe symptoms in immunocompromised hosts. Thus, it appears more likely that these various herpesviruses are serving as a marker for immune suppression, and that the immune system might play a role in the etiology or progression of PD. Individuals susceptible to herpesvirus infections might have defects in genes regulating pro-inflammatory and/or anti-inflammatory cytokines or acquired deficits in adaptive immunity (T-cell-mediated immune responses) required to respond adequately to the infection [26, 27]. Furthermore, HZ is more common in persons 60 and older due to waning T-cell-mediated immune responses [28]. However, individuals with PD may be less susceptible to active infections from herpesviruses due to a heightened inflammatory response. In fact, there is growing evidence that neuroinflammation may be critical for PD progression [26, 29, 30]. Our current findings on herpesviruses, together with other inverse associations between PD and both immunosuppressant use [20] and medical conditions [23, 31] treated with immunosuppressants, support the continued investigation of immunosuppression as a possible means for reducing risk of PD or slowing its progression. Alternatively, we cannot rule out the possibility that the antiviral medications used to treat herpesviruses reduce PD risk. These medications inhibit viral DNA synthesis required for replication of herpesviruses, but it is possible that use of antivirals to treat herpesvirus may reduce aberrant protein aggregation in PD [5]. Lastly, there could be non-causal explanations for the inverse association observed between alpha herpesviruses and PD.

We acknowledge several limitations to our study. Since Medicare data are only population-based when considering those 65 and older, and, since our data were limited to recent years prior to PD diagnosis, we could not fully investigate the time course of herpesvirus infection and lifetime use of anti-herpetic medications as they relate to the initiation of PD. Nevertheless, our secondary analysis in the independent dataset in which we were able to assess lifetime history and apply a 20-year lag provides compelling confirmatory evidence supporting our findings for HZ. This study did not have information on HSV, but a similar PD study that observed results for HZ found no evidence of an association for HSV [4]. Another limitation is that it is possible that some individuals with a code for PD did not have PD, but were only suspected as having PD. This is likely, given the relatively large number of PD cases identified in our study. However, our results remained similar when we restricted to the most certain PD cases. We also do not have access to laboratory data, and therefore we cannot fully investigate the type of HSV and seropositivity/seroprevalence (e.g., presence of antibodies, antibody titers, detection of HSV in the blood or cerebrospinal fluid or mucosal surfaces) of herpesviruses in order to determine if herpesvirus infections were primary (new) or chronic infections. With that said, studies with measurements in biospecimens inherently were relatively small and collected the biospecimens after the development of PD, limiting inference [9–11]. Our study only included herpesvirus claims prior to the diagnosis of PD, and we applied progressively greater lagging, which strengthens the interpretation of the PD-herpesvirus associations we observed. Additionally, true incident herpesvirus is difficult to ascertain given that individuals may be asymptomatic or might not seek medical care during the initial infection [32]. Similarly, we did not have lifetime medical histories, and likely missed herpesvirus infections for that reason, as well. However, this source of exposure measurement error should be similar in PD cases and controls, which would bias associations toward null, and therefore cannot account for the inverse associations we observed. Moreover, an inverse association for HZ was also found in the two population-based studies with lifetime medical histories [3, 4]. In fact, the associations we observe may actually indicate that those with relative immunosuppression have a lower risk of PD. Despite these limitations, this study provides additional evidence that herpesviruses may impact PD risk, potentially through modulation of the immune system.