Vaccine 33 (2015) 367–373
Contents lists available at ScienceDirect
Vaccine journal homepage: www.elsevier.com/locate/vaccine
Influenza vaccine effectiveness against laboratory confirmed influenza in Greece during the 2013–2014 season: A test-negative study Theodore Lytras a,b,c,∗ , Athanasios Kossyvakis d , Angeliki Melidou e , Maria Exindari e , Georgia Gioula e , Vasiliki Pogka e , Nikolaos Malisiovas e , Andreas Mentis d a
Department of Epidemiological Surveillance and Intervention, Hellenic Centre for Disease Control and Prevention, Athens, Greece Centre for Research in Environmental Epidemiology (CREAL), Barcelona, Spain c Department of Experimental and Health Sciences, Universitat Pompeu Fabra (UPF), Barcelona, Spain d National Influenza Reference Laboratory for Southern Greece, Hellenic Pasteur Institute, Athens, Greece e National Influenza Reference Laboratory for Northern Greece, Medical School, Aristotle University of Thessaloniki, Greece b
a r t i c l e
i n f o
Article history: Received 22 July 2014 Received in revised form 24 October 2014 Accepted 6 November 2014 Available online 18 November 2014 Keywords: Epidemiology Influenza Influenza-like illness—ILI Molecular methods Surveillance Vaccine effectiveness
a b s t r a c t Background: In 2013–2014 Greece experienced a resurgence of severe influenza cases, coincidental with a shift to H1N1pdm09 predominance. We sought to estimate Vaccine Effectiveness (VE) for this season using available surveillance data from hospitals (including both inpatients and outpatients). Methods: Swab samples were sent by hospital physicians to one of three laboratories, covering the entire country, to be tested for influenza using RT-PCR. The test-negative design was employed, with patients testing positive serving as cases and those testing negative serving as controls. VE was estimated using logistic regression, adjusted for age group, sex, region and calendar time, with further adjustment for unknown vaccination status using inverse response propensity weights. Additional age group stratified estimates and subgroup estimates of VE against H1N1pdm09 and H3N2 were calculated. Results: Out of 1310 patients with known vaccination status, 124 (9.5%) were vaccinated, and 543 patients (41.5%) tested positive for influenza. Adjusted VE was 34.5% (95% CI: 4.1–55.3%) against any influenza, and 56.7% (95% CI: 22.8–75.7%) against H1N1pdm09. VE estimates appeared to be higher for people aged 60 and older, while in those under 60 there was limited evidence of effectiveness. Isolated circulating strains were genetically close to the vaccine strain, with limited evidence of antigenic drift. Conclusions: These results suggest a moderate protective effect of the 2013–2014 influenza vaccine, mainly against H1N1pdm09 and in people aged 60 and over. Vaccine coverage was very low in Greece, even among groups targeted for vaccination, and substantial efforts should be made to improve it. VE can and should be routinely monitored, and the results taken into account when deciding on influenza vaccine composition for next season. © 2014 Elsevier Ltd. All rights reserved.
1. Introduction Seasonal influenza vaccination is recommended every year, because of the antigenic change in circulating influenza viruses and the short duration of the immunity induced. The vaccine provides usually moderate protection, which can vary substantially from
Abbreviations: ICU, Intensive care unit; ILI, Influenza-like illness; HA, hemagglutinin; HCDCP, Hellenic Centre for Disease Control and Prevention; NIRL, National Influenza Reference Laboratory; RT-PCR, Reverse-Transcriptase Polymerase Chain Reaction; VE, Vaccine Effectiveness. ∗ Corresponding author at: Hellenic Centre for Disease Control and Prevention, Agrafon 3-5, 15123 Marousi, Athens, Greece. Tel.: +30 6946814271. E-mail address: [email protected] (T. Lytras). http://dx.doi.org/10.1016/j.vaccine.2014.11.005 0264-410X/© 2014 Elsevier Ltd. All rights reserved.
year to year [1]. Factors responsible for this variation include the closeness of the antigenic match between vaccine and circulating strains, and the timing of the seasonal influenza outbreak relative to vaccination [2]; other relevant, non-vaccine-related factors include the degree of influenza virus circulation and co-circulation of other respiratory pathogens. Therefore, many countries are regularly performing observational studies to evaluate influenza Vaccine Effectiveness (VE) [2–7]. Such studies usually employ the test-negative design, a variant of the case-control design in which, from a given population seeking medical care for influenza-like illness (ILI), those who test positive for influenza serve as cases and those who test negative serve as controls. The test-negative design is simple and convenient, and controls for the different healthcare-seeking behaviour between vaccinated and unvaccinated participants [8].
368
T. Lytras et al. / Vaccine 33 (2015) 367–373
Recommendations on which individuals should be vaccinated against influenza vary between countries [9]. In Greece, influenza vaccination is recommended for all persons aged 60 and over, patients with certain chronic conditions, pregnant women, healthcare workers and carers of infants or elderly people (http://goo.gl/wp4AIG). After a relatively mild 2012–2013 influenza season during which both H3N2 and H1N1pdm09 strains were in roughly equal circulation, in 2013–2014 Greece experienced a resurgence in severe influenza cases coincidental with a shift to H1N1pdm09 predominance. Among cases with laboratory-confirmed influenza reported up to May 22nd, 2014 there were 145 deaths and 338 intensive-care unit (ICU) admissions, compared to 49 deaths and 108 ICU admissions in the previous season [10]. Thus the 2013–2014 influenza season was comparable in severity to the first post-pandemic season 2010–2011 [11], and was featured prominently in many media reports regarding the high death rate and the reported low uptake of vaccination [12]. Against this background, we decided to utilize the available surveillance data to estimate the effectiveness of influenza vaccination for the 2013–2014 season, using a test-negative design. As secondary objectives, we sought to determine how early in the season we could have a reliable VE estimate, explore any differences in VE among age groups and by type of influenza strain, and describe the genetic features of representative strains that circulated in Greece during the season.
