Background

Influenza is a major public health problem that can cause hospitalisations, and it is related with respiratory failures [1, 2]. The Global Influenza Hospital Surveillance Network (GIHSN) is an international public-private collaboration that started in 2012. The GIHSN goals are to improve understanding of influenza epidemiology, quantifying the circulation of the different types and subtypes of influenza, in order to measure the effectiveness of seasonal influenza vaccines and better inform public health policy decisions. We conduct a prospective, active surveillance, hospital-based epidemiological study that collects epidemiological and virological data from those sites that are included in the network. Each season results are presented in annual meetings and, since 2012, have been published [3,4,5,6], with the agreement of the Principal Investigators of all concerned sites. The implementation and data collection for the last season (2016–2017) was led by the Foundation for the Promotion of Health and Biomedical Research of Valencia Region (FISABIO), a regional public health institution in Valencia, Spain, and funded by the Foundation for Influenza Epidemiology. Fifteen sites in fourteen countries participated in the GIHSN in the season 2016–2017. Among them, there were 12 sites (St. Petersburg, Moscow, Kazakhstan, Czech Rep., Canada, Romania, Turkey, Spain, Tunisia, Suzhou/Shanghai, India and Mexico) from Northern Hemisphere countries not situated in the tropics and three sites (Ivory Coast, Peru and South Africa) from the tropics or the Southern Hemisphere. Since Peru and Ivory Coast only reported two positive cases for influenza in the influenza season, the analysis was performed without the data from these countries, and therefore, results are reported for 13 sites. A common core protocol and standard operating procedures are used for all participating sites, in order to allow comparisons among countries, and analyse results of all sites.

Methods

This study aims to determine the frequency of influenza-related hospitalisations in different countries, by circulating strains and age groups, to study risk factors for influenza-associated hospitalisations and estimate Influenza Vaccine Effectiveness (IVE) by site, age group and strain. Each site had one or more hospitals that recruited patients for the study, between October 2016 and May 2017 in Northern Hemisphere sites, except China, whose patients were recruited between June and September. For Southern Hemisphere sites, patients were recruited between May and November. Patients were included in the study if they presented any of the admission diagnoses included in the protocol, and only if they signed the informed consent to participate in the study. Among them, we selected for the study only those who were residents in the predefined hospital catchment’s area in the previous past 6 months, who were not institutionalised, who hadn’t been discharged from other hospitalisation in the last 30 days, and who had symptoms possibly related to influenza in 7 days or less prior to admission (Fig. 1). We also excluded patients who had previously tested positive for influenza in the current season, and also patients for whom the difference between the date of the onset of symptoms and the date of swabbing was 10 days or more (that is, those admitted after the 7th day after the onset of symptoms+maximum delay in swabbing). For patients 5 years old or more, they must also have complied with the Influenza-Like Illness (ILI) definition, detailed in European Centre for Disease Prevention and Control (ECDC) protocols, according to the decision of the Commission of the European Union of 8 August 2012 [7]. Patients enrolled outside the influenza epidemic period of each of the participating sites were also excluded. Influenza seasons were previously determined by each site, following recommendations of previous studies [8]. This methodology has been used in the GIHSN since the beginning of the network, and has been previously described [9]. For patients under 14 years old, nasal and/or nasopharyngeal swabs were collected, whereas, for patients 14 years old or more, pharyngeal and/or nasopharyngeal swabs were taken. Reverse transcription-polymerase chain reaction (RT-PCR) was used, according to each site’s protocol, in order to detect influenza virus; viral subtyping was performed in order to identify A(H1N1)pdm09, A(H3N2), B/Yamagata-lineage, and B/Victoria-lineage strains in the positive specimens.

