To evaluate the indirect mortality benefits of the Japanese schoolchildren vaccination program, we studied seasonal influenza-related excess mortality rates among Japanese and American seniors from 1978 to 2006. We found that crude excess P&I mortality rates in Japanese seniors increased by 93% after the schoolchildren vaccination program was discontinued in 1994. A large fraction of this increase was due to more frequent circulation of the severe influenza A/H3N2 virus subtype in recent years. After controlling for circulating subtype, trends in population demographics and baseline risk of death, we estimated that the schoolchildren vaccination program conferred a 36% protective benefit against influenza-related mortality to Japanese seniors (95% CI:17–51%). On average, we estimate that this program was associated with 992 excess P&I deaths averted per season (95% CI: 355–1,825). In addition to this statistical analysis, we developed an age-structured transmission model to estimate the indirect mortality benefits of schoolchildren vaccination (Information S1
). The model predicted a 26–52% reduction in influenza-related mortality amongst senior populations, consistent with our statistical findings (Figs. S1
To control for changes in the severity of circulating influenza viruses, we estimated trends in influenza-related mortality in the US, where influenza vaccination was targeted to high-risk groups and coverage among seniors has gradually increased to ~65%. We did not find significant changes in influenza-related mortality rates among US seniors during the study period (1978–2006), consistent with previous work 
In 2001, Reichert et al. reported that the Japanese schoolchildren vaccination program prevented between 10,000 and 12,000 excess P&I deaths per season in the entire population, but the authors did not analyze mortality data specific to seniors 
. Our estimate of 992 excess P&I deaths (95% CI: 335–1,825) averted among seniors is substantially lower. Reichert et al. estimated mortality reductions by comparing excess P&I mortality rates in 1990 to those in the 1960 s. Notably, influenza-related excess mortality rates in the 1950 s–60 s were declining sharply in all countries due to socio-economic changes. Excess mortality rates also declined during the years immediately following the 1968 A/H3N2 pandemic, likely due to the population-wide acquisition of natural immunity to these viruses over time (rather than to vaccination) 
. In addition, this study and others have shown that excess mortality estimates are very sensitive to the frequency of A/H3N2 virus circulation, which was not controlled for in the Reichert et al. study 
. Therefore, we believe our evaluation of the schoolchildren vaccination program using mortality data specific to seniors, and including recent years for comparison, is more prudent and allows adjustment for important confounders.
Our results can also be compared to those of a recent Canadian cluster-randomized controlled trial, in which immunization of ~80% of schoolchildren conferred 61% indirect protection against clinical influenza infection to unvaccinated individuals (95% CI: 8–83%) 
. Our analysis suggests a lower indirect mortality benefit of 36% (95% CI: 17–51%), although the confidence intervals of the two studies are large and overlap. Our study's lower point estimate could be explained by lower vaccination rates among Japanese schoolchildren, or lower efficacy of historical vaccines 
. We note that our age-structured transmission model corresponds well with the Canadian findings, predicting a ~60% reduction in influenza-related mortality among seniors at high vaccine coverage (Fig. S2
). Further research should focus on combining epidemiological data with mathematical transmission models to evaluate different vaccination strategies.
Several limitations of our study should be mentioned. We concentrated on P&I mortality, despite the fact that the impact of influenza is not limited to solely P&I 
. The Japanese schoolchildren vaccination program was associated with a non-significant reduction in excess all-cause mortality in seniors (not shown), which could be explained by a lack of specificity and unaccounted baseline time trends 
. Although it would have been interesting to investigate trends in other causes of death traditionally linked to influenza, such as respiratory and cardiac diseases, these data were not available to us. Additionally, our study was ecological and prone to confounders, which is why we introduced the US comparison as a control. Furthermore, we adjusted for the circulation of more virulent A/H3N2 viruses, accounted for population aging by studying age-specific mortality rates and used sensitivity analyses to explore the robustness of our findings. We also supplemented our data on vaccine doses distributed with age-specific vaccine coverage data for selected years 
. Finally, we used mortality data from summer periods, free of the effect of influenza, to adjust for time trends in the background risk of death among seniors in 1978–2006 due to differences in access to care, underlying co-morbidities or other causes. Patterns identified in adjusted excess mortality estimates were always in the same direction as those identified in crude mortality estimates.
We considered two competing explanations for the increase in excess influenza-related mortality observed in Japanese seniors after the schoolchildren vaccination program was discontinued: the growing impact of nursing homes and antiviral treatment. While the number of “Nursing Homes for the Elderly” increased 5 fold between 1980 and 2006 in Japan, the percentage of three-generation households decreased from 16.2 to 9.1% 
. However, the overall proportion of elderly persons living in nursing facilities remained low throughout the study period in Japan and could not account for the substantial increases seen in influenza-related mortality rates in recent years.
Oseltamivir use has increased substantially in Japan since 2003 in all age groups, making Japan the country with the highest annual level of oseltamivir use per capita, comprising >70% of the world's consumption in 2009 
. The decline in influenza-related mortality in the Japanese elderly observed after 2000 may be due in part to the routine use of oseltamivir and in part to increasing vaccination rates amongst the elderly, other high-risk groups, and children aged 6–13 years (, Table S1
). While it is too soon to precisely evaluate to effect of oseltamivir use on influenza-related mortality in the Japanese population, high oseltamivir usage was limited to the last 3 years of our study and would only bias our analyses towards null hypotheses.
In conclusion, numerous studies have reported that school-aged children have high influenza attack rates and play an important role in the transmission of influenza within schools, families and communities 
. Importantly, children respond well to vaccination and vaccinating this age-group is cost-effective, regardless of indirect benefits to unvaccinated contacts 
. Here, we have shown that the Japanese schoolchildren vaccination program was associated with significant reductions in influenza-related excess P&I mortality among the Japanese elderly. Our estimates of indirect vaccination benefits are conservative because they focus on P&I mortality, a fraction of the total mortality burden of influenza and because the Japanese population has not entirely escaped influenza vaccination or antiviral treatment since the schoolchildren vaccination program was discontinued. Our findings fit well with an accumulating body of theoretical and experimental evidence suggesting that high vaccination coverage of children can contribute to reductions in morbidity and mortality among non-immunized community members 
While the societal structure of Japan is markedly different from that of the US, we believe that our findings can extend to the US population for several reasons. First, several community-scale studies in the US have indicated that vaccinating schoolchildren against influenza confers herd immunity in unvaccinated community members 
. Second, the introduction of several pediatric vaccines has produced substantial herd protective effects on the population level, most notably the pneumococcal vaccine 
. In particular, the US-introduction of a vaccine targeting seven types of pneumococcal disease in young children in 2000 has substantially reduced the rates of carriage and invasive disease amongst people aged >50 years 
. Our findings support vaccination of school-aged children, a group included in the most recent ACIP recommendations, which encourage yearly seasonal influenza vaccination for all persons aged ≥6 months in the US 
. While seasonal influenza vaccine coverage remained low in US children aged 6 months – 17 years during the 2008–2009 epidemic season, they reached ~44% towards the end of 2009 
. It will be interesting to compare influenza-related disease trends in the US with those in other countries as vaccine coverage continues to increase in pediatric populations.