Influenza epidemiology and influenza vaccine effectiveness during the 2014–2015 season: annual report from the Global Influenza Hospital Surveillance Network

The Global Influenza Hospital Surveillance Network (GIHSN) has established a prospective, active surveillance, hospital-based epidemiological study to collect epidemiological and virological data for the Northern and Southern Hemispheres over several consecutive seasons. It focuses exclusively on severe cases of influenza requiring hospitalization. A standard protocol is shared between sites allowing comparison and pooling of results. During the 2014–2015 influenza season, the GIHSN included seven coordinating sites from six countries (St. Petersburg and Moscow, Russian Federation; Prague, Czech Republic; Istanbul, Turkey; Beijing, China; Valencia, Spain; and Rio de Janeiro, Brazil). Here, we present the detailed epidemiological and influenza vaccine effectiveness findings for the Northern Hemisphere 2014–2015 influenza season. Electronic supplementary material The online version of this article (doi:10.1186/s12889-016-3378-1) contains supplementary material, which is available to authorized users.


Introduction
Every year, between 5 % and 10 % of adults and 20 -30 % of children have symptomatic influenza illness [1,2], and 3 to 5 million individuals suffer from severe influenza, leading to 250,000 to 500,000 deaths [2][3][4]. Influenza illness can result in hospitalization and death, mainly among high-risk groups but also in a substantial proportion of previously healthy individuals [5]. In recent years, especially after the 2009 pandemic season, influenza surveillance has been expanded, as recommended by the World Health Organization (WHO), to include additional epidemiological data [6].
The Global Influenza Hospital Surveillance Network (GIHSN) is an international public-private collaboration initiated in 2012 by Sanofi Pasteur and the Fundación para el Fomento de la Investigación Sanitaria y Biomédica de la Comunitat Valenciana (FISABIO), a regional public health institution in Valencia, Spain. The aim of the GIHSN is to improve understanding of influenza epidemiology to better inform public health policy decisions. It is the first global network focusing exclusively on severe cases of influenza requiring hospitalization. The GIHSN runs a prospective, active surveillance, hospital-based epidemiological study to collect epidemiological and virological data for the Northern and Southern Hemispheres over several consecutive seasons. A standardised protocol and standard operating procedures are shared between sites allowing comparison and pooling of results [7]. The GIHSN is coordinated by FISABIO and is made up of several country sites affiliated with national health authorities. Each site coordinates several hospitals in its region. The network currently includes 27 hospitals coordinated by 7 sites in 6 countries (St. Petersburg and Moscow, Russian Federation; Prague, Czech Republic; Istanbul, Turkey; Beijing, China; Valencia, Spain; and Rio de Janeiro, Brazil).
The surveillance data collected by the GIHSN are used to describe the circulating strains related to severe disease, estimate the burden of severe influenza disease, and evaluate the benefit of influenza vaccination to prevent severe disease. Results have been published from the network's first two seasons, 2012-2013 [5,8] and 2013-2014 [9]. In this report, we describe the influenza epidemiology and vaccine effectiveness results from the GIHSN during the 2014-2015 influenza season. Complete data from the Southern Hemisphere was not available at the time of the meeting or during analysis and writing, so only data provided by sites in the Northern hemisphere during the 2014-2015 season are presented.

Summary of overall methodology
As described in detail elsewhere [7], patients admitted in the participating hospitals are included, after written consent, if they are residents in the predefined hospital's catchment area, present with an acute illness possibly related to influenza, are not institutionalised, and the onset of symptoms was within 7 days of admission. Swabs are collected from patients meeting the inclusion criteria and tested by reverse transcription-polymerase chain reaction (RT-PCR) for influenza (Fig. 1). Influenza-positive samples are sub-typed by RT-PCR to identify A(H1N1)pdm09, A(H3N2), B/Yamagata-lineage, and B/Victoria-lineage strains. Vaccine effectiveness is assessed using a testnegative design in which vaccine coverage is compared between admissions with and without laboratoryconfirmed influenza.

