A systematic review and meta-analysis of risk factors for postherpetic neuralgia.

Patients with herpes zoster can develop persistent pain after rash healing, a complication known as postherpetic neuralgia. By preventing zoster through vaccination, the risk of this common complication is reduced. We searched MEDLINE and Embase for studies assessing risk factors for postherpetic neuralgia, with a view to informing vaccination policy. Nineteen prospective studies were identified. Meta-analysis showed significant increases in the risk of postherpetic neuralgia with clinical features of acute zoster including prodromal pain (summary rate ratio 2.29, 95% confidence interval: 1.42-3.69), severe acute pain (2.23, 1.71-2.92), severe rash (2.63, 1.89-3.66), and ophthalmic involvement (2.51, 1.29-4.86). Older age was significantly associated with postherpetic neuralgia; for individual studies, relative risk estimates per 10-year increase ranged from 1.22 to 3.11. Evidence for differences by gender was conflicting, with considerable between-study heterogeneity. A proportion of studies reported an increased risk of postherpetic neuralgia with severe immunosuppression (studies, n = 3/5) and diabetes mellitus (n = 1/4). Systemic lupus erythematosus, recent trauma, and personality disorder symptoms were associated with postherpetic neuralgia in single studies. No evidence of higher postherpetic neuralgia risk was found with depression (n = 4) or cancer (n = 5). Our review confirms a number of clinical features of acute zoster are risk factors for postherpetic neuralgia. It has also identified a range of possible vaccine-targetable risk factors for postherpetic neuralgia; yet aside from age-associated risks, evidence regarding risk factors to inform zoster vaccination policy is currently limited.

1. Introduction

Postherpetic neuralgia (PHN) is pain after an acute episode of herpes zoster (commonly known as shingles) continuing beyond rash healing.[16] The pain has been described as a constant burning or stabbing sensation, and some individuals experience allodynia (pain triggered from light contact with nonpainful stimuli).[47] Symptoms can persist for months or even years, and the condition can profoundly affect a patient's quality of life.[12,24] PHN is the most common complication of zoster; an estimated 12.5% of patients with zoster aged ≥50 years have PHN 3 months after zoster onset, and the proportion affected increases sharply with age.

Postherpetic neuralgia is often refractory to treatment.[10,20,40] Despite decades of research, evidence for the efficacy of administering antivirals at first appearance of the rash in reducing PHN incidence is unconvincing.[7] However, an effective live-attenuated vaccine is now available providing protection against zoster and might be used to protect those most likely to develop PHN and other complications of zoster.[30,36] Apart from age, other often reported risk factors for PHN relate largely to characteristics of the acute zoster episode, particularly, the severity of acute pain and rash at initial zoster presentation; however, the evidence has not been systematically reviewed.[14,31,32,42,48] Furthermore, as these are not vaccine-targetable, there is interest in identifying risk factors for PHN, which can be identified before the zoster episode, to inform zoster vaccination policy.

This article aims to systematically collate and summarise the epidemiological literature on risk factors for PHN including clinical features of acute zoster and those which are “vaccine-targetable.”

2. Methods

2.1. Study selection

2.1.1. Search terms

We searched all published journal articles in MEDLINE and Embase between 1950 and February 3, 2014. We searched for articles containing PHN terms and risk factor analysis terms (Box 1 for full details). The search strategy used both subject heading and text word searches. Initial search terms were updated after searching the reference lists of relevant articles. To capture relevant grey literature, the New York Academy of Medicine Grey Literature Report (www.greylit.org), the Electronic Theses Online Service through the British Library (http://ethos.bl.uk), and the ISI Conference Proceedings Citation Index (http://isknowledge.com) were searched for the terms: “PHN” or “postherpetic neuralgia” or “postherpetic neuralgia,” within the keywords or title (Box 1).

[ { (“Postherpetic neuralgia”[exploded MeSH] OR PHN[Title or abstract] OR “postherpetic neuralgia”[Title or abstract] OR “post herpetic neuralgia”[Title or abstract] OR post-herpetic neuralgia[Title or abstract] OR “postherpetic pain”[Title or abstract] OR “post herpetic pain”[Title or abstract] OR post-herpetic pain [Title or abstract] OR ((“Neuralgia”[exploded MeSH] OR “Pain”[exploded MeSH] OR neuralgia [Title or abstract] OR pain [Title or abstract]) AND (“Herpes zoster”[exploded MeSH] OR zoster[Title or abstract] OR shingles[Title or abstract] OR zona[Title or abstract] OR VZV[Title or abstract]))

AND (“Risk factors”[exploded MeSH] OR “Epidemiologic studies”[exploded MeSH] OR “Odds ratio”[exploded MeSH] OR “Multivariate analysis”[exploded MeSH] OR “Logistic Models”[exploded MeSH] OR “Prevalence”[exploded MeSH] OR “Incidence”[exploded MeSH] OR “odds ratio”[Title or abstract] OR “risk ratio”[Title or abstract] OR “relative risk”[Title or abstract] OR “risk factor”[Title or abstract] OR risk[Title or abstract] OR predict*[Title or abstract] OR correlat*[Title or abstract] OR etiol*[Title or abstract] OR aetiol*[Title or abstract] OR prevalence[Title or abstract] OR incidence[Title or abstract] OR rate*[Title or abstract]) }

