The association between influenza vaccine effectiveness and egg-based manufacturing technology: literature review and US expert consensus

Abstract

Background

Influenza is associated with significant disease burden in the US and is currently best controlled by vaccination programs. Influenza vaccine effectiveness (VE) is low and may be reduced by several factors, including egg adaptations. Although non-egg-based influenza vaccines reportedly have greater VE in egg-adapted seasons, evidence for egg adaptations’ reduction of VE is indirect and dissociated, apart from two previous European consensuses.

Methods

This study replicated the methodology used in a 2020 literature review and European consensus, providing an updated review and consensus opinion of 10 US experts on the evidence for a mechanistic basis for reduction of VE due to egg-based manufacturing methods. A mechanistic basis was assumed if sufficient evidence was found for underlying principles proposed to give rise to such an effect. Evidence for each principle was brought forward from the 2020 review and identified here by structured literature review and expert panel. Experts rated the strength of support for each principle and a mechanistic basis for reduction of VE due to egg-based influenza vaccine manufacture in a consensus method (consensus for strong/very strong evidence = ≥ 3.5 on 5-point Likert scale).

Results

Experts assessed 251 references (from previous study: 185; this study: 66). The majority of references for all underlying principles were rated as strong or very strong supporting evidence (52–86%). Global surveillance, WHO candidate vaccine virus selection, and manufacturing stages involving eggs were identified as most likely to impact influenza VE.

Conclusion

After review of extensive evidence for reduction of VE due to egg-based influenza vaccine manufacture, influenza experts in the US joined those in Europe in unanimous agreement for a mechanistic basis for the effect. Vaccine providers and administrators should consider use of non-egg-based influenza vaccine manufacture to reduce the risk of egg adaptations and likely impact on VE.

Keywords:

• Influenza
• influenza vaccine
• vaccine effectiveness
• egg adaptations
• antigenic drift
• vaccine development

Introduction

In the 2017/2018 season, influenza was estimated to be responsible for 710,000 hospitalizations and 52,000 deaths in the US alone¹. Influenza vaccination is widely regarded as the most successful strategy for reducing morbidity and mortality of the disease and has been endorsed by virtually all specialty societies and public health agencies. However, influenza vaccine effectiveness (VE) is low. Mean influenza VE in the US between 2009 and 2020 was estimated at approximately 43%², with a meta-analysis suggesting lowest VE for H3N2³. Older estimates have been higher but are also suboptimal⁴. Low influenza VE has been related to host factors that are difficult to control from a public health standpoint, such as older age⁵, serial vaccination⁶, low immunocompetence and underlying comorbidity⁷. Antigenic mismatch between vaccine and circulating strains may also reduce influenza VE and has been proposed to occur due to antigenic drift⁸ and egg adaptations, host-adaptive antigenic changes in viral strains during egg-based influenza vaccine manufacture^9–18, the most common means of vaccine production in the US¹⁹.

Antigenic drift is a naturally occurring evolutionary phenomenon involving selection of genetic mutations adaptive to host immunity, with mutations in circulating viral strains contributing to antigenic mismatch when affecting the conformation of proteins at antibody binding sites²⁰. Occurrence of such antigenic drift after pre-season recommendation of vaccine virus strains by the WHO, usually in February in the northern hemisphere, may contribute to low VE in a season even if recommendations accurately predict dominant strains from previous season influenza activity and global surveillance reports²⁰. Antigenic mismatch has also been proposed to arise during later stages of egg-based influenza vaccine manufacture, such as reassortment of recommended wild-type vaccine strains by WHO Collaborating Centres for adaptation to high growth in eggs to optimize manufacturing yield²¹. Reassorted strains are then provided to influenza vaccine manufacturers as candidate vaccine viruses (CVVs). Growth of seed vaccines by inoculation and incubation of CVVs in eggs may provoke adaptation of virus strains to avian rather than human host cells²² before harvest and purification to provide monovalent strains or combined polyvalent vaccine batches for release and distribution. Differences between human (a-2,6 type) and avian (avian a-2,3 type) cell sialic acid residues, to which influenza virus hemagglutinin binds, may result in antigenic mismatch between candidate viruses and circulating strains^9,18,23–26. In seasons affected by egg adaptations, non-egg-based influenza vaccines have shown greater effectiveness than egg-based and were associated with lower health and economic burden^27,28.