Epidemiological analysis
Epidemiological and virological data were collected from 7 coordinating sites and a total of 27 hospitals in 6 countries (Additional file 1). Briefly, eligible admissions included non-institutionalised residents in the predefined catchment areas of the participating hospitals, hospitalised in the last 48 h, and with presenting illness potentially associated with influenza (Additional file 2 and Additional file 3). The study activities were performed over influenza circulation periods defined using pre-specified criteria (Additional file 3). Nasopharyngeal swabs (all subjects), pharyngeal swabs (subjects ≥14 years) or nasal swabs (subjects <14 years) were tested by semiquantitative RT-PCR for influenza A (subtypes H3 and H1pdm09) and B (Yamagata and Victoria lineages). The distribution of hospital admission according to RT-PCR result was described by site and risk group. Secondary outcomes included hospital admissions by subtype for influenza A(H1N1)pdm09, A(H3N2), and B-lineage, by site and risk group. The significance of differences among groups or categories was estimated by the likelihood ratio test, t-test, or nonparametric tests as required. A P-value <0.05 was considered to indicate statistical significance. To describe the major determinants for admission with influenza (vs. influenza-negative admission), a stepwise logistic regression model was fitted by including all risk factors at P < 0.2. Adjusted odds ratios (aORs) for RT-PCR-positive vs. RT-PCR-negative admissions in the presence of major risk factors of interest were Fig. 1 Overview of the methodology used by the GIHSN estimated by multivariate logistic regression using minimal sufficient adjustment sets of covariates identified as confounders by causal diagrams. To account for the possible effect of study site, data were fitted to a random effects logistic regression model including site as a cluster variable. Likelihood ratio tests were used to check for the potential effect of clustering by site [10]. The adjusted effect of site in the probability of influenza with admission was estimated. Heterogeneity in the effects of risk factors by influenza strain and site were quantified using the I 2 test. Heterogeneity was defined as an I 2 > 50 % [11,12]. Further details are provided elsewhere [5,7,8].

Influenza vaccine effectiveness analysis
Influenza vaccine effectiveness (IVE) was estimated as (1 ˗ OR) × 100, where the OR compared the vaccine coverage rate between influenza-positive and influenza-negative patients. Patients were considered vaccinated if they had received the current season's influenza vaccine at least 14 days before symptom onset. The types of vaccines used at each site are summarised in Additional file 4. IVE overall (irrespective of vaccine type) was determined in patients who had been swabbed within 7 days of the onset of ILI symptoms. Records for which outcome, exposure, or confounding variables were missing were excluded from the multivariate IVE analyses. The adjusted IVE was estimated by logistic regression using a random effects model with study site as a shared parameter for the pooled analysis and including week of symptom onset as a continuous variable, and age group, sex, hospitalisation in the previous 12 months, presence of chronic conditions, and smoking habits as potential confounding factors. A P-value <0.05 was considered to indicate statistical significance. Heterogeneity in IVE estimates was assessed using the I 2 . Potential sources of heterogeneity, including coordinating site, age, and influenza subgroup were examined in ad-hoc analyses. Heterogeneity was defined as low if I 2 statistic <25 %, moderate if 25 -49 %, and high if ≥50 %. Further details of the methodology are described elsewhere [8].

Results
Epidemiology of influenza in the GIHSN during the 2014-2015 influenza season Patients included in the epidemiology analysis Twenty thousand five hundred fifty-one eligible admissions were identified between November 16, 2014 and May 23, 2015, of which 9614 met the selection criteria and were included (Table 1). Based on RT-PCR, 2177 (23 %) were positive for influenza. Major reasons for exclusion included no ILI symptoms before admission (15 %), previous admission fewer than 30 days from the current episode (13 %), admission more than 7 days after the onset of symptoms (6 %), recruitment outside periods of continuous admissions with influenza (6 %).

Influenza viruses identified in admissions
In the 2177 included influenza-positive patients, A(H3N2) (n = 1243; 57 %) was the most commonly identified type of influenza, followed by B/Yamagatalineage (n = 623; 29 %), A(H1N1)pdm09 (n = 121;6 %), A not subtyped (106; 5 %), B lineage not determined (n = 97; 5 %), and B/Victoria-lineage (n = 11; 0.5 %) ( Table 1 and Fig. 2a and b). Mixed influenza infections were found in 24 cases. Influenza B lineage not determined were considered B/Yamagata-lineage for 39 cases in Turkey and four in Valencia. Due to the mixed circulation of B/Yamagata and B/Victoria lineages in Moscow, this assumption was not applied at that site to cases where B-lineage was not determined.

