The impact of reactive mass vaccination campaigns on measles outbreaks in the Katanga region, Democratic Republic of Congo

The Katanga region in the Democratic Republic of Congo (DRC) has been struck by repeated epidemics of measles, with large outbreaks occurring in 2010–13 and 2015. In many of the affected health zones, reactive mass vaccination campaigns were conducted in response to the outbreaks. Here, we attempted to determine how effective the vaccination campaigns in 2015 were in curtailing the ongoing outbreak. We further sought to establish whether the risk of large measles outbreaks in different health zones could have been determined in advance to help prioritise areas for vaccination campaign and speed up the response. In doing so, we first attempted to identify factors that could have been used in 2015 to predict in which health zones the greatest outbreaks would occur. Administrative vaccination coverage was not a good predictor of the size of outbreaks in different health zones. Vaccination coverage derived from surveys, on the other hand, appeared to give more reliable estimates of health zones of low vaccination coverage and, consequently, large outbreaks. On a coarser geographical scale, the provinces most affected in 2015 could be predicted from the outbreak sizes in 2010–13. This, combined with the fact that the vast majority of reported cases were in under-5 year olds, would suggest that there are systematic issues of undervaccination. If this was to continue, outbreaks would be expected to continue to occur in the affected health zones at regular intervals, mostly concentrated in under-5 year olds. We further used a model of measles transmission to estimate the impact of the vaccination campaigns, by first fitting a model to the data including the campaigns and then re-running this without vaccination. We estimated the reactive campaigns to have reduced the size of the overall outbreak by approximately 21,000 (IQR: 16,000–27,000; 95% CI: 8300–38,000) cases. There was considerable heterogeneity in the impact of campaigns, with campaigns started earlier after the start of an outbreak being more impactful. Taken together, these findings suggest that while a strong routine vaccination regime remains the most effective means of measles control, it might be possible to improve the effectiveness of reactive campaigns by considering predictive factors to trigger a more targeted vaccination response.


Introduction
There have been repeated outbreaks of measles in the Democratic Republic of stop measles transmission during epidemics, targeting more than 25 health zones. 23 The time interval between the outbreak starting in different parts of Katanga 24 and the vaccination response implemented varied. Previously, modelling studies in 25 Niger have demonstrated that even late vaccination intervention in response to an 26 outbreak could prevent a large number of cases, though early intervention will al-27 ways have a larger impact [4,5,6,7]. However, this may be context-specific and  Modelling measles with mass vaccination campaigns 73 We modelled measles transmission at the level of health zones using a stochastic zone i, I it was drawn from a negative binomial distribution with mean λ it S i(t−1) 77 and shape m, allowing for overdispersion of transmission, or superspreading [12]: where S i(t−1) and I i(t−1) are the number of people susceptible and infected, 79 respectively, at time t−1, and λ it is the force of infection experienced by susceptibles 80 in health zone i at time t: where N i is the population size of health zone i, R 0 is the basic reproduction 82 number.

83
When a mass vaccination campaign was conducted, the number of susceptible 84 people immunised was calculated by multiplying the number of doses administered 85 5 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. was not certified by peer review) (which The copyright holder for this preprint this version posted October 10, 2019. . https://doi.org/10.1101/19003434 doi: medRxiv preprint with the proportion of the population still susceptible S it /N i , and a campaign efficiency factor e i , estimated as part of the inference procedure described below. This 87 factor comprises both vaccine efficacy and the efficiency in targeting susceptible 88 children, which were not identifiable separately. With a perfect vaccine and random 89 distribution, this would take a value of 1. If vaccines were preferentially given to 90 susceptibles, it would take values of greater than 1 (subject to vaccine efficacy). If 91 vaccines were preferentially given to already immune children, it would take values 92 of less than 1.