OR “Postherpetic neuralgia/etiology”[exploded MeSH]

OR ((“Neuralgia/etiology”[MeSH] OR “Pain/etiology”[MeSH]) AND (herpes zoster[exploded MeSH] OR “zoster”[Title or abstract] OR “shingles”[Title or abstract] OR “zona”[Title or abstract] OR “VZV”[Title or abstract])) ]

AND “Humans”[MeSH]; limited to articles in language: ENGLISH

[ { (“Postherpetic neuralgia”[exploded subject heading] OR “PHN”[Title or abstract] OR “postherpetic neuralgia”[Title or abstract] OR “post herpetic neuralgia”[Title or abstract] OR “post-herpetic neuralgia”[Title or abstract] OR “postherpetic pain”[Title or abstract] OR “post herpetic pain”[Title or abstract] OR post-herpetic pain [Title or abstract] OR ((“Neuralgia”[exploded subject heading] OR “Pain”[exploded subject heading] OR “neuralgia” [Title or abstract] OR “pain” [Title or abstract]) AND (herpes zoster[exploded subject heading] OR “zoster”[Title or abstract] OR “shingles”[Title or abstract] OR “zona”[Title or abstract] OR “VZV”[Title or abstract]))

AND (“Risk factor”[exploded subject heading] OR “Epidemiology”[exploded subject heading] OR “Odds ratio”[exploded subject heading] OR “Multivariate analysis”[exploded subject heading] OR “Statistical model”[exploded subject heading] OR “Prevalence”[exploded subject heading] OR “Incidence”[exploded subject heading] OR “odds ratio” OR “risk ratio” OR “relative risk” OR “risk factor” OR “risk” [Title or abstract] OR “risk factor”[Title or abstract] OR “predict*”[Title or abstract] OR “correlat*”[Title or abstract] OR “etiol*”[Title or abstract] OR “aetiol*”[Title or abstract] OR “prevalence”[Title or abstract] OR “incidence”[Title or abstract] OR “rate*”[Title or abstract]) }

OR “Postherpetic neuralgia/etiology”[exploded subject heading]

OR ((“Neuralgia/etiology”[subject heading] OR “Pain/etiology”[subject heading]) OR (herpes zoster[exploded subject heading] OR “zoster”[Title or abstract] OR “shingles”[Title or abstract] OR “zona”[Title or abstract] OR “VZV”[Title or abstract])) ]

AND “Humans”[subject heading]; limited to language: ENGLISH

New York Academy of Medicine Grey Literature Report: PHN OR postherpetic neuralgia OR title:(postherpetic AND neuralgia) OR title:PHN

ISI Conference Proceedings Citation Index: [ {TS=(PHN or “postherpetic neuralgia” or “post herpetic neuralgia”) AND TS=(risk or epidem* or “odds ratio” or rate)} OR {TI=(PHN or “postherpetic neuralgia” or “post herpetic neuralgia”)}] AND TI=(risk or epidem* or “odds ratio” or rate) AND LANGUAGE: (English).

Note: In both databases the subject heading terms are arranged in a hierarchy with more specific linked subheadings arranged beneath wider terms. Exploding a subject heading indicates that the search includes all results below that heading.

2.2. Inclusion and exclusion criteria

Criteria were developed in an iterative process after preliminary searches. We included studies based on original data from analytical epidemiological studies, among adults (18 years+) with zoster. Postherpetic neuralgia had to be a study outcome and an age-adjusted effect estimate was required. We included risk factors, which were either (1) clinical features of the acute zoster episode or (2) vaccine-targetable, defined as risk factors identifiable before the onset of the zoster rash. Studies assessing only age as a risk factor were required to treat age as a continuous exposure (ie, linear on a log scale) such that its effects on PHN risk could be reported per 10-year increase. Studies assessing genes as risk factors for PHN were not required to have an age-adjusted effect measure, because allele frequencies are not typically associated with age.

We omitted studies assessing antiviral therapy as a determinant of PHN as they have been recently summarised in a Cochrane Systematic Review[7]; we also omitted studies assessing other PHN treatments (such as acupuncture and corticosteroids). We excluded studies examining risk factors for PHN within a general population sample (where patients with PHN were compared with non-zoster controls) because the risk of PHN in the general population comprises 2 parts; first, the risk of zoster and second, the risk of developing PHN among those with zoster. In these studies, it is impossible to disentangle whether any identified risk factors are simply predictive of zoster itself, or whether they are specifically risk factors for getting PHN. We also excluded studies restricted to specific clinical subgroups of patients with zoster, such as individuals with HIV, because their risk factors for PHN may differ. We restricted to English articles only; however, we did not place any restriction on study location or publication status.