The 2014-2015 influenza season at the GIHSN sites
Influenza was detected over a span of 27 weeks, from week 47 of 2014 to week 20 of 2015, with the peak at week 7 of 2015 (Fig. 2). The earliest start of the influenza season was reported in Beijing, where influenza-positive admissions occurred over a span of 23 weeks in two waves, the first due to A(H3N2) and the second due to B/Yamagata-lineage (Fig. 2). The latest influenza-positive admission was in St. Petersburg, where continuous weekly admissions with influenza were observed over a span of 19 weeks.

Main characteristics of included patients
Overall, all age groups were represented. Approximately one-third of included admissions were patients less than 5 years of age, one-third were 5 to 64 years of age, and one-third were 65 years of age or older ( Table 2). More than half of the included patients were male (n = 5417; 56 %). Most (n = 5867; 61 %) did not have an underlying chronic condition, and most (n = 6756; 70 %) had not been hospitalised in the 12 months before the current episode.

Site-related characteristics of included patients
Patients included in St. Petersburg were younger than patients included at other sites ( Table 2). The difference    For Turkey and Valencia, all B not subtyped were assumed to be B/Yamagata lineage based on virus circulation at these sites. This assumption was not applied for Moscow because of a mixed pattern of influenza B circulation in age of included patients was especially marked when comparing St. Petersburg with the Czech Republic and Valencia. Patients were most frequently young adults in Moscow and the Czech Republic. Ages were homogeneously distributed in Turkey and Beijing (P = 0.9480). By contrast, in Valencia most (62 %) admissions were in elderly patients (≥65 years).
Patients without comorbidities represented 88 % of admissions in St. Petersburg, 89 % in Moscow, 71 % in Beijing, 37 % in Czech Republic, 30 % in Turkey, and 33 % in Valencia. Of the different chronic conditions, cardiovascular disease, respiratory disease, and diabetes were the most common, and their relative importance at each site corresponded to the proportion of patients with one or more underlying chronic condition.
In Moscow, among admissions in patients with known risk factors for influenza, pregnant women represented the majority of admissions (n = 291; 96 %). Obese patients represented 10 -15 % of admissions in St. Petersburg, Moscow, the Czech Republic, and Beijing, whereas 18 % in Turkey and 26 % in Valencia were obese. The proportion of who never smoked ranged from 47 -53 % in adult (≥18 years) admissions and was similar across sites (p = 0.1520). The overall proportion of current smokers, however, differed, with the highest rate (34 %) in St. Petersburg and Moscow (33 %), followed by Czech Republic (25 %), Beijing (18 %), Valencia (16 %), and Turkey (13 %) (p < 0.0001). For elderly adults, functional impairment status was mild or minimal in 83-97 % of included admissions at all sites except Beijing, where 30 % of admissions in elderly patients had moderate to severe functional impairment. Rates of influenza vaccination were below 5 % for all sites except Beijing (11 %) and Valencia (48 %).

Heterogeneity between sites
The proportion of samples with positive results differed between sites, from as low as 12 % for Turkey to as high as 75 % for the Czech Republic (Table 1; p < 0.0001 by test of homogeneity for equal odds). This difference persisted after excluding pregnant women and excluding the two sites with extreme

Risk of admission with influenza according to age and sex and variability by influenza virus
Influenza positivity was related to age. Overall, influenza-positive admissions tended to be older than influenza-negative admissions (Table 3). Admissions positive for A(H1N1)pdm09 were younger than those negative for influenza, those positive for A(H3N2), and those positive for B/Yamagata-lineage. Also, admissions positive for A(H3N2) were older than influenza-negative admissions, those positives for A(H1N1)pdm09, and those positive for B/Yamagata-lineage (Table 3 and Fig. 3). After adjusting for sex, occupational class, comorbidity, influenza vaccination, time to swab, and the clustering effect of site, heterogeneity due to strain was significant for admissions in subjects ≥5 years of age due to a decrease in aOR with age for admission with A(H1N1)pdm09 (Table 4 and Additional file 7). After excluding admissions with A(H1N1)pdm09, the aOR for admission with influenza was homogeneous for elderly patients but heterogeneous for patients 5-64 years of age (I 2 = 75-77 %) due to a higher aOR for admissions with B/Yamagata-lineage than for A(H3N2) (Additional file 7).