2. Materials and methods As part of influenza surveillance activities in Greece, nasopharygeal swabs taken by hospital physicians were sent for Real-Time Reverse-Transcriptase Polymerase Chain Reaction (RT-PCR) testing to one of three laboratories, which cover the entire country: (1) the National Influenza Reference Laboratory (NIRL) of Southern Greece (Hellenic Pasteur Institute, Athens), (2) the NIRL of Northern Greece (2nd Laboratory of Microbiology, Medical School, Aristotle University of Thessaloniki), and (3) the Department of Microbiology, Medical School, National and Kapodistrian University of Athens. Physicians from all public and private hospitals could send at their discretion samples from both inpatients and outpatients, provided the patient’s clinical presentation was suggestive of ILI. Swabs were sent to these laboratories along with a standardized paper notification form, filled in by the physician; the form is available at the Hellenic Centre for Disease Control and Prevention (HCDCP) website (http://goo.gl/O30IoU), and contains basic demographic and clinical information including age, sex and influenza vaccination status. Laboratories marked the test result (positive/negative, plus influenza type and subtype for positive samples) on this reporting form, and forwarded it by fax to the Department of Epidemiological Surveillance and Intervention of the HCDCP. This was the data source utilized for this study. Patients under one year of age were excluded from the analysis, since age was recorded in completed years and the seasonal influenza vaccine is not recommended in infants under 6 months of age according to the Greek National Childhood Immunization Schedule (http://goo.gl/nKmT0E). To avoid bias, the study period was limited to weeks with laboratory-confirmed influenza cases. The study followed the test-negative design; patients with a positive RT-PCR for influenza were classified as cases and those testing negative as controls. The Odds Ratio (OR) for influenza vaccination was calculated, similar to a case control study, and VE was derived as one minus OR, expressed as a percentage. Because cases and controls are sampled in the same process, a situation conceptually similar to incidence density sampling in case-control studies,
exposure OR is an unbiased direct estimator (i.e. not an approximation) of the Incidence Rate Ratio (RR) for the corresponding hypothetical cohort [13]. However, because outcome status and the ratio of cases to controls are not known in advance and do not affect recruitment, a test-negative study is fundamentally different than a traditional case-control study [8,14,15]. For univariate analyses, the Mann–Whitney test was used to compare continuous variables and the Fisher’s exact test to compare categorical variables. All p-values are two-sided. In addition to calculating a crude VE, logistic regression was used to estimate the VE adjusted for age group, sex, region (North or South – the latter including samples sent to the University of Athens) and month of sample collection. To further adjust the analysis for participants with unknown (not reported) vaccination status, we used inverse response propensity weights in which “response” was defined as “known vaccination status” and response odds were modeled using logistic regression as a function of age group, sex, region and RTPCR test result. Additional subgroup VE estimates for H1N1pdm09 and H3N2 influenza subtypes were calculated. The effect of age on VE was examined by including appropriate interaction terms in the models and performing likelihood ratio testing. Furthermore, we calculated “rolling” weekly estimates of VE for the duration of the influenza season, in order to assess how early we could obtain stable estimates. All analyses were performed using the R software environment, version 3.1.0 [16]. Finally, in order to assess vaccine-virus match at the genetic level, a temporally and geographically representative sample of circulating H1N1pdm09 and H3N2 strains was selected from the study patients, and sequencing of the hemagglutinin (HA) gene was performed. The HA1 domain contains the receptor-binding cavity as well as most of the antigenic sites of the HA molecule. HA1 amplification was performed by one-step RT-PCR and the purified products sequenced, as previously described [17]. The obtained sequences were used to build phylogenetic trees, and compare the circulating influenza strains with the vaccine and other reference strains. All sequences were submitted to the GISAID database.
3. Results From November 2013 to May 2014 there were a total of 1370 unique patient swab samples sent to the three laboratories, for which the reporting form had been forwarded to the HCDCP: 873 from the NIRL South, 472 from the NIRL North and 25 from the University of Athens. Fig. 1 shows the distribution of these patients over time, grouped by lab result, along with the corresponding community ILI rate (as recorded by a sentinel surveillance system operated by the HCDCP). The first confirmed influenza cases occurred in week 01/2014; 21 influenza-negative patients that had been sampled in earlier weeks were excluded from further analysis. 39 samples were from infants under one year old and were also excluded; none of these infants was reported to have been vaccinated against influenza. Of 1310 patients finally analyzed, 712 (54.4%) were male and 598 (45.6%) female, and in 1090 (83.2%) their vaccination status was known (Table 1). 124/1090 patients (11.4%) were vaccinated against influenza while 966/1090 (88.6%) were not. In those aged 60 and over, vaccination coverage was 18.4% (84/456 patients), while in those aged 18–60 it was 6.3% (40/634 patients). The date of vaccination was reported in just 13/124 patients (10.5%), the majority of whom (8 cases) were vaccinated in October 2013. A positive RT-PCR test for influenza was found in 543/1310 patients (41.5%); of those 41 (7.6%) were vaccinated, 457 (84.2%) were unvaccinated, and in 45 (8.3%) vaccination status was unknown. The distribution of influenza types and subtypes was not significantly different between vaccinated and unvaccinated
T. Lytras et al. / Vaccine 33 (2015) 367–373
80 60 40
100
4 /2 20
18
/2
01
01
4
4 01
4
/2 16
01
4 01
/2 14
/2 12
/2 10
/2 08
01
01
4
4
4 01
4
/2
01 /2
06
01 04
/2 02
52
/2
01
4
3
3 01 /2
01
3 50
3 /2 48
46
/2
01
01
3
/2 44
01
3 01
/2 42
/2 40
3
0
20
50
ILI cases per 1,000 consultations
150
negative type B A/H3N2 A/H1N1pdm09 type A (not further subtyped)
0
Number of individual patient samples
200
100
369
Week number Fig. 1. Time distribution of reported individual patient swab samples, by test result and by influenza type/subtype, and corresponding weekly ILI rate; season 2013–2014.
