Effectiveness of the live zoster vaccine during the 10 years following vaccination: real world cohort study using electronic health records

Study setting

KPNC is an integrated healthcare delivery system with 4.3 million members, of whom 1.6 million are 50 years or older. KPNC’s diverse population is similar demographically to the overall Northern California population. KPNC’s electronic medical records contain data for diagnoses, healthcare visits, hospital admissions, immunizations, prescriptions, and laboratory tests. KPNC provided live zoster vaccines free of charge. Starting in July 2013, an electronic medical records prompt targeted members aged 60 years or older for vaccination with the live zoster vaccine. The institutional review board at KPNC approved this study.

Study design and study population

Our study population is described elsewhere.1213 In brief, we conducted a prospective cohort study with follow-up from 1 January 2007 to 31 December 2018 of KPNC members who were eligible by age for live zoster vaccine. Vaccine eligibility was based on US licensure dates for people aged 60 years or older (25 May 2006, with study entry starting 1 January 2007) and for people aged 50-59 years (24 March 2011). We restricted study entry to people with continuous membership since becoming eligible by age for live zoster vaccine. We excluded individuals who received live zoster vaccine before study entry, or who had an HZ diagnosis in the year before study entry. We followed study participants until the first occurrence of: HZ diagnosis, disenrollment from the health plan, a second dose of live zoster vaccine, receipt of recombinant zoster vaccine, death, or end of study (31 December 2018).

Vaccination status was a time varying covariate. All people started follow-up unvaccinated. If vaccinated, their status was updated annually on the anniversary of vaccination (first year, second year, etc).

Outcomes

We identified incident cases based on diagnoses, prescriptions, and laboratory tests, validating samples of cases by chart review. We defined an incident HZ case as the first encounter during follow-up with an HZ diagnosis (International Classification of Diseases (ICD)-9 code 053.xx or ICD-10 code B02.xx) with an antiviral prescription or positive varicella zoster laboratory test. Approximately 84% of all first HZ diagnoses met these criteria and chart review (n=200) showed 98% as incident HZ cases. We therefore considered all HZ cases identified with this definition as incident without chart review. Among HZ cases, we then identified the subsets with postherpetic neuralgia, HZO, or admission to hospital for HZ.

We defined postherpetic neuralgia cases based on postherpetic neuralgia diagnoses from an encounter or prescription between 90 days and 1 year after the initial HZ diagnosis.13 After chart review of 200 such potential cases, we included those with postherpetic neuralgia diagnoses in both an encounter and a prescription without additional review, while the remainder underwent chart review to ascertain case status.

We defined HZO cases based on HZO diagnoses (ICD-9 053.2x and ICD-10 B02.3x) at an ophthalmology visit within 30 days of the initial HZ diagnosis (94% of all HZO diagnoses were recorded in ophthalmology visits). Chart review of 40 of these potential cases confirmed that 100% were HZO, therefore, we included all without further review.

Admission to hospital with a principal diagnosis of HZ had to be within 30 days after the initial HZ diagnosis. Among all admissions to hospitals for HZ within one year of the initial diagnosis, 94% occurred within the first 30 days.

Statistical analysis

We calculated the incidence of HZ, postherpetic neuralgia, HZO, and admission to hospital for HZ by vaccination status overall and by age group (50-59, 60-69, 70-79, and ≥80 years). We also calculated the percentage of HZ cases who had postherpetic neuralgia, HZO, or admission to hospital for HZ.

For each outcome, we examined the vaccine effectiveness of live zoster vaccine using Cox regression. Cox models were specified with a calendar timeline stratified by birth year to adjust for confounding associated with age as well as calendar time. For each day on which a case occurred, a risk set was formed, including the case along with all people in follow-up that day who were born the same year as the case. All models were also adjusted for covariates including sex, race/ethnicity, and time varying factors, including influenza vaccination, visit frequency, comorbidities, and immunocompromise status.12

We used a Cox model to estimate vaccine effectiveness in relation to the number of years since receipt of live zoster vaccine. For the HZ outcome, we fitted a model with 12 time varying binary (yes/no) vaccination indicators to denote—at each time point during follow-up—either the time since vaccination (eg, 1-29 days, 30 days to less than one year, one to less than two years, two to less than three years, etc) or that the individual was unvaccinated (reference group). We estimated vaccine effectiveness against HZ for each year following vaccination. For postherpetic neuralgia, HZO, and admission to hospital for HZ, the models were similar to that for HZ except, due to fewer cases, we included fewer indicators of time since vaccination (eg, 1-29 days, 30 days to less than one year, one to less than three years, three to less than five years, etc). For each outcome, we estimated a hazard ratio for each time since vaccination interval (beginning on day 30 to allow time for an immune response) in comparison to the unvaccinated group. We then estimated vaccine effectiveness as 1 minus the hazard ratio estimate, scaled as a percentage.

For each outcome, we also calculated two summary measures of vaccine effectiveness across more than 10 years of follow-up time. One measure that we refer to as “overall vaccine effectiveness” summarises vaccine effectiveness in the usual way, as if the hazard ratio was not changing over time. This measure is the same summary measure commonly referred to simply as vaccine effectiveness in most clinical trials and observational studies. This summary measure can be problematic if vaccine effectiveness wanes because it gives more weight to the earlier years postvaccination when vaccine effectiveness is higher (since people who were vaccinated later in the study period can only contribute earlier years postvaccination). Our other measure, “average vaccine effectiveness”, averages all the time specific hazard ratios across the 10 years after vaccination. This measure indicates the average percentage reduction in incidence among individuals who live 10 years after vaccination. To estimate overall vaccine effectiveness, we included a single vaccination indicator (yes/no) in the Cox model. To estimate average vaccine effectiveness, we included multiple indicators of time since vaccination in the Cox model and weighted the indicators in accordance with their relative durations. We calculated average vaccine effectiveness for the 10 years after vaccination, as well as for the first three, five, and eight years. We used the term overall vaccine effectiveness only when needed to contrast with average vaccine effectiveness. Otherwise, we simply use the term “vaccine effectiveness”, as in other studies.

For each outcome, we also used Cox regression to estimate vaccine effectiveness in subgroups defined by age group at vaccination (50-59, 60-69, 70-79, and ≥80 years), sex, race or ethnic group (white, black, Asian or Pacific Islander, Hispanic, American Indian or Alaskan Native, or other or unknown), and immunocompromise status at vaccination (none, low, or high immunocompromise).

Analyses were done with SAS 9.3. We used the Lexis macro to partition person time (http://bendixcarstensen.com/Lexis/Lexis.sas).