Data source

The CPRD provides quality-assured anonymised primary care patients’ clinical, administrative and lifestyle data representative of the UK population and covering approximately 7% of general practices from England.[16, 17] Additionally, ~75% of CPRD general practices based in England can be linked at an individual level to hospitalisation (Hospital Episode Statistics, HES) data,[18] which provides information on all admissions to NHS hospitals, and at the Lower-layer Super Output Area (LSOA) level [19] to deprivation data (Index of Multiple deprivation (IMD) score).[19, 20] The English IMD score is a composite measure of relative deprivation for small geographic areas (LSOA), which cover an average population of 1500 individuals or 650 households.[19] This score is derived from using seven domains of deprivation: education, employment, income, health and disability, crime, housing and living environment with no definitive cut-off points for defining deprivation.[19] The validity of various diagnoses recorded in CPRD was reported as high in a systematic review spanning a 21-year study period.[21]

Study population

This 2-year cohort study from England spanned the period from 01/09/2013 to 31/08/2015, the first two years after the zoster vaccine was introduced. To maintain patient anonymity, CPRD data provide only year of birth for adult patients. This posed a problem in how to identify individuals who were eligible for zoster vaccination, which is determined by their age on a specific date. The common convention of using the mid-year (1^st July) to assign study participants’ day and month of birth would wrongly classify some individuals as eligible for zoster vaccination. Importantly, the resulting unvaccinated group would comprise a mixture of individuals with possibly differing social factors: a) those eligible for vaccination who chose not to be vaccinated and b) those ineligible on the grounds of age, thus potentially resulting in biased effect estimates. To address this, we selected all individuals born in 1943 (or 1934 for catch-up cohort), who would have been eligible for vaccination at some point during the 2-year follow-up period as follows: those born in January-August 1943 would be eligible for the vaccine in 2013/14 or in 2014/15 if born September-December 1943; and determined vaccine uptake for the 2-year study period. The study population therefore comprised individuals born in 1943 (the routine cohort) and in 1934 (the catch-up cohort), who were alive and registered on 01/09/2013 (the start of the national programme) with a CPRD general practice that had agreed to linkage to HES and IMD data and that met CPRD’s quality assurance criteria.[20] Start of follow up was on 01/09/2013 and a minimum of five months of follow-up was required from then (i.e. from September until the end of January, coinciding with the main part of the seasonal influenza vaccination season),[22] to ensure that individuals had sufficient opportunity to receive zoster vaccination. Individuals who had any immunosuppressive conditions or therapies at the start of follow up, that were contraindications to receiving the live zoster vaccine,[4] were excluded from analyses of the socio-demographic determinants of vaccine uptake but included in descriptive analyses of inadvertent zoster vaccinations amongst immunosuppressed individuals. All individuals with zoster vaccine codes prior to the start of national programme and start of the study (01/09/2013) were also excluded.[23, 24] End of follow-up was defined as the earliest of: (a) the end of the study (31/08/2015), (b) individuals’ transfer out date from the practice, (c) individuals’ date of death, or (d) the date data were last collected from the practice.[20]

Defining the outcome

Zoster vaccination was determined in five different data files in CPRD: using product codes in patients’ therapy files, immunisation codes in their immunisation files and Read codes (S1 Table) in their clinical, referral and test files.[20] Additional immunisation and Read codes provided further information on whether the vaccine was advised, refused or administered. When vaccination data appeared in more than one file, we used an algorithm to assign vaccination date for each individual and handle conflicting information; details are provided in the S1 Text and S1 Fig.

