Seasonal Influenza Vaccine Effectiveness in Preventing Laboratory-Confirmed Influenza in Primary Care in Israel, 2016–2017 Season: Insights Into Novel Age-Specific Analysis

Abstract

Background

The 2016–2017 influenza season in Israel was dominated by the circulation of influenza A(H3N2). Influenza vaccine is recommended for the entire population in Israel aged >6 months. The inactivated influenza vaccine was chosen for use this season.

Methods

We estimated the 2016–2017 end-of-season influenza vaccine effectiveness (VE) in preventing influenza-like illness due to influenza A(H3N2), using the test-negative design. Age-specific VE was estimated using a moving age window and weekly analysis.

Results

During the 2016–2017 season, 1267 samples were collected; 467 (36.9%) were positive for influenza, with 97.9% A(H3N2), 0.2% A(H1N1)pdm09, and 1.9% B. A total of 1088 individuals were found eligible to be included in VE assessment. All vaccinated individuals included in the VE assessment received the inactivated influenza vaccine. Adjusted VE against influenza A(H3N2) was 29.0% (95% confidence interval [CI], 0.3%–49.5%). Age group–specific adjusted VE was 69.2% (95% CI, 19.4%–88.3%) for children aged 5–17 years and 58.8% (95% CI, .8%–82.9%) for adults aged 45–64 years. Other age groups demonstrated lower VE estimates that were not statistically significant. Adjusted VE estimates using a moving window of 15 years and weekly VE analyses provided a more defined understanding of age-specific VE during the 2016–2017 season.

Conclusions

Estimating VE using a moving age window as well as weekly VE analysis may provide more detailed information regarding the relationship between VE and age.

influenza, vaccine, vaccine effectiveness

Vaccination is the primary approach recommended to prevent influenza [1]. Ongoing genetic and antigenic changes among influenza viruses require continuing analyses of the circulating viruses [2] to determine influenza vaccine composition each year [3]. In recent years, influenza vaccine effectiveness (VE) studies have been conducted by various groups [4]. These studies demonstrated that influenza VE estimates can vary substantially, by season [5], type/subtype [4], and geographic location [4].

Israel is a country of 8.7 million, located in the western part of Asia, adjacent to the southeastern coast of the Mediterranean Sea. Influenza activity in Israel is seasonal, taking place generally from December through March [6] in patterns comparable to those of Western Europe.

Israel’s Ministry of Health endorses influenza vaccination for the whole population aged >6 months [7]. The influenza vaccines are administered through outpatient clinics of 4 healthcare services, with the inactivated influenza vaccines offered free of charge. Following the controversy regarding the effectiveness of live attenuated influenza vaccine in the 2015–2016 season [8, 9], the inactivated influenza vaccine (trivalent or quadrivalent) was selected by the Israel Ministry of Health to be the vaccine of choice for the 2016–2017 season.

An outpatient sentinel influenza surveillance system was established in Israel 20 years ago and consists of primary care clinics located in all 7 districts of Israel and staffed by family physicians, pediatricians, and internists. During the 2016–2017 influenza season, 26 sentinel clinics took part in influenza surveillance.

The 2016–2017 influenza season in Israel started early and peaked in the beginning of January 2017, demonstrating moderate intensity in the community and marked by the dominance of influenza A(H3N2) [10].

In this study, we estimated the 2016–2017 VE against laboratory-confirmed influenza in patients with influenza-like illness (ILI) attending the primary care sentinel clinics in Israel, using the test-negative design. We applied a novel analysis to better understand age-specific VE estimates during this season.