patients (Fisher’s exact test: p = 0.15), with A/H1N1pdm09 being the predominant strain. However, there was a substantial proportion of influenza type A cases that was not further subtyped (139/498, 27.9%). 767/1310 patients (58.5%) were negative for influenza; of those 83 (10.8%) were vaccinated, 509 (66.4%) were unvaccinated, and in 175 (22.8%) vaccination status was unknown. Patients testing positive for influenza were more likely to have reported their vaccination status, and were on average younger than those testing negative (median and interquartile range: 50 years (34–65) vs 62 years (42–75) respectively, p < 0.001). Comparing the odds of vaccination among those testing positive and those testing negative for influenza, a crude VE estimate of 44.9% (95% CI: 18.6% to 63.2%) was calculated. Adjustment for age group, sex, region and month of sample collection using logistic regression resulted in a lower VE estimate of 33.1% (95% CI: −1.2% to 55.7%). Further adjustment for unknown vaccination status with
the use of inverse response propensity weights minimally changed the estimated VE to 34.5% (95% CI: 4.1% to 55.3%). In similar fashion, subgroup VE estimates were calculated using only H1N1pdm09 cases and H3N2 cases, compared with all testnegative controls. We also split the study sample into those aged 60 years and over (n = 714), i.e. the age group targeted for vaccination, and those under 60 years old (n = 596); stratified VE estimates were then calculated after refitting all models to include an interaction term between vaccination and age stratum. These results are shown on Table 2. Although not formally statistically significant (p = 0.08), the test for interaction was suggestive of a higher VE in those aged ≥60 against all influenza (49.8%, 95% CI: 17.5–69.5%), with no evidence of effectiveness in those under 60. Also, the vaccine appeared to offer substantially higher protection against H1N1pdm09 (VE = 56.7%, 95% CI: 22.8–75.7%), whereas against H3N2 there was very limited evidence of effectiveness. Overall, the
Table 1 Characteristics of patients with medically attended ILI, by influenza PCR test result. Test-negative controls (%)
Influenza cases (%)
73 (10) 103 (13) 184 (24) 407 (53)
74 (14) 124 (23) 156 (29) 189 (35)
28 (11) 70 (27) 75 (28) 91 (34)
18 (18) 11 (11) 24 (24) 49 (48)
425 (55) 342 (45)
287 (53) 256 (47)
130 (49) 134 (51)
61 (60) 41 (40)
Month of sample collection December January February March April
4 (1) 75 (10) 246 (32) 329 (43) 113 (15)
0 (0) 66 (12) 258 (48) 184 (34) 35 (6)
0 (0) 49 (19) 121 (46) 88 (33) 6 (2)
0 (0) 9 (9) 43 (42) 39 (38) 11 (11)
Region Northern Greece Southern Greece
273 (36) 494 (64)
179 (33) 364 (67)
112 (42) 152 (58)
64 (63) 38 (37)
Seasonal influenza vaccination No Yes Unknown
509 (66) 83 (11) 175 (23)
457 (84) 41 (8) 45 (8)
227 (86) 14 (5) 23 (9)
81 (79) 10 (10) 11 (11)
Age group 1–15 16–39 40–59 ≥60 Sex Male Female
A/H1N1pdm09 (%)
A/H3N2 (%)
370
T. Lytras et al. / Vaccine 33 (2015) 367–373
Table 2 Results for Vaccine Effectiveness (VE)a .
All participants Aged ≥ 60 yearsb Aged < 60 yearsb Test for interactionc a b c
All influenza
Limited to H1N1pdm09
Limited to H3N2
34.5% (4.1–55.3%) 49.8% (17.5–69.5%) −3.2% (−93.4% to 45.0%) p = 0.08
56.7% (22.8–75.7%) 67.6% (29.6%–85.1%) 31.0% (−69.7% to 71.9%) p = 0.21
28.3% (−42.8% to 64.0%) 47.8% (−23.9% to 78.0%) −46.5% (−351.8% to 52.5%) p = 0.17
All estimates adjusted for age group, sex, region and month of sample collection, and weighed by inverse response propensity. Stratified estimate from model with interaction. Likelihood-ratio test between models with and without interaction.
50 0 -50
Vaccine Effectiveness) (%
100
vaccine appeared to have the most benefit against H1N1pdm09 for people aged 60 and over (VE = 67.6%, 95% CI: 29.6–85.1%). We ran the main model (for all influenza types, all ages and with inverse response propensity weights) repeatedly using only the patients reported up to each given week, from week 05/2014 to 20/2014 (end of the influenza surveillance season). The weekly VE estimates and corresponding 95% CIs are shown on Fig. 2. Already from week 08/2014, before the season peak, the calculated VE approximated the final end-of-season estimate and did not vary substantially in subsequent weeks. Furthermore, the HA gene was sequenced in a sample of 34 H1N1pdm09 and 12 H3N2 influenza strains isolated from the study patients. The obtained H1N1pdm09 sequences had a 98.9–100% amino acid identity with each other, and 97.7–98.2% identity with the vaccine strain (A/California/07/2009). All the examined H1N1pdm09 strains clustered within subgroup 6B (Fig. 3), defined by P84S, D97 N, K163Q, S203 T, S185 T, A256 T, K283E and I321 V amino acid substitutions compared to the vaccine strain. However, substitutions in antigenic sites were limited, as only S162 N, S163Q (on antigenic site Sa) and S185 T (on antigenic site Sb) could be identified. On the other hand, our H3N2 sequences had a 98.8–100% amino acid identity with each other, and 98.6–99.2% identity with the vaccine strain (A/Texas/50/2012). All but one examined H3N2 strains clustered within phylogenetic subgroup 3C.3, featuring N145S and R142G amino acid substitutions on the antigenic site A with respect to the vaccine strain (Fig. 4). Additional substitutions, N122D on antigenic site A and L157S on antigenic site B, were identified in two strains. According to these results, there
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
Week number (2014) Fig. 2. Weekly adjusted VE estimates and 95% confidence intervals (against any influenza).