Exposure variables

The socio-demographic factors of interest were identified based on our previously developed methodology of using CPRD linked to HES and IMD data.[15] The factors of interest, in addition to age and gender, included ethnicity, immigration status, care home residence, marital status, cohabitation (defined as two individuals living as a couple) and living alone; code lists are provided in S2 Table. The latter three social factors provided complementary information about an individual’s living arrangements. Religion was not examined as our previous work has shown it to be poorly recorded in CPRD data (<3% of older individuals).[15] For binary socio-demographic variables, individuals without relevant codes were considered not to have that characteristic. Ethnicity (five categories: White, South Asian, Black, Others and Mixed) and immigration status (binary) were defined as factors that did not vary with time. Time-varying factors included marital status (six categories: single, widow, married, divorced, separated, partner uncategorised/other), cohabitation, living alone and care home residence (binary variables). All the time-varying factors were determined at the start of follow up, with any changes by the end of the 2-year follow-up period quantified and described. Actual IMD scores are not made available to researchers by CPRD, to avoid identification of patients’ area of residence, but all LSOA-level IMD scores in England are ranked (with scores in 2015 ranging from 0.47 to 92.6) and then divided into quintiles for research use. Deprivation data (IMD quintile at LSOA level: 2015, quintile 1 representing least deprived and quintile 5 the most deprived quintile of deprivation) for both patient- and practice-LSOA-level l were available. Practice- LSOA-level IMD quintiles were used if patient-level data were unavailable.

Other variables

At the initiation of the zoster vaccination programme, general practitioners (GPs) were encouraged to co-administer zoster with seasonal influenza vaccine (SIV).[25] We therefore investigated the concurrent administration of zoster vaccine with SIV. This was achieved by identifying specific product codes, immunisation type codes and Read codes in CPRD (S3 Table) during the SIV campaign season (September-March)[26] of 2013/14 and 2014/15. Individuals who received SIV or/and zoster vaccine were quantified.

We also identified, throughout the study period, individuals who had immunosuppressive conditions or treatments that were contraindications to receipt of zoster vaccine. This was done to identify those who were eligible to receive the live zoster vaccine for the main analysis, and to describe the extent of inadvertent administration of zoster vaccine to those with contraindications. The immunosuppressive conditions and drugs included were those listed in the UK Green Book; code lists (S2 Table) and algorithms used to identify these are described in the S4 Table.[4]

A past history of zoster was also ascertained using zoster or PHN codes from both CPRD and HES (S5 Table) occurring before the start of follow-up (01/09/2013), or a first zoster code of PHN occurring during follow-up.

Analyses

A complete case analysis using multivariable logistic regression was used to determine the association of socio-demographic factors with zoster vaccine uptake, using adjusted odds ratios (aOR) with 95% confidence intervals (CI). Logistic regression models were chosen to address the problem of potential misclassification of individuals for vaccine eligibility based on their date of birth and therefore the lack of information on person-time at risk for vaccination. The exposure and outcome characteristics of individuals excluded from complete case primary analysis because of missing data were described.

Gender and being a part of the routine or catch-up cohort (born in 1943 and 1934, respectively) were considered to be a priori confounders of the socio-demographic factors of interest, as well as potential determinants of zoster vaccine uptake. An existing conceptual framework[27] was adapted to postulate the hierarchical inter-relationships between distal and proximate social factors with the outcome (S6 Table).[28] Using this framework, socio-demographic factors were added sequentially, as long as data sparsity or multicollinearity were not encountered. The three multivariable models were as follows: Model 1 adjusted for a priori confounders (gender and year of birth) immigration status and ethnicity; Model 2 additionally adjusted for deprivation, and Model 3 additionally adjusted for the remaining socio-demographic factors. Standard errors of the coefficients were compared in successive analyses to assess multicollinearity between socio-demographic factors. Likelihood ratio tests were conducted for hypothesis testing unless otherwise indicated. Linear trends, if appropriate, were also examined for ordered categorical variables such as patient- and practice-LSOA-level IMD. The catch-up programme is different from the routine programme, and differences between the two age groups (routine and catch-up) in the effect of one sociodemographic factor (gender) on vaccine uptake have been reported by the national surveillance data.[6, 7] We therefore investigated whether the effects of the social factors varied between the two age cohorts by adding interaction terms to the models.

In sensitivity analyses, we repeated multivariable analyses: 1) restricted to individuals who were followed up for the entire study period of two years; 2) adding the social factors that had been excluded due to multicollinearity issues; 3) replacing the status of the time-varying factors to that ascertained at the end of follow-up instead of at the start of follow-up, and 4) including past history of zoster in the final multivariable model as a potential mediator of some of the socio-demographic factors.