METHODS

Study Design and Population

VE against laboratory-confirmed influenza was estimated using the test-negative design [11, 12] and calculated as (1 – odds ratio) × 100, expressed as a percentage. The study period lasted from 2 October 2016 until 18 March 2017 (the influenza surveillance season), and was carried out in 26 outpatient sentinel clinics throughout Israel. Combined nasal–throat samples [13] were obtained from patients aged ≥6 months who met the following case definition of ILI: a temperature of ≥37.8°C, together with ≥1 of the following symptoms: coryza, cough, sore throat, and muscle ache [14]. Discretion has been given to providers to include other symptoms or signs deemed relevant. The combined nasal–throat swab kits were supplied by the Israel Central Virology Laboratory. Each clinic was asked to send up to 10 samples per week.

A questionnaire was completed for each sample gathering the following information: date of birth, sex, date of symptom onset, date of sample collection, the presence of chronic medical conditions, 2016–2017 influenza vaccination status, date of vaccination, and type of vaccine used. For children <9 years of age, the receipt of a second vaccine dose, if indicated (for those receiving the influenza vaccine for the first time), and the date of the second dose were also recorded.

An individual was defined as vaccinated if the 2016–2017 influenza vaccine was received ≥14 days prior to the onset of illness. Among children aged 6 months to 9 years receiving the influenza vaccine for the first time, only those who received 2 doses, with the second dose administered ≥14 days prior to onset of illness, were considered vaccinated. Samples obtained >7 days after the beginning of symptoms and <14 days after vaccination were excluded from the analysis.

VE was estimated only for influenza A(H3N2) due to the dominance of this subtype in the population during the 2016–2017 season.

Laboratory Methods

The combined nasal–throat samples obtained from sentinel ILI patients were transported weekly, in cooling containers, to the Israel Central Virology Laboratory, and tested for influenza A and B viruses as well as the subtype by real-time polymerase chain reaction as previously described [15–17].

Nucleotide sequencing of the hemagglutinin (HA) gene from a subset of influenza A(H3N2) viruses circulating in Israel during the 2016–2017 season was performed as described elsewhere [17, 18].

Statistical Analysis

Demographic characteristics were compared by means of the Pearson χ² test. The odds ratios for crude VE calculation and 95% confidence intervals (CIs) were calculated utilizing a univariate logistic regression model. Adjustment for age group, sex, calendar month of sample collection, days from disease onset to swab, and influenza underlying chronic conditions was performed using a multivariable logistic regression model. VE was estimated for the entire study group and for specific age groups, for the entire season and for each week of the season. The VE estimates for each week included all cases and controls recruited by that week.

To gain a more detailed insight into VE at different ages, we developed a dynamic method of VE calculation. Dynamic VE is a nondiscrete stratification and visualization method for calculating VE for a subgroup of the data according to certain variable(s). VE was calculated for each 15-year window starting at the first year of age and moving gradually by 1 year.

We applied locally weighted scatterplot smoothing (LOWESS) [19] to the dynamic VE data, to visualize more clearly the trend of VE through the various ages and to better understand the relationship between VE and age.

We also applied cubic spline function to the adjusted VE data with knots placed at discrete 15-year age intervals [5, 19, 20]. This function has commonly been used to demonstrate changes in VE along a spectrum of variables [5, 20]. In this study, we used this function to demonstrate the changes in VE along the age spectrum in comparison to the dynamic VE.

Statistical analyses were carried out using SAS version 9.1.3 (SAS Institute, Cary, North Carolina) and R version 3.4.0 (R Foundation for Statistical Computing, Vienna, Austria) software.

Ethical Considerations

Sentinel ILI surveillance in Israel, inclusive of the viral detection component, is performed in accordance with the Public Health Ordinance enacted in Israel and does not necessitate informed consent.

RESULTS

Seasonality and Virus Circulation

Based on weekly rates of outpatient physician visits due to ILI per 10000 Maccabi Healthcare Services members, the 2016–2017 season coincided with the epidemic threshold on epidemiologic week 49 (Figure 1) [21]. Compared with data since 1998 (excluding the 2009 influenza pandemic), the 2016–2017 season peaked 3 weeks earlier than the multiyear median (Figure 1).

Figure 1.