2.3. Selecting studies

The titles and abstracts of all identified articles were assessed. If a study was deemed to potentially fulfil the inclusion criteria, full-text versions were retrieved and assessed. Reference lists of all retrieved articles were searched. To assess how reliably the study eligibility criteria were applied, a second author (R.F.) applied the inclusion criteria to a random 10% sample of all articles, and agreement between the primary allocation and the sample allocation was tested using Cohen's kappa statistic.[29] A kappa score of 1 denotes full agreement, and kappa values greater than 0.75 indicate excellent agreement.[44]

2.4. Data extraction

Extraction tables were piloted by S. L. Thomas and H. J. Forbes and then applied to remaining studies. Data (listed in Appendix, available online as Supplemental Digital Content at http://links.lww.com/PAIN/A132) were extracted by H. J. Forbes for each study. Authors were contacted for missing information (see appendix for template e-mail to corresponding authors, available online as Supplemental Digital Content at http://links.lww.com/PAIN/A132). When individual studies used multiple definitions of PHN, results classifying PHN as pain at 3 months after zoster onset (or that closest to 3 months) were extracted for the main analysis, as this is the most widely used definition of PHN.[12,17,30,36,45] Results from other PHN definitions were extracted for the Appendix (available online as Supplemental Digital Content at http://links.lww.com/PAIN/A132).

2.5. Assessing risk of bias

The risk of bias assessment was based on the Cochrane Collaboration approach, in which each study is assessed separately for prespecified bias domains (see Appendix for further details available online as Supplemental Digital Content at http://links.lww.com/PAIN/A132).

We also considered the validity of each study based on the sampling of patients with zoster, numbers declining to participate, and their characteristics, particularly the percentage developing PHN.

2.6. Data analysis

When at least 2 studies were deemed to be capturing the same risk factor within similar populations, we assessed between-study heterogeneity using the Cochrane Q statistic and the I2 statistic, with I2 > 50% used as a threshold indicating moderate heterogeneity. In the absence of heterogeneity, we planned to combine the estimates and produce a summary relative risk using fixed effects meta-analysis. However, for some risk factors, there was significant between-study statistical heterogeneity; therefore, we performed posthoc analysis to help ascertain the possible reasons for heterogeneity. This included rerunning the meta-analysis removing studies at high risk of bias and comparing I2 values between clinical and methodological subgroups to evaluate potential sources of heterogeneity.[22] For this latter analysis, summary estimates from subgroups were formally compared using meta-regression; we compared subgroups according to (1) mean age of the study population (≥60 years vs <60 years), (2) definition of PHN (pain at 4 months vs pain at 3 months), (3) ascertainment of PHN (self-reported vs ascertained from medical records), (4) whether immunosuppressive patients were included or excluded, and (5) sources of study population (primary care vs other).

We also created a funnel plot to determine the risk of publication bias; gender was the only risk factor assessed in sufficient studies to be suitable for assessment (age effects were reported in different units making it unsuitable). The odds ratios (OR), representing the effect estimate of gender on PHN, were plotted against the standard error of the log odds,[41] representing the precision of the estimate, and symmetry was assessed visually (as there were too few studies to perform a formal test).[43] Statistical analyses were performed in STATA (version 13.1).

3. Results

The initial search identified 3614 articles. After removing duplicates, 2559 titles and abstracts were screened. Of these, 116 full-text articles were retrieved, 19 of which were included in the review (Fig. 1). Excluded studies are listed in the Appendix (Table A1), available online as Supplemental Digital Content at http://links.lww.com/PAIN/A132.

Figure 1

Flow diagram describing study selection.

Agreement between reviewers over the application of the inclusion criteria was very good (kappa score, 0.88). From the 10% sample of articles double screened, 1 study was not agreed on; the second reviewer initially selected this study[11] for inclusion; however, both reviewers subsequently agreed this extra article replicated a study already selected.[13]

3.1. Study characteristics and findings

Study characteristics are described in Table 1. There were 18 cohort studies and 1 case-base study (a modified case–control study, where the risk ratio is estimated by sampling controls from those at risk at the start of follow-up).[39] Study sizes ranged from 55 to 34,280, and 17/19 studies had less than 1000 participants at baseline. Zoster diagnoses were predominantly based on clinical opinion. Definitions of PHN were presence of pain 3 months after rash onset in 10 studies, although other definitions from 1 to 6 months were used. The percentage of patients with zoster developing PHN ranged from 2.6% to 67.3%. Mean age of study participants (available in 9 studies) ranged from 52.3 to 67.7 years. Studies were all from high-income countries.

Table 1

Studies assessing vaccine-targetable risk factors for postherpetic neuralgia nested within a population of patients with zoster: study characteristics.

Study findings are summarised in Table 2 and Figures 2 and 3. Data were collected on clinical features of the acute episode including pain (15), rash extent and location (14), rash duration (9), sensory dysfunction (3), and other clinical features (11), and also vaccine-targetable risk factors including age and gender (18 studies), severe immunosuppression (5), other physical comorbidities such as autoimmune conditions (4), diabetes (6), cancer (5), recent physical trauma (1), psychological comorbidities (4), and other risk factors (9).

