The broad-scale coverage goals for measles vaccination, such as those adopted by the World Health Assembly in response to the Millennium Development Goals, have resulted in significant reductions in the burden of measles disease and childhood mortality [5
]. The overall reduction of measles incidence is predicted to lead to shifts in the underlying dynamics of measles transmission and persistence that may impact the broader goal of eradication. The same classical models that have guided much of current vaccination policy (i.e. the 95% herd immunity threshold and the critical community size; [10
]) also predict that vaccination should affect the age distribution of measles cases [10
], should increase the rate of stochastic local fadeout [43
] and increase the annual variability in incidence at the local scale [21
]. Here, we illustrate, using aggregate national data, that these classic predictons do appear to hold, in general, but are mediated by local variation in birth rate. Thus, the achievement of measles reduction or elimination goals at the local (national or lower) scale is likely to require programmatic and vaccine coverage goals that are tailored to local demographic conditions.
Simons et al.
] recently showed that mortality due to measles had decreased by 74 per cent from 2000 to 2010; this reduction was achieved both due to a reduction in overall incidence and to an increase in the age distribution of cases from the high-risk under-5 years age class. Here, we have shown that the mean age of measles infection is broadly correlated with overall first-dose measles vaccination coverage, which would predict further reductions in the overall case-fatality rate with increased vaccination. However, the rate at which the mean age at infection increases with vaccination coverage is negatively correlated with birth rate. Thus, in many low-income countries, where birth rates are high, we may predict to continue to see measles infection concentrated in young children who are at high risk for mortality.
Herd immunity is a central concept to the proposition that eradication is logistically feasible in the absence of perfect immunization at 100 per cent coverage [44
]. As incidence declines, however, the increasing likelihood of stochastic local fadeout can obscure the evaluation of herd immunity as the absence of local transmission does not necessarily imply sufficient population level immunity. A single dose of measles vaccine administered at 9–12 months is only expected to achieve immunity in 84 per cent of individuals [45
], thus, a single dose alone is not able to achieve the classic 95 per cent immunity threshold. Here, we find that there is no empirical support for a threshold first-dose vaccination coverage level at which measles is likely to fade out at the national-level; neither was there a significant correlation between first-dose coverage and the rate of measles fade out. Some of the variation in observed measles fade out may be due to differences in the administration of both first and second dose opportunities through SIAs, though the effect was weak. Though one would expect a clearer relationship between vaccination and fade out, this may reflect the challenges in using reported vaccination coverage as a measure of population immunity.
The proportion of stochastic fadeout (months with less than 10 cases) is strongly negatively correlated with both population size and the size of the unvaccinated birth cohort; i.e. the number rather than the proportion of new susceptibles in the population. Lloyd-Smith et al.
] found that local persistence in wildlife populations failed to show a threshold with population size, but did increase gradually with demographic turnover rates. The results here suggest that local persistence of measles correlates with both population size and demographic rates. This suggests that broad-scale vaccine coverage goals are unlikely to have the same impact on the interruption of measles transmission in all demographic settings. Indeed, these results suggest that in order to achieve increased rates of local stochastic extinction of measles, target vaccine coverage should be scaled positively with either population size or the size of the birth cohort.
The rate of local stochastic fadeout is likely to feed back on the mean age of measles infection. In times of local measles fadeout, natural infection is removed as a source of immunity and those individuals missed by immunization programmes (due to access or refusal) will age until the next reintroduction of measles. Thus, in areas of high rates of local fadeout, we would predict to see an increase in mean age of infection relative to the predictions due to vaccination alone. Several examples of resurgent measles outbreaks in areas following periods of local elimination (Sao Paulo in 1996 [47
], Burkina Faso in 2009 [9
], Malawi in 2010 [8
], France 2012 [7
]) have been characterized by broad age distributions which are likely to result from the interaction of the age-shift due to vaccination and the ageing of the remaining susceptible population during times of local stochastic fadeout. This interaction between births and local fade out may further explain the stronger effect of birth rate than vaccination coverage on mean age of measles infection ().
Here, we have presented an evaluation of national-scale patterns as a function of both vaccine programme performance, as measured by first-dose measles coverage, and demographic rates using a broad, comparative analysis. The combined impact of vaccination and demography on susceptible recruitment has a powerful impact on measles dynamics, which underlines the key role of herd immunity in limiting epidemics [48
]. Comparative analyses across a range of other pathogens underline this point [49
Aggregate patterns of measles age distribution and persistence at the national scale are likely to obscure the complexities of local, sub-national transmission dynamics and vaccination coverage (due to access or refusal). Further, additional local variation in human movement [51
], school attendance [52
] and seasonal migration [22
] are likely to affect both the age at infection and the local persistence of measles infection. Covariation of these heterogeneities with local clustering of vaccination rates and refusal is a critical area for future work as measles vaccination continues to ramp up. The relevance of these local-scale drivers reinforces, rather than detracts, from the overall conclusion of this comparative analysis.
(a) Measles modelling and vaccination policy
In general, the development and evaluation of measles control and elimination strategies has relied on two main sources of information—country experience and modelling. Modelling can be used to generalize the lessons from individual country settings for the evaluation of novel policies. McLean & Anderson [11
] used dynamic models to illustrate that optimal vaccine targeting depends on the local demographic rates and that a single age-target for routine immunization was always preferential to a two-stage introduction strategy. Since then, models have been used to evaluate the impact of the introduction of and timing of a second dose of measles vaccine [53
], the effect of periodic SIAs [54
] and the efficacy of outbreak response vaccination [55
]—all leading to revisions in global measles policy [57
]. Detailed case studies have been useful in validating the predictions of theoretical models such as the role of seasonality and birth rate in producing locally erratic outbreak dynamics and the maintenance of regional persistence through metapopulation dynamics [21
]. Finally, models and surveillance data have been explicitly combined to develop policies that are grounded in both general theoretical understanding and local dynamics. Gay et al.
] combined serological surveillance with model predictions to quantify outbreak risk in the UK and recommend enhanced vaccination. Where permitted by local logistics, such synthesis of serological estimation of populations at risk with models provide an especially powerful tool for targeting measles control. Where serological data are not available, cases surveillance can serve as a proxy; Simons et al.
] developed models that include national surveillance data in global evaluation of progress towards the global measles mortality reduction targets.
The dynamics of measles infection and persistence, particularly as incidence declines due to the success of control measures, will be mediated by locally specific demographic forces. Our analysis here indicates that there is broad-scale correspondence between the theoretical predictions and the patterns observed across all countries. We note, however, that the individual variation from the expected relationship may reflect either local deviation from the model assumptions—e.g. due to population growth, the role of migration, local vaccine programme performance, vaccine refusal. In practice, identifying these variations from theoretical predictions could be used to identify regions that exceed programmatic expectations or to bolster elimination efforts in regions that are falling behind. Thus, while broad-scale programmatic goals have been highly successful in attaining goals for the proportional reduction of incidence and mortality, achieving the definitive goal of eradication will probably require the integration of surveillance data and quantitative models to develop control measures that are custom-tailored to local conditions.