^{1}

^{*}

^{2}

The authors have declared that no competing interests exist.

Conceived and designed the experiments: JL CJEM. Performed the experiments: JL CJEM. Analyzed the data: JL CJEM. Contributed reagents/materials/analysis tools: JL CJEM. Wrote the paper: JL CJEM.

Despite a safe and effective vaccine, rubella vaccination programs with inadequate coverage can raise the average age of rubella infection; thereby increasing rubella cases among pregnant women and the resulting congenital rubella syndrome (CRS) in their newborns. The vaccination coverage necessary to reduce CRS depends on the birthrate in a country and the reproductive number, R_{0}, a measure of how efficiently a disease transmits. While the birthrate within a country can be known with some accuracy, R_{0} varies between settings and can be difficult to measure. Here we aim to provide guidance on the safe introduction of rubella vaccine into countries in the face of substantial uncertainty in R_{0}.

We estimated the distribution of R_{0} in African countries based on the age distribution of rubella infection using Bayesian hierarchical models. We developed an age specific model of rubella transmission to predict the level of R_{0} that would result in an increase in CRS burden for specific birth rates and coverage levels. Combining these results, we summarize the safety of introducing rubella vaccine across demographic and coverage contexts.

The median R_{0} of rubella in the African region is 5.2, with 90% of countries expected to have an R_{0} between 4.0 and 6.7. Overall, we predict that countries maintaining routine vaccination coverage of 80% or higher are can be confident in seeing a reduction in CRS over a 30 year time horizon.

Under realistic assumptions about human contact, our results suggest that even in low birth rate settings high vaccine coverage must be maintained to avoid an increase in CRS. These results lend further support to the WHO recommendation that countries reach 80% coverage for measles vaccine before introducing rubella vaccination, and highlight the importance of maintaining high levels of vaccination coverage once the vaccine is introduced.

In vaccination policy, rubella is an unusual case because introduction of a safe and effective vaccine can lead to an increase in severe disease. This is because the most severe outcome of rubella infection, congenital rubella syndrome (CRS), occurs in the newborns of pregnant women infected in the first trimester of pregnancy. When a vaccine to a disease is introduced at levels insufficient to eliminate the disease, the result may be to increase the average age of infection. For many diseases this is a good thing, because older children and adults tend to experience less severe outcomes than young children. However, for rubella, an increase in the average age of infection may lead to an increased risk of rubella among women of child bearing age, and hence an increase in CRS. Caution in the introduction of rubella-containing vaccine is supported by observations of suspected vaccination-associated transient increases in the CRS burden in Greece and Costa Rica

Epidemic theory and empirical observation show that the average age of infection for a vaccine preventable disease conferring lifelong immunity is predominantly determined by the population birthrate, the level of vaccine coverage and the transmissibility of the disease. In most populations the birthrate is known to some degree of accuracy through census data. Because rubella vaccine is most often distributed as part of a bivalent measles-rubella (MR) vaccine or trivalent measles-mumps-rubella (MMR) vaccine, the coverage that will be obtained upon introduction of rubella vaccine is known with reasonable certainty based on current measles vaccine coverage. However, transmissibility, generally characterized by the basic reproductive number, R_{0}, is not so easily measured. R_{0} is defined as the number of individuals a single infectious individual is expected to infect in a fully susceptible population, and is a result of biological, environmental and social factors. Because R_{0} depends on factors other than pathogen biology, no one value can be used across countries and settings. Estimates in the literature of rubella’s R_{0} range from 2 (Edmunds et al. 2000) to 12 (Cutts et al. 2000) _{0} is generally measured indirectly; hence there may be substantial uncertainty as to its value, even in a particular setting. This puts public health officials in a quandary. The two pieces of the puzzle they know, birthrate and vaccine coverage, are useless without knowing a value whose measurement requires time, resources and expertise. For instance, in a country with a birthrate of 33 per 1,000 and 60% vaccine coverage, introduction of rubella vaccine will decrease CRS cases if R_{0} is 6.8 or lower, while CRS will increase of R_{0} is greater than 6.8.