Of all influenza subtypes, A/H3N2 appears subject to the most extensive and frequent antigenic changes arising from genetic mutation. In the US 2014/2015 season, over 80% of characterized circulating A/H3N2 virus strains varied from vaccine strains, and VE for A/H3N2 was only 13%²⁹. Low A/H3N2 VE in 2012/2013³⁰, 2016/2017^31,32 and 2017/2018³³ seasons in the US has been attributed to egg adaptations. During the 2016/2017 season, egg-adapted A/H3N2 strains lost a glycosylation site, reducing their likelihood of eliciting antibodies effective against circulating strains due to the role of the site in binding influenza hemagglutinin^31,32. Antibody-mediated inhibition of circulating virus hemagglutination by ferret sera exposed to vaccine-homologous strains suggests poorer antigenic match for egg-propagated strains³⁴. In the 2017/2018 season in the US, predominated by A/H3N2, 48.2% and 77.3% of A/H3N2 viruses tested were well-inhibited by ferret antisera raised against egg-propagated A/Hong Kong/4801/2014 and A/Singapore/INFIMH-16–0019/2016 reference viruses, respectively. This was much lower than the 93.4% well-inhibited by ferret antisera raised against cell-propagated A/Michigan/15/2014³³. A 2021 expert consensus estimated that egg adaptations were likely to reduce VE against A/H3N2 by up to 16% in the 18–64 years age group, and an average of 9% across all strains in all age groups³⁵.

Seasonal and geographic variation in circulating strains, WHO strain selection, and agent-host and immuno-epidemiologic factors³⁶ make it difficult to study the impact of egg adaptations on VE due to uncertainty of the presence and degree of effect in a given season. As a result, most findings on the topic remain dissociated. In addition to the consensus conducted in 2021³⁵, a structured literature review and consensus conducted with European experts in 2020 aimed to overcome lack of direct evidence for reduction of VE due to egg adaptations during influenza vaccine manufacture³⁷. Findings suggested a mechanistic basis for the relationship between egg adaptations and low VE. Since then, research on antigenic drift, egg adaptations, influenza vaccine strain selection and manufacturing methods has increased. The aim of this study was to update the 2020 European structured literature review and consensus on the mechanistic basis for an association between reduced influenza VE and egg-based manufacturing technologies with US experts to determine support for the effect with twice as many experts representing opinion in a second major vaccine market.

Method

This study was conducted between June 2021 and January 2022 and replicated the methodology used in a 2020 literature review and European consensus³⁷. The aim of this body of work was to unite existing dissociated findings on which the hypothesis of reduced influenza VE due to egg-based manufacturing methods rests. The objective of this study was to provide an updated review and consensus of 10 US experts to answer the research question, “Is there a mechanistic basis for an association between reduced influenza VE and egg-based manufacturing technologies, particularly against influenza A/H3N2?” A secondary objective was to interpret these findings in light of the study conducted with European experts in 2020 to determine whether further support for the mechanistic basis was apparent.

In both studies, a mechanistic basis for reduced influenza VE due to egg-based manufacture was assumed if sufficient support was identified for underlying principles proposed to give rise to such an effect. Component principles were defined as antigenic drift, stage(s) in egg-based influenza vaccine manufacture and candidate virus selection, egg adaptation changes in influenza vaccine manufacturing processes and influenza vaccine manufacturing stages most likely to reduce VE (Figure 1). Evidence for each principle was brought forward from the previous 2020 study and combined with evidence identified here by structured literature review and expert panel (AC, DHC, SD, NE, VCH, RJh, ACN, MV, RV and LBA). Experts assessed strength of support for component principles to determine extent of consensus on a mechanistic basis for reduced VE due to egg-based influenza vaccine manufacture. Expert rating of the evidence was interrogated using statistical methods.

Figure 1. The research question and component principles.

