Effectiveness of Trivalent Inactivated Influenza Vaccine in Children Estimated by a Test-Negative Case-Control Design Study Based on Influenza Rapid Diagnostic Test Results

We assessed vaccine effectiveness (VE) against medically attended, laboratory-confirmed influenza in children 6 months to 15 years of age in 22 hospitals in Japan during the 2013–14 season. Our study was conducted according to a test-negative case-control design based on influenza rapid diagnostic test (IRDT) results. Outpatients who came to our clinics with a fever of 38°C or over and had undergone an IRDT were enrolled in this study. Patients with positive IRDT results were recorded as cases, and patients with negative results were recorded as controls. Between November 2013 and March 2014, a total of 4727 pediatric patients (6 months to 15 years of age) were enrolled: 876 were positive for influenza A, 66 for A(H1N1)pdm09 and in the other 810 the subtype was unknown; 1405 were positive for influenza B; and 2445 were negative for influenza. Overall VE was 46% (95% confidence interval [CI], 39–52). Adjusted VE against influenza A, influenza A(H1N1)pdm09, and influenza B was 63% (95% CI, 56–69), 77% (95% CI, 59–87), and 26% (95% CI, 14–36), respectively. Influenza vaccine was not effective against either influenza A or influenza B in infants 6 to 11 months of age. Two doses of influenza vaccine provided better protection against influenza A infection than a single dose did. VE against hospitalization influenza A infection was 76%. Influenza vaccine was effective against influenza A, especially against influenza A(H1N1)pdm09, but was much less effective against influenza B.


Introduction
Influenza vaccination is the most effective method of preventing influenza virus infection and its potentially severe complications, and vaccine efficacy from randomized control trials [1][2][3] and vaccine effectiveness (VE) from observational studies [4][5][6][7] in healthy children has been reported to be 40%-70%. However, VE varies considerably with time, place, and degree of antigenic distance between the vaccine strain and circulating strain. Although VE has generally been interpreted in the context of vaccine matches with circulating strains, low VE has recently been reported to be related to mutations in the egg-adapted H3N2 vaccine strain rather than to antigenic drift in circulating viruses [8,9].
Low VE against influenza A/H3N2 was reported during the 2012-13 season, especially in the elderly (-11% and 9%) [10,11], even though the vaccine and epidemic strains matched. By contrast, during the 2013-2014 season, influenza A(H1N1)pdm09 virus caused major epidemics in the United States (U.S.) and Canada, and high VE (60% to 75%) against influenza A (H1N1)pdm09 virus was reported [12,13]. The vaccine and epidemic strains also matched in the 2013-2014 season. However, since there was a marked change in VE between the 2012-2013 season and 2013-2014 season because of the difference in epidemic influenza subtypes, it has become very important to estimate VE every season to monitor the performance of current influenza vaccine.
Although in many countries influenza vaccination is recommended starting at 6 months of age, there are few studies evaluating the effectiveness of trivalent inactivated influenza vaccine (TIV) in young children. TIV has been reported not to reduce the influenza A infection attack rate in 6-24 month-old children [14]. By contrast, a recent study revealed TIV was highly effective in children aged <24 months [15].
Indirect protection of the elderly by influenza vaccination of children and adults has recently been highlighted [16][17][18][19], and the indirect protection of the elderly is now thought to be more effective than direct protection. Assessing VE in children, especially in schoolchildren, is essential to achieving indirect protection of the elderly, because schools are the most efficient amplifiers of influenza epidemics in the community. The U.S. and Canada are aiming for universal immunization ranging from children 6 months of age to the elderly [20].
Influenza A(H1N1)pdm09 was the main strain in the 2013-2014 season in Japan for the first time in 3 seasons [21]. Other epidemic viruses were influenza B and subtype A(H3N2). The number of patients/week/sentinel (epidemic index) exceeded 1.0 at the national level (a sign of the start of the epidemic season) in week 51 of 2013, and it was maintained at that level for 21 weeks till week 19 of 2014. The epidemic peaked in week 5 of 2014, when the incidence was 34.4 cases/sentinel.
In the 2013-2014 season the prefectural and municipal public health institutes reported a total 8230 isolation/detections. The influenza viruses isolated/detected in the 2013-2014 season consisted of A(H1N1)pdm09 (43%), subtype A(H3N2) (21%), and type B (36%). A(H1N1) pdm09 became dominant for the first time since the 2010-2011 season [21]. The ratio of Yamagata lineage type B viruses to Victoria lineage type B viruses was 7:3.
It is easy to perform a very large VE study in Japan, because all children with influenza-like illness who are seen in hospitals and clinics are tested by an influenza rapid detection test (IRDT) [28]. In this study we used the results of the IRDT as a basis for estimating VE by a test-negative control design in children who had received TIV during the 2013-2014 season.

