Study design
The study was conducted under a retrospective cohort design using the same claims database and analytical methodology as in our previous pediatric study of influenza vaccination[5]. The claims database was provided by Japan Medical Data Center Co., Ltd. (Tokyo, Japan) and included more than 3 million enrollees of employee health care insurance plans, run mainly by large-scale private enterprises [6]. Deidentified claims data of employees and their dependents including diagnoses, procedures, and pharmacy prescriptions were collected from the health insurance plans. Study duration was the epidemic period, starting from 1 October 2013 to 31 May 2014 [7].
Ethical approval for the study protocol and waiver for informed consent to participate were provided by the Keio University Faculty of Pharmacy Ethics Committee for Research Involving Humans (No. 160118–1), in accordance with local ethical guidance for medical research involving human subjects.
Influenza vaccination
Seasonal trivalent inactivated and nonadjuvanted influenza vaccination is available for children aged 6 months or older and adults in Japan. The recommended dosage per season is two doses of 0.25 ml for those aged 6–24 months, two doses of 0.5 ml for those aged 3–12 years, and one dose of 0.5 ml for those over 12 years. Vaccines for the 2013/2014 season contained two influenza A strains (A/California/7/2009(X-179A)pdm09, A/Texas/50/2012(X-223)) and one influenza B strain (B/Massachusetts/2/2012(BX-51B)). Circulating viral subtypes of influenza in the 2013/2014 season were A/H1N1pdm09 (43%), A/H3N2 (21%), B/Yamagata (24%), B/Victoria (9%), and B/untyped (4%) [8].
The influenza vaccination period in Japan starts at the beginning of October and the vaccination schedule is managed by the Ministry of Health, Labour and Welfare [5]. The Japanese National insurance scheme, which provides basic universal coverage of all medical costs for all nationals, does not cover influenza vaccine. Japanese nationals have to pay the cost of vaccination out-of-pocket, except for the elderly aged 65 years or older, whose vaccination is provided at local government cost by public health service centers. Children and other adults receive influenza vaccination at community clinics or hospitals at their own cost. Several employment-based health insurance plans operated by private companies subsidize influenza vaccination and the scope of subsidy differs by the age group covered and fee. Accordingly, vaccination status and dates were identified from records in the health plans for the respective influenza vaccination subsidies.
Study population
Subjects were employees and their dependents aged 1 to 64 years as of October 1st 2013. Enrollees were eligible if their health plan provided subsidies for influenza vaccination. Because the Japan Pediatric Society does not recommend influenza vaccination for infants aged less than 1 year, potentially biasing pediatrician preferences for who should be vaccinated in this age group [9], we excluded infants aged ≤1 year from the evaluation of vaccine effectiveness. Some health plans do not provide a subsidiary program for influenza vaccination for particular age groups, such as adults aged 20 years or older. We therefore excluded subjects who were not eligible for subsidies or whose enrollment in the health plan started on or after October 2012 or ended before May 2014, to eliminate potential misclassification of vaccination status during the 2012/2013 and 2013/2014 seasons. Because seniors aged over 64 years were strongly recommended to receive vaccination and were often supported by subsidies from local governments rather than by health insurance plans, we also excluded subjects aged over 64 years to avoid misclassification of influenza vaccination status. We further excluded patients with prolonged hospitalization, on the basis that the extended period they spent in a managed environment may have affected their probability of being vaccinated or their risk of virus exposure. Patients with prolonged hospitalization were identified as those with a record of hospitalization for ≥24 days/month over ≥7 months.
The risk period for outcome events in vaccinees was from 14 days after the date of vaccination to 31 May 2014. Subjects who experienced an outcome event within 13 days after vaccination were excluded to ensure that the effects of vaccination were accurately assessed [10]. For non-vaccinees, the whole study period was the risk period for effectiveness. Subjects who got vaccinated after having experienced an outcome event were classified as non-vaccinees and censored at the time of outcome diagnosis.