Risk of admission with influenza according presence of comorbidity
Similar proportions of influenza-positive admissions (882/2177; 41 %) and influenza-negative admissions (2865/7437; 39 %) had one or more chronic underlying condition (p = 0.0940) ( Table 3). After excluding pregnant women, 42 % of influenza-positive admissions had comorbidity compared to 39 % of influenza-negative admissions (p = 0.006) (data not shown). The aOR for admission with influenza was 1.5 (95 % CI, 1.3-1.7) for patients with comorbidities, although the values were heterogeneous by strain (I 2 = 63 %) (  Fig. 4, and Additional file 9). The aOR was heterogeneous for the risk of admission with influenza in patients with COPD (aOR 1.4 [95 % CI, 1.2-1.7]; I 2 = 58 %) due to lower aOR for admission with A(H1N1)pdm09 (Additional file 9). Point values for aORs were above 1.0 for admission with influenza for patients with renal, neuromuscular, or autoimmune disease, but 95 % CIs overlapped 1.0. In patients with active neoplasms, the overall aOR for influenza-positive admission was heterogeneous and not significant (I 2 = 56 %; aOR = 1.3 [95 % CI, 0.9-1.8]),    The probability of laboratory-confirmed influenza was higher in included pregnant women than included same age non-pregnant women (54 % vs. 24 %; p < 0.0001; data not shown). After taking into account clustering by site (and not considering data from Beijing), the crude OR of admission with influenza was 3.5 (95 % CI, 2.2-5.3) ( Table 4). This crude estimated OR was higher in pregnant women with associated comorbidity (OR 7.1 [95 % CI, 3.1-16.2]), with moderate evidence of an interaction between comorbidity and pregnancy before adjustment (p = 0.0659) and a significant interaction after adjustment (p < 0.0001). Taking into account the modifying effect of associated comorbidity, the aOR for admission with influenza in pregnant women was 4.3 (95 % CI, 2.7-6.9) in presence of associated comorbidity and 2.1 (95 % CI, 1.4-3.0) for pregnant women with no comorbidity. In both cases, the values were homogenous (I 2 = 0 %) for A(H3N2), A(H1N1)pdm09, and B/Yamagata-lineage infections.
The probability of admission with influenza was higher in all three trimesters for pregnant women without associated comorbidities than for non-pregnant women in the same age group without comorbidity. In pregnant women with comorbidities, the risk of admission with influenza was highest in the first trimester ( Fig. 5 and Additional file 10).