Additionally, inadvertent zoster vaccination amongst individuals with immunosuppressive conditions or therapies was also described.

Data were analysed using Stata 14 (StataCorp, College Station, TX, USA).

Ethics

Approval was sought and obtained from the Observational/Interventions Research Ethics Committee of the London School of Hygiene and Tropical Medicine (Reference: 11910) and from the Independent Scientific Advisory Committee (ISAC) of the Medicines and Healthcare products Regulatory Agency (reference:16_168). The ISAC protocol was made available to the reviewers of this paper.

Primary analysis

A slight majority of the participants in the primary analysis (Table 1) were female. A higher proportion was born in the year 1943 (the main target group), were from a White ethnic background, and were cohabiting (living as a couple) and/or not living alone. Data for marital status were missing for 37.1% participants (Table 1) and practice-LSOA-level IMD was used to replace the missing patient-LSOA-level IMD for 0.07% (n = 26) participants. A past history of zoster was present for 11.4% of participants (Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Baseline characteristics of the study population: Comparison of individuals with minimum follow-up of 5 months and 24 months.

https://doi.org/10.1371/journal.pone.0207183.t001

Of the total participants considered for the primary analysis, zoster vaccine was administered to 18,499 (52.4%) individuals. Uptake amongst the main target group (those born in 1943) was 53.4% compared to 50.8% amongst the catch-up cohort (individuals born in 1934). Nearly half (n = 17,527) of the participants received both zoster vaccine and SIV (Fig 2); of these, only 36.8% (n = 6455) got both vaccines on same date (Fig 2), however the majority (86.8%, n = 16,066) received zoster vaccination during the influenza campaign period (September-March)[26] of 2013/2014 and 2014/2015. Amongst 79.8% (n = 28,192) of the participants who received SIV, 73.3% (n = 20,685) received SIV in both the 2013/2014 and 2014/2015 campaign periods while 22.4% (N = 6,323) received SIV only in 2013/2014, and 4.2% (N = 1,184) only in 2014/2015 season (data not shown).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Zoster and seasonal influenza vaccine uptake amongst study participants.

https://doi.org/10.1371/journal.pone.0207183.g002

A decision was made to drop marital status from the multivariable analyses, due to a large proportion of individuals with missing data for this variable. Thus, in the subsequent complete case analysis, only patients with missing ethnicity data (N = 3884, 11%) were excluded. This resulted in a final study population of 31,449 (89%) individuals, amongst whom the zoster vaccine uptake was 53.5% (Fig 1). Comparison of individuals included (n = 31,449) and excluded (n = 3884) from the complete case analysis due to missing ethnicity data is available in S7 Table. Briefly, excluded individuals were more likely to be in the main target cohort, and to be from less deprived patient- and practice-level deprivation areas, and were less likely to be care home residents, to have evidence that they were an immigrant or to have past history of zoster. The excluded group was also less likely to be vaccinated for zoster.

Time-varying exposures at the start and end of follow-up remained largely unchanged for 98.8% individuals included in the complete case analysis (S8 Table).

a) Minimally adjusted model.

In the analysis adjusted for a priori confounders (gender and year of birth), there was strong evidence of an association between higher zoster vaccine uptake and male gender, with uptake 10% higher compared to females (Table 2). There was also evidence of lower vaccine uptake amongst the catch-up cohort, immigrants, those of non-White ethnicity, care home residents, those living alone and those not cohabiting, with reduced odds of between 12% (being in the catch-up cohort) to 46% (Black ethnicity) (all p values <0.001, Table 2)). There was also strong evidence for a linear trend (p for trend <0.0001) for decreasing vaccine uptake with increasing patient-LSOA-level deprivation (IMD) score, the most deprived group having 34% lower odds of uptake compared to the least deprived group. Non-linear decreases in uptake were seen for practices in more deprived areas (Table 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Multivariable analyses- social factors associated with zoster vaccine uptake complete case analysis: Individuals with minimum follow-up of 5 months^a.

https://doi.org/10.1371/journal.pone.0207183.t002

b) Multivariable analyses.

