Population-Level Effect of Cholera Vaccine on Displaced Populations, South Sudan, 2014

Following mass population displacements in South Sudan, preventive cholera vaccination campaigns were conducted in displaced persons camps before a 2014 cholera outbreak. We compare cholera transmission in vaccinated and unvaccinated areas and show vaccination likely halted transmission within vaccinated areas, illustrating the potential for oral cholera vaccine to stop cholera transmission in vulnerable populations.

To estimate the population as risk in Juba county, we used UN OCHA data for the population from April 2014 (http://www.unocha.org/south-sudan/) and then subtracted the estimated (case-weighted) PoC site population. While it is clear the entire Juba population is not at risk for cholera, with the limited data available on population distribution and demographics within the city, it is difficult to estimate the true size of the at risk population. Following Ali et al (1), we assumed that only those without access to improved sanitation (likely to overlap with those who also have access to safe drinking water) as measured by the UNICEF/WHO Joint Monitoring Program (84%, http://www.wssinfo.org/) were at risk. This resulted in a final at risk population in Juba of 387,512. Thus, we estimated the attack rate to be 10,000 × 2,071 387,512 = 53.4 per 10,000. It is worth noting that if the entire population of Juba County (minus the camps in this calculation) were assumed to be at risk, the attack rate would then be 10,000 × 2,071 461,324 = 44.9 per 10,000, which is lower than that estimated in the camps.
Only a single point estimate for the population size, based on biometric registration data from July 2014, was available (from IOM) for the Malakal camp. Population data from Wau-Shilluk based on survey data from the same month was available based on use of the displacement tracking matrix methodology (http://southsudan.iom.int/wpcontent/uploads/2014/08/DTM-Report-Round-IV.pdf and http://www.iomsouthsudan.org/tracking//dtm).
Age specific attack rates were estimated for the Juba locations. The age distribution for the Juba community was assumed to be equivalent as that for the entire country as estimated by the U.S. Census Bureau

Potential Explanations for the Observed Age Distribution of Suspected Cases in Juba Populations
In the Juba camps, we observed a far different age-distribution of suspected cholera cases than in the community (main text; Technical Appendix Figure

Differences in Historical Cholera Exposure
One possible explanation for the high attack rates in children in the Juba camps is that the IDPs came from a population with a different historical exposure pattern to cholera from people in the community. The median age of suspected cases in Upper Nile State (6 years old), one location where some IDPs came from, was significantly lower than that in Central Equatoria State (Camps and Community combined, 24 years old) (Technical Appendix Figure 4). If this observed age distribution was due to the immune landscape as opposed to differential careseeking behavior, differences in suspected case definitions based on age or differences in the

Possible Co-circulation of a Childhood Diarrheal Pathogen in the Camps
Exploring the proportion of rapid diagnostic test (RDT) positive suspected cases among those under-5 and those over-5 in the community in camps can provide us some additional insight into what may (or may not) have contributed to our estimates of high under-5 attack rates in the camps compared to the community. Among those tested with RDTs (Crystal VC, Span Diagnostics), we found that a higher proportion were cholera positive in the camps compared to the community (Technical Appendix Table), suggesting that the suspected case definition in the camps may have been more specific. We also see that the proportion of RDT-positive cases between under-5s and over-5s did not significantly differ within each setting (using 2-sample test of independent proportions as implemented in R using prop.test). This provides evidence against the hypothesis that another diarrheal pathogen circulated in the camps mostly among children leading to inflated suspected cases in children compared to adults in the camps. While interesting, these results should be interpreted with caution as it is unclear what the criteria were for the use of RDT in the camps and the community, and this likely does not represent a true random sample of the suspected cases in each population.

Differences in Age-Specific Vaccine Coverage in Juba Camps
The LQAS survey referenced in the main text did not have a sufficient sample size to precisely estimate coverage by age within the Juba camps. However, they did collect age data on participants and they estimate that 100% of those under 5 received at least 1-dose and 80% received 2 doses in Tongping and UN House combined (n = 15). These are consistent with other OCV campaigns where coverage in young children has typically been high (2).

Estimation of Rt
We estimated the time-varying reproductive number using methods similar to that of Wallinga and Teunis (3). Since not all cases had a reported symptom onset date, we used the empirical distribution of the time from (self-reported) symptom onset to admission (Technical Appendix Figure 5) to impute the symptom onset dates for those individuals with missing or obviously inconsistent data (e.g., a symptom onset date after admission date).
To further support our findings that there were far fewer days (both as a number and as a proportion of epidemic days) where Rt>1, we used the bootstrap resamples to estimate the number of days for each location that Rt>1. We then tested the differences between comparison areas with a Wilcoxan Rank Sum test, with a null hypothesis that the number of days with Rt>1 was the same in each population. As we might expect more days in larger populations (like Juba compared to the camps), we also treated each bootstrap as a binomial observation to test whether the proportion of days with Rt>1 differed between the two populations using a simple logistic regression model where the dependent variable was an indicator for Rt>1(one observation for each day in each location) and the dependent variable was location (performed separately for the two comparison groups). As reported in the main text, we found that the probability of any day having a reproductive number greater than unity was significantly larger in unvaccinated camps than vaccinated camps.