Table 2

Association between PHN and various risk factors (defined as either vaccine-targetable or clinical features of the acute zoster episode): risk factors, adjusted effect measure and 95% confidence interval (CI) by study.

Figure 2

Summary of associations between postherpetic neuralgia and clinical features of acute zoster. 1Composite score ranges from 0-100 numerical pain ratings and McGill Pain Questionnaire Present Pain 2Intensity ratings of average and worst shingles pain. ²Intensity of pain using the Short Italian questionnaire, from 0-10. 3³Temperature differences are between normal and affected skin. 4Percentage of body surface area thermal asymmetry (≥3 vs <3%). †Risk factors too varied to combine in meta-analyses. •Not included in summary RR (either because study has already contributed to meta-analysis, or exposure definition is not in-keeping with other studies). *Studies reporting RR (rather than OR) are not included in meta-analysis. CI, confidence interval; DN4, Neuropathic pain questionnaire with 4 questions; NPSI, Neuropathic pain symptom inventory score; OR, odds ratio; RR, rate ratio; SF-12, short-form 12; VAS, visual analogue scale ranging from 0 (non pain) to 100 (worst pain ever experienced); VZV, varicella zoster virus; ZBPI, Zoster brief pain inventory interference score.

Figure 3

Summary of associations between postherpetic neuralgia and vaccine-targetable risk factors from identified studies. *Only 10/20 studies reported age such that the effect estimate could be converted into 10-year increases. Of the remaining 10 studies; 8 reported an increased risk of PHN with greater age, 1 showed no effect all, and 1 did not report an age-effect. **Studies reporting RRs rather than ORs not included in meta-analysis as RR can underestimate OR when outcome becomes common. ***Effect estimate from study may be erroneous therefore the study is not included in the meta-analysis: Parruti 2010 CIs are too narrow, and Opstelten 2002 confidence also too narrow. 1Using high-dose oral corticosteroids or other immunosuppressive drugs, having invasive cancer or HIV/AIDS. 2Undefined, however included HIV or currently being treated for cancer. 3Connective tissue disease, HIV infection or organ allograft. 4Better health: measured using continuous physical component summary score (higher scorer score reflects worse health). 5Poorer health: measured using continuous variable of total number of medical conditions. †Risk factors too varied to combine in meta-analyses. ‡The large study by Jih et al. (N = 34,280) dominated the pooled relative risk contributing to 99·1% of the model. Other risk factors investigated as predictors of PHN, but not included in the final model, included; surgical intervention, hepatitis-C virus infection, hypertension, neurological disorders, allergy, family history of CHD, angina, chronic obstructive pulmonary disorder, education, alcohol abuse, familial status, years of education and race. APOE, alipoprotien E; CI, confidence interval; OR, odds ratio; RA, rheumatoid arthritis; RR, rate ratio; SLE, systemic lupus erythematosus.

3.2. Clinical features of acute zoster episode as risk factors

3.2.1. Pain

3.2.1.1. Prodrome

Eleven cohort studies and the case-base study collected data on prodromal pain, ie, pain appearing before rash onset. Seven included prodromal pain in the final age-adjusted model and 5 reported effect estimates, with each giving a point estimate above 1. We obtained a pooled effect estimate of 2.29, 95% confidence interval (CI): 1.42 to 3.69 (Pheterogeneity = 0.716; I2 = 0.0%) in fixed effect meta-analysis. A cohort study among 533 immunocompetent patients reported a shorter prodrome (≤3 days) before rash onset was associated with reduced risk of PHN (adjOR: 0.49, 95% CI: 0.24-0.99).

3.2.1.2. Severe acute pain during zoster

Twelve cohort studies investigated severe acute pain as a risk factor for PHN. Although definitions of severe acute pain varied among studies, eg, pain scoring ≥4 using the Neuropathic Pain Questionnaire[5] and pain scoring ≥5 on the Visual Analogue Scale,[9] 8 reported it as a binary variable enabling us to pool estimates; there was good evidence that severe acute pain was associated with increased risk of PHN (rate ratio [RR]: 2.23, 95% CI: 1.71-2.92, Pheterogeneity = 0.649; I2 = 0.0%).

3.2.1.3. Allodynia

Allodynia was investigated in 3 cohort studies. One study reported a greater than 4-fold increased risk of PHN with brush (adjOR: 5.89, 95% CI: 1.50-23.1) and stretch-evoked allodynia (adjOR: 4.13, 95% CI: 0.98-17.50)[19]; however, small numbers (N = 93) led to wide CIs. A study among hospital patients treated in a pain clinic found no effect of allodynia (definition unclear; adjOR: 0.82, 95% CI: 0.24-2.81), whereas a final cohort study similarly reported no evidence of effect.[5,26] A summary estimate was not calculated because of the varying definitions of allodynia.

3.2.1.4. Pain interferes with daily functioning

Pain interfering with daily functioning at zoster onset was assessed in 3 cohort studies. The first, among 1358 individuals, reported a 1-unit increase in zoster brief pain inventory interference score was associated with 18% increase in PHN risk (adjOR: 1.18, 95% CI: 1.05-1.31).[13] Two other cohort studies reported binary (yes or no) data on pain interference. One found over 2-fold increased risk,[27] whereas the other found no evidence of association; the summary estimate of these 2 studies suggested there was strong evidence that pain interfering with daily functioning was associated with PHN (summary RR: 2.10, 95% CI: 1.27-3.48).

