While the burden of CRS may be small relative to the overall disease burden of children in certain regions, the low cost of the rubella vaccine, the relatively low transmission rate of the infection compared with measles and the ability to deliver vaccine without administering an extra shot mean that it may be one of the diseases most effectively tackled. Conversely, once immunization against rubella is implemented, ceasing vaccination is likely to result in a considerably increased CRS burden; so long-term maintenance of coverage must be assured.
The classic concern for rubella vaccination has been that insufficient vaccination coverage of children may lead to an increase in the burden of CRS because of an increase in the average age of infection, insufficiently offset by reduced incidence. However, evidence, including high average age of rubella infection [20
], in many countries suggests that rubella may frequently have a low R0
(although see [34
]). In this situation, reducing incidence requires relatively lower coverage (particularly if the birth rate is not too high [6
]), and problems may rather emerge from stochastic dynamics in a metapopulation and waiting times following local extinction before re-introduction [20
]. Given an average age of infection for rubella in South Africa suggesting that R0
is not excessively high (, even in the face of a likely downward bias due to sampling targeted at measles that classically has a much lower age of infection), and the relatively low birth rate (around 20 per 1000 per year https://www.international.ipums.org/international/
), this was the focus of our analysis here.
We first characterized seasonal dynamics and the metapopulation structure of rubella transmission in South Africa. Susceptible reconstruction suggested changes in reporting rates compatible with the occurrence of a measles outbreak from 2003 to 2005 [36
] and a larger outbreak in 2009 and 2010 [37
]. Seasonal dynamics followed school terms, as frequently observed for immunizing childhood infections [19
] including rubella [20
] and district population size proved to be the key determinant of movement between districts (as measured via the duration of rubella fade-outs, see above), with larger populations producing both more infected emigrants and attracting more infected immigrants (in line with previous work [24
]). This conclusion may be affected by the fact that spatial variation in reporting rates is likely, with a higher index for reporting and testing in some provinces (e.g. Gauteng, Western Cape and KwaZulu-Natal and perhaps Northern Cape) than in others (Limpopo, Mpumalanga, North West, Eastern Cape), a pattern likely to be correlated to some extent with urbanization and population density; although this did not emerge strongly in estimates of coverage obtained via susceptible reconstruction (). Incorporating Euclidean distance between districts or ‘cost of travel’ did not improve the model, but this might be partly because the spatial scale of case data available was too coarse relative to the range of actual travel decisions (districts often cover large areas, and many settlements, thus the representation of each with distances or costs of travel between single district centroids is likely to be an oversimplification). This observation brings to the fore an important caveat relative to the broader predictions of our model: if district population sizes do not actually reflect population sizes relevant to epidemiological dynamics, then predictions relative to changing CRS burdens across the region in response to vaccination will be affected. For all our model predictions, caveats linked to heterogeneity in reporting rates should also be considered.
The missing element for developing a simulation of rubella is then the structure of transmission over age. We used two approaches (i) we fitted a smooth WAIFW to the observed age-incidence data combined with data on the demographic structure of the population (see the electronic supplementary material, appendix S2) and (ii) we used the average WAIFW indicated by diary studies across Europe (both shown in the electronic supplementary material, figure S2). The former cannot capture the complexity of changing contacts between adults and children, but conversely, it is not clear that the latter is appropriate in the social and cultural context of South Africa. However, both provided similar qualitative conclusions, and are likely between them to capture a large part of the possible range of transmission structures. Empirical surveys that could ascertain the validity of these approaches are desirable, however; and again, model predictions will be vulnerable to misspecification of contacts over age.
Combining parameters obtained from the rubella incidence data, with the chosen transmission WAIFW, we developed a model to characterize the effect of vaccination on the burden of CRS. Overall, our results indicate that a global reduction in infectious individuals moving between districts may increase the burden in smaller populations below the CCS if these are unvaccinated, because of longer waiting times following extinction (), making the interaction between population movement, coverage and the CCS a key equity question. However, the current vaccination coverage results overall in a reduction in the burden of CRS in South Africa over a 30 year time horizon (). The absence of a relationship between district population size and reported coverage for measles (e.g. in 2000, the correlation between vaccine coverage and log population size was ρ
= 0.05, d.f. = 52, p
> 0.1; similar results in other years) and relatively low variance in coverage (σ2
= 0.03 across all years; and that or lower for each individual year) is likely to be a key contributing factor. It is, however, key to note that this prediction is vulnerable to errors of model misspecification, in particular the relevance of the spatial scale of the data may affect conclusions, but uncertainties in reporting and questions relative to the age-structure of transmission will also play a role; and our estimate of CRS incidence of approximately 200 CRS cases per year () may consequently be either an over- or underestimate (previous analyses based on sero-prevalence suggest a value of around 600 [39
An interesting point revealed by the simulations is that it is apparent that the CCS may have been overestimated in previous work [33
), given under-reporting (); in reality the CCS for rubella might be much closer to that of measles (see the electronic supplementary material, figure S8). Although, theory predicts a slightly larger value for rubella given that its R0
is generally lower than that of measles [41
], this theoretical analysis did not incorporate seasonality in transmission, and troughs between major outbreaks due to seasonality are likely to determine extinction probability and increase with the R0
of the infection (see the electronic supplementary material, figure S8).
In the absence of broad stochastic effects on age-structure driven by extinction and re-colonization linked to the CCS, one might expect the response of rubella in South Africa to vaccination to be in line with previous work that suggested the 80 per cent cut-off rule [42
]. However, our predictions are in line with more recent work suggesting that with a birth rate around 20 per 1000 per year and an R0
around 6, increases in CRS over a 30 year time horizon are unlikely (although, again, model misspecification remains a potential concern). In this context, an important public health dimension of the introduction of the rubella vaccine is the degree to which it is prevalent in the private sector [45
], i.e. for individuals who receive for example MMR vaccination in an unregulated market where all vaccines are available [39
]. Assuming no spatial heterogeneity in coverage, our model suggests that such private-sector vaccination with levels of coverage between 20 and 30 per cent should be particular cause for concern (). Previous analyses suggest coverage of around 15–20% in such unregulated markets [39
Figure 6. Assuming homogeneous coverage over the whole country, and using the POLYMOD WAIFW, predicted CRS burden over 30 years for the whole country, for values of coverage between 0 and 0.9 showing the median and range of three simulations for each point. This (more ...)
To conclude, data available for South Africa shed light on basic aspects of rubella epidemiology (CCS, connectivity, seasonality), but also highlight areas of consideration in a public health setting, including metapopulation-induced changes in age-incidence, which can lead to public health equity issues. The methods we have developed could be use to explore the impact of heterogeneous vaccination in other connectivity contexts and for other infections with an age-specific impact, such as mumps. Interestingly, our model predictions are broadly positive relative to the introduction of routine rubella vaccination in South Africa, despite the relatively low measles vaccine coverage levels explored, with possible relevance to a number of countries in the region. Of course, these conclusions rest on the model assumptions (detailed above), and the data available. Model misspecification is always a risk, and key areas for future research include further detail on the age-transmission profile of rubella in developing and middle-income country settings.