Effectiveness of seasonal trivalent inactivated influenza vaccine in preventing influenza hospitalisations and primary care visits in Auckland , New Zealand , in 2013

N Turner (n.turner@auckland.ac.nz)1, N Pierse2, A Bissielo3, Q S Huang3, S Radke1,3, M G Baker2, M A Widdowson4, H Kelly5,6, on behalf of the SHIVERS investigation team7 1. The University of Auckland, Auckland, New Zealand 2. University of Otago, Wellington, New Zealand 3. Institute of Environmental Science and Research, Wellington, New Zealand 4. Centers for Disease Control and Prevention, Atlanta, GA, United States 5. Australian National University, Canberra, Australia 6. Victorian Infectious Diseases Reference Laboratory, Melbourne, Australia 7. Members of the team are listed at the end of the article


Introduction
Influenza infection causes a major burden of illness in adults and children [1,2].Seasonal trivalent influenza vaccines (TIVs) are effective in preventing a range of laboratory-confirmed outcomes [3], but effectiveness varies by severity and season, the presence of comorbidities and age [4,5].
The SHIVERS (Southern Hemisphere Influenza Vaccine Effectiveness, Research and Surveillance) study has allowed estimation of vaccine effectiveness (VE) against influenza illness requiring hospitalisation since 2012 and against influenza illness requiring primary care (general practice) since 2013.
In New Zealand, seasonal non-adjuvanted inactivated trivalent influenza vaccine is available annually free of charge to all adults aged 65 years and over, pregnant women and all those over six months of age with chronic medical conditions that are likely to increase the severity of the infection.Influenza vaccines are also available on the private market for all others over six months of age.Two commercial vaccine products were available in the New Zealand market in 2013: Fluarix (GlaxoSmithKline) and Fluvax (bioCSL).Both vaccines contained A/California/7/2009 (H1N1)-like virus, A/Victoria/36/2011 (H3N2)-like virus and B/Wisconsin/1/2010-like virus (belonging to B/ Yamagata/16/88 lineage).
Using the case test-negative design, we estimated the effectiveness of seasonal trivalent inactivated influenza vaccine in preventing laboratory-confirmed influenza in patients hospitalised with severe acute respiratory infections (SARI) and in patients presenting to general practice with an influenza-like illness (ILI) during the 2013 influenza season, which is from March to September in New Zealand.

Methods
Ethics approval for the study was obtained from the Northern A Health and Disability Ethics Committee (NTX/11/11/102 AM02).

Study design
In both hospital and community settings, we conducted a study using a standard case test-negative design [6], drawing on an urban population of approximately 838,000 people in Central, South and East Auckland [7].
For community cases, we recruited 18 sentinel general practices with 103,884 enrolled patients.Patients in these sentinel practices were broadly representative of the ethnically diverse urban population of Auckland by age and sex distribution, but with more Pacific people (27% in the practices compared with 15% in the source population) and slightly fewer people of Asian descent (14% versus 19%, respectively) [7].
The practices recruited individuals aged six months and older who presented to a general practitioner or practice nurse with ILI, defined as a history of fever or measured fever of ≥38 °C and cough, with onset during the preceding 10 days [8,9].
All patients presenting to one of the sentinel general practices with suspected respiratory infections were screened by the general practitioner or nurse for ILI.All identified ILI cases were entered on an electronic form in the practice management system and a nasopharyngeal or throat swab was collected for influenza virus testing from all consenting patients.
For hospitalised patients, we enrolled individuals aged six months and older who were admitted with SARI to one of the four public hospitals covering the whole population in this catchment area.On the basis of the World Health Organization (WHO) definition, SARI was defined as hospitalisation with a patient-reported history of a fever or a measured temperature of ≥38 °C and cough with onset within the past 10 days [10].A confirmed case of influenza was defined as a SARI or ILI patient with a positive laboratory result for any influenza virus detected by real-time reverse transcription polymerase chain reaction (RT-PCR), while non-cases (controls) were patients with SARI or ILI who tested negative for all influenza viruses.SARI patients were identified following screening of patients admitted with respiratory disease by dedicated research nurses.Overnight admissions of patients with respiratory symptoms were screened by the nurses on the following day.All patients satisfying the SARI case definition were invited to participate in the study.Patients who gave verbal consent completed a case report form and provided a nasopharyngeal swab or aspirate for influenza virus testing.
Patients who were identified at seven days post onset of symptoms were excluded from the ILI and SARI analysis, based on the pattern of shedding, which peaks in  the first three days, then declines steadily until days 7-9 [9].We also excluded patients with incomplete data for vaccination status or age, all infants under 6 months of age, children aged under 9 years who were only given one dose of trivalent inactivated influenza vaccine, patients who were vaccinated less than 14 days before admission to hospital or presentation to general practice.For patients with multiple episodes, the first influenza virus-positive episode was used for the analysis, or the first illness episode if there was no influenza virus-positive episode.
The influenza season was defined as starting when there were two consecutive weeks with two or more cases and ending when there were no consecutive weeks with two or more cases.Analysis was thus undertaken from 3 June to 10 November 2013 for SARI cases and from 13 May to 27 October 2013 for ILI cases (Figures 1 and 2).