3.2.2. Rash severity and location

3.2.2.1. Severe rash

Rash severity data were collected in 8 studies. Five included it in their final age-adjusted model (although one did not report an OR); when combined in meta-analysis, severe rash was strongly associated with PHN risk (summary RR: 2.63, 95% CI: 1.89-3.66, Pheterogeneity = 0.892; I2 = 0.0%).

3.2.2.2. Ophthalmic involvement

A total of 13 studies recorded information on zoster location. Only 3 studies reported an effect estimate for ophthalmic involvement. Each of these 3 studies reported a point estimate above 1, yet the CIs crossed 1. When combining in a meta-analysis, there was evidence that ophthalmic zoster was associated with over twice the risk of PHN, compared with nonophthalmic zoster (summary RR: 2.51, 95% CI: 1.29-4.86, Pheterogeneity = 0.782; I2 = 0.0%).

3.2.3. Rash duration at presentation

Longer rash duration at presentation of zoster showed some evidence of being associated with reduced risk of PHN. A study on 598 immunocompetent patients showed for everyday the rash was present since presentation in primary care; there was over 20% reduced risk of PHN (adjOR: 0.78, 95% CI: 0.64-0.97).[35] Three other cohort studies estimated the risk of PHN for everyday from onset to diagnosis; point estimates were all below 1 (yet CIs were wide).[9,25,27] The summary estimate from meta-analysis showed a small reduction in PHN risk with everyday since rash onset (0.93, 95% CI: 0.86-0.99).

3.2.4. Other

One study assessed pinprick hypaesthesia (or numbness) as a risk factor for PHN: it was associated with a 7-fold increased risk of PHN (adjOR: 7.72, 95% CI: 2.00-29.90).

3.3. Vaccine-targetable risk factors

3.3.1. Age

Eighteen studies assessing the effects of age showed an increased risk of PHN with greater age. When possible, we summarised the effect of a 10-year increase in age on PHN risk (n = 9). The point estimates ranged from 1.22 to 3.11 per 10 years; the meta-analysis showed strong evidence of between-study heterogeneity (Pheterogeneity = 0.029; I2 = 55.1%). A small study (N = 249) showing an increased risk of PHN with a 10-year increase in age (adjRR: 1.22, 95% CI: 1.00-1.48) was excluded from the meta-analysis as the effect was reported as a risk ratio. In posthoc analysis, there was some weak evidence that the effect of age was associated with age of the study population (P value from meta-regression = 0.08; specifically the effect of age on PHN risk seemed higher in studies where the mean age was ≥60 years) (Appendix Table A2, available online as Supplemental Digital Content at http://links.lww.com/PAIN/A132). There was no evidence that the effect of age on PHN risk varied by definition of PHN (P = 0.52), ascertainment of PHN (P = 0.14), immunosuppression status (P = 0.23), or sources of study population (P = 0.18).

3.3.2. Gender

Of 9 studies reporting the age-adjusted association between gender and PHN, some suggested an increased risk of PHN among females,[9,25,38] others a decreased risk,[2,5] whereas others found no evidence of an association.[2,6,8,23,33,37] These conflicting results were supported by strong evidence of between-study heterogeneity (Pheterogeneity < 0.001; I2 = 73.9%). In posthoc analysis, the effect of female gender seemed protective in studies in which the mean age was ≥60 years, compared with among studies with mean age <60 years, for which female gender increased the risk of PHN; heterogeneity was reduced within these subgroups (<1% in both) (Appendix Table A2, available online as Supplemental Digital Content at http://links.lww.com/PAIN/A132). There was no evidence that the effect of gender on PHN risk varied by definition of PHN (P = 0.45), ascertainment of PHN (P = 0.83), immunosuppression status (P = 0.25), or sources of study population (P = 0.97). These analyses were limited by 4/7 studies in meta-analysis of gender having at least 1 bias domain assigned high-risk.

3.3.3. Severe immunosuppression

A cohort study among patients with zoster≥18 years found immunosuppression (including HIV, currently treated for cancer, or exposed to high-dose corticosteroids) was more common in patients with PHN (15%, n = 3/20) than without (7.3%, n = 6/82); but the sample size was too small to be conclusive.[27] Another cohort study among patients ≥50 years of age reintroduced 12 patients with immunosuppression excluded from the main analysis (defined as using high-dose oral corticosteroids/other immunosuppressive drugs, having invasive cancer or HIV/AIDS); these patients had an increased risk of PHN after adjustment for confounders (adjRR: 1.98, 95% CI: 1.14-3.45).[13] Finally, the case-base study in the United States found connective tissue disease, HIV, or organ allograft was associated with 10-fold increased risk of PHN, although the CI was wide (adjOR: 9.5, 95% CI: 2.0-45.0).[8] Two studies specifically assessed HIV: one excluded HIV from the final multivariable analyses,[38] whereas another found over 50% decreased risk of PHN among patients with HIV (antiretroviral treatment status not reported) (adjRR: 0.48, 95% CI: 0.26-0.86).[23] The latter study also reported strong evidence of an increased risk of PHN with lymphoma/leukaemia (adjRR: 1.74, 95% CI: 1.32-2.28).