Structured literature review

The review was guided by the Cochrane Handbook for Systematic Reviews of Interventions where applicable. The aim of the review was to generate a list of references for each component principle. A protocol, including search strategy, was developed a priori by two authors (RJa, RW) independently and was reviewed by two authors (KA, AS). The search strategy was the same as in the previous consensus conducted with European experts in 2020³⁷ but was limited to studies published between 2019 and 2021, as references identified by the previous review were carried over. Search strings and inclusion/exclusion criteria were developed using the PICOS framework (Supplemental Table 1). The search was limited to studies on human influenza published in English without geographical limitation. Separate searches were completed for each component principle with specific limiters (Supplemental Table 1). Searches were conducted in PubMed and the Cochrane Library with date of last access June 2021. Working independently, two research analysts screened titles and abstracts with disputes settled by discussion. One analyst screened full-text articles for eligibility and extracted data on author, year of publication, title and journal into an extraction table developed a priori. This was quality checked by the other analyst with disputes settled by discussion. Assessment of study quality was not carried out at this stage, as experts assessed quality of evidence in the consensus exercise.

Expert panel

Eleven US influenza vaccine specialists were recruited to the expert panel using purposive sampling via professional networks, publication scanning, and through a global pharmaceutical company’s US medical department. Experts were healthcare professionals or virologists in the US with a special interest in influenza vaccination evidenced by publications on influenza burden and/or experience as a medical committee chair, clinical commissioning group lead, or in other senior roles that require specialist knowledge of influenza vaccination.

Assessment of expert opinion

Opinion of the expert panel was assessed in the two-stage Jandhyala method, which combines literature review and a consensus exercise to develop highly accurate disease-specific construct measures and core datasets in contexts with paucity of data or where generic measures are inaccurate^38–42. It has been used for research questions that cannot be addressed by traditional methods, such as in this case, where the effect occurs unpredictably in the real-world context such that direct observation is impracticable in the usual course of planned research^35,37.

The Jandhyala method was delivered in two rounds using the qualitative online survey from the consensus study conducted with European experts in 2020³⁷ with updates to include evidence identified in the present study. Awareness Round (1) determined experts’ baseline knowledge of the component principles and supporting evidence. Participants gave free-text responses to open-ended questions asking for their opinion on each component principle and a list of references to support their stance. Experts could conduct independent searches of the literature during survey completion. Responses and references were coded by a research analyst, and after removing duplicates and applying inclusion criteria, expert-generated references were added to those from the structured literature review and the previous European study.

Consensus Round (2) assessed consensus opinion on the support for each component principle and the mechanistic basis for the effect. Text extracts from the structured literature review and expert-generated references were presented to the panel for appraisal. Text extracts were selected from full-text articles identified as most relevant to answering the research question by a research analyst and quality checked by a systematic reviewer experienced in influenza virology. To manage the large number of references, publications were divided into five groups; each group was presented to two experts. Experts had access to full-text articles and were asked to rate the strength of each reference as support for the respective component principle on a 5-point Likert scale from “not evidence” to “very strong evidence.”

Cumulative frequency and evidence rating for each reference were recorded and analyzed from all survey answers. A reference was considered supportive of a component principle if the median agreement level on the Likert scale was ≥ 3.5. Summary statistics were generated with equal weighting given to all references and presented to experts in a final round in which they were asked whether cumulative evidence supported a potential mechanistic basis for an association between egg-based influenza vaccine manufacturing and reduced VE.

Statistical analyses

To determine relative growth in the body of knowledge on each component principle, average publication rate per annum of references included in the review was assessed. Experts’ rating of the evidence was assessed visually by density plots. To account for observed findings, rating scores were subjected to statistical analysis by expert and time of publication. Significant difference between paired US experts’ rating of the same evidence was determined by two-sample t-test. Experts’ median graded scores for each reference were computed for literature published from 1969–2018 and 2019–2021, and a Wilcoxon rank sum test with continuity correction determined significant difference between the calendar ranges. STATA version 14 was used for all statistical analyses.

Results

Due to attrition of one expert for reason of unavailability, 10 experts completed the study, comprising two virologists, one expert in public health, one expert in pharmacy-based immunization practices, and six infectious disease specialists based throughout the US.