was no evidence of substantial antigenic drift based on the criterion of at least four substitutions on at least two antigenic sites [18]. 4. Discussion Our findings suggest that vaccination for the 2013–2014 season was moderately protective against influenza, mostly against H1N1pdm09, with little or no effect against H3N2. We also demonstrated that reliable VE estimates can be achieved early in the influenza season and provide useful public health information, in agreement with previous studies [19,20]. Our adjusted overall VE estimate of 34.5% was lower than the one previously reported from Canada (71%, 95% CI: 54–81%) for the same season [5], and much closer to the estimates from Navarre, Spain [21]. Dissimilar study populations may account for this difference, as well as waning of the immunity provided by the vaccine, given that the influenza season in Greece peaked several weeks later than in the other European and northern hemisphere countries [22]. The degree of influenza circulation in the community, in relation to other respiratory viruses, may also play a role. We did not find evidence of antigenic change that could explain the suboptimal VE observed, especially as regards the H3N2 strains. However, according to a recent report, circulating H3N2 strains similar to ours have reacted poorly in hemagglutinin inhibition (HI) assays with post-infection ferret antiserum raised against the eggpropagated vaccine virus, though more effectively if the vaccine virus was exclusively propagated in tissue culture cells [23]. The same effect was observed during the 2012–2013 season, and low VE against H3N2 for that season has recently been ascribed to mutations in the egg-adopted H3N2 vaccine strain rather than antigenic drift in circulating viruses [24]. This offers a highly plausible explanation for the low VE observed this season against H3N2, both in our study and in Spain, despite the close match of the examined circulating viruses to the vaccine reference strain. In addition, two new phylogenetic groups of H3 HA genes have been identified recently, that are possible H3N2 antigenic variants (John McCauley, personal communication, 18 June 2014). These rare strains co-circulated in Europe during 2013–2014 along with other more commonly identified 3C.2 and 3C.3 group viruses. Although we did not identify any strains harboring the same amino acid substitutions, we cannot rule out the possibility that the presence of such strains may have affected VE. Another notable finding is the observed higher VE in those aged 60 and over, compared to those under 60. Although the difference did not reach statistical significance (p = 0.08 against all influenza), it merits some further consideration. The effectiveness of influenza vaccination in older adults is overall a controversial issue [1,25–27], and specifically for the 2013–2014 season the Spanish study found somewhat lower VE in older adults [21], while the Canadian study found similar VE in all age groups [5]. This season was marked by increased circulation of H1N1pdm09, which has been observed to proportionally affect older people less than younger ones [28], a phenomenon that may be explained by qualitatively different immunologic response in older people due to long-term immunological memory [29]. Such potential differences may be
T. Lytras et al. / Vaccine 33 (2015) 367–373
371
Fig. 4. Phylogenetic relationships of Greek influenza A/H3N2 strains with recent European reference and vaccine strains based on the HA gene, 2013–2014 season. The evolutionary history was inferred using the Neighbor-Joining method. The percentage of replicate trees in which the associated taxa clustered together in the bootstrap test (1000 replicates) is shown next to the branches. Only values above 70% are shown.
Fig. 3. Phylogenetic relationships of Greek influenza A/H1N1pdm09 strains with recent European reference and vaccine strains based on the HA gene, 2013–2014 season. The evolutionary history was inferred using the Neighbor-Joining method. The percentage of replicate trees in which the associated taxa clustered together in the bootstrap test (1000 replicates) is shown next to the branches. Only values above 70% are shown.
heterogenous across different populations, and might account for the apparent variation in VE for older people among studies. Most previous studies of influenza VE have been performed in a community setting, as part of a sentinel surveillance scheme; we used data from hospital physicians, which included outpatients as well as inpatients, and obtained fairly reasonable estimates. The high percentage of patients testing positive for influenza (41.5%) also indicates that hospital physicians were sufficiently reliable in diagnosing ILI. A strength of the current study is the use of RT-PCR to detect influenza, which is a highly sensitive and specific test. Although the test-negative design can produce valid estimates even with suboptimal test sensitivity [15], studies that use Rapid Influenza Diagnostic Tests (RIDTs) [30] or serology may suffer from bias due to misclassification, and can respectively underestimate or overestimate true VE. We also included a fairly large sample, which covered the entire country, and was exclusively distributed along
372
T. Lytras et al. / Vaccine 33 (2015) 367–373
the period of influenza circulation; this improves the validity of our findings. Although we did adjust for calendar time (month of sample collection), it is not a confounder provided that vaccination occurred before patient recruitment [31], as was indeed the case based on the very limited information available in our study. As a sensitivity analysis, we re-fitted all models without adjustment for calendar time; in all cases, estimates for VE remained within a few percentage points (data not shown), indicating no significant confounding by calendar time. Limitations of the current study include the unavailability of information on hospitalization status (inpatient or outpatient) and on comorbidities, as well as the paucity of information on time of vaccination. Therefore we could not assess the effect of influenza vaccination on severe disease, nor any potential waning of immunity. Date of symptom onset was also not available for most cases; this is an additional limitation, as the sensitivity of RT-PCT declines with increasing time from symptom onset to swabbing. Vaccination status was unknown (not reported) in 16.8% of the study population; although we used inverse response propensity weights to adjust the analysis, the possibility that unknown vaccination status may be associated with other unknown confounders cannot be ruled out. Finally, due to the very low vaccination coverage in this population (