Participant information
For all ILI patients, variables extracted from the electronic form and patient management system included age, sex, ethnicity, chronic medical conditions and current smoking status, socio-economic status as identified by the New Zealand deprivation status (a meshblock measure reflecting eight dimensions of deprivation distributed into deciles) [11] and a subjective assessment of obesity by the clinician as body mass index measurements were not consistently available.
Similar information was collected for all SARI patients, but for this group we also collected the following: a patient-or caregiver-reported measure of dependence (which assessed requirement for assistance with normal activity or full dependency on nursing care); a measure based on long-term use of oxygen that we labelled 'frailty'; a history of chronic medical conditions; and a self-defined, standardised functional wellbeing health status score from a national survey [12], combining fair or poor well-being versus all other more positive well-being scores.
In New Zealand, almost all influenza vaccinations are administered in general practices.For ILI cases, vaccination status was taken from the general practice record.SARI vaccination status for the 12 months before hospitalisation was determined by self-report.

Laboratory methods
Nasopharyngeal swabs, aspirates and other respiratory samples were collected according to hospital or general practice standard procedures.Samples were tested using the United States Centers for Disease Control and Prevention (CDC) real time RT-PCR protocol [13] or the AusDiagnostic PCR protocol [14].The AusDiagnostic assay had a sensitivity of 100% and  specificity of 96.6% when the United States CDC method was used as the comparator [15].RT-PCR assays detected influenza virus types A and B and subtyping was performed for type A. All influenza virus PCR-positive samples were forwarded to the National Influenza Centre and characterised antigenically using established methods [14].

Statistical analysis
Univariate chi-squared tests were used to compare characteristics of patients who were influenza virus positive (cases) and negative (controls).A multivariate logistic model restricted to the test-negative controls was used to calculate an adjusted odds ratio for the propensity to be vaccinated for a range of patient characteristics possibly associated with vaccination (Table 1) and used in a previous study [16].The results from this propensity model are presented as odds ratios.This propensity model was then applied to the complete dataset to generate individual propensity scores for vaccination.In order to ensure these propensity scores were linear with respect to influenza virus positive status, we used the cubic spline for these scores as an adjustment variable for estimating VE.For both SARI and ILI, we calculated the crude VE, adjusting only for the timing of presentation relative to the influenza season (defined as weeks from the peak), and the adjusted VE, which included the timing of presentation and the cubic spline of the fitted values of the propensity model.VE estimates were calculated against both SARI and ILI, by influenza virus type and subtype and by age group (6 months-17 years, 18-64 years and ≥65 years).
For all patient characteristics, other than age and vaccination status, each missing data point was imputed as the baseline (referent) value for the corresponding variable.The baseline values were: non-Māori, non-Pacific ethnicity, female, not in New Zealand deprivation groups 8,9 or 10 (the three lowest deprivation deciles), not pregnant, current non-smoker, without chronic disease, not obese, with self-rated health average or better, not using long-term oxygen and living without assistance).All male patients and all female patients not aged 13-55 years were assumed to be not pregnant.Overall, 53 SARI (5%) cases and 1 ILI (0.1%) case had any imputed covariates.Sensitivity analyses were performed using only those individuals with complete data.
As a further sensitivity analysis, we also compared the overall VE estimate from the propensity-adjusted model with an epidemiological model that included covariates that were assessed as potential confounders and/ or effect modifiers (Table 1) and a statistical model, derived from the epidemiological model, where only covariates that were significant at the 0.05 level were included in the final model.

Results
Characteristics and vaccination status of participants ILI and SARI patients in this study by influenza virus status are shown in Figures 1 and 2.