3.3.4. Other physical comorbidities

3.3.4.1. Overall physical health

One study measured overall health status at zoster presentation using the physical component summary score and found a decreased risk of PHN with better physical health.[5] The second study summed total number of reported medical conditions and found no evidence of association with PHN.[27]

3.3.4.2. Autoimmune conditions

A large cohort study among 34,280 patients with zoster identified in Taiwanese electronic health insurance records identified 284 patients with systemic lupus erythematosus (0.83%), who were more than twice as likely to develop PHN (adjRR: 2.27, 95% CI: 1.75-2.94).[23] Another smaller study (N = 837) using electronic medical records from the Netherlands collected data on lupus and rheumatoid arthritis; however, they were not included in the final model (numbers not reported).[33]

3.3.4.3. Diabetes

Three cohort studies reported point estimates for the association between diabetes and PHN ≥1 in multivariable analyses; however, there was insufficient evidence to confirm an association.[8,26,33] A larger cohort study among 34,280 patients with zoster did find evidence of an increased risk (adjRR: 1.35, 95% CI: 1.25-1.47).[23] There was no evidence of between-study heterogeneity for studies reporting age-adjusted diabetes effects (Pheterogeneity = 0.564; I2 = 0.0%); the pooled effect estimate was 1.36 (95% CI: 1.25-1.47) in the fixed effect meta-analysis; however, the large study (N = 34,280) dominated the pooled relative risk (contributing 99.1% to the model).

3.3.4.4. Cancer

Five studies investigated cancer and its relationship with PHN; 3 excluded it from the final model.[26,28,38] Breast and liver cancer were investigated in a single study, but were not associated with PHN in the final adjusted model.[23] The case-base study found 13.5% of PHN cases and 4.7% of non-PHN controls had a cancer diagnosis 180 days before zoster[8]; after adjustment, cancer was associated with a reduced risk of PHN (adjOR: 0.1, 95% CI: 0.02-0.9); however, the CIs were wide. A meta-analysis for cancer effect estimates was not conducted as they involved different cancer sites.

3.3.4.5. Recent physical trauma

The only study to investigate this risk factor reported over 2-fold increased risk of PHN associated with experiencing trauma at the zoster site (contusions, burnings, wounds, and multiple traumas) within 6 months before study enrolment.[38]

3.3.4.6. Other

Other physical conditions investigated as predictors of PHN, but not included in the age-adjusted models included surgical intervention,[38] hepatitis-C virus infection,[38] hypertension,[26,38] neurological disorders,[38] allergy,[38] family history of coronary heart disease,[38] angina,[26] and chronic obstructive pulmonary disorder.[33]

3.3.5. Psychological comorbidities

These were assessed as risk factors for PHN in 4 studies. Two cohort studies assessed a range of psychological comorbidities; only personality disorder symptoms (adjOR: 1.09, 95% CI: 1.01-1.18),[27] and lower levels of trust in health care (adjOR: 1.01, 95% CI: 1.00-1.03)[35] showed a small association with PHN in multivariable analyses. Neither depression nor anxiety was included in multivariable analyses.[5,27,35,38]

3.3.6. Other risk factors

A cohort study found alipoprotein E-ε3 was more common and alipoprotein E-ε4 less common among female patients with zoster and PHN, suggesting that this host genetic factor may influence the risk of PHN.[50] One study found evidence that current/former smoking was associated with greater risk of PHN (adjOR: 2.08, 95% CI: 1.22-3.55)[38] whereas another included it in their final model, but did not report the association.[2] One study suggested a low state of varicella zoster virus (VZV)-specific cell-mediated immunity, evidenced from reduced response to VZV skin-test, was associated with greater risk of PHN.[2] Studies investigating education,[13,27,38,46] race,[27] being married,[5,27,38] being in work,[13] consultation rate,[8,33] or alcohol abuse[38] did not select these risk factors in their final model.

Nine of the 19 studies had 2 or more definitions of PHN. Briefly, studies additionally defined PHN as pain at 1[6,8,13,21,33,35,38] (n = 7) and 6 months[9,21,28] (n = 3) after zoster onset; there were no major differences in study findings using these alternative definitions, except older age was a stronger risk factor for pain persisting 6 months, compared with 2 or 3 months, after zoster (Appendix Table A3, available online as Supplemental Digital Content at http://links.lww.com/PAIN/A132), indicating older age may be a risk factor for long-term PHN.