Risk of admission with influenza and complications by strain
Intensive care unit (ICU) admissions, extracorporeal membrane oxygenation, and mechanical ventilation were more frequent for influenza-negative than for influenzapositive admissions (p ≤ 0.002), whereas rates of inhospital death were similar (p = 0.3460) ( Table 5). By strain, the point estimate of rates of ICU admission and extracorporeal membrane oxygenation were higher in admissions with A(H1N1)pdm09, although differences were not significant. In contrast, rates of in-hospital death were significantly higher in admissions with A(H3N2) (p = 0.0080). Less than 4 % of admissions in these categories experienced a severe outcome. Finally, length of stay did not differ between influenza-positives and influenza-negative admissions for influenza overall or between strains ( Table 5).
Exacerbation of chronic obstructive pulmonary disease, respiratory failure, exacerbation of asthma, and bronchiolitis were more frequently reported for influenza-positive admissions than for influenza-negative admissions (Table 5). These were associated with A(H3N2), except in the case of bronchiolitis, where the proportions for admission with all three strains (A(H3N2), A(H1N1)pdm09, and B/Yamagata-lineage) were higher than the proportion for influenza-negative  Table 5). Figure 6 shows the estimated marginal probabilities by strain and age for severe outcomes after adjusting by sex, comorbidity, calendar time, age, and clustering by site. We found several non-significant associations: A(H1N1)pdm09 was associated with intensive care unit admission and shock; A(H3N2) was One or more underlying conditions or individual comorbidities vs. no comorbidity. aORs were adjusted for sex, occupational social class group, obesity, influenza vaccination, time to swab, calendar time, and site as a clustering factor e aOR adjusted for sex, age, occupational social class group, influenza vaccination, time to swab, calendar time, and site as a clustering factor f Women 15-45 years of age included in Moscow, St. Petersburg, Czech Republic, Turkey and Valencia. aOR adjusted for smoking habits, time to swab, calendar time, comorbidity, and site as a clustering factor. For results stratified by comorbidity, aORs were adjusted by the same covariates and were estimated taking into account the interaction between pregnancy and comorbidity associated with an increased probability of COPD exacerbation, respiratory failure, cardiovascular complications, and death; B/Yamagata-lineage was related to respiratory failure; and all three strains were related to death at both extremes of age (Fig. 6). We found similar non-significant associations for complications when influenza-negative admissions were included (Additional file 11). Overall, 1709 of 2002 (85 %) influenza vaccinations among study patients were both self-reported and confirmed from registries. Self-report captured 156 of 2002 vaccinations (8 % overall, 67 % in Moscow, 8 % in Turkey, 1 % in Beijing, and 7 % in Valencia; data not shown). Another 137 patients (7 % overall, 12 % in St. Petersburg, 42 % in Turkey, and 7 % in Valencia; data not shown) with vaccination records failed to self-report vaccination.

Influenza vaccine effectiveness in the GIHSN during the 2014-2015 influenza season Patients included in the influenza vaccine effectiveness analysis
The proportion of participants with comorbidity was significantly higher in vaccinated than in nonvaccinated admissions (82 % vs. 30 %, p < 0.001) ( Table 7). Vaccination was also more common among elderly (median age = 79 years for vaccinated patients vs. 15 years for non-vaccinated patients, p < 0.001), obese patients (28 % obese for vaccinated patients vs. 15 % for non-vaccinated patients, p < 0.001), elderly patients with impairment or minimal functional impairment (28 % impaired for vaccinated patients vs. 15 % for non-vaccinated patients, p < 0.001), patients with outpatient visits (83 % for vaccinated patients vs. 72 % for non-vaccinated patients, p < 0.001), and patients admitted to a hospital in the previous 12 months (38 % for vaccinated patients vs. 29 % for non-vaccinated patients, p < 0.001) ( Table 7). Three

Influenza vaccine effectiveness
Against all-age influenza-related hospitalisation, the overall crude IVE was −1 % (95 % confidence interval   (Table 6). IVE estimates were consistently higher for recipients of the 2012-2013 influenza vaccine, the 2013-2014 influenza vaccine, or both vaccines than for recipients of only the current season's vaccine, although confidence intervals overlapped (Additional file 12).
Statistical heterogeneity across sites in the estimates of IVE against influenza-related hospitalisation was relatively low, with site-specific adjusted point estimates ranging from -27 -35 % [I 2 = 0 %; P = 0.835) (Additional file 13).
Sensitivity analyses were performed to assess the effects of excluding pregnant women, participants vaccinated within 14 days before symptom onset, and without medical vaccination records. In all cases, IVE estimates remained similar to those of primary analysis (Additional file 14). Further sensitivity analyses using various statistical methods to account for potential data