Fig. 1
figure 1

Overview of the methodology used by the GIHSN

Full size image

We performed a test-negative study [10] in order to compare positives (cases) and negatives (controls) for influenza and estimate Influenza Vaccine Effectiveness (IVE). Odds Ratios were used to estimate IVE, comparing cases and controls of patients depending on the vaccination status. Patients were considered vaccinated if they received an influenza vaccine in the current season, at least 15 days before the onset of symptoms. Patients with contra-indication to influenza vaccination were excluded from the IVE analysis, but were included in the analysis regarding influenza circulation. Vaccination status was ascertained either by recall or by vaccination registries. Adjusted odds ratios (aOR) were calculated using a logistic regression model including sex, occupational social class, obesity status, pregnancy, underlying conditions, general practitioner (GP) consultations in last 3 months, smoking habits, days from onset of symptoms to swabbing as fixed effects, age and epidemiological week of admission using cubic splines, and site as a cluster variable, in order to consider sites variability [11]. IVE was calculated as (1-aOR) × 100. The same factors were used to adjust IVE by strain or age group. The variables relative to the Barthel Index (in patients 65 years old or older) and the previous hospitalisations in the last year were initially considered to be included in the model, but were excluded from the final model as they were not statistically significant considering all variables mentioned above. The model did not include the number of consultations at the GP in the last 3 months to estimate IVE in Canada, as this site did not provide information for this variable. Severe outcomes were also studied, defining them as an influenza positive patient admitted to ICU during the hospitalisation, or with COPD exacerbation, respiratory failure, any cardiovascular complication, shock or death during hospitalisation. Heterogeneity was studied, using the I2 test, and considering that heterogeneity was relevant if I2 ≥ 50% [12, 13].

Results

Included patients: distribution, characteristics and influenza positives and negatives

There were 21,967 eligible admissions between October 1, 2016 and November 9, 2017. However, only 10,140 patients complied with the conditions described above, and had laboratory results, hence only these were included in the analysis. Among them, 2895 (28.6%) tested positive for influenza, and 7245 (71.4%) tested negative for influenza (Table 1). The most common reason of exclusion was the fact that patients didn’t have ILI symptoms in the 7 days previous to admission. It is important to note that 2/3 of all included patients in the GIHSN came from 4 sites (St. Petersburg, Moscow, Canada and Valencia). These 4 sites also have the highest numbers of influenza positive cases, including 77.8% of all influenza positives in the GIHSN, and 84.3% of the A(H3N2) influenza positives among all participant sites. A (H3N2) was the predominant strain this season, being detected in 63.6% of all influenza positive cases (1840 patients), followed by B/Victoria, with 21.3% among the influenza positive cases (618 patients) (Table 1). Influenza A(H3N2) was detected throughout the season, whereas B/Victoria started to increase in the second week of 2017 in the Northern Hemisphere, and in the 31st week of 2017 in the Southern Hemisphere, approximately in the middle of the season in each Hemisphere (Fig. 2).

Table 1 Patients included and excluded in the current analyses, inclusion criteria and influenza laboratory results
Full size table
Fig. 2
figure 2

Influenza-associated admissions by epidemiological week and virus type/subtype

Full size image

In the Northern Hemisphere, there was a significant increase in the number of influenza cases in week #49 of 2016, with a peak in the number of positive cases during the second week of 2017 and starting to descend at the eighth week of 2017. Influenza B/Victoria started to increase clearly in the second week of 2017, as A(H3N2) started to descend. 70.3% of all influenza cases were positive for influenza A, whereas 29.7% were positive for influenza B, with a clear different distribution among sites.

A(H3N2) was predominant in all sites, except in Mexico, where the predominant strain was A(H1N1)pdm09, and Romania and India with a predominance of B/Victoria-lineage. Both B lineages circulated during this season, with geographical differences, so in Canada, Czech Republic, Turkey, Tunisia, Mexico and South Africa, B/Yamagata was more often detected, while the B/Victoria was elsewhere. Influenza B cases generally appeared as a second influenza wave (Fig. 3). In Valencia, no cases were positive for influenza B.