Figure 3
Flowchart of all selected, recruited and tested patients with influenza-like illness and severe acute respiratory infection for influenza vaccine effectiveness analysis, New Zealand, 2013 influenza season Of the 482 ILI patients who tested influenza virus positive, 44 (9%) were vaccinated, compared with 177/1,013 (17%) who tested negative (Figure 3).The proportion vaccinated did not change throughout the season.In those excluded because of incomplete laboratory tests, self-reported vaccination in the previous 12 months among SARI cases was 49/135 (36%), slightly less than the proportion of included SARI cases, 454/1,042 (44%).In the ILI cases, the proportion vaccinated was much higher, at 8/21 (38%), in those excluded because of incomplete laboratory tests compared with the ILI cases included, of whom 9% (44/482) were vaccinated.
Influenza-positive cases and influenza-negative controls were compared across a range of patient characteristics.SARI and ILI patients who were aged 6 months to 5 years or over 80 years, and those presenting outside the influenza season were less likely to test positive.In comparison with the community patients, the hospitalised patients were more likely to be vaccinated, to be older, to live in a deprived area, to be of Māori or Pacific ethnicity, to be a current smoker and to be obese (Table 1).Details on pregnancy were poorly recorded but less than 3% (30/1,042) of other data fields were missing for both SARI and ILI patients.
Although vaccination was more common in SARI patients, the same factors affected the propensity to be vaccinated in persons with ILI or SARI.The adjusted odds ratios for the association of various patient characteristics with the likelihood of vaccination showed that older age groups and those with chronic diseases were most likely to be vaccinated (Table 3).In contrast, there was no statistically significant difference in the likelihood of vaccination by ethnicity, sex, deprivation score, pregnancy, obesity, self-rated health, smoking, assisted living or the timing of the admission relative to the influenza season (Table 3).

Vaccine effectiveness
The VE against all circulating influenza virus strains, adjusted only for the number of weeks from the peak of the influenza season, was 32% (95% confidence interval (CI): 7 to 50) for influenza-confirmed SARI and 56% (95% CI: 37 to 70) for influenza-confirmed ILI (Table 4).After also adjusting for the propensity to be vaccinated, the estimated VE was 52% (95% CI: 32 to 66) for SARI and 56% (95% CI: 34 to 70) for ILI.Thus, adjusting for the propensity to be vaccinated had more effect on the VE estimate for SARI than for ILI.For ILI, the crude and adjusted VE point estimates were the same.
Adjusting for only the variables that were significant in the model (p<0.05)resulted in a VE estimate of 54% (95% CI: 33 to 69) for SARI and 59% (95% CI: 41 to 72) for ILI.When restricting the analysis to within four days of onset of symptoms, the adjusted VE for ILI was 48% (95% CI: 3 to 68) and for SARI 57% (95% CI: 36 to 71).When analysed by restricting to a shorter period around the peak (weeks 28-40), the VE for ILI was 46% (95% CI: 16 to 65) and for SARI 53% (95% CI: 27 to 50).
For both SARI and ILI influenza-positive cases, the vaccination rate was constant over time.

Table 1
Characteristics of study participants with influenza-like illness and severe acute respiratory infection, New Zealand, 2013 influenza season* a Unless otherwise indicated.b A meshblock measure reflecting eight dimensions of deprivation distributed into deciles.c Subjective assessment of obesity by the clinician.d A self-defined, standardised functional well-being health status score.e Defined as currently on long-term oxygen use.f Requirement for assistance with normal activity or full dependency on nursing care, as reported by the patient or caregiver.

Table 2
Vaccinated and unvaccinated influenza cases by virus type and subtype in hospitalised and community study participants, New Zealand, 2013 influenza season* One SARI case and four ILI cases tested positive for both influenza A and B viruses.Not all cases of influenza A were subtyped.The number of subtypes does not add up to the number of all influenza A viruses identified. a

Table 3
Characteristics of non-influenza virus-positive study patients with severe acute respiratory infection and influenza-like illness (controls) and their association with influenza vaccination status, New Zealand, 2013 influenza season OR: adjusted odds ratio compared with referent group: female, aged 46-64 years, non-Māori, non-Pacific ethnicity, not in the New Zealand deprivation measure of the three lowest deciles (8,9 or 10), not pregnant, current non-smoker, without chronic disease, not obese, with self-rated health average or better, not on long-term oxygen use, living without assistance and admitted to hospital for severe acute respiratory infection during the peak influenza season.
a A meshblock measure reflecting eight dimensions of deprivation distributed into deciles.b Subjective assessment of obesity by the clinician.c A self-defined, standardised functional well-being health status score.d Defined as currently on long-term oxygen use.e Requirement for assistance with normal activity or full dependency on nursing care, as reported by the patient or caregiver.f Admission or presentation before 1 June 2013.