Discussion
According to data collected by active surveillance within the GIHSN sites, the 2014-2015 influenza season was characterised by a predominance of A(H3N2) and B/Yamagata-lineage, and to a lesser extent, A(H1N1)pdm09, while B/Victoria-lineage was relatively rare. Reports of severe influenza, defined as hospitalisation with laboratory (i.e., PCR)-confirmed influenza, spanned 6 months and affected all ages, although influenza-related admissions were most common in older individuals. Among patients with laboratory-confirmed influenza, those with A(H1N1)pdm09 were younger than those with A(H3N2) or B/Yamagata-lineage, whereas those with B/Yamagata-lineage were most frequently young and middle-aged adults. This pattern of influenza circulation is consistent with that reported by the WHO [13]. Likewise, the age distribution of the A(H1N1)pdm09, A(H3N2) and B/Yamagata-lineage strains agrees with others' reports [14,15]. According to our data, comorbidity increased the risk of admission with influenza, irrespective of the strain involved. This was also the case for pregnant women. Furthermore, the combination of pregnancy and comorbidity increased the risk of admission several-fold, suggesting an interaction. Remarkably, however, nearly 60 % of eligible admissions with influenza were patients without known risk factors.
The probability of ICU admission and shock were higher in patients infected with A(H1N1)pdm09 than with other strains. Also, A(H3N2) infection was associated with respiratory failure and cardiac complications, whereas B/Yamagata-lineage was associated with an increased probability of respiratory failure. Influenza infection overall was associated with in-hospital death at both age extremes. These findings agree with other reports [15][16][17], although there may be differences in the absolute percentage of admissions with influenza in patients with comorbidity, patterns of severity, lengths of hospital stay, rates of ICU admission, use of supportive measures, or estimates of in-hospital death rates [15,18,19].
Although vaccination coverage was low at the participating sites (2.8-48 %; average 20.9 %), we found that vaccination conferred a low to moderate protective effect (adjusted IVE = 22 %). This protective effect was greater for adults ≥65 years of age than for adults <65 years of age and was greater for B/Yamagata-lineage than for A(H3N2).
The low influenza vaccine effectiveness for the 2014-2015 season is similar to others' reports and appears to be due mostly to a mismatch between the main A(H3N2) circulating strain and the vaccine strain [20][21][22][23]. Across all strains, the IVE was lower in young patients, although only age-specific estimates for A(H3N2) were significantly different due to few cases of B/Yamagata-lineage and A(H1N1)pdm09 and a higher IVE in patients vaccinated during the 2012-2013, 2013-2014, or both seasons than in those vaccinated during the 2014-2015 season, a finding also reported by others [24]. This lower IVE in young patients, however, contrasts with previous reports where the opposite was found [25]. Thus, there appears to be variability in the interference or protection conferred by vaccination in previous seasons. This could be explained by the differences between the various strains circulating in different seasons and their distance from the vaccine strains, combined with inhibition of the immunological response when the vaccine strains are similar to those in previous seasons' vaccines [26].

Limitations and considerations
Our results are to be interpreted with caution due to the heterogeneity and bias of multi-centric observational studies. We assumed heterogeneity in the circulating strains, socio-demographic diverse populations observed, their health care seeking behaviour, the characteristics of the different health care systems involved, the types of participating hospitals, and by calendar time along the season. We took account of this heterogeneity by thoroughly describing the season, the sites, and included admissions, as well as by quantifying the heterogeneity of our estimates. In this way, we are able to visualise the relative impact of the different influenza strains on diverse risk factors, including age, comorbidity, pregnancy, and obesity [12]. Furthermore, we restricted our analysis to periods with influenza circulation [27], took into account risk by calendar date [28], as well as the clustering effect of site [10] by adjusting and modelling and, finally, compared PCR-detected influenza-positive admissions with influenza-negative admissions. We consider this a reasonable approach for describing the effect of influenza in individuals according to their risk profile [29]. In addition, to reduce bias and to allow us to describe the severe consequences of community-acquired influenza, we accepted only data from patients admitted within 7 days of onset of ILI symptoms and for whom swabbing was performed within 48 h of admission.
Even with a large dataset as the one accrued annually by the GIHSN sites, small numbers are a limitation. Splitting the data by strain and risk group can decrease group sizes, so that sufficient power is available only for detecting large differences (i.e., OR ≥2). This limitation can be only dealt with by increasing the number of participating sites and by pooling data across influenza seasons. In fact, the GIHSN continues to grow, and data pooling across seasons is underway.
Most hospital studies rely on the criteria of the physician providing care for influenza confirmation and employ historical database searching [15,17,18,[30][31][32]. This combined with different case definitions and laboratory methods can complicate comparisons between sites and seasons and between different studies. Our approach of using active surveillance, a shared core protocol, and PCR confirmation of influenza avoids these limitations. This approach has very recently begun to be employed by others and for other respiratory viruses [33].