Epidemiology and burden of illness of seasonal influenza among the elderly in Japan: A systematic literature review and vaccine effectiveness meta‐analysis

Abstract Background Elderly populations are particularly vulnerable to influenza and often require extensive clinical support. In Japan, nationwide passive surveillance monitors seasonal influenza but does not capture the full disease burden. We synthesized existing evidence on the epidemiology, vaccine effectiveness (VE), and economic burden of seasonal influenza in the elderly population. Methods PubMed, EMBASE, and ICHUSHI were searched for articles on seasonal influenza in Japan, published between 1997 and 2018, in English or Japanese. Grey literature was also assessed. A random‐effects meta‐analysis characterized VE of influenza vaccines among studies reporting this information. Results Of 1,147 identified articles, 143 met inclusion criteria. Reported incidence rates varied considerably depending on study design, season, study setting and, most importantly, case definition. In nursing homes, the maximum reported attack rate was 55.2% and in the 16 articles reporting mortality rates, case fatality rates varied from 0.009% to 14.3%. Most hospitalizations were in people aged >60; healthcare costs were partially mitigated by vaccine administration. Meta‐analysis estimated overall VE of 19.1% (95% CI: 2.3% ‐ 33.0%) with a high proportion of heterogeneity (I2: 89.1%). There was a trend of lower VE in older people (40.1% [−57.3‐77.2] in the <65 group; 12.9% [−8.0‐29.8] in those 65; P = .21). Conclusions Despite differences between studies that make comparisons challenging, the influenza burden in elderly Japanese is significant. While vaccines are effective, current vaccination programs offer suboptimal protection. Health economic data and cost‐effectiveness analyses were limited and represent areas for policy‐relevant future research.


| INTRODUC TI ON
Seasonal influenza is an acute respiratory illness caused by influenza type A and B viruses which are clustered into seasonal outbreaks typically lasting 8-10 weeks from late autumn to early spring. 1 These annual epidemics are responsible for approximately 3 to 5 million cases of severe illness and 290 000 to 650 000 respiratory deaths worldwide, 2 and thereby place unwelcome pressure on healthcare systems. Morbidity and mortality are disproportionately high among the elderly, the very young, and people with certain chronic diseases who are therefore targeted for influenza vaccination. 3 In common with other temperate countries, Japan suffers seasonal outbreaks and reports the second-highest number of cases in  5 Information on influenza mortality is scarce but in a global modeling estimate, the influenza-associated respiratory mortality rate in Japan was estimated at 0.2, 3.5, and 27.5 per 100 000 individuals aged <65 years, 65-74 years, and ≥75 years, respectively. 7 Considering Japan's aging population -the proportion aged ≥65 is projected to rise from 28.4% in 2019 to 35.3% in 2040 -without more effective prevention, the influenza disease burden is likely to increase. 8 National influenza surveillance data, maintained by the National Institute of Infectious Disease (NIID) and others, are available in Japan, but as with other countries these systems are designed for epidemic detection, to monitor epidemiological trends and to detect circulating influenza viruses rather than to fully measure disease burden. 9 Dedicated epidemiological studies are therefore often conducted to improve the understanding of disease incidence, severity, and risk factors, to inform health policy.
Annual vaccination against seasonal influenza using quadrivalent influenza vaccines (QIV), which include both influenza B lineages, is recommended for elderly and other high-risk population groups including those with chronic diseases, residents receiving homeand facility-based care, their regular contacts and healthcare workers. [10][11][12] One limitation to this preventative health strategy is that the vaccine effectiveness (VE) of influenza vaccination has been shown to decline in individuals aged over 65-70 years old, a consequence of age-associated immune dysfunction (immunosenescence), 13 which is one reason the WHO has advocated for the development of improved influenza vaccines. 14 Antiviral therapies are commonly used in Japan both therapeutically and, in some instances, for influenza prophylaxis, particularly in residential care and similar facilities in which influenza outbreak risk is highest. 15 Assessing the cost-effectiveness of vaccination programs requires both disease costs and information on vaccine performance at averting them. Current influenza vaccination programs have generally been shown to be cost-effective in Japan, 16 but robust economic data may be needed as inputs in future health economic analyses of new vaccines. Globally, influenza vaccine VE is affected by virus type and subtype, antigenic match between vaccine and circulating strains, the age and health status of vaccine recipients, and the time between vaccine receipt and infection. 17 Due to the heterogeneity of VE measurement and associated challenges in interpreting these data, meta-analytical approaches have been employed to understand vaccine performance and its variation across populations, seasons, and strains. 18 We conducted a systematic review of literature published in English or Japanese to synthesize evidence describing seasonal influenza epidemiology, prevention, and health economics in Japan, and to identify important gaps. We then conducted a meta-analysis to estimate reported VE and describe sources of heterogeneity in vaccine performance. TA B L E 1 Characteristics of studies examining influenza incidence and vaccine effectiveness in different age groups according to the studied season, target population, outcome measure, and vaccine effectiveness