Contents lists available at ScienceDirect
Vaccine journal homepage: www.elsevier.com/locate/vaccine
Influenza vaccine effectiveness against laboratory confirmed influenza in Greece during the 2013–2014 season: A test-negative study Theodore Lytras a,b,c,∗ , Athanasios Kossyvakis d , Angeliki Melidou e , Maria Exindari e , Georgia Gioula e , Vasiliki Pogka e , Nikolaos Malisiovas e , Andreas Mentis d a
Department of Epidemiological Surveillance and Intervention, Hellenic Centre for Disease Control and Prevention, Athens, Greece Centre for Research in Environmental Epidemiology (CREAL), Barcelona, Spain c Department of Experimental and Health Sciences, Universitat Pompeu Fabra (UPF), Barcelona, Spain d National Influenza Reference Laboratory for Southern Greece, Hellenic Pasteur Institute, Athens, Greece e National Influenza Reference Laboratory for Northern Greece, Medical School, Aristotle University of Thessaloniki, Greece b
a r t i c l e
i n f o
Article history: Received 22 July 2014 Received in revised form 24 October 2014 Accepted 6 November 2014 Available online 18 November 2014 Keywords: Epidemiology Influenza Influenza-like illness—ILI Molecular methods Surveillance Vaccine effectiveness
a b s t r a c t Background: In 2013–2014 Greece experienced a resurgence of severe influenza cases, coincidental with a shift to H1N1pdm09 predominance. We sought to estimate Vaccine Effectiveness (VE) for this season using available surveillance data from hospitals (including both inpatients and outpatients). Methods: Swab samples were sent by hospital physicians to one of three laboratories, covering the entire country, to be tested for influenza using RT-PCR. The test-negative design was employed, with patients testing positive serving as cases and those testing negative serving as controls. VE was estimated using logistic regression, adjusted for age group, sex, region and calendar time, with further adjustment for unknown vaccination status using inverse response propensity weights. Additional age group stratified estimates and subgroup estimates of VE against H1N1pdm09 and H3N2 were calculated. Results: Out of 1310 patients with known vaccination status, 124 (9.5%) were vaccinated, and 543 patients (41.5%) tested positive for influenza. Adjusted VE was 34.5% (95% CI: 4.1–55.3%) against any influenza, and 56.7% (95% CI: 22.8–75.7%) against H1N1pdm09. VE estimates appeared to be higher for people aged 60 and older, while in those under 60 there was limited evidence of effectiveness. Isolated circulating strains were genetically close to the vaccine strain, with limited evidence of antigenic drift. Conclusions: These results suggest a moderate protective effect of the 2013–2014 influenza vaccine, mainly against H1N1pdm09 and in people aged 60 and over. Vaccine coverage was very low in Greece, even among groups targeted for vaccination, and substantial efforts should be made to improve it. VE can and should be routinely monitored, and the results taken into account when deciding on influenza vaccine composition for next season. © 2014 Elsevier Ltd. All rights reserved.
1. Introduction Seasonal influenza vaccination is recommended every year, because of the antigenic change in circulating influenza viruses and the short duration of the immunity induced. The vaccine provides usually moderate protection, which can vary substantially from
Abbreviations: ICU, Intensive care unit; ILI, Influenza-like illness; HA, hemagglutinin; HCDCP, Hellenic Centre for Disease Control and Prevention; NIRL, National Influenza Reference Laboratory; RT-PCR, Reverse-Transcriptase Polymerase Chain Reaction; VE, Vaccine Effectiveness. ∗ Corresponding author at: Hellenic Centre for Disease Control and Prevention, Agrafon 3-5, 15123 Marousi, Athens, Greece. Tel.: +30 6946814271. E-mail address: [email protected] (T. Lytras). http://dx.doi.org/10.1016/j.vaccine.2014.11.005 0264-410X/© 2014 Elsevier Ltd. All rights reserved.
year to year [1]. Factors responsible for this variation include the closeness of the antigenic match between vaccine and circulating strains, and the timing of the seasonal influenza outbreak relative to vaccination [2]; other relevant, non-vaccine-related factors include the degree of influenza virus circulation and co-circulation of other respiratory pathogens. Therefore, many countries are regularly performing observational studies to evaluate influenza Vaccine Effectiveness (VE) [2–7]. Such studies usually employ the test-negative design, a variant of the case-control design in which, from a given population seeking medical care for influenza-like illness (ILI), those who test positive for influenza serve as cases and those who test negative serve as controls. The test-negative design is simple and convenient, and controls for the different healthcare-seeking behaviour between vaccinated and unvaccinated participants [8].
368
T. Lytras et al. / Vaccine 33 (2015) 367–373
Recommendations on which individuals should be vaccinated against influenza vary between countries [9]. In Greece, influenza vaccination is recommended for all persons aged 60 and over, patients with certain chronic conditions, pregnant women, healthcare workers and carers of infants or elderly people (http://goo.gl/wp4AIG). After a relatively mild 2012–2013 influenza season during which both H3N2 and H1N1pdm09 strains were in roughly equal circulation, in 2013–2014 Greece experienced a resurgence in severe influenza cases coincidental with a shift to H1N1pdm09 predominance. Among cases with laboratory-confirmed influenza reported up to May 22nd, 2014 there were 145 deaths and 338 intensive-care unit (ICU) admissions, compared to 49 deaths and 108 ICU admissions in the previous season [10]. Thus the 2013–2014 influenza season was comparable in severity to the first post-pandemic season 2010–2011 [11], and was featured prominently in many media reports regarding the high death rate and the reported low uptake of vaccination [12]. Against this background, we decided to utilize the available surveillance data to estimate the effectiveness of influenza vaccination for the 2013–2014 season, using a test-negative design. As secondary objectives, we sought to determine how early in the season we could have a reliable VE estimate, explore any differences in VE among age groups and by type of influenza strain, and describe the genetic features of representative strains that circulated in Greece during the season.