Here, we present an analysis aimed at helping policy makers, program funders and other stakeholders reason about the utility of introducing rubella vaccination in specific settings while taking into account the uncertainty in the underlying transmission dynamics of the disease. We develop a framework for presenting our results that aims to be intuitive and easy to use by a non-technical audience, while not obscuring the technical details from those who are interested. This approach may also serve as a basis for decision making in other settings where substantial uncertainty exists.

For a given R_{0}, birthrate, and vaccine coverage we simulated 30 years of rubella incidence using a previously described age structured TSIR model _{0} and birthrate but no vaccination. The model assumes mild seasonal forcing of transmission

For any particular birthrate and vaccination level, there exists a critical threshold of R_{0}. If the true value of R_{0} is below this threshold, then the number of cases of CRS will decrease if vaccination is introduced with the specified coverage; while if the true value of R_{0} is above this threshold, then vaccination will lead to an increase in cases. To determine this threshold for each birthrate and vaccine coverage, we performed a binary search of possible values of R_{0}

We estimated R_{0} based on the age distribution of infection using laboratory-confirmed rubella case data collected as part of WHO measles surveillance in 40 different countries in Africa from 2002–2009 (from Table 1 in Goodson et at al. 2011 _{0}.

Age specific rubella case reports were grouped into 5 age classes: less than 1 year of age, 1–4 years of age, 5–9 years of age, 10–14 years of age and 15 or more years of age.

For each country

The force of infection,

Where

Parameters (_{0} were estimated using Bayesian Markov Chain Monte Carlo (MCMC) methods with non-informative priors (two chains 1,000,000 iterations, 500,000 iteration burn in). Convergence was assessed by visual examination of chains and posterior distributions and an _{0}s in a random country was determined by integrating

The data on the distribution of R_{0}s and the threshold value of R_{0} are combined to make a figure summarizing our confidence that rubella vaccination would result in a reduction of CRS cases. The figure is a grid, where each cell represents a particular combination of birthrate (indicated by the column) and vaccine coverage (indicated by the row). In each cell we print the R_{0} threshold value calculated as described above. The cell is shaded to reflect our confidence that the true R_{0} is below the threshold given the estimate distribution of R_{0}s. That is, our confidence that CRS will decrease if the vaccine is introduced. Each cell is colored on a gradient from red to yellow to green, where red designates a high confidence that CRS cases would increase if rubella vaccine were introduced, yellowing shades represent decreasing confidence in an increase in CRS, and green represents 95% confidence that CRS cases would decrease if a vaccine were introduced.

We considered scenarios where there was only routine rubella vaccination among children and infants, administered as part of a countries measles vaccination program, and where rubella vaccine was administered in combination with supplemental immunization activities (SIAs). We considered SIAs with 60% coverage conducted every 4 years targeting 1–4 year olds, and every 4 years SIAs combine with a kickoff campaign in 1–14 year olds conducted in the first year of rubella vaccination.

We considered two different scenarios of population mixing. In the first we assume that age groups mix evenly, and individuals are no more likely to be infected by a member of a different age group than their own. In the second, we assume that there is assortative mixing and differences in the frequency of infectious contact by age. We assume that assortativity and contact frequency are proportional to what was measured in POLYMOD

All statistical analyses were done using R 2.15 (

The median of the estimated R_{0} distribution for rubella is 5.2 (_{0} will be a particular value in a given setting, we are 90% confident that R_{0} will be between 4.0 and 6.7, and 50% confident will be between 4.7 and 5.7 (

(A) Cumulative distribution function, the dotted line represents 95^{th} percentile. (B) Probability distribution. (C) Individual country estimates; points indicate point estimates, gaps between points and solid lines the inter quartile range, and the range of the solid lines indicates the 95% credible interval for each country.