| Meta-analysis methodology and statistical analysis
Studies describing the relative risk (RR) of developing influenza in vaccine recipients vs non-recipients were meta-analyzed to esti- Self-reported influenza was collected in surveys.
*Presented influenza/ILI attack rates or estimated number of incidence exclusively for elderly population aged ≥50. † Reported VE was used in meta-analysis. If not explicitly reported, VE was calculated using study data.

TA B L E 1 (Continued)
non-respiratory outcomes, and those relying on serological criteria or self-reporting as study outcomes, were excluded. RR estimates for laboratory-confirmed influenza were used where available.
A random-effects meta-analysis which incorporates betweenand within-study variance to account for seasonal and other variations across studies, was performed to combine estimates. 20 To explore heterogeneity, a meta-regression approach assessed whether study characteristics (subject age; study setting, design or circulating influenza virus subtypes) were explanatory of the overall relative risk. RR estimates were presented as a forest plot and used to estimate VE using the formula RR = 1-VE. The risk of publication bias from small studies was assessed using the approach proposed by Egger. 21 Analyses were conducted in Stata v15.0, using the Admetan package. 22

| RE SULTS
A total of 1,147 studies were identified. After reviewing titles and abstracts, 367 full text articles and six grey literature articles were reviewed for relevance, of which 143 met the inclusion criteria. A PRISMA diagram of the selection process is shown in Figure 1.

| Seasonal influenza incidence and hospitalization rate
Of the 74 studies that presented data on seasonal influenza incidence, 36 reported LCI including those using rapid diagnostic kits.
The remaining 38 studies reported non-LCI influenza outcomes such as influenza-like-illness (ILI

| Seasonal influenza incidence in comorbid populations
End-stage renal disease requiring hemodialysis was the most commonly studied comorbidity associated with influenza, probably because of the recommendation for influenza vaccination in the

| Disease severity
Influenza infections most often resulted in severe and hospitalized outcomes in the elderly Japanese population: from 2011 to 2018, over 60% of influenza hospitalizations occurred in people aged >60. 6,30,45 Severity and disease progression were also described as a function

| Mortality rate and excess mortality
Twenty articles reported mortality rates associated with seasonal influenza, 16

| Vaccine effectiveness
Twenty-six studies focused on VE (

| Meta-analysis of vaccine effectiveness
After exclusions and stratification for age a total of 21 age-specific . None of these characteristics was significantly predictive of VE at the 95% confidence level.

| Antiviral use in elderly patients
Fourteen observational studies reported on the effectiveness of antivirals (amantadine, oseltamivir, laninamivir, peramivir, and zanamivir) among elderly patients to prevent or treat various influenza-related outcomes (Table 1). When used prophylactically in hemodialysis patients, oseltamivir prevented influenza (attack rate; 6.5% in untreated vs 0% in treated group, P < .01) 34 but post-exposure administration of amantadine did not show significant reductions in fever duration following ILI in 1999. 75

| Health economic burden and costeffectiveness of preventive measures
We identified only one health economic study which directly as-   baloxavir, a new antiviral drug, and no distinction was made between time periods when trivalent and quadrivalent influenza vaccines (before/after 2015/2016 season) were in use. A meta-analysis found considerable heterogeneity between VE studies that examined VE, and we employed only a simple meta-regression approach which failed to identify strong deterministic variables. Administration of antiviral prophylaxis in institution-based VE studies was also not captured, leading to incorporation of potentially biased VE estimates. The meta-analysis was also subject to methodological limitations: we used crude RRs where adjusted, age-stratified effect sizes were unavailable and made assumptions to convert ORs to RRs.
Most included studies were observational cohorts, a design vulnerable to confounding based on the baseline health status of individuals and systematic biases affecting our results cannot be excluded. 92 Here, a large difference in VE according to circulating type or subtype was not observed, but we used national-level surveillance data which may not have been reflective of local epidemiology at the sites where studies were conducted.

ACK N OWLED G EM ENTS
The authors would like to thank Naoki Kawai of Japan Physicians Association, and Sandra S. Chaves, Yasushi Koizumi and Clotilde El Seblain, of Sanofi Pasteur, for careful review and feedback.