Weekly rates of outpatient physician visits due to influenza-like illness (ILI) per 10000 Maccabi Healthcare Services members for the 2016–2017 season, compared with multiyear percentile analysis of ILI rates for the preceding seasons since 1998. The 2009 pandemic influenza was excluded from the multiyear percentile calculation.

A total of 1267 nasal–throat samples were collected from ILI patients at the sentinel clinics. Of those, 467 (36.9%) samples were positive for influenza, with 457 (97.9%) samples positive for influenza A(H3N2), 1 (0.2%) positive for A(H1N1)pdm09, and 9 (1.9%) positive for influenza B (Figure 2).

Figure 2.

Weekly distribution of influenza-positive samples collected from primary care sentinel clinics during the 2016–2017 influenza season.

Molecular characterization of the HA gene from a convenience sample of 40 influenza A(H3N2) viruses from sentinel patients demonstrated that all belonged to the 3C.2a clade. However, they demonstrated 3 groups of mutations when compared to the 2016–2017 influenza A(H3N2) vaccine component. A detailed account of the genetic divergence of the 2016–2017 influenza A(H3N2) is described elsewhere [17].

Vaccine Effectiveness

Due to clear dominance of influenza A(H3N2) in the 2016–2017 season, we estimated VE against this subtype only. Of the 1267 samples obtained, 179 samples were excluded because they did not meet the evaluation inclusion criteria (Figure 3). The characteristics of cases and controls included in the evaluation are shown in Table 1.

Figure 3.

Flowchart of patients with influenza-like illness from 2016–2017 primary care sentinel clinics in Israel. Exclusion of individuals applied only to the vaccine effectiveness analysis.

Table 1.

Characteristics of Influenza A(H3N2) Cases and Controls, 2016–2017 Season

Variable	Controls (n = 674), No.	Cases A(H3N2) (n = 414), No.	Total (N = 1088), No.	P Value
Age group, y
0.5–4	225	50	275	<.01
5–17	85	117	202
18–44	216	140	356
45–64	101	64	165
≥65	47	43	90
Sex
Male	317	208	525	.33
Female	357	206	563
Background disease
No	556	329	885	.24
Yes	118	85	202
Days between symptom onset and swabbing
0–1	311	187	498	.65
2–4	323	207	530
5–7	40	20	60
Vaccinated
No	529	344	873	.08
Yes	145	70	215

All but 1 vaccinated individual received the trivalent injected egg-grown 2016–2017 Northern Hemisphere vaccine (split or inactivated) containing the A/California/7/2009 (H1N1)pdm09-like virus, the A/Hong Kong/4801/2014 (H3N2)-like virus, and the B/Brisbane/60/2008-like virus. One individual received the injectable influenza quadrivalent Northern Hemisphere vaccine containing also the B/Phuket/3073/2013-like virus.

The adjusted end-of-season VE against laboratory-confirmed influenza A(H3N2) in sentinel ILI patients for the 2016–2017 season for all ages was 29.0% (95% CI, .3%–49.5%; Table 2). Age group–specific adjusted VE was 69.2% (95% CI, 19.4%–88.3%) for children aged 5–17 years and 58.8% (95% CI, .8%–82.9%) for adults aged 45–64 years (Table 2). Adjusted VE estimates for the other age groups were lower, particularly for those aged ≥65 years, and were not statistically significant (Table 2).

Table 2.

Influenza A(H3N2) Case and Control Vaccination Status and Vaccine Effectiveness Estimates, 2016–2017