Outcome definition
We used the same definitions for the incidences of influenza diagnosis and complications as the ones used for the preceding pediatric study, which were based on the International Statistical Classification of Diseases and Related Health Problems 10th Revision (ICD-10) codes [5]. A primary diagnosis of influenza was based on the ICD-10 codes J101, J110, J111, and J118. To test the robustness of the definition based on these codes in sensitivity analyses, three different definitions of influenza incidence were developed, as follows: 1) combination of the ICD-10 codes defined above with records of the use of a rapid-testing kit, identified by claim records for the influenza virus antigen (high-sensitive) test; 2) combination of these ICD-10 codes with a prescription for antiviral drugs, as determined using the Anatomical Therapeutic Chemical (ATC) J05B4 classification of the European Pharmaceutical Marketing Research Association; or 3) the combination of the J101 code “influenza due to identified seasonal influenza virus” and use of a rapid-testing kit. Secondary outcomes of influenza complications included pneumonia (J12-J18) and respiratory tract diseases (RTD: J00-J22, apart from the above influenza codes). There were both defined using ICD-10 codes only. Cases requiring hospitalization were additionally defined for secondary outcomes as subjects hospitalized within 3 days before or after the date of influenza diagnosis (hospitalization with influenza); subjects hospitalized within 7 days of the diagnosis date of RTD (hospitalization with RTD); and emergency hospitalization with influenza or pneumonia. We identified emergency hospitalization using claim codes of the national health insurance medical fee schedule for emergency hospitalization (A205 or A300) [11].
Confounding factors
Covariates considered for inclusion in adjustment of potential confounders in evaluating the effectiveness of vaccination included influenza vaccination status during the prior season (2012/2013 influenza season), age at October 1 2013, employment status, gender, number of other dependents aged 0 to 15 years covered under the same insurance number (number of children aged 0 to 15 in a family), preceding onset of influenza in family members during the current influenza season, history of high-risk medical conditions, emergency hospitalization, and number of outpatient visits during or outside office hours in the prior influenza off-season (June to September). High-risk medical conditions were defined using the definition of the US Centers for Disease Control and Prevention (CDC) [12]. Family members could be identified because employees and their dependents shared the same insurance number. “Preceding onset of influenza in family members” was considered to be the risk of second infection that subjects were exposed to when a family member received any influenza diagnosis code before his/her first influenza. This was considered in both the primary analysis and also in the secondary analysis for hospitalization using influenza as outcome.
Statistical analysis
Subject characteristics in the four age groups (1–4 years, 5–12 years, 13–19 years and 20–64 years) were summarized with descriptive statistics. Between-group comparisons were tested with the Mann-Whitney test for continuous variables and the chi-square test for categorical variables.
The primary analysis was to estimate the effectiveness of influenza vaccination in preventing the onset of influenza. The secondary analysis aimed to estimate the effect of influenza vaccination on influenza complications outcomes. First, we calculated odds ratios (ORs) of outcome events for influenza vaccination in the 2013/2014 season and other covariates using conventional multivariate logistic regression. To adjust confounding related to influenza vaccination for whole subjects and respective age groups, we next used a doubly robust method (DR) which combines a logistic regression model with inverse probability treatment weighting (IPTW) by propensity score (PS) to calculate the ORDRs [13]. PSs were calculated as those for the probability of being vaccinated in the 2013/2014 season by considering the covariates mentioned above and the presence of children aged 0 to 15 years in a family, but not considering preceding onset of influenza in family members or the number of children aged 0 to 15 years in a family. In the analysis for whole age subjects, age effect was modeled by linear tail-restricted cubic spline functions with 5 knots, based on percentiles on the basis that treating age effect as a linearity of age effect could not be assumed [14]. Vaccinees having a PS < 0.1 as well as non-vaccinees with PS > 0.9 were excluded because these subjects with opposite, extreme PS values were reported to have the potential to cause bias in the IPTW estimates by being excessively weighted with reciprocals of PSs for vaccinees and of (1-PS) for non-vaccinees [15]. C-statistics were calculated to ascertain the validity of PSs. When stratified by age group, actual age was not considered in calculating PSs or ORDR.
In addition to the DR method, Cox hazard regression analysis was also conducted as a sensitivity analysis, using the time-dependent covariates of current year vaccination and preceding onset of influenza among siblings. Some studies investigating the effects of influenza vaccine on elderly mortality using electronic health record databases have suggested a potential bias towards apparent greater effectiveness [16,17,18]. The presence of this bias was examined with a Cox hazard model by investigating vaccine effectiveness during the 2013 pre-epidemic period, with minimum influenza circulation between October 1st and December 16th, the latter date being the day when the National Statistics announced the beginning of the epidemic season [19].
All statistical analyses were conducted using SPSS version 25.0 (IBM Corp., Armonk, NY) or SAS version 9.4 (SAS Institute Inc., Cary, NC). Tests were 2-tailed and had a significance level of 0.05.