After additionally adjusting the minimally adjusted model for immigration status and ethnicity (Multivariable Model 1, Table 2), no appreciable changes in effect estimates were observed, except lower uptake amongst immigrants was no longer apparent after adjustment for ethnicity, with no evidence of collinearity between the two variables. As patient- and practice-LSOA-level IMD were considered to be correlated, and as social factors relevant at an individual level were more of interest, only patient-LSOA-level IMD was added to Model 2. No noticeable changes in effect estimates between Model 1 and Model 2 were observed, and the strong evidence of linear trend of lower uptake with increasing deprivation score seen in minimally adjusted analysis was still evident (p<0.001) in this model. Similarly, living alone status and cohabitation were closely correlated, and so living alone was added to the final multivariable model (Model 3), along with care home residence. Again, the previously observed associations of lower uptake with living alone and residing in care home persisted in this model; individuals living alone and those residing in care home had 15% and 36% decreased odds of uptake, respectively (Table 2). The effect estimates for other variables were unchanged.

There was evidence that the effect of male gender, patient-LSOA-level IMD and care home residence varied with age (Table 3). Analyses showed that the increase in vaccine uptake among males was restricted to the catch-up cohort, and that the effects of care home status and (to a lesser extent) increasing deprivation on lower vaccine uptake were more marked in the catch-up cohort compared to the routine cohort.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Interaction between age and social factors.

https://doi.org/10.1371/journal.pone.0207183.t003

Sensitivity analyses

There were no appreciable differences between the baseline characteristics of individuals with follow-up for the entire 2-year study period and those included in primary analysis (follow-up period of ≥5 months) (Table 1). The results of multivariable analyses for those with longer follow up were similar to those from primary analysis except there was some attenuation in the association of ethnicity with vaccine uptake; individuals of Other ethnicities had 31% reduced odds (versus 39% in primary analysis) of uptake compared to those of White ethnicity in this model (Table 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Sensitivity analyses: Social factors associated with zoster vaccine uptake complete case analysis (Model 3).

https://doi.org/10.1371/journal.pone.0207183.t004

Substitution of cohabitation status instead of living alone, practice- LSOA-level IMD instead of patient-LSOA-level IMD, excluding individuals with immunosuppressive conditions at the end as opposed to start of follow-up (S9 Table) and determining care home status at the end instead of start of follow-up (Table 4 sensitivity analyses) did not change the findings. Individuals with a past history of zoster had 12% higher odds of uptake (Table 4), but inclusion of past zoster in the multivariable model made little difference to the other effect estimates.

To assess the impact of excluding individuals from the complete case analysis, a further exploratory analysis was conducted, by re-running the minimally adjusted analysis for the entire study population (N = 35,333) with follow-up ≥5 months including those with missing data on ethnicity (S10 Table). Comparing the minimally adjusted model with that of primary analysis revealed no noticeable differences in effect estimates. As the zoster vaccination programme is delivered through primary care in England, the effect of clustering at practice-level was also examined in an additional sensitivity analysis, using a multivariable (Model 3 in Table 2) logistic regression model with random effects (S11 Table). Although there was evidence for clustering within practices (with a p value <0.001 for the likelihood ratio test of an intra-cluster correlation coefficient of 0), the effect estimates in the random effects model were very similar to those from the primary analysis except for weaker evidence of reduced uptake among two of the non-White ethnicities (South Asian and Mixed ethnicity, S11 Table).

Inadvertent zoster vaccinations

Of the 19,330 in the total study population who received vaccination (Fig 1), 3% (n = 596) received zoster vaccine whilst immunosupressed.[4] Of these 596 patients, 28 (4.7%) patients had more than one immunosuppressive condition at the time of vaccination. The maximum number of immunosuppressive conditions at time of vaccination was three. The most common immunosuppressive condition (n = 445) during which the patients received zoster vaccine was cancer chemotherapy or radiotherapy, followed by patients taking other immunosuppressive medications (n = 69), patients with leukemia, lymphoma, myeloma, other plasma cell dyscrasias (n = 49), treatment with immunosuppressive dose of oral corticosteroid (n = 28), cellular immune deficiency (n = 25), solid organ transplant (n = 9) and HIV (n = 1). None of the patients received zoster vaccination during the immunosuppressive phase of a stem cell or bone marrow transplant.