Age, y	Adjustment/Stratification	Cases			Controls			VE, %	(95% CI)
		All	Vaccinated	(%)	All	Vaccinated	(%)
All ages	Crude	414	70	(17)	674	145	(22)	25.8	(–1.8 to 45.9)
	Adjusteda	414	70	(17)	674	145	(22)	29.0	(.3–49.5)
<65	Crude	371	42	(11)	627	120	(19)	46.1	(21.3–63.0)
	Adjusteda	371	42	(11)	627	120	(19)	50.9	(26.6–67.1)
0.5–4	Adjustedb	50	11	(22)	225	58	(26)	39.5	(–36.2 to 73.1)
5–17	Adjustedb	117	12	(10)	85	17	(20)	69.2	(19.4–88.3)
18–44	Adjustedb	140	10	(7)	216	15	(7)	12.5	(–108.7 to 63.7)
45–64	Adjustedb	64	9	(14)	101	30	(30)	58.8	(.8–82.9)
≥65	Adjustedb	43	28	(65)	47	25	(53)	–116.6	(–507.4 to 22.7)

Abbreviations: CI, confidence interval; VE, vaccine effectiveness.

a

Adjustment for age as a categorical variable (age categories: 0.5–4, 5–17, 18–44, 45–64, and ≥65 years), sex, calendar month of sample collection, days from disease onset to swab, and influenza underlying chronic conditions was done as outlined in the Methods.

b

Adjusted for age (as a continuous variable) sex, calendar month of sample collection, days from disease onset to swab, and influenza underlying chronic conditions.

The dynamic adjusted VE estimates are presented in Figure 4. The highest VE estimates were observed among individuals aged 0.5–20 and 44–61 years (Figure 4A). Within the age range 0.5–20 years, the VE estimate of each 15-year window was statistically significant. Within the 44- to 61-year age range, only the VE estimate of the 46- to 60-years age window was statistically significant (Figure 4A). The lowest VE estimates (of <25%) were observed for the ages 20–49 years and ≥53 years. The low VE of the those aged 20–49 years corresponded with low vaccination rates of ≤10%, while the low VE among individuals aged ≥53 years occurred in the presence of substantially higher vaccination rates (Figure 4A).

Figure 4.

A, Dynamic adjusted vaccine effectiveness (VE) according to age with window width of 15 years. Circle position represents VE values; circle size represents the number of individuals (both cases and controls) included in each age window; lite-colored center indicates 95% confidence interval (CI) lower value >0 (statistically significant VE). Solid (blue) line represents vaccine coverage of the entire age range group; dashed (green) line represents vaccine coverage of cases; dotted (red) line represents vaccine coverage of controls. Graphic demonstration of moving VE was truncated at age window 75–89 years as the VE and 95% CIs demonstrated a general decline. B, Locally weighted scatterplot smoothing of dynamic VE with 0.999 CI (taking in consideration multiple comparisons of 76 points). C, Cubic spline smoothing of 15-year interval data. Blue dots representing the VE estimate of the interval age group used for cubic spline analysis are demonstrated in B and C.

Figure 4B demonstrates the application of LOWESS to the data presented in Figure 4A. The application of LOWESS enables a clearer visualization of the trend of VE over the various ages and allows a better understanding of the relationship between VE and age.

Figure 4C demonstrates the results of cubic spline function applied to the adjusted VE estimates, with knots at discrete 15-year age intervals. As shown, the dynamic VE (without and with LOWESS) reveals more detailed information of VE estimates at different ages compared with the cubic spline function.

Weekly VE Estimates

Adjusted weekly VE estimates for all ages and for individuals <65 years of age are demonstrated in Figure 5. The adjusted VE point values were above zero in all the weeks presented in Figure 5, and the 95% CIs generally narrowed with season progression (Figure 5A and 5B). The weekly adjusted VE for individuals <65 years of age were higher than the adjusted VE for all ages from week 50 onward (Figure 5B), and were statistically significant from week 51 onward (Figure 5B). For individuals of all ages, the weekly adjusted VE estimates were statistically significant only on weeks 52, and borderline statistically significant on weeks 6, 10, and 11 (Figure 5A).

Figure 5.

Weekly adjusted vaccine effectiveness (VE) estimates. A, Weekly estimates and confidence intervals (CIs) of adjusted 2016–2017 influenza VE for all ages. B, Weekly estimates and CIs of adjusted 2016–2017 influenza VE for age <65 years. The number of individuals included in each weekly analysis appears above the upper limit of the weekly 95% CI.