Fig. 3
figure 3

Admissions with influenza by site, epidemiological week and virus type/subtype

Full size image

Influenza B was mainly observed in the youngest, and was the predominant strain in the age group 5–17 years old. Among the two influenza B lineages, in general B/Victoria was detected more often than B/Yamagata, except in the age group 50–64 years (Fig. 4).

Fig. 4
figure 4

Percentages of influenza-associated admissions by age group and type/subtype

Full size image

The distribution of influenza cases among the age groups was clearly different among sites, but differences were mainly due to the characteristics of the participating hospitals for each site. Tunisia and Czech Republic only recruited patients 18 years old or older, while Suzhou/Shanghai only enrolled patients under 18 years old. In Moscow, the majority of influenza positives were pregnant women (which represented the 49.4% of the included patients), and therefore, the highest number of influenza positives among the different age groups was situated in the age group 18–49 years old in this site. Influenza positive cases were mainly found in patients 65 years old or older in Valencia and Canada, but 89.8% of the included patients from Canada were 50 years old or older. In St. Petersburg and South Africa, due to the characteristics of the patients of the participating hospitals (mainly children) there were more influenza positive cases in the youngest groups (Fig. 5).

Fig. 5
figure 5

Admissions with influenza by site, age group and virus type/subtype

Full size image

25.8% of the included patients were previously hospitalised in the same year and 36.6% of the included patients had at least one underlying condition, but this percentage varied among sites, in Canada, for example, more than 90% of the included patients had at least one underlying condition, whereas in St. Petersburg, this percentage was lower than 10% and in Turkey was 48.2%, but these percentages could be related to the age distribution of the included patients in each site. Among the different comorbidities, the most common were cardiovascular (20.7% of the included patients), diabetes (10.4%) and chronic obstructive pulmonary disease (COPD) (9.9%). Obesity was also found in more than 14% of the included patients, being more relevant in Canada (29.6%), Valencia (26.3%) and Czech Republic (23.4%). Moscow was the site with the highest number of pregnant women among all sites (800 pregnant in Moscow among 940 pregnant women in all sites), being 49.4% of the included patients in this site. In Kazakhstan, pregnant women represented 22.6% of the included patients. The Barthel Index in those over 65 years showed that almost 90% of these subjects were not dependent or had a mild dependence. 68.3% of the patients who tested negative for influenza were swabbed from 0 to 4 days after symptoms started, but this percentage was 78.4% for influenza positive cases (p-value< 0.0001).

Vaccination coverage differed among sites. Patients were considered as vaccinated if vaccination was at least 15 days before symptoms onset (Table 2). Targeted patients for vaccination criteria were different among sites (Additional file 1: Complementary Table S1). Vaccination coverage was 11.1% among the influenza positives and 18.4% among the influenza negatives overall. Cardiovascular diseases, renal impairment, chronic obstructive pulmonary disease and diabetes were the most common comorbidities among influenza positives (Table 3). Seasonality had also a clear geographical distribution. Sites in higher latitudes had, generally, an earlier start of the influenza season.

Table 2 Characteristics of included patients overall and by site
Full size table
Table 3 Characteristics of included patients according to RT-PCR result
Full size table

Patients with a qualified occupation had a higher risk of being admitted with influenza. Patients with a swab taken 8–9 days after symptoms onset appeared with less risk of being admitted with influenza, suggesting a decrease in the influenza viral load for these patients (Table 4).

Table 4 Subject characteristics and risk of admission with influenza
Full size table

Pregnant women had a 3 times higher risk of having influenza at admission than non-pregnant. Also subjects with diabetes had 1.19 times higher risk of being an influenza case. On the other hand, patients with COPD or neoplasm had lower risk of testing positive for influenza. Despite there was a high number of admissions with cardiovascular diseases (CVD), no difference in the risk of influenza was found in these patients. (Fig. 6).

Fig. 6
figure 6

Adjusted Odds Ratio (aOR) and number of admissions with influenza according to comorbidity

Full size image

During pregnancy, the risk of testing positive for influenza was higher during the third trimester than in the first trimester, and also if they had any comorbidity in the first trimester (Fig. 7).