If only routine vaccination is used, countries with vaccine coverage greater than 80% can be highly confident in a reduction in CRS if they introduce rubella vaccine, while those with vaccine coverage less than 40% and a birthrate of 37 per 1,000 or higher are likely to see an increase in CRS cases if they introduce rubella vaccine (

(A) Routine vaccination only, assuming even mixing across all population age groups. (B) Routine vaccination only, assuming assortative mixing and heterogeneities in contact between age groups. (C) Routine vaccination supplemented with SIAs of 1–4 year olds with 60% coverage every 4 years (assortative mixing). (D) Routine vaccinations and SIAs supplemented with a catch-up campaign covering 1–14 year olds with 60% coverage conducted when rubella vaccine is introduced. White circle shows the cell most closely corresponding to Guinea-Bissau. The square shows the cell most closely corresponding to Guinea Bissau. The diamond indicates the cell most closely corresponding to Somalia.

When routine rubella vaccination is supplemented by SIAs with 60% coverage conducted every four years, starting the year of vaccine introduction, we are confident in a decrease in CRS cases when routine vaccination levels are 60% or higher. Even if routine vaccination is as low as 50%, we are unlikely to see an increase in CRS cases from the introduction of the rubella vaccine. However, populations with low coverage and a high birthrate remain in the zone where an increase in CRS is likely. Substantial additional benefits can be realized by kicking off a rubella vaccination program with a large catch-up campaign covering 1–14 year olds at 60% coverage. In this scenario, only in the settings combining the highest birthrates with the lowest vaccination rates is an increase in CRS cases is likely.

To see how these results play out in real world situations, consider three countries with very different demographics and measles vaccine coverage: Nepal, Guinea-Bissau and Somalia (

Overall, our results support the WHO recommendation that countries introduce rubella vaccine into their regular vaccination program if they can maintain coverage above 80% through a combination of routine vaccination and SIAs

Of the 40 countries included in our R_{0} estimation, 19 had routine MCV vaccination rates of 80% or greater, and would be able to introduce rubella vaccination by the WHO criteria with no supplemental campaigns (

By attempting to be general across a wide range of countries and settings, this work necessarily makes many generalizations and has several limitations. Data on rubella incidence is based on the analysis of suspected measles cases, and relies on surveillance systems that differ markedly by country and may be biased towards detecting rubella in particular age groups. Estimating the age distribution of rubella cases from measles surveillance data may bias the estimated age of infection downwards. However, this will tend to increase the estimate of R_{0}, thereby increasing the predicted required coverage and resulting in more conservative predictions. Likewise, while most countries considered in this analysis have a pyramidal age structure, some do not (_{0} across countries. While differences in age structure may affect estimates of R_{0}, the critical threshold appears to be insensitive to drivers of age structure other than birth (i.e., mortality

Previous work has, for the most part, either ignored the possible effect of assortative mixing between age groups on the introduction of rubella vaccine

Even if a country is within the range where introducing rubella vaccine is predicted to result in a reduction in CRS, countries must carefully consider their individual situation. Administrative coverage estimates may overestimate actual vaccine coverage

The threshold R_{0} identified for each birth rate/vaccination coverage combination is such that vaccination reduces the cumulative burden of CRS over 30 years, but makes no predictions about transient increases in the CRS burden. Previous reports of CRS increases in Greece and Costa Rica likely reflect this pattern

Our projections also ignore the effect of local disease dynamics on CRS burden. Local extinction of rubella may lead to an increase in the CRS burden by allowing individuals to age into childbearing years without exposure to the infection, up until the point where rubella is re-introduced

Rubella vaccination is part of a renewed focus on vaccination through

The authors would like to thank Bryan Grenfell and William J. Moss for their assistance and valuable suggestions. This work grew in part out of valuable interactions with the SAGE Working Group on Rubella.