2. Materials and methods As part of influenza surveillance activities in Greece, nasopharygeal swabs taken by hospital physicians were sent for Real-Time Reverse-Transcriptase Polymerase Chain Reaction (RT-PCR) testing to one of three laboratories, which cover the entire country: (1) the National Influenza Reference Laboratory (NIRL) of Southern Greece (Hellenic Pasteur Institute, Athens), (2) the NIRL of Northern Greece (2nd Laboratory of Microbiology, Medical School, Aristotle University of Thessaloniki), and (3) the Department of Microbiology, Medical School, National and Kapodistrian University of Athens. Physicians from all public and private hospitals could send at their discretion samples from both inpatients and outpatients, provided the patient’s clinical presentation was suggestive of ILI. Swabs were sent to these laboratories along with a standardized paper notification form, filled in by the physician; the form is available at the Hellenic Centre for Disease Control and Prevention (HCDCP) website (http://goo.gl/O30IoU), and contains basic demographic and clinical information including age, sex and influenza vaccination status. Laboratories marked the test result (positive/negative, plus influenza type and subtype for positive samples) on this reporting form, and forwarded it by fax to the Department of Epidemiological Surveillance and Intervention of the HCDCP. This was the data source utilized for this study. Patients under one year of age were excluded from the analysis, since age was recorded in completed years and the seasonal influenza vaccine is not recommended in infants under 6 months of age according to the Greek National Childhood Immunization Schedule (http://goo.gl/nKmT0E). To avoid bias, the study period was limited to weeks with laboratory-confirmed influenza cases. The study followed the test-negative design; patients with a positive RT-PCR for influenza were classified as cases and those testing negative as controls. The Odds Ratio (OR) for influenza vaccination was calculated, similar to a case control study, and VE was derived as one minus OR, expressed as a percentage. Because cases and controls are sampled in the same process, a situation conceptually similar to incidence density sampling in case-control studies,
exposure OR is an unbiased direct estimator (i.e. not an approximation) of the Incidence Rate Ratio (RR) for the corresponding hypothetical cohort [13]. However, because outcome status and the ratio of cases to controls are not known in advance and do not affect recruitment, a test-negative study is fundamentally different than a traditional case-control study [8,14,15]. For univariate analyses, the Mann–Whitney test was used to compare continuous variables and the Fisher’s exact test to compare categorical variables. All p-values are two-sided. In addition to calculating a crude VE, logistic regression was used to estimate the VE adjusted for age group, sex, region (North or South – the latter including samples sent to the University of Athens) and month of sample collection. To further adjust the analysis for participants with unknown (not reported) vaccination status, we used inverse response propensity weights in which “response” was defined as “known vaccination status” and response odds were modeled using logistic regression as a function of age group, sex, region and RTPCR test result. Additional subgroup VE estimates for H1N1pdm09 and H3N2 influenza subtypes were calculated. The effect of age on VE was examined by including appropriate interaction terms in the models and performing likelihood ratio testing. Furthermore, we calculated “rolling” weekly estimates of VE for the duration of the influenza season, in order to assess how early we could obtain stable estimates. All analyses were performed using the R software environment, version 3.1.0 [16]. Finally, in order to assess vaccine-virus match at the genetic level, a temporally and geographically representative sample of circulating H1N1pdm09 and H3N2 strains was selected from the study patients, and sequencing of the hemagglutinin (HA) gene was performed. The HA1 domain contains the receptor-binding cavity as well as most of the antigenic sites of the HA molecule. HA1 amplification was performed by one-step RT-PCR and the purified products sequenced, as previously described [17]. The obtained sequences were used to build phylogenetic trees, and compare the circulating influenza strains with the vaccine and other reference strains. All sequences were submitted to the GISAID database.
3. Results From November 2013 to May 2014 there were a total of 1370 unique patient swab samples sent to the three laboratories, for which the reporting form had been forwarded to the HCDCP: 873 from the NIRL South, 472 from the NIRL North and 25 from the University of Athens. Fig. 1 shows the distribution of these patients over time, grouped by lab result, along with the corresponding community ILI rate (as recorded by a sentinel surveillance system operated by the HCDCP). The first confirmed influenza cases occurred in week 01/2014; 21 influenza-negative patients that had been sampled in earlier weeks were excluded from further analysis. 39 samples were from infants under one year old and were also excluded; none of these infants was reported to have been vaccinated against influenza. Of 1310 patients finally analyzed, 712 (54.4%) were male and 598 (45.6%) female, and in 1090 (83.2%) their vaccination status was known (Table 1). 124/1090 patients (11.4%) were vaccinated against influenza while 966/1090 (88.6%) were not. In those aged 60 and over, vaccination coverage was 18.4% (84/456 patients), while in those aged 18–60 it was 6.3% (40/634 patients). The date of vaccination was reported in just 13/124 patients (10.5%), the majority of whom (8 cases) were vaccinated in October 2013. A positive RT-PCR test for influenza was found in 543/1310 patients (41.5%); of those 41 (7.6%) were vaccinated, 457 (84.2%) were unvaccinated, and in 45 (8.3%) vaccination status was unknown. The distribution of influenza types and subtypes was not significantly different between vaccinated and unvaccinated
T. Lytras et al. / Vaccine 33 (2015) 367–373
80 60 40
100
4 /2 20
18
/2
01
01
4
4 01
4
/2 16
01
4 01
/2 14
/2 12
/2 10
/2 08
01
01
4
4
4 01
4
/2
01 /2
06
01 04
/2 02
52
/2
01
4
3
3 01 /2
01
3 50
3 /2 48
46
/2
01
01
3
/2 44
01
3 01
/2 42
/2 40
3
0
20
50
ILI cases per 1,000 consultations
150
negative type B A/H3N2 A/H1N1pdm09 type A (not further subtyped)
0
Number of individual patient samples
200
100
369
Week number Fig. 1. Time distribution of reported individual patient swab samples, by test result and by influenza type/subtype, and corresponding weekly ILI rate; season 2013–2014.