3.4. Assessment of bias

Our assessment of bias found 8/19 studies with at least 1 prespecified domain with a high risk of bias, 8 studies with at least 1 domain of medium risk, and 3 studies with only low or unclear risk of bias. Residual confounding by age was the most common source of potential bias, affecting 7/19 studies requiring age-adjustment (Table 3). Studies using electronic health care records were at greatest risk of reporting bias; specifically ascertainment bias, where outcome ascertainment relies on patients returning to their GP and higher general practice (GP) attendance could have increased the chance of PHN diagnosis.[8,23,26,33] Of the cohort studies, 5 experienced loss to follow-up of greater than 10% (Table 3). See Appendix Table A4 for detailed note on the bias assessment (available online as Supplemental Digital Content at http://links.lww.com/PAIN/A132).

Table 3

Assessment of bias for individual studies ().

The funnel plot gave a relatively symmetric pattern, suggesting that there was little indication of publication bias (Fig. 4). The distribution did not suggest that more extreme findings were being selectively published.

Figure 4

Assessment of publication bias for gender as a risk factor for postherpetic neuralgia. Funnel plot of the log odds ratio plotted against the standard error of the log odds ratio for seven studies reporting the effect of female gender on PHN risk (dotted line represents pseudo 95% confidence limits).

The sampling methods and patient characteristics of some studies suggest their external validity may be limited; characteristics of included patients indicate a nonrepresentative sample in some studies (Coen et al. reported that 20% of the study population was immunosuppressed[8] and in 3 studies over 30% of the cohort developed PHN[28,46,50]); 1 study used convenience sampling,[6] thus not all population members had an equal probability of being selected; and the number or characteristics of eligible patients refusing to participate were unclear in most studies.

4. Discussion

4.1. Summary of evidence

Our systematic review identified 19 prospective studies investigating risk factors for PHN. There was good evidence that clinical features of acute zoster including prodromal pain, severe acute pain, severe rash, and ophthalmic involvement were associated with increased risk of PHN. Rash duration at zoster presentation was less strongly associated with PHN. Regarding vaccine-targetable risk factors, older age was consistently associated with PHN. The evidence for gender as a risk factor was conflicting. Immunosuppression and diabetes were significantly associated with PHN in a few, but not all studies. Systemic lupus erythematosus, recent trauma, and personality disorder symptoms were associated with PHN; however, evidence came from single studies only. No studies found evidence suggesting that depression or cancer was associated with increased risk of PHN. Most studies had small sample sizes reducing their power to detect associations. Our review highlights that we have a good understanding of which clinical features of zoster predict PHN, yet there is a need for better evidence on common and potentially easily vaccine-targetable risk factors for PHN prevention.

4.2. Interpreting the findings

It is believed that several pathophysiological mechanisms may contribute to the development of PHN. Acute zoster infection undoubtedly results in nerve damage to both the peripheral and central nervous system, yet the nature of the damage and specific mechanism resulting in persistent pain are not fully understood.[4] There are 2 (nonmutually exclusive) hypotheses for its development; the first is that persistence of VZV after acute zoster, at higher levels than during latency, causes continued pain; and the second, that after acute zoster infection, there is increased neuronal excitability and alteration of pain perception caused by neural damage.[1,20]

The variety of possible risk factors for PHN identified in the review may reflect these different mechanisms.[3] The finding that greater rash severity and greater acute pain are associated with increased risk of PHN supports the notion that greater neural damage caused by more severe infection contributes to the development of PHN.[15] That longer rash duration was associated with reduced risk of PHN initially seems inconsistent with the finding that more severe zoster rash is associated with PHN. However, late presentation might indicate patients had milder zoster not immediately demanding medical attention. Either way, this finding is unlikely to be due to the duration of the rash itself. Patients with ophthalmic zoster seem at greater risk of PHN, although it is not clear whether concerns about eye complications cause them to react differently, rather than the increased risk being driven by a biological mechanism.[20] Ageing undoubtedly causes a waning of cell-mediated immunity and may cause increased levels of the virus after zoster reactivation, potentially causing PHN. Other risk factors for PHN identified here are also associated with reduced cell-mediated immunity, including severe immunosuppression, systemic lupus erythematosus, and smoking. Trauma at the site of the rash may induce local changes facilitating reactivation of herpes zoster (HZ) and greater nerve damage leading to increased risk of PHN. However, the aetiological mechanism(s) by which these risk factors affect the development of PHN remains largely unknown.

4.3. Limitations of the selected studies

The included studies had some limitations. Many had small sample sizes, and we were unable to combine some results in a meta-analysis. Furthermore, many tested a number of risk factors; the associations observed may occur by chance due to testing multiple exposures. Most studies based zoster diagnosis on clinical opinion rather than serological or virological testing; this may have led to misclassification of patients with zoster; however, clinical diagnosis is typically reliable.[34]

Some studies may have been affected by specific biases. Age is a very strong predictor of PHN and yet 7/18 studies assessing age adjusted for it as a binary or categorical variable with wide age intervals, potentially causing residual confounding by age. Loss to follow-up affected 5/19 studies, and if loss to follow-up is associated with both PHN and the risk factor, bias could have been introduced.[18] Patients with PHN may be more likely to return for follow-up as they require continued care, and patients with particular risk factors may also return to their GP more commonly, making bias due to loss to follow-up likely. Ascertainment bias may have affected studies using routinely collected health care data. Here, spurious associations between PHN and medical conditions requiring regular contact with health care professionals may arise. One such study adjusted for health care utilisation[8] and still found a positive association with PHN and certain immunosuppressive disorders, suggesting the effect cannot be driven solely by ascertainment bias. Finally, not all studies adjusted for clinical features of the acute zoster episode,[2,21,23] and results may be subject to residual confounding.

