Effectiveness of the MF59‐adjuvanted trivalent or quadrivalent seasonal influenza vaccine among adults 65 years of age or older, a systematic review and meta‐analysis

Abstract Background Standard‐dose seasonal influenza vaccines often produce modest immunogenic responses in adults ≥65 years old. MF59 is intended to elicit a greater magnitude and increased breadth of immune response. Objective To determine the effectiveness of seasonal MF59‐adjuvanted trivalent/quadrivalent influenza vaccine (aTIV/aQIV) relative to no vaccination or vaccination with standard or high‐dose egg‐based influenza vaccines among people ≥65 years old. Methods Cochrane methodological standards and PRISMA‐P guidelines were followed. Real‐world evidence from non‐interventional studies published in peer‐reviewed journals and gray literature from 1997 through to July 15, 2020, including cluster‐randomized trials, were eligible. Two reviewers independently extracted data; risk of bias was assessed using the ROBINS‐I tool. Results Twenty‐one studies conducted during the 2006/07–2019/20 influenza seasons were included in the qualitative review; 16 in the meta‐analyses. Meta‐analysis of test‐negative studies found that aTIV reduced medical encounters due to lab‐confirmed influenza with pooled estimates of 40.7% (95% CI: 21.9, 54.9; I 2 = 0%) for non‐emergency outpatient visits and 58.5% (40.7, 70.9; I 2 = 52.9%) for hospitalized patients. The pooled estimate of VE from case‐control studies was 51.3% (39.1, 61.1; I 2 = 0%) against influenza‐ or pneumonia‐related hospitalization. The pooled estimates for the relative VE of aTIV for the prevention of influenza‐related medical encounters were 13.9% (4.2, 23.5; I 2 = 95.9%) compared with TIV, 13.7% (3.1, 24.2; I 2 = 98.8%) compared with QIV, and 2.8% (−2.9, 8.5; I 2 = 94.5%) compared with HD‐TIV. Conclusions Among adults ≥65 years, aTIV demonstrated significant absolute VE, improved relative VE compared to non‐adjuvanted standard‐dose TIV/QIV, and comparable relative VE to high‐dose TIV.


| INTRODUC TI ON
It is estimated that the global disease burden from influenza includes up to a billion infections, 3-5 million cases of severe disease, and 290,000-650,000 deaths annually. 1 Adults 65 years of age or older are particularly vulnerable to the complications resulting from influenza infections with higher rates of influenza-associated complications and hospitalizations than younger people. [2][3][4] Older adults make up the majority of influenza-related deaths, with up to 90% of deaths occurring in this age group. 5 Vaccination is the most effective method of preventing influenza infections and subsequent complications. 6 The World Health Organization (WHO) recommends annual influenza vaccination for adults aged 65 or older. 7 Vaccine efficacy has been estimated to be 60% in adults aged 18-65 years, but decreases in older age groups 8,9 and is influenced by virus, vaccine, and host-related factors. A metaanalysis of test-negative design studies estimated vaccine effectiveness to be 51% for adults younger than 65 years of age compared to 37% for adults aged 65 years or older. 10 In older adults, the natural biological aging process results in immunosenescence and a suboptimal immune response to vaccination in our most vulnerable population. 11,12 Rapid antigenic drift can occur resulting in an antigenic mismatch between the influenza strains included in a vaccine and the actual circulating strains. 13 Strategies to improve the effectiveness of influenza vaccines among older adults have included the use of adjuvants or increased antigen content in the vaccine.
The squalene-based adjuvant MF59® is an oil-in-water emulsion that has been shown to increase antigen uptake, macrophage recruitment, and lymph node migration and to broaden the spectrum of antibody recognition of hemagglutinin epitopes. 14,15 When added to seasonal influenza vaccines, MF59 elicits a greater magnitude and wider breadth of immune response to vaccination thereby improving protection compared to conventional inactivated influenza vaccines. 16 with additional studies being published since that review was conducted, particularly in recent seasons. This systematic review and meta-analysis aim to identify and synthesize the available body of evidence to date.
The objective was to determine the effectiveness of seasonal MF59-adjuvanted trivalent or quadrivalent seasonal influenza vaccine (aTIV/aQIV) relative to no vaccination or vaccination with standard or high-dose egg-based influenza vaccines among adults 65 years of age or older using real-world evidence through a systematic review of the literature and meta-analysis of comparable data.