Identification and availability of evidence for component principles

Modified PRISMA flow charts for all component principles (Figure 2) and each component principle (Supplemental Figures 1–4) show identification of references reviewed in this study. Of the 288 identified by this review, 35 were included. Unique references from the previous consensus (n = 185) and US expert panel (n = 31) were added to these, giving 251 overall. Mean references provided by US experts were generally equivalent except for component principle four (CP1: n = 6.4; CP2: n = 4.1; CP3: n = 5.3; CP4: n = 3.6). The 251 references were divided among five groups (n = 47–55 references each). Number of references was highest for component principle one with stepwise sequential decrease through subsequent component principles (CP1: n = 165, 66%; CP2: n = 44, 17%; CP3: n = 29, 12%; CP4: n = 13, 5%). Most references comprised historical modelling of genetic and antigenic data, studies that genetically and antigenically characterized the virus, and surveillance and effectiveness reports by national and global agencies. Mean publication rate per annum of references included in this study showed substantial growth between 2004 and 2021; greatest increase was observed in component principle one, antigenic drift, with sequential decrease observed in each subsequent component principle (Supplemental Figure 5).

Figure 2. Modified PRISMA flowchart for all component principles.

Figure 3. Expert rated stages in influenza vaccine manufacture most likely to impact VE. Strongly agree (dark blue, score 3.5–4): global surveillance and WHO CVV selection; Strongly agree (blue, score 4–4.5): manufacturing stages involving eggs; agree (light blue, score 2–2.5): purification, batching, quality control/release.

Figure 4. US (blue) and European (green) expert rating of evidence for component principles one through four. Dark blue: unique references added since the previous consensus in 2020.

Expert assessment of evidence for component principles

The majority of references for all component principles were rated as strong or very strong supporting evidence (CP1: 87/165 (53%); CP2: 38/44 (86%); CP3: 20/29 (69%); CP4: 9/13 (69%). Experts scored evidence from the literature reviews as stronger than evidence sourced by fellow experts, except for component principle four. Abbreviated lists of the highest ranked references for each component principle can be found in Supplemental Table 2. Of stages in influenza vaccine manufacture, experts identified global surveillance, WHO CVV selection, and manufacturing stages involving eggs as most likely to impact VE (Figure 3). Experts gave unanimous support for a mechanistic basis for an association between reduced influenza VE and egg-based manufacturing technologies.

Comparison between US experts

Two-sample t-test showed no significant differences between paired US experts rating the same evidence in each assigned group (Table 1). Density plots for rating of references published in 1969–2018 and 2019–2021 showed varying distribution of scores between the time periods for some experts (E1, E3, E6, E8, E9) with a tendency towards lower scores for more recent references (Supplemental Figure 7). However, Wilcoxon rank sum test with continuity correction showed no significant differences between the time periods (W = 59, p-value = 0.4779).

Table 1. Two sample t-test of difference between two experts reviewing the same literature.

Experts Grouped	t-statistic, degrees of freedom	p Value
E1, E2	t = −0.0017359, df = 280	0.9986^a
E3, E4	t = −0.0029235, df = 328	0.9977^a
E5, E6	t = 0.0017208, df = 304	0.9986^a
E7, E8	t = 0.00044045, df = 298	0.9996^a
E9, E10	t = −0.00019286, df = 286	0.9998^a

^a No statistical difference.

Comparison with the 2020 European review and consensus

Patterns of evidence rating and availability were largely similar to the 2020 European consensus apart from lower ratings for US experts for component principle one (Figure 4). Matched comparison of US and European expert rating of references for component principle one confirmed a lower percentage rated as strong or very strong for US experts (US: 69/123, 56%; Europe: 95/123, 77%). An overall agreement rate of 60% (74/123) between US and European experts was observed (agree: strong or very strong n = 58/123, moderate n = 11/123, weak n = 5/123; disagree: n = 48). Of the 48 disagreed references, US experts rated more as weak or moderate than European experts (US w/m, Europe s/vs: n = 37; US s/vs, Europe w/m: n = 11). With statistical within-group comparisons for US experts, this suggests a slight tendency for US experts to rate some evidence for antigenic drift as lower than European experts. Number of references identified for other component principles were insufficient for robust direct comparison.