Strengths and limitations

Strengths of the study include the large sample size, and linkages to hospitalisation data which provided additional information about socio-demographic factors as well as zoster vaccine contraindications. For multivariable analysis, a hierarchical approach using conceptual framework, enabled appropriate interpretation of effect estimates.[28] The NHS zoster vaccination programme is administered via general practice only and thus capture of vaccination is likely to be good. Additionally, our previous methodological study investigating ascertainment of socio-demographic factors in linked CPRD data showed that the distribution of factors such as ethnicity, not living alone, cohabitation and care home residence for individuals aged ≥65 years was comparable to the distribution in the 2011 English Census.[15]

The study limitations include potential misclassification of both socio-demographic factors and the zoster vaccination recording. Our categorisation of individuals without a relevant codes for a binary socio-demographic variable as not having the characteristic could have introduced errors; if this misclassification was non-differential with respect to vaccine uptake, this may have underestimated the effect estimates in this study. However, our previous methodological study showed good capture of socio-demographic factors in CPRD data when compared to the 2011 English Census.[15] Similarly, certain immunosuppressive treatments such as biological agents, which are mainly prescribed in a hospital setting, may not be completely captured in these primary care data. An additional issue is that some individuals may have changed exposure status over time. Ideally, person-time at risk to estimate vaccination rates would have been preferable but unavailability of complete dates of birth in CPRD precluded this ascertainment. However, factors ascertained at start of follow-up remained unchanged for 98.8%-99.9% individuals at the end of follow-up and sensitivity analyses to assess the impact of the time-varying factors reassuringly revealed similar results to the primary analysis. There was also potential for bias resulting from the complete case analysis (owing to missing ethnicity information) which led to exclusion of data from 11% of the study population. Further assessment of this revealed that effect estimates from the minimally adjusted models for individuals for the entire study population (after dropping ethnicity) had no appreciable differences to those obtained from the complete case analysis. Lack of recording of marital status prevented investigation of its association with zoster vaccine uptake, but other closely related variables for living arrangements such as cohabitation and living alone were available.

Other factors indicative of socio-economic status of an individual and associated with vaccine uptake amongst older individuals, as reported by our systematic review,[12] such as income, education, and occupation could not examined in this study as these factors were unavailable in CPRD. However, the association of deprivation using IMD quintiles for both patient- and practice-LSOA-level could be assessed in this study.

It is also feasible that individuals who see their GPs more frequently may have greater opportunity to be offered vaccination. Thus, it is possible that healthcare utilisation may be a potential mediating variable for the association between socio-demographic factors such as deprivation, male gender (in the catch-up cohort) or ethnicity and zoster vaccine uptake. If this were the case, adjusting for this mediating variable could attenuate the association between some of the socio-demographic factors and vaccine uptake.

Although we were able to examine whether there was evidence of clustering at practice-level, lack of information on practice-level variables (other than practice-LSOA-level-IMD) prevented further investigation of this. It is possible that variations in certain general practice-level characteristics, for example staffing levels, opening hours, and call/recall processes, could have affected implementation of the zoster vaccine programme for all zoster vaccine-eligible patients in that practice. These general practice-level data were unavailable in CPRD and therefore we could not determine to what extent the lower uptake we found in specific groups related to lack of opportunity to be vaccinated rather than individual choice not to come forward for vaccination. However, it is notable that practice-level clustering had little impact on most of the effect estimates studied other than some of the estimates for ethnicity. Future studies assessing the effects of these factors on vaccine uptake may provide a better understanding for the reasons of lower uptake amongst specific population groups.