| ME THODS
Cochrane methodological standards and preferred reporting items for systematic reviews and meta-analysis protocols (PRISMA-P) guidelines were employed. The protocol was registered with the international prospective register of systematic reviews, PROSPERO 19 (CRD42020177747).

| Eligibility criteria
Eligible study designs included prospective or retrospective noninterventional cohort, case-control, and test-negative design studies. Cluster-randomized controlled trials were eligible if the vaccine was assigned at the facility level leaving the vaccination of the individual at the discretion of the patient or attending healthcare provider. Reports published from the year of aTIV first's licensure (1997) through to the final search (July 15, 2020) were eligible for inclusion.
Only original research was considered. Effect estimates based on meta-analytic approaches were excluded. Information from peerreviewed journals as well as gray literature was included if it was available in English, French, Italian, or Spanish.

| Search
We used database subject terms and text words for FLUAD or influenza vaccines. The following search terms were included in the electronic databases: FLUAD terms (ie, fluad or MF59 or MF59 or aTIV or aQIV or chiromas or gripguard or Influpozzi Adiuvato or allV3 or allV4), influenza virus terms (ie, influenza vaccines, influenza, influenza*, flu, human, quadrivalent, influenza virus, influenza a virus, h1n1 subtype, h3n2 subtype, influenza b virus), vaccine terms (ie, vaccin* or immuni* or innoculat*), and types of vaccines (ie, adjuvant* or squalene* or emulsion*). Table S1 details the search strategy.

| Study selection and data collection
Duplicate references were removed using EndNote TM prior to review. Two reviewers independently assessed titles and abstracts to identify potential literature for full-text review. The reviewers came to a consensus about eligibility when selections were discrepant.
After downloading the available full-text reports, the two reviewers conducted a review of the full text. Reference lists were search to identify other eligible reports.
The two reviewers independently extracted data using predefined fields. Extractors came to agreement on the eligibility of reports and the data abstracted through consensus. Authors of abstracts were asked to provide posters, presentations, and publications of data presented at conferences/meetings. If more than one source was available, data from the more complete of the sources were extracted.

| Risk of bias assessment
Risk of bias was assessed at the outcome level using the risk of bias in non-randomized studies of interventions (ROBINS-I) tool. 20 Reviewers discussed discrepant results to reach consensus. The ROBINS-I results were used to inform the overall assessment using the GRADE approach. 20,21 Funnel plots/Egger's test of bias was not used due to the paucity of similar studies 22 (all meta-analyses included <10 studies).

| Synthesis of results
Random effects models were specified a priori for meta-analyses in anticipation of heterogeneity in vaccine effectiveness due to viral, vaccine, and host factors. Heterogeneity was assessed using the I 2 statistic. Effect estimates were not pooled for <3 comparable studies nor for different study designs. 22 Laboratory-confirmed influenza and clinically diagnosed (not laboratory confirmed) influenza were not pooled due to differences in the sensitivity and specificity of these diagnostic criteria.
Separate meta-analyses were conducted for absolute and relative VE. Odds ratio (OR) was pooled with relative risk (RR) and incidence rate ratio (IRR) when the rare disease assumption applied 23 or if incidence-density sampling was used in the study reporting the OR.
VE estimates based on medical encounters in different clinical settings (ie, hospital, ED, OP) were pooled for relative VE (only) if there were three or more studies that were otherwise comparable. VE estimates were pooled across seasons and countries to allow for the evaluation of general trends regardless of differences that may have impacted VE estimates. Only adjusted estimates were reported due to the potential for confounding and/or bias due to population differences. 24 The primary adjustment method was used when specified by the authors; otherwise, data were pooled based on comparability of adjustment methods. Pooled estimates were reported regardless of high heterogeneity (ie, I 2 ≥ 75% and p < 0.05) but were investigated using subgroup analyses where possible. All pooled effect size estimates include 95% confidence intervals.