4.4. Strengths and limitations of the review

This is the first study to systematically review the literature on risk factors for PHN; although clinical features of acute zoster have been acknowledged as risk factors for PHN, this is the first to summarise age-adjusted results and pool them in a meta-analysis. We undertook a comprehensive search of several databases using multiple keywords and indexed subject headings. The reliability of study selection criteria was confirmed by double screening of 10% of the articles.

There are some important limitations to this review. There is no consensus over the exact definition of PHN; in this review, PHN definitions ranged from pain persisting 1 to 6 months after rash onset, with some studies assessing any pain, whereas others required severe pain. A full assessment of risk factors by different PHN classifications was not possible here because of too few studies.

Between-study variability prevented us from pooling the effects of age and gender on PHN; there was some evidence that age of the study population contributed to the observed heterogeneity. However, these analyses were limited by the small number of studies and may have reduced our power to detect associations. Variability may be due to different adjustment for confounders or some studies reporting biased effect estimates, eg, due to PHN measurement error or loss to follow-up. Studies also used different definitions for certain clinical features of acute zoster, such as severe acute pain and severe rash, potentially giving some heterogeneity to the results.

Our search strategy may have missed some studies; however, we used multiple databases (including grey literature) and searched reference lists of selected articles, to minimise this issue. As with any literature review, studies finding no effects may have gone unpublished. Our funnel plot did not demonstrate any evidence of publication bias with respect to assessing gender as a risk factor for PHN. However, publication bias may affect other risk factors differently, and there were not enough studies per risk factor to assess this for other exposures. Finally, non–English-language articles were excluded because of resource limitations; however, the authors believe it is unlikely to have led to the omission of any major articles in the area.

4.5. Implications

Zoster vaccination offers a way of preventing this debilitating complication by preventing zoster itself, but is currently expensive; therefore, targeting the vaccine toward groups at high-risk of PHN may be beneficial. The vaccine is currently licensed in certain countries in the European Union, United States, and Australia.[49] It is targeted at older age groups and contraindicated in patients with severe immunosuppression. As older age is the only indisputable risk factor that vaccination policies can use, this approach seems reasonable. If patients with severe immunosuppression are at increased risk of PHN as suggested by this review, in addition to being at greater risk of zoster itself, there is even more need to identify alternative strategies to prevent zoster in these groups.

This review has highlighted our lack of understanding of vaccine-targetable risk factors for PHN, and the need to perform studies exploring suggested associations. Such studies would need to be generalizable to a wide group, by recruiting patients aged 18 and over and including immunosuppressed patients, to examine the risk of PHN by age and immunosuppression status. Other desirable features would include recruiting a large number of individuals to achieve greater power to help detect small effects, collecting data on all known and possible risk factors for PHN, actively following up patients with zoster to allow persistent pain to be identified for the entire cohort at the same time and reducing loss to follow-up to avoid differential ascertainment of PHN. Finally, at the analysis stage, detailed adjustment for age using either a continuous or finely categorised age variable would reduce residual confounding by age.

5. Conclusions

This study confirms that features of the acute zoster episode, including prodromal pain, severe rash, severe acute pain, and ophthalmic involvement are risk factors for PHN. Our current understanding of vaccine-targetable risk factors for PHN is however limited. There are some suggestions that immunosuppression, systemic lupus erythematosus, diabetes, and recent trauma may be associated with greater risk of PHN. Increasing age is the only established risk factor for PHN that has been quantified with sufficient rigour as to usefully inform vaccine policy. Larger studies with greater power to detect associations, and studies addressing the limitations of previous research, may elucidate some of the unknown risk factors for PHN.

Conflict of interest statement

The authors have no conflicts of interest to declare.

This article presents independent research supported by an NIHR Clinician Scientist Fellowship (to Doctor Langan, grant number: NIHR/CS/010/014). The study was approved by ISAC (application no. 11,028) and the London School of Hygiene and Tropical Medicine [application no. 5930]. K. Bhaskaran is supported by an NIHR postdoctoral fellowship (PDF-2011-04-007). L. Smeeth is supported by a Wellcome Trust senior research fellowship in clinical science. S. L. Thomas was supported by an NIHR Career Development Fellowship (CDF-2010-03-32). All authors performed this research independently of the funding bodies. The findings and conclusions in this report are those of the authors and do not necessarily represent the views of the UK Department of Health.

The funding source had no role in the study design; data collection, analysis, or interpretation of the data or writing of the report. H. J. Forbes has access to all studies identified from the initial search. The corresponding author has full access to all the data in the study and had final responsibility for the decision to submit for publication.

The authors do not have a commercial or other association that might pose a conflict of interest.