patients (Fisher’s exact test: p = 0.15), with A/H1N1pdm09 being the predominant strain. However, there was a substantial proportion of influenza type A cases that was not further subtyped (139/498, 27.9%). 767/1310 patients (58.5%) were negative for influenza; of those 83 (10.8%) were vaccinated, 509 (66.4%) were unvaccinated, and in 175 (22.8%) vaccination status was unknown. Patients testing positive for influenza were more likely to have reported their vaccination status, and were on average younger than those testing negative (median and interquartile range: 50 years (34–65) vs 62 years (42–75) respectively, p < 0.001). Comparing the odds of vaccination among those testing positive and those testing negative for influenza, a crude VE estimate of 44.9% (95% CI: 18.6% to 63.2%) was calculated. Adjustment for age group, sex, region and month of sample collection using logistic regression resulted in a lower VE estimate of 33.1% (95% CI: −1.2% to 55.7%). Further adjustment for unknown vaccination status with
the use of inverse response propensity weights minimally changed the estimated VE to 34.5% (95% CI: 4.1% to 55.3%). In similar fashion, subgroup VE estimates were calculated using only H1N1pdm09 cases and H3N2 cases, compared with all testnegative controls. We also split the study sample into those aged 60 years and over (n = 714), i.e. the age group targeted for vaccination, and those under 60 years old (n = 596); stratified VE estimates were then calculated after refitting all models to include an interaction term between vaccination and age stratum. These results are shown on Table 2. Although not formally statistically significant (p = 0.08), the test for interaction was suggestive of a higher VE in those aged ≥60 against all influenza (49.8%, 95% CI: 17.5–69.5%), with no evidence of effectiveness in those under 60. Also, the vaccine appeared to offer substantially higher protection against H1N1pdm09 (VE = 56.7%, 95% CI: 22.8–75.7%), whereas against H3N2 there was very limited evidence of effectiveness. Overall, the
Table 1 Characteristics of patients with medically attended ILI, by influenza PCR test result. Test-negative controls (%)
Influenza cases (%)
73 (10) 103 (13) 184 (24) 407 (53)
74 (14) 124 (23) 156 (29) 189 (35)
28 (11) 70 (27) 75 (28) 91 (34)
18 (18) 11 (11) 24 (24) 49 (48)
425 (55) 342 (45)
287 (53) 256 (47)
130 (49) 134 (51)
61 (60) 41 (40)
Month of sample collection December January February March April
4 (1) 75 (10) 246 (32) 329 (43) 113 (15)
0 (0) 66 (12) 258 (48) 184 (34) 35 (6)
0 (0) 49 (19) 121 (46) 88 (33) 6 (2)
0 (0) 9 (9) 43 (42) 39 (38) 11 (11)
Region Northern Greece Southern Greece
273 (36) 494 (64)
179 (33) 364 (67)
112 (42) 152 (58)
64 (63) 38 (37)
Seasonal influenza vaccination No Yes Unknown
509 (66) 83 (11) 175 (23)
457 (84) 41 (8) 45 (8)
227 (86) 14 (5) 23 (9)
81 (79) 10 (10) 11 (11)
Age group 1–15 16–39 40–59 ≥60 Sex Male Female
A/H1N1pdm09 (%)
A/H3N2 (%)
370
T. Lytras et al. / Vaccine 33 (2015) 367–373
Table 2 Results for Vaccine Effectiveness (VE)a .
All participants Aged ≥ 60 yearsb Aged < 60 yearsb Test for interactionc a b c
All influenza
Limited to H1N1pdm09
Limited to H3N2
34.5% (4.1–55.3%) 49.8% (17.5–69.5%) −3.2% (−93.4% to 45.0%) p = 0.08
56.7% (22.8–75.7%) 67.6% (29.6%–85.1%) 31.0% (−69.7% to 71.9%) p = 0.21
28.3% (−42.8% to 64.0%) 47.8% (−23.9% to 78.0%) −46.5% (−351.8% to 52.5%) p = 0.17
All estimates adjusted for age group, sex, region and month of sample collection, and weighed by inverse response propensity. Stratified estimate from model with interaction. Likelihood-ratio test between models with and without interaction.
50 0 -50
Vaccine Effectiveness) (%
100
vaccine appeared to have the most benefit against H1N1pdm09 for people aged 60 and over (VE = 67.6%, 95% CI: 29.6–85.1%). We ran the main model (for all influenza types, all ages and with inverse response propensity weights) repeatedly using only the patients reported up to each given week, from week 05/2014 to 20/2014 (end of the influenza surveillance season). The weekly VE estimates and corresponding 95% CIs are shown on Fig. 2. Already from week 08/2014, before the season peak, the calculated VE approximated the final end-of-season estimate and did not vary substantially in subsequent weeks. Furthermore, the HA gene was sequenced in a sample of 34 H1N1pdm09 and 12 H3N2 influenza strains isolated from the study patients. The obtained H1N1pdm09 sequences had a 98.9–100% amino acid identity with each other, and 97.7–98.2% identity with the vaccine strain (A/California/07/2009). All the examined H1N1pdm09 strains clustered within subgroup 6B (Fig. 3), defined by P84S, D97 N, K163Q, S203 T, S185 T, A256 T, K283E and I321 V amino acid substitutions compared to the vaccine strain. However, substitutions in antigenic sites were limited, as only S162 N, S163Q (on antigenic site Sa) and S185 T (on antigenic site Sb) could be identified. On the other hand, our H3N2 sequences had a 98.8–100% amino acid identity with each other, and 98.6–99.2% identity with the vaccine strain (A/Texas/50/2012). All but one examined H3N2 strains clustered within phylogenetic subgroup 3C.3, featuring N145S and R142G amino acid substitutions on the antigenic site A with respect to the vaccine strain (Fig. 4). Additional substitutions, N122D on antigenic site A and L157S on antigenic site B, were identified in two strains. According to these results, there
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
Week number (2014) Fig. 2. Weekly adjusted VE estimates and 95% confidence intervals (against any influenza).