| Study selection
The applied search criteria yielded a total of 6153 potentially relevant records. Duplicate records (n = 1779) were removed by the research librarian, and the titles and abstracts of the remaining 4367 records were screened for potential eligibility by two independent screeners. Seven additional records were identified through hand searches and included in the screening.
The study identification, screening, and eligibility assessment process are summarized visually in the PRISMA diagram ( Figure 1).

| 815
Following the exclusion of 4287 records during title and abstract screening, 87 full-text records were assessed for eligibility. Reports were excluded because they were review articles, systematic reviews, literature reviews, or meta-analyses, the vaccine of interest (aTIV/aQIV) could not be separated from other vaccines or was not clearly delineated, they were a protocol or interim report, a duplicate record, an erratum to a record that already incorporated the changes, or the age criterion for participants was not strictly met.
Twenty-six records from 21 studies were deemed eligible for inclusion. Consensus was reached for all records selected by the reviewers.

| Risk of bias
All studies except one were considered at moderate risk of bias using the ROBINS-I tool (Table S3) Estimates from four studies that reported adjusted VE estimates for aTIV in preventing OP office visits due to lab-confirmed influenza (any strain) ranged from 16.2% to 58.1%, with a pooled VE estimate of 40.7% (21.9, 54.9; (I 2 = 0%, p=0.44) [28][29][30][31] (Figure 2A). The fifth study reported VE against influenza A only (−1.1%). 26 Estimates from the three studies that reported adjusted VE estimates for aTIV in hospitalized patients ranged from 48.3% to 75.6%, with a pooled estimate of 58.5% (40.7, 70.9; I 2 = 52.9%, p = 0.12) 25,27,30 ( Figure 2B). The results suggest that aTIV was effective in the prevention of laboratoryconfirmed influenza in both OP and hospital clinical settings. the studies evaluated cases diagnosed in a hospital setting with controls that were either hospital- 33,34,36 or community-based. 35,37 One case-control study 37 was excluded from the meta-analysis since the study pertained to a different outcome (pneumonia, stroke, or myocardial infarction) and comparator (TIV) than the other four studies. As shown in Figure 3, estimates from the four studies reporting VE estimates for aTIV in preventing hospitalization for influenza or pneumonia ranged from 48% to 87.8%, with a pooled estimate of 51.3% (39.1, 61.1; I 2 = 0, p = 0.42). 33-36