Fig. 7
figure 7

Predicted probability of having an admission with influenza in pregnant and non-pregnant women by trimester

Full size image

There were no significant statistical differences among influenza positives and negatives for those who were admitted to ICU or who received mechanical ventilation or those who died while they were hospitalised, and differences for those with extracorporeal membrane oxygenation could be due to sparse numbers of patients who received extracorporeal membrane oxygenation. Apart from influenza, the main discharge diagnosis was pneumonia, either for influenza-negatives or influenza-positives (Table 5).

Table 5 Influenza severity and complications 232 by RT-PCR results
Full size table

Probabilities of most common severe outcomes by strain by age and influenza strains are displayed in Fig. 8. This probability had an upward trend up to 80 years old after a shock. The probability point estimates of having any cardiovascular complication increased greatly from 90 years old for those who had influenza. Similar trends were found for each individual strain for these discharge diagnoses.

Fig. 8
figure 8

Predicted probability of severe outcome by strain

Full size image

Vaccination coverage was 9% or higher for targeted groups only in 4 sites (Fig. 9), and only 6 sites had at least 20 patients vaccinated among the patients targeted for vaccination. The IVE analysis was restricted to the sites with the highest vaccination coverage in targeted groups for vaccination having at least 20 patients vaccinated in these groups. These sites were Valencia, Canada, St. Petersburg, Mexico, Moscow and Turkey.

Fig. 9
figure 9

Vaccination coverage in targeted groups by site

Full size image

The IVE analysis, therefore, will be carried out in these six sites and globally. Vaccination coverage in pregnant women was 0% in Kazakhstan among the included patients, and in Moscow, only 1.3% (10 out of 800) of the admitted pregnant women received the vaccine at least 15 days before symptoms onset, therefore, adjusted IVE could not be estimated for pregnant women.

Vaccination coverage was higher in patients older than 65 years and in patients with two or more comorbidities. Among immunized women 15 to 45 years old, 19 of 47 were pregnant (40.4%), and among all vaccinated patients, 26.7% were obese.

Of the subjects vaccinated, 78.0% were also vaccinated in season 2015–2016 and 67.2% were vaccinated in season 2014–2015. However, 8.0% of the unvaccinated patients in the current season were vaccinated in the season 2015–2016, and 6.6% in the season 2014–2015 (Table 6).

Table 6 Characteristics of patients included in the primary analysis by vaccination status
Full size table

IVE estimates for included patients

In the selected sites for IVE estimates, vaccination coverage was 11.7% among the influenza positives and 22.2% among the influenza negatives. The overall IVE was 27.24% (95% CI 15.62 to 37.27%) in targeted groups for vaccination. Table 7 shows IVE for different strains, Fig. 10 by study country.

Table 7 IVE for all cases and for targeted groups only by age and strain
Full size table
Fig. 10
figure 10

Adjusted Influenza Vaccine Effectiveness by site

Full size image

IVE was statistically significant for all strains except for A(H1N1)pdm09 due to the limited sample size, and the point estimate was higher for both influenza B lineages, even using the trivalent vaccine (Fig. 11). Heterogeneity among influenza types/subtypes was relevant (I2 = 57.4%).

Fig. 11
figure 11

Adjusted Influenza Vaccine Effectiveness by strain

Full size image

This season IVE estimate was higher in patients 85 years old or older (51.17% [95% CI: 35.13 to 63.24]). IVE was also high and statistically significant for patients 2 to 4 years old (49.37% [95% CI: 21.60 to 67.30]) (Fig. 12). Heterogeneity among the different age groups was relevant (I2 = 69%).