Study enrollment and location
Children (6 months to 15 years of age) with a fever of 38°C or over who received an IRDT in an outpatient clinic of one of 22 hospitals in Japan (in Gunma, Tochigi, Saitama, Tokyo, Chiba, Kanagawa, and Shizuoka prefectures) between November 9, 2013 and March 31, 2014 were enrolled in this study.

Case and control patient identification
All participants who visited any of 22 hospitals located in the Kanto region in Japan between November 6 2013 and March 2014 were included in the study. Case patients were identified as patients who were IRDT-positive, and control patients were identified as patients who were IRDT-negative during the same period. Both the case patients' and control patients' medical charts were reviewed, and information regarding symptoms, influenza vaccination, number of vaccine doses (one or two), influenza complications and hospitalizations, gender, age, comorbidities, and treatment with neuraminidase inhibitors (NAIs) was collected and recorded. We excluded children for whom definite information on influenza vaccination was unavailable.
In Japan, two 0.25 ml doses of vaccine 2 to 4 weeks apart are recommended for children 6 months to 2 years of age, and two 0.50 ml doses of vaccine 2 to 4 weeks apart are recommended for children 3-12 years of age. Only one 0.5 ml dose of vaccine is recommended for children over 13 years of age and over. The children in the vaccinated group in this study were immunized at our hospitals or elsewhere.
Test-negative case-control design VE was estimated by a test-negative case-control design in which a patient who presented with a fever of 38°C or over and was IRDT-positive for influenza virus was considered a case, and a patient who presented with a fever of 38°C or over and was IRDT-negative was considered a control. VE was defined as "1-OR (odds ratio)", and OR was calculated as (no. influenza-positive among vaccinated patients x influenza-negative among unvaccinated) ___________________________________________ (no. influenza-negative among vaccinated patients x no. influenza-positive among unvaccinated patients).
We calculated VE and adjusted VE as shown in Statistical Analysis.

Statistical Analysis
The statistical analysis was performed by using the SPSS. Ver. 22 software program (IBM, USA) and "Excel Tokei (Statistics) 2012 for Windows" software program (Social Survey Research Information Co., Ltd., Tokyo, Japan). VE was adjusted for age group (6-11 months, 1-2 years, 3-5 years, 6-12 years and 13-15 years), comorbidity (yes or no), area of the Kanto region of Japan, i.e., (north area: Gunma Prefecture and Tochigi Prefecture; middle area: Saitama Prefecture, Tokyo Prefecture, and Chiba Prefecture; and south area: Kanagawa Prefecture and Shizuoka Prefecture), and month of onset of illness.
We also estimated VE according to the number of doses of vaccine administered, the phase of the season, and area of the Kanto region where the hospitals were located, and we assessed VE in preventing hospitalization. The Breslow Day test was used to assess the homogeneity of the odds ratios in several 2 × 2 contingency tables.

Ethics
This study was approved by The Keio University Ethics Committee in 2013 (No. 20130216) and the Institutional Review Board (IRB) at each hospital. Eligible patients and their guardians (usually parents) were informed about the study objects and methods verbally at the outpatient departments. We recorded the necessary information from patients and the guardians using a standardized questionnaire sheets, when we obtained the consent to be enrolled to this study. The requirement for obtaining written consent was waived by IRBs because testing patients with IRDT is a standard practice in Japan.

Characteristics of the enrollees
A total of 4970 children were enrolled in this study, but 243 were subsequently excluded from the analysis for the following reasons: 117 were <6 months old or >15 years old; 73 had a fever <38°C; 22 were examined twice during one episode (the data for the first visit for each patient was deleted); 30 had an unclear influenza vaccination history; and since one was both influenza A and B positive, there was a possibility of being false-positive. Of the remaining 4727 patients who were eligible for the analysis in this study (Table 1). 876 had influenza A, 66 of whom were confirmed to have A(H1N1)pdm09 infection, the other 810 patients had influenza A subtype unknown; and 1405 patients had influenza B. Of the 4727 subjects of the analysis, 2446 were IRDT-negative. Table 1 shows the characteristics of the enrollees. The comorbidities consisted of respiratory comorbidities (n = 422), including asthma, neurological comorbidities (n = 128), including epilepsy, cardiac comorbidities (n = 57), allergic comorbidities (n = 49), renal comorbidities (n = 30), endocrinological comorbidities (n = 28), immunological comorbidities (n = 4), and other comorbidities (n = 96).