Discussion

There is a need for more effective influenza vaccines. Findings support the hypothesis that egg adaptations may contribute to variation in VE, but evidence for the effect is predominantly indirect and dissociated. Lack of direct evidence was addressed by a review and consensus of European experts in 2020³⁷. This update included 66 references published since and validated findings with experts from the US, a major global vaccine market. The majority of evidence for each component principle was rated as strong or very strong, and US experts unanimously agreed there is potential for a mechanistic association between egg-adaptations and reduced VE.

Component principle one, antigenic drift, had the highest number of references rated as strong or very strong evidence and the highest growth in publication rate since 1969 of all component principles. This was supported by previous findings. Antigenic drift was first identified in 1965 and has been well-evidenced, resulting in a generally accepted understanding of theory and mechanism^43–45. US experts had a tendency to rate some references less highly than European experts. This may be due to a preference for certain types of evidence. For example, US experts rated historical variation in the genetic diversity of various subtypes^46–49 as most supportive of antigenic drift. While European experts also rated these references as strong or very strong, they were not among their most highly rated references, which included greater variation in type of evidence than US experts. Both groups rated surveillance evidence across various geographical regions, including phylodynamic reports and antigenic mapping, as highly supportive^50–53.

For component principle two, experts agreed that eight steps in egg-based manufacture and CVV selection provided a level of detail appropriate for the purpose of this research (Figure 3). This component principle had fewer references, likely because manufacturing processes are not generally addressed by scientific research. Results were largely in line with the European consensus despite some minor variation in references cited. This was expected, as egg-based manufacturing has remained unchanged in recent years. US experts rated highly reports by governing bodies⁵⁴, analyses of egg- and cell-based manufacturing methods^55,56, and accounts of the manufacturing process and its challenges^57–62. Findings emphasized the need for processes to maximize strain specificity and antigenic match.

The most highly rated references for component principle three, egg-adaptation changes in the influenza vaccine manufacture process were in line with the European consensus and included antigenic analyses comparing cell- and egg-grown viruses^26,63, studies showing altered antibody receptor binding arising from egg-adapted genetic mutations^33,64,65, and studies that linked genetic variation in egg-adapted strains to reduced VE in the 2012/2013 season in Canada²⁵.

For component principle four, experts agreed that global surveillance, strain selection, and manufacturing steps involving eggs were the most likely to impact VE. The first two stages here may influence the antigenic match of CVVs to circulating strains, and manufacturing steps involving eggs include proliferation and reassortment processes that may result in egg-adapted virus strains^56,62. Similar references were rated by US and European experts as the strongest support for this component principle. Of these, one showed antigenic mismatch between a circulating strain and the receptor binding sites of patients immunized with an egg-adapted strain⁶. The other showed the need for more effective influenza vaccines, the complex and multi-factorial nature of VE and challenges in developing solutions⁵. Few solutions have been identified to control the impact of host factors on influenza VE, and the ability of global surveillance and strain selection to ensure VE is limited due to antigenic drift⁵². Thus, avoiding manufacturing steps involving eggs may be the most feasible way to improve influenza VE.

This study adds to evidence challenging established vaccine production methods and supports the widespread adoption of newer vaccine-manufacturing technologies that can provide a better match to circulating strains. Methods for manufacture of cell-based, recombinant, and next-generation mRNA influenza vaccines avoid egg-based methods and likely associated reduction in VE. Influenza vaccines using this technology are in development^66–69. European experts have estimated that egg adaptations may reduce VE by up to 16%³⁵. A 5% absolute increase in VE may have prevented an additional 1,050,000 illnesses and 25,000 hospitalizations in the US during a single season⁷⁰. Avoiding egg-based influenza vaccine manufacture may therefore substantially decrease disease burden and healthcare-associated costs. Despite this, available evidence in support of egg-adaptations was approximately four times lower than that for antigenic drift despite similar estimated reduction of influenza VE³⁵, suggesting a need for further research. Findings here suggest that further updates and replications of this review and consensus may be useful.

Limitations

Strength of evidence against component principles was not explicitly measured by the Likert scale used for expert rating, as the variable of interest in this study was support for component principles. This was unlikely to impact overall findings, as evidence ‘against’ a principle has tended not to disprove the phenomena itself but to preference one effect over others as a reason for low VE in certain seasons. In this study, mean scores suggested that such evidence was rated ‘weak’ for the disfavoured principle (for example, Skowronski et al. 2014 for component principle one). Publication bias could be considered a factor in higher relative frequency of supportive evidence.