DISCUSSION

The 2016–2017 influenza season in Israel started early and was characterized by marked dominance of the influenza A(H3N2) subtype. The overall end-of-season adjusted VE against laboratory-confirmed influenza A(H3N2) in sentinel ILI patients was in the moderate-to-low range, with the age groups 5–17 years and 45–64 years demonstrating the highest VE (Table 2).

The relatively high VE among children 5–17 years old, coupled with the fact that children are thought to be instrumental in propagating respiratory pathogens [22, 23], strongly supports vaccinating children against influenza. In this regard, a recent study demonstrated that influenza vaccination was associated with decreased mortality among children with laboratory-confirmed influenza [24]. Furthermore, several studies suggested that vaccination of children against influenza confers indirect protection in certain settings such as closely connected communities [25] and households [26, 27].

The low VE with wide CI in children aged 0.5–4 years should be interpreted with caution. Substantial differences in social exposure and behaviors exist among children of different ages within this age group. Specifically, the older children within the 0.5- to 4-years age group are more likely to attend day care and have a wider social circle compared with the younger children of this age group [22]. As a result, they are more likely to contract disease from individuals outside their family. Thus, VE of older children in the youngest age group may resemble VE results observed for the 5- to 17-years age group. A larger number of children in the 0.5- to 4-years age group is necessary to examine this phenomenon.

The relatively high VE among adults aged 45–64 years stands in sharp contrast to the low VE found among adults aged 18–44 years. The lower influenza vaccination rate among those aged 18–44 years (Table 2) may have led to reduced precision of its VE estimate.

The adjusted VE was particularly low in those aged ≥65 years. These results are particularly interesting in view of the substantially higher vaccination rates in this age group, among cases and controls, compared with the other age groups (Table 2 and Figure 4A).

Our end-of season influenza A(H3N2) VE estimate for all ages is consistent with that of the United States Flu VE network (34%, statistically significant) [28]. Furthermore, similar to Israel, the US Flu VE Network age-specific VE against influenza A(H3N2) demonstrated low VE without statistical significance in the age groups 18–49 and ≥65 years, while showing higher and statistically significant VE in the other age groups [28]. The low VE estimates we found in those aged ≥65 years were also consistent with the findings of other studies [4, 29–31].

These low VE estimates in those aged ≥65 years are of concern due to the fact that individuals in this age group are at increased risk of developing complications and are thus in greater need of vaccine-induced protection.

Estimating VE for predetermined age groups is informative; however, it may not provide sufficient information due to the discrete nature of these age groups. Furthermore, some age groups may not be sufficiently large to allow an accurate VE estimation. To address these issues, we developed a dynamic VE estimation method. This method is described and used here for the first time. The dynamic VE method appears useful in providing details about VE changes along smaller age intervals, compared with the age-group VE presented in Table 2. A method that has commonly been used to demonstrate changes in VE along a spectrum of variables, such as age, is the cubic spline function [5, 20]. By comparing the dynamic VE to the cubic spline function of the adjusted VE estimates (Figure 4A–C), we were able to show that the dynamic VE method reveals more detailed VE estimates at various ages.

Our 2016–2017 analysis demonstrates differences in VE among different age groups, suggesting that the overall modest end-of-season VE estimate may not to be useful by itself. Our data support the analysis of VE according to age, including the use of dynamic VE analysis, while considering additional factors such as influenza vaccination rates.

Most VE studies provide end-of-season VE, and some provide midseason VE. However, a week-by-week account of adjusted VE can be very useful for within-season decision making, such as recommendation for antiviral use. Our weekly analysis indicated rather early that the influenza vaccine was less effective in individuals aged ≥65 years throughout the season.

The variability in amino acid substitutions observed in Israel during the 2016–2017 season [17] is consistent with the variability shown in studies originating from Canada, Influenza-Monitoring Vaccine Effectiveness network (I-MOVE), and the United Kingdom [30, 32, 33]. However, the size of our study did not allow VE estimation against influenza A(H3N2) based on genetic group.