VE against influenza
Influenza vaccine was effective against influenza virus infection overall (Table 2). It was more effective against influenza A infection than against influenza B infection (62% vs. 32%; p < 0.001, Breslow-Day test) and very effective against A(H1N1)pdm09, against which its VE was as high as 77% (95% confidence interval [CI]: 59 to 87).

Number of doses of vaccine
Two doses of influenza vaccine provided better protection against influenza A infection than only one dose in the group of children 6 months to 12 years of age ( Table 3). The OR of two doses versus one dose was 0.72 (95% CI: 0.52 to 0.99). No significant difference in protection between the two doses was found against influenza B infection.

Protection against hospitalization
Influenza vaccination was effective in preventing hospitalization (Table 4), especially for influenza A virus infection (76%, 95% CI: 51 to 88). VE was as high as 90% (95% CI: 54 to 98) in preventing hospitalization for influenza A(H1N1)pdm09 infection. By contrast, the TIV was not effective in preventing hospitalization for influenza B virus infection. The reason for hospitalization was known in 14 (15%) of the 92 hospitalizations, and it was pneumonia in 5 cases, encephalopathy in 4 cases, asthma in 2 cases, mild consciousness disturbance in 1 case, seizure clusters in 1 case, and co-infection with Mycoplasma pneumoniae in 1 case.

VE by month of onset of illness
VE was lower in the late phase of the influenza epidemic (Table 5), decreasing from 59% (95% CI: 49 to 66) in the 3-month period November, December, and January to 39% (95% CI: 29 to 47) in the 2-month period February and March. VE against influenza B was only 31% (95% CI: 19 to 40) in the 2-month period February and March. The number of influenza B patients was much higher in February and March, i.e., in the late phase of the epidemic.

VE according to area and hospital
There were no significant differences between VE against influenza A according to area, but VE against influenza B was significantly higher in the north area than in the middle area or