| Effectiveness against influenza illness, not laboratory-confirmed: cohort design
Eight studies used administrative data sources for medical care (clinic, ED, or hospital visits) to assess the effectiveness of aTIV relative to other influenza vaccines. Three of the studies were conducted in Italy using hospital catchment areas or OP rosters 38,41,42 and five studies were conducted in the USA using Medicare 39,40 or other medical and pharmacy claims-based information and medical records. [44][45][46][47]51 These studies were conducted in the 2006/07 through 2018/19 influenza seasons. Six studies compared aTIV to TIV, [38][39][40][41][42]44,47 five to QIV, [38][39][40]44,47 and five to high-dose TIV. 39,40,44,47,51 Six of the eight studies reported on VE against medical encounters for influenza with or without pneumonia in various clinical settings including: OP; hospital or ED; or OP, hospital, or ED.
The relative VE estimate for the prevention of influenzarelated medical encounters (hospitalization, ED visit, or OP visit) comparing aTIV to TIV ranged from −11.9% to 33%. As shown in Figure 4A, the pooled relative VE estimate showed a benefit of aTIV relative to TIV at 13.9% (4.2, 23.5) but with considerable heterogeneity (I 2 = 95.9%, p < 0.01). 39,41,42,44,47 The relative VE of aTIV compared to QIV ranged from −6.6% to 36.3% with a pooled estimate of 13.7% (3.1, 24.2; I 2 = 98.8%, p < 0.01), indicating a benefit of aTIV over QIV 39,40,44,47 (Figure 4B). The relative VE comparing aTIV to HD-TIV for reducing any medical encounters due to influenza and/or pneumonia ranged from −14.9% to 16.6% in five studies. 39,40,44,47,51 The pooled estimate from four studies with similar outcomes was not different for aTIV compared with HD-TIV at 3.2% (−2.5, 8.9), although there was considerable heterogeneity (I 2 = 94.5%, p < 0.01) between studies 39,40,44,47 (see Figure 4C). The Van Aalst 43,51 study was not included in the metaanalysis since the outcome used (any respiratory condition ICD10: J) was broader than the outcome for others studies (influenza with or without pneumonia). In a sensitivity analysis, the relative VE of aTIV vs HD-TIV remained non-significant when the Van Aalst study was included in the meta-analysis.

| Cluster-randomized trials
A cluster-randomized trial of 823 nursing homes in the USA with almost 53,000 long-term stay residents was conducted during the 2016/17 influenza season. 49 39 Cocchio et al, 42 and Mannino et al 41 restricted their analyses to peak periods of influenza activity, for example. In the absence of other potential sources of bias, it is expected that restricting to a period of higher influenza activity would increase the specificity of the outcome definition resulting in a more accurate, but potentially less precise, estimate due to reduction in sample size. 53 It is likely that the differences in the statistical heterogeneity in meta-analyses of absolute compared to relative VE analyses were However, high precision does not necessarily equate with high accuracy and therefore the observed heterogeneity could be the result of residual confounding, true variability in the underlying effects, or a combination of the two.
The use of random-effects models for the meta-analyses was specified a priori in anticipation of the different sources of heterogeneity described above. As future studies are published, additional subgroup analyses may become feasible and may allow for the use of additional methods like meta-regression to better evaluate sources of heterogeneity. Future meta-analyses could estimate the width of the underlying distribution with the use methods like Bayesian meta-analysis which can generate a prediction interval that aims to estimate the width of the distribution.

| CON CLUS IONS
The MF59-adjuvanted trivalent influenza vaccine was effective in preventing influenza in adults 65 years of age or older. Compared to standard-dose egg-based QIV and TIV, aTIV was significantly more effective in preventing influenza-related medical encounters (illnesses or hospitalizations). The effectiveness of aTIV was comparable to high-dose TIV in preventing influenza-related medical encounters. High heterogeneity was observed for all relative VE analyses with seasonal variation, distinct populations, and diverse outcomes potentially playing a role in the variability of the effect sizes coupled with precise effect estimates. As such, further research is needed to confirm the findings for relative VE.

PEER R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/irv.12871.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.

Brenda L. Coleman
https://orcid.org/0000-0002-7144-4827 F I G U R E 4 Forest plot of adjusted aTIV relative VE estimates compared with (A) standard-dose TIV, (B) standard-dose QIV, and (C) high-dose TIV for preventing influenza with/without pneumonia. Adults 65 years or older, cohort study design studies (hospital, ED, or non-ED outpatients). *Cocchio may have included some QIV in later seasons. aTIV, adjuvanted trivalent inactivated vaccine; ED, emergency department; GP, general practitioner; I sq, I 2 ; rVE, relative vaccine effectiveness; TIV, trivalent inactivated vaccine; VE, vaccine effectiveness. Pooling weight based on DerSimonian and Laird random-effects meta-analysis