was no evidence of substantial antigenic drift based on the criterion of at least four substitutions on at least two antigenic sites [18]. 4. Discussion Our findings suggest that vaccination for the 2013–2014 season was moderately protective against influenza, mostly against H1N1pdm09, with little or no effect against H3N2. We also demonstrated that reliable VE estimates can be achieved early in the influenza season and provide useful public health information, in agreement with previous studies [19,20]. Our adjusted overall VE estimate of 34.5% was lower than the one previously reported from Canada (71%, 95% CI: 54–81%) for the same season [5], and much closer to the estimates from Navarre, Spain [21]. Dissimilar study populations may account for this difference, as well as waning of the immunity provided by the vaccine, given that the influenza season in Greece peaked several weeks later than in the other European and northern hemisphere countries [22]. The degree of influenza circulation in the community, in relation to other respiratory viruses, may also play a role. We did not find evidence of antigenic change that could explain the suboptimal VE observed, especially as regards the H3N2 strains. However, according to a recent report, circulating H3N2 strains similar to ours have reacted poorly in hemagglutinin inhibition (HI) assays with post-infection ferret antiserum raised against the eggpropagated vaccine virus, though more effectively if the vaccine virus was exclusively propagated in tissue culture cells [23]. The same effect was observed during the 2012–2013 season, and low VE against H3N2 for that season has recently been ascribed to mutations in the egg-adopted H3N2 vaccine strain rather than antigenic drift in circulating viruses [24]. This offers a highly plausible explanation for the low VE observed this season against H3N2, both in our study and in Spain, despite the close match of the examined circulating viruses to the vaccine reference strain. In addition, two new phylogenetic groups of H3 HA genes have been identified recently, that are possible H3N2 antigenic variants (John McCauley, personal communication, 18 June 2014). These rare strains co-circulated in Europe during 2013–2014 along with other more commonly identified 3C.2 and 3C.3 group viruses. Although we did not identify any strains harboring the same amino acid substitutions, we cannot rule out the possibility that the presence of such strains may have affected VE. Another notable finding is the observed higher VE in those aged 60 and over, compared to those under 60. Although the difference did not reach statistical significance (p = 0.08 against all influenza), it merits some further consideration. The effectiveness of influenza vaccination in older adults is overall a controversial issue [1,25–27], and specifically for the 2013–2014 season the Spanish study found somewhat lower VE in older adults [21], while the Canadian study found similar VE in all age groups [5]. This season was marked by increased circulation of H1N1pdm09, which has been observed to proportionally affect older people less than younger ones [28], a phenomenon that may be explained by qualitatively different immunologic response in older people due to long-term immunological memory [29]. Such potential differences may be
T. Lytras et al. / Vaccine 33 (2015) 367–373
371
Fig. 4. Phylogenetic relationships of Greek influenza A/H3N2 strains with recent European reference and vaccine strains based on the HA gene, 2013–2014 season. The evolutionary history was inferred using the Neighbor-Joining method. The percentage of replicate trees in which the associated taxa clustered together in the bootstrap test (1000 replicates) is shown next to the branches. Only values above 70% are shown.
Fig. 3. Phylogenetic relationships of Greek influenza A/H1N1pdm09 strains with recent European reference and vaccine strains based on the HA gene, 2013–2014 season. The evolutionary history was inferred using the Neighbor-Joining method. The percentage of replicate trees in which the associated taxa clustered together in the bootstrap test (1000 replicates) is shown next to the branches. Only values above 70% are shown.
heterogenous across different populations, and might account for the apparent variation in VE for older people among studies. Most previous studies of influenza VE have been performed in a community setting, as part of a sentinel surveillance scheme; we used data from hospital physicians, which included outpatients as well as inpatients, and obtained fairly reasonable estimates. The high percentage of patients testing positive for influenza (41.5%) also indicates that hospital physicians were sufficiently reliable in diagnosing ILI. A strength of the current study is the use of RT-PCR to detect influenza, which is a highly sensitive and specific test. Although the test-negative design can produce valid estimates even with suboptimal test sensitivity [15], studies that use Rapid Influenza Diagnostic Tests (RIDTs) [30] or serology may suffer from bias due to misclassification, and can respectively underestimate or overestimate true VE. We also included a fairly large sample, which covered the entire country, and was exclusively distributed along
372
T. Lytras et al. / Vaccine 33 (2015) 367–373
the period of influenza circulation; this improves the validity of our findings. Although we did adjust for calendar time (month of sample collection), it is not a confounder provided that vaccination occurred before patient recruitment [31], as was indeed the case based on the very limited information available in our study. As a sensitivity analysis, we re-fitted all models without adjustment for calendar time; in all cases, estimates for VE remained within a few percentage points (data not shown), indicating no significant confounding by calendar time. Limitations of the current study include the unavailability of information on hospitalization status (inpatient or outpatient) and on comorbidities, as well as the paucity of information on time of vaccination. Therefore we could not assess the effect of influenza vaccination on severe disease, nor any potential waning of immunity. Date of symptom onset was also not available for most cases; this is an additional limitation, as the sensitivity of RT-PCT declines with increasing time from symptom onset to swabbing. Vaccination status was unknown (not reported) in 16.8% of the study population; although we used inverse response propensity weights to adjust the analysis, the possibility that unknown vaccination status may be associated with other unknown confounders cannot be ruled out. Finally, due to the very low vaccination coverage in this population (