Discussion
In this large study of over 4700 children 6 months to 15 years of age the overall influenza VE for prevention of laboratory-confirmed medically attended influenza illness was 46% (Table 2). High VE was shown in the influenza A group (63%), and VE was as high as 77% in the group with confirmed A(H1N1)pdm09 infection. Our results were consistent with reports in other countries where A(H1N1)pdm09 was the main epidemic virus [12,13]. In the 2013-2014 season in Japan, both A(H3N2) and A(H1N1)pdm09 were co-circulating, but A(H1N1)pdm09 was the dominant strain [21]. VE against influenza A was over 72% in the group of children 1-5 years of age (72% in the 1-2-year-old group, 73% in the 3-5-year-old group) and was higher than VE in the 6-12-yearold group (58%) ( Table 2). By contrast, VE against A(H1N1)pdm09 increased with the age groups, from 67% in the 1-2-year-old group, to 84% in the 3-5-year-old group, and 90% in the 6-12-year-old group. Most of the children in the 6-15-year-old group had been infected in the A(H1N1)pdm09 pandemic in 2009-2010 [30] and probably had sufficient residual immunity against A(H1N1)pdm09 from five years before. The pre-epidemic measurement of hemagglutination inhibition (HI) titers showed that 60%-70% of the children 5-15 years of age in Japan had HI titers to A(H1N1)pdm09 that were over 1: 40 [31]. Because no antigenic changes in A (H1N1)pdm09 have been reported, the majority of children over 6 years of age who were IRDT-positive for influenza A probably had an A(H3N2) infection, and that would have led to the lower effectiveness against influenza A in this age group. In other studies VE has been estimated to be low in patients infected with A/H3N2 [8,9]. VE has not been demonstrated in children 13-15 years of age, because the number of children tested was small and/or they had mainly contracted influenza A/H3N2. VE against influenza B in the present study was low, only 26% (Table 2). Lower vaccine efficacy against influenza B than against influenza A has been postulated in Japan, especially in young children [4]. Low VE against influenza B probably reflects the characteristic epidemic pattern of influenza B in Japan: influenza B virus epidemics usually start in February, after the end of the influenza A epidemic, and continue through March to April. Since children in Japan usually receive influenza vaccine in October and November, the level of antibody titer generated in response to the vaccine may be lower in March when influenza B epidemics peak. Our data showed that VE against influenza A and B was lower in February and March ( Table 5). The lower VE against influenza B was probably attributable to waning immunity in response to the vaccine, as has been reported in regard to the effectiveness of vaccines against influenza A H3N2 [26,32], and the low VE against influenza B was partly attributable to the mixed epidemic of influenza B lineages, i.e., of B/Victoria and B/Yamagata, in the 2013-2014 season in Japan [21], although B/Yamagata was both the main epidemic strain and the vaccine strain. Blineage mismatches present a greater obstacle to vaccine efficacy in children [3]. In the near future, quadrivalent vaccines that contain both B-lineage antigens will be introduced in Japan, thereby reducing B-lineage mismatches. Low HI titer responses to influenza B strain are of particular concern in children who receive influenza vaccine [3], because influenza B infection is common in children and as serious clinically as influenza A infection. NAIs are also known to be less effective against influenza B [33]. In this study influenza vaccine was not effective against either influenza A (30%, 95% CI -85, 74) or influenza B (-45%, 95% CI -684, 75) in 6-to 11-month-old infants, but it was effective in children over 1 year old ( Table 2). Indirect protection attributable to the mass vaccination program of schoolchildren in Japan in the 1960s to 1980s protected young children 1-4 years of age against influenza encephalopathy [17,19]. Young children did not receive influenza vaccine in the 1960s to 1980s. On the other hand, traditional cohort studies to estimate VE may have yielded excessively high VE for infants. One study reported 42%-69% VE in infants [34]. Since siblings and/or parents in families in which an infant has received influenza vaccine have usually also received influenza vaccine, thereby providing highly effective indirect protection to their infants whether indirect protection exists in the background should be noted when interpreting reports on VE in children.
Two doses of influenza vaccine have been reported to be necessary to provide sufficient protection in children, especially in young children [6,7,35,36]. However, the results of the present study showed that only one dose of influenza vaccine was effective in protecting against influenza A (52%) and influenza B (27%). Two doses of influenza vaccine were needed to optimize protection only against influenza A in the present study (one dose vs. two doses, OR = 0.72) ( Table 3). However, a second dose did not have any additive effect against influenza B.
A meta-analysis of VE in children showed no convincing evidence that influenza vaccine can reduce mortality, hospitalizations, or serious complications [37]. However, the results of the present study demonstrated that influenza vaccination was effective in reducing hospitalization of children infected with influenza A infection by as much as 76% and of children infected with A(H1N1)pdm09 infection by as much as 90% (Table 4). On the other hand, no effectiveness in preventing hospitalization for influenza B was shown. By contrast, in the 2002-2008 seasons VE for preventing hospitalization with influenza A and B in children 6 months to 5 years old was reported to be 71% and 72% [34]. There is a recent report of a study showing that influenza vaccination was associated with a three-quarters reduction in the risk of lifethreatening influenza illness in children 6 months to 17 years of age [38]. Moreover, VE against hospitalization with laboratory-confirmed influenza A and B has been estimated to be 61.7% [39].
The results of this study showed clear differences between VE against influenza B according to area (Table 6). In the 2013-2014 season, the ratio of Yamagata lineage viruses to Victoria lineage viruses was 7:3 nationwide [21]. Since the vaccine virus was the Yamagata lineage in the 2013-2014 season, Yamagata lineage viruses may have been more epidemic in the north area than in the other areas of the Kanto region.
The limitations of this study need to be considered. Unlike most previous test-negative case-control design studies based on PCR data, this study was based on the results of IRDTs. Suzuki et al. [40] found no difference between VE estimated on the basis of IRDT results on the basis of PCR data. However, the use of IRDTs in test-negative studies may result in underestimations of VE. Orenstein et al. [41] reported that the bias toward underestimating true VE introduced by low test specificity increases and lesser degree, by low test sensitivity. In a casecontrol study on influenza VE based IRDT results [5], it was postulated that virus shedding is greatest during the first days of influenza infection and therefore, patients who were tested during the early phase of the illness were more likely to test positive. In our study, about 90% of the enrollees received IRDT within 48 hours of onset of illness (Table 1).
On the other hand, the sensitivity of IRDTs is low when patients are tested within 12 hours after the onset of influenza illness because of the low virus infection titers in the upper respiratory tract in the early phase of infection [42]. We therefore selected patients who were tested 12 hours or more after the onset of influenza illness, and when we estimated the VE of the influenza vaccine in that group of patients, we found that it was slightly higher (+4% to +8%) ( Table 2). When estimating VE by a test-negative case-control design based on IRDT results, it is better to select patients tested 12 hours or more after the onset of influenza-like illness.
Almost all children with influenza-like illness in Japan receive an IRDT, and if positive, they are treated with NAIs [28]. This diagnosis and treatment system established in Japan worked well during the pandemic caused by A(H1N1)pdm09 and resulted in the very low fatality rate in Japan during that pandemic [30]. In future epidemics in Japan, VE estimated by a test-negative case-control design based on IRDT results from various parts of Japan will be reported rapidly. The large number of cases in Japan makes it possible to estimate VE with considerable precision, and the most appropriate vaccination policy should be established based on the data obtained.