Inclusion of only US experts may limit the generalizability of these findings; however, the experts in this study doubled the 10 who completed the previous European consensus. Though purposive sampling may have introduced potential sampling bias, expert affiliations were sufficiently diverse to consider them representative of the population of US influenza vaccination experts. As with all consensus methods, there was a risk of oversimplification of the topic of study due to inherent methodological limitations⁷¹. However, expert consensus was appropriate for the research question due to the need to unite dissociated and methodologically diverse findings. Finally, the study did not investigate the impact of egg-adaptation changes on different strains of influenza virus, even though reduction of influenza VE due to egg-adaptations may vary by strain³⁵. Further research could usefully address this topic.

Conclusion

Experts from the US agreed that there is a mechanistic basis for an association between reduced influenza VE and egg-based manufacturing technologies. Evidence rated as strong or very strong by the US experts for the four component principles suggested that, of the steps identified as most likely to impact influenza VE, steps involving eggs were the only ones that were both likely to reduce VE and be immediately manageable through human intervention. There is now US and European expert agreement for the increased risk of reduced influenza VE resulting from egg-based manufacturing techniques. Increasing the use of non-egg-based manufacturing that avoids egg-adaptations is a currently available strategy that may improve influenza VE.

Transparency

Declaration of funding

This research was carried out by Medialis Ltd, a medical affairs consultancy and contract research organisation. CSL Seqirus, Inc. funded this work but did not participate in the design of the study; the collection, analysis or interpretation of the data; the writing of the manuscript, apart from review and approval; or the decision to publish the results. This study received no external grants.

Declaration of financial/other relationships

AC, SD, NE, VCH, ACN, and RV declare no conflicts of interest. KA and AS are paid employees and shareholders of CSL Seqirus, Inc. RJa and RW are employees of Medialis Ltd. The expert panel (AC, DHC, SD, NE, VCH, RJh, ACN, MV, RV, LBA) were remunerated by CSL Seqirus, Inc. for their involvement in this study. DHC has completed consulting work for CSL Seqirus, Inc and is a recipient of investigator-initiated grants to his university from Pfizer to study pneumococcal vaccines and Sanofi Pasteur and CSL Seqirus, Inc. to study influenza vaccines. RJa is the founder and CEO of Medialis Ltd., and RW is a paid employee of Medialis Ltd., which was commissioned and funded by CSL Seqirus, Inc. to conduct the research. RJh is a consultant for AstraZeneca, CSL Seqirus, Inc. and Dynavax, and he receives an editorial stipend from the Pediatric Infectious Diseases Society and royalties from UpToDate. MV is on an advisory board for Sanofi, which makes influenza vaccine; has done clinical trials with Pfizer, GSK, Sanofi, and others; and has received honorarium from CSL Seqirus, Inc. The Jandhyala method was developed by Dr Jandhyala (RJa) but is free of commercial licensing restrictions and while used as part of proprietary methodology, is not a direct means of commercial gain for the author.

Peer reviewers on this manuscript have received an honorarium from CMRO for their review work. A reviewer on this manuscript also disclosed that they occasionally consult for CSL Seqirus Vaccines plc. and Sanofi Pasteur who both manufacture egg-based influenza vaccine. The former also manufacture cell-grown vaccines and the latter recombinant vaccines. Peer reviewers on this manuscript have no other relevant financial relationships or otherwise to disclose.

Author contributions

Conceptualization, RJa and RW; Formal analysis, RJa and RW; Investigation, AC, DHC, SD, NE, VCH, RJh, ACN, MV, RV; Methodology, RJa and RW; Writing – original draft, RJa and RW; Writing – review & editing, AC, KA, DHC, SD, NE, VCH, RJa, RJh, ACN, AS, MV, RV and RW. All authors were involved in final approval of the published version and agree to be accountable for all aspects of the work.

Acknowledgements

The authors would like to acknowledge Dr Lauren B. Angelo for her contributions as an expert participant in the literature review process. Employees of Medialis Ltd. provided biostatistics, medical writing and research analytics for this work as part of their roles.