Fig. 12
figure 12

Adjusted Influenza Vaccine Effectiveness by age group

Full size image

Discussion

The GIHSN included sites from the two hemispheres in the 2016/17 season. However, Ivory Coast and Peru were not included in the epidemiology study or in the IVE study due to the low influenza cases detected. This season was characterized by a predominance in the circulation of A(H3N2) virus, and a second wave of B/Victoria. However, A(H1N1)pdm09 was predominant in Mexico. B/Yamagata-strain, which was not included in the vaccine, also circulated in some areas.

Influenza A(H1N1)pdm09 was mainly found in Mexico. A low vaccination coverage was seen in most of the GIHSN sites.

The GIHSN represents an opportunity to analyse the epidemiology of hospitalized influenza cases, and an assessment of the vaccine effectiveness worldwide. However, there are some limitations that should be mentioned:

  • Although the same protocol was developed, the adaptation to different countries or sites produced some heterogeneity in the results, as previously reported in the network [3].

  • In general vaccination coverage was low in most sites, even among high risk groups.

  • Other factors as number of cases per site, and variability in the vaccination coverage, increased the heterogeneity in the reporting and analysis.

All of these limitations contributed to the complexity of the interpretation of the results.

In the northern hemisphere, the season differed by latitude [14], and this may have implications in the calendar of the vaccination campaigns.

Patients tested for influenza 8 to 9 days after symptoms onset had a higher proportion of samples negative for influenza than patients tested within the first 7 days after symptoms onset, as that viral load decreases with increasing time since infection, [15]. However, there were a few cases in our study as we collected all cases whose admission was in the 7 days after ILI symptoms started, and any delay in approaching the patient could result in a late swabbing.

Among inpatients with COPD, there was not a higher risk of testing for influenza. As all the cases were hospitalized, this result cannot be interpreted as COPD not being a risk factor for influenza hospitalization, as any other respiratory infection may decompensate the respiratory condition and force an admission. Besides vaccination coverage is higher in subjects with chronic conditions [16] and therefore, protection from the vaccine may also impact on our finding.

The risk of testing positive for influenza in diabetic patients was slightly higher than non-diabetic patients, as it also happened in previous seasons [3, 4]. Pregnancy also increased the probability of having influenza in women, particularly if they had at least one comorbidity in the first trimester.

Despite differences in the characteristics of the included patients relative to the age or pregnancy status, heterogeneity in the IVE analysis among the 6 sites with the highest numbers of vaccinated patients was low. Point estimates of the overall IVE from a two-step pooling was 27.2% (95% CI: 15.62 to 37.27) in hospitalized, which is higher than that reported in Europe for hospitalised patients [17], that ranged from 2.4 to 7.9%, depending on the age group, and lower to that estimated by the US CDC, which was 40% (95% CI: 32 to 46) [18].

Pooled Influenza vaccine effectiveness showed protection against all influenza virus that circulated, although for A(H1N1)pdm09 did not reach statistical significance, as the circulation of the virus was low except in Mexico. There was a significant effectiveness against both B lineages, even though most of the vaccines used were trivalent, i.e. only contained the B/Victoria linage, following recommendations of the World Health Organisation (WHO) for trivalent vaccines in the Northern Hemisphere [19]. Although antigenically different, there has been shown some degree of cross-protection among both B lineages.

Conclusion

The GIHSN provides an opportunity to analyse influenza epidemiology and vaccine effectiveness worldwide. In the 2016/17 season, A(H3N2) was the predominant influenza strain this season (first wave), followed by B/Victoria (second wave). Influenza A(H1N1)pdm09 was mainly found in Mexico. A low vaccination coverage was seen in most of the GIHSN sites.

Differences in the distribution of influenza cases among the age groups were mainly due to the characteristics of the participating hospitals. Pregnant women had higher risk of testing positive for influenza, as occurred with diabetics, however this difference was not seen in COPD subjects.

Overall IVE was low to moderate 27.24 (95% CI 15.62 to 37.27) in this season. A moderate to high effectiveness was seen for both influenza B lineages, and a non-significant low effectiveness for Influenza A(H1N1)pdm09.