Comparison with other studies

Our finding of higher uptake in the main target group compared to the catch-up cohort reflects the findings from the national annual zoster vaccine coverage data for England over the same time period, and vaccine coverage among the individuals in our study who were followed up for the entire study duration (and thus had fuller capture of uptake of vaccination) is comparable to the coverage estimates in the national data.[6, 7] Higher uptake amongst males in the catch-up cohort observed in this study also was shown in the national data for 2013–2015 for reasons that are currently unexplained.[6, 7] This is in contrast to findings from North America that have reported a higher zoster vaccine uptake amongst females.[29–31] The majority of zoster vaccinations (87%) in this study occurred during the influenza season, suggesting that opportunistic targeting of the eligible population for zoster vaccine during SIV programme might have played a role. The national annual zoster vaccination data also supports this finding.[6, 7] Our finding of a linear relationship between increasing level of deprivation and lower zoster vaccine uptake also confirms the earlier analyses of the national data, which found a similar trend but was restricted to examining deprivation at the general practice-level;[14] studies from the US and Canada have also reported the lower zoster vaccine amongst individuals with lower income.[30–32] Higher zoster vaccine uptake amongst individuals of White ethnicity, as seen here, has been reported from other zoster vaccine studies from high income countries.[14, 30, 32–35]

Several of our findings are novel with respect to zoster vaccine uptake. These include lower uptake among individuals who were living alone or not cohabiting; this echoes studies showing lower uptake of seasonal influenza vaccine among those living alone in older European populations.[12] Our finding that immigration status was not independently associated with zoster vaccine uptake after adjusting for ethnicity is in contrast with previous findings of lower uptake among immigrant populations for seasonal influenza and pneumococcal vaccine in Spain and Israel, although none of these studies adjusted additionally for ethnicity.[36–43] The lower uptake of zoster vaccine among care home residents adds to a number of studies investigating uptake of other vaccines such as influenza and pneumococcal vaccines which have reported higher and lower uptake respectively, amongst care home residents.[44–48]

Interpretation of findings and implications

This study demonstrates that in a public funded healthcare system, vaccination inequalities exist during a crucial period of programme initiation, and identifies socio-demographic groups that could be targeted with tailored interventions to increase zoster vaccine uptake. Of particular interest is the finding of lower uptake among care home residents; we have shown recently that individuals in care homes are at higher risk of developing zoster,[49] and so are a group with possible double health inequity (of both zoster burden and zoster vaccine uptake). Lower vaccine uptake among these residents could be due to lack of awareness amongst care home staff about the newly introduced programme and issues around getting consent. The reasons cited for differential (higher or lower) uptake of other vaccines among care home residents have included presence of vaccination polices in care homes, staff awareness, vaccination consent from the residents, location and care home ownership (public versus private).[44, 47, 48] The potential double health inequity amongst care home residents highlights a need for more rigorous targeting of these individuals to mitigate health inequality.

Similarly, targeting of older individuals who live alone may be needed to encourage zoster vaccination. Individuals cohabiting or living with their relatives may be more motivated by their social networks to get vaccinated. Secondly, higher disease awareness amongst these individuals, by witnessing the debilitating effect of zoster in their relatives, may also increase uptake. This was examined in a US study that reported higher zoster vaccine uptake amongst individuals in the three months after occurrence of zoster in their partners, reflecting a short term effect of disease awareness.[50] However, our finding that adjustment for a past personal history of zoster made little difference to effect estimates of social factors suggests that social networks may have a longer-lasting effect on encouraging vaccine uptake in older individuals, or that the occurrence of zoster in partners versus self may have a different effect on uptake. The lower uptake of vaccine among those of non-White ethnicity, but the attenuation of this association after restricting to individuals with longer follow up, suggests that there might be delay in uptake amongst some ethnic groups. There may be a lack of zoster disease awareness among some ethnic groups because of lower lifetime risk of zoster, which may be due to genetic causes, social mixing patterns that limit contacts with varicella and thus boosting of varicella-zoster virus immunity, and late onset of varicella among those born overseas.[33, 50, 51] Alternatively, the lower uptake may reflect existing healthcare inequalities. The lack of association of immigration status with zoster vaccine uptake after adjusting for ethnicity could be due to confounding by ethnicity, or simply to lack of power to detect an effect—the number of individuals identified as immigrants in the study population was relatively small (1.4%; n = 512) of which zoster vaccine uptake was observed amongst only 229 individuals. It is also feasible that in England, where national zoster vaccination programme is available free-of-charge, vaccination inequalities were not observed for older immigrant populations.