The 2016–2017 end-of-season VE results against influenza A(H3N2) from Israel and midseason VE estimates from Northern Hemisphere countries [30, 34] are higher than those obtained for the 2014–2015 season, when the influenza A(H3N2) vaccine component was found to confer little or no protection against influenza in community settings [11, 35, 36]. Whereas during the 2014–2015 season, a significant drift occurred among circulating influenza A(H3N2) as compared with the influenza A(H3N2) vaccine component [36], the 2016–2017 influenza A(H3N2) viruses circulating in Israel demonstrated fewer amino acid substitutions in antigenic sites in each of the variants, compared with the influenza A(H3N2) vaccine component [17].

Selection of influenza A(H3N2) vaccine components is challenging as mutations occur in this influenza virus subtype more frequently compared with influenza A(H1N1) or influenza B [37]. In fact, a recent meta-analysis demonstrated that VE estimates against influenza A(H3N2) tend to be lower than VE against influenza A(H1N1) or influenza B [4].

The divergence of influenza A(H3N2) HA amino acid substitutions during the 2016–2017 season, and the inability to predict which amino acid substitution group will become dominant globally, suggests a need for a continuous evaluation of the different genetic groups, in terms of amino acid substitutions, antigenic comparability, global spread, and their effect on VE.

Our study has several limitations. In recent years, various studies examined the effect of prior influenza vaccinations on influenza VE in community settings [38–41]. A 2016–2017 midseason VE evaluation from northern Spain demonstrated that VE results varied based on the history of prior influenza vaccinations, with history of ≥3 prior influenza vaccine doses affecting negatively the current year VE [42]. A recent systematic review and meta-analysis did not find evidence that influenza vaccination in the prior season negatively affected current season VE [43]. Unfortunately, history of prior influenza vaccination among individuals included in our VE analysis was not available for the current study.

The use of statins was recently reported to be associated with reduced VE against influenza [44, 45] and low hemagglutination inhibition antibody titers following influenza vaccination [45, 46]. Information regarding medication use among the study individuals was not available for the current study.

A relatively small number of individuals aged ≥65 years were included in our study, as well as in other VE studies [8, 47, 48]. Furthermore, based on health management organization databases in Israel, relatively small numbers of individuals aged ≥65 years with ILI have been observed in primary care clinics during the influenza seasons [6, 49]. These findings raise the possibility that many individuals belonging to this age group may suffer rapid clinical deterioration upon developing respiratory symptoms, requiring them to obtain prompt hospital care rather than visit their primary physician. The small numbers of individuals in the ≥65 age group from our study may make the VE results of this age group difficult to generalize.

We are continuing our efforts to expand our capacity for accurate and timely analysis of VE estimates. These efforts are particularly important due to the unique geographic location, climate conditions, and diverse population of Israel.

Notes

Acknowledgments. We are grateful to Esti Turgeman, Moriel Doron, Eti Iluz, and Lior Chanoch for providing dedicated technical support, and to Anneke Ifrah for English-language editing.

Potential conflicts of interest. All authors: No reported conflicts of interest. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.

Israel Influenza Surveillance Network (IISN)

Muhamed Affawi, Arkadi Akerman, Yoav Alkan, Shlomo Amsel, Galab Asala, Lev Dinkin, Akiva Fradkin, Michael Ginzburg, Ali Haj-Daud, Kamil Hashivon, Yael Hess, Ella Kalminsky, Angela Kozminsky, Yoseph Laks, Tali Levenstein, Alexander Lustman, Nadia Mansour Washahi, Nir Marcus, Oded Mazor, Idit Meshulach, Margarita Neimark, Shiri Perga-Menzov, Karen Rechavi, Nirit Segal, Eva Shlank, Rephael Singer, Paul Slater, Ronen Yunes, and Ran Zivner.