93
During a two-week span, half of vaccinations were modelled to be administered 94 before transmission occurred and half afterwards. While the measles vaccine takes 95 2 weeks to come into effect, it provides potentially high level of protection from 96 72 hours after administration [13,14,15]. We therefore assumed that vaccination 97 starts to fully immunise a child instantaneously.

98
For the counterfactual scenarios of how the outbreaks would have evolved with-99 out a reactive mass vaccination, we simulated the model from the time of the mass 100 vaccination campaigns, but without reducing the number of susceptibles as a con-101 sequence of vaccination. We then drew samples from the joint distribution of tra-102 jectories and observations, to obtain alternative trajectories of observed cases. To 103 evaluate the impact of the campaigns, we calculated the reduction in the number of 104 cases observed in each of the trajectories. If this yielded a negative difference (i.e., if 105 random sampling yielded alternative trajectories with more cases than the observed 106 ones), we treated the impact as 0 (i.e., same number of cases in both scenarios).

107
Selection of health zones for fitting and estimating populations 108 The health zones selected for the dynamic model were ones that reported more  Model fitting and counterfactual scenarios 126 The model was fitted simultaneously to the eight selected health zones. The likeli-127 hood of observing bi-weekly incidence D it in health zone i at time t was taken to 128 follow a negative binomial distribution with fixed overdispersion ϕ.
where ρ is the proportion of cases that is reported, $µ is the rate of background 130 reporting of measles, either due to cases that were not part of the epidemic or 131 misclassification, for example of rubella cases, and ϕ is the reporting overdispersion. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. was not certified by peer review) (which The copyright holder for this preprint this version posted October 10, 2019. . https://doi.org/10.1101/19003434 doi: medRxiv preprint to be the start of the time series), were all estimated as part of the inference procedure, as well as likely trajectories of the state variables. The reporting rate ρ i and 138 initial number infectious I i0 was allowed to vary between health zones. The prior 139 distribution on the mean reporting rate was weakly informed by a coverage survey 140 that was conducted in Kabalo. The initial proportion immune r i0 was estimated 141 with a mean and lower bound given by the vaccination coverage per health zone v i 142 estimated in [8]. Informed or regularising prior distributions of the parameters to 143 be estimated are shown in Table 1.
The model was fitted to the data using a particle filter in combination with    11 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. was not certified by peer review)   185 To investigate the impact of the mass vaccination campaign in more detail, we 186 fitted a dynamic model to the case trajectories in 8 health zones (Fig. 6). We   to susceptible children.

201
There was heterogeneity in impact between health zones. The greatest abso- 14 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. was not certified by peer review) (which The copyright holder for this preprint this version posted October 10, 2019. . https://doi.org/10.1101/19003434 doi: medRxiv preprint as well as potential for improvement.

218
The predictability of outbreaks is related to the quality of the available data. We In the health zones modelled, the case-fatality ratio in the reported data was 1.2%, 253 suggesting that around a hundred infant lives were probably saved by the campaigns.

254
Our transmission model suffered from several limitations. We did not have access  The fact that another big outbreak could happen so soon after the last suggests a 289 17 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. was not certified by peer review) (which The copyright holder for this preprint this version posted October 10, 2019. . https://doi.org/10.1101/19003434 doi: medRxiv preprint rapid increase in susceptibles that have not been served by the routine vaccination 290 programme, and strengthening this should be a priority. At the same time, it is 291 clear that the mass vaccination campaigns only prevent part of the observed cases, 292 partly because of unavoidable delays in confirming an outbreak and launching a 293 campaign. Preventive strategies based on predictive models have a potential to 294 have a much greater impact if they can prevent outbreaks altogether, but their use 295 is based on the predictive potential of the models used. We found that vaccination 296 estimates based on a spatial model applied previously to vaccination survey data was 297 a good predictor of outbreak size at the relatively fine level of health zones. There is 298 enormous promise in using such estimates to guide strategic immunisation activities 299 and close any existing gaps in immunity. As has been proven many times over,