(task) Mapping the burden of cholera in sub-Saharan Africa and implications for control: an analysis of data across geographical scales - The Lancet

> All Africa RS

4 comment to Schroeder article.

cholera, biosecurity, data, mapping

> http://www.thelancet.com/journals/lancet/article/PIIS0140-6736(17)33050-7/fulltext <http://www.thelancet.com/journals/lancet/article/PIIS0140-6736(17)33050-7/fulltext>
>
> Mapping the burden of cholera in sub-Saharan Africa and implications for control: an analysis of data across geographical scales
>
> In our analyses, we included 279 datasets representing epidemiological time series of 1–7 years in length. These datasets include data for 2283 different, though sometimes nested, locations in 37 countries in sub-Saharan Africa (appendix). Our study area (ie, excluding Djibouti and Eritrea) had a population of 988, 919 755 people. Within our study area, 27·5% (272 423 332 of 988 919 755) of the population were classed as urban—marginally lower than the 2016 UN and World Bank estimate of 38%.14, 15 Across sub-Saharan Africa (excluding Djibouti and Eritrea), a mean of 141 918 cholera cases (95% CrI 141 538–146 505) were reported annually, although year-to-year variation was substantial, as indicated by the calculated CVs (data not shown). We estimated that 21 695 308 people (19 757 461–23 654 156) in sub-Saharan Africa live in 20 km × 20 km grid cells in which average annual cholera incidence is greater than one in 1000 people (high incidence; figure 1 </cms/attachment/2119058242/2088621572/gr1.jpg>). 61·1% (54·2–66·4) of this population live in rural areas, which is similar to that in the sub-Saharan African population overall (62%).15 65 471 944 people (55 335 356–70 896 060) live in grid cells with moderate incidence, and 126 253 539 people (116 107 402–139 827 994) live in grid cells with mild incidence. 543 554 976 people (482 466 484–588 554 996) live in areas with negligible incidence of reported cholera (<1 per 1 million people per year), constituting 55·0% of the study area population.
>
> </cms/attachment/2119058242/2088621572/gr1.jpg>
> Figure 1
>
> Mean annual cholera incidence
>
> (A) Mean annual cholera incidence per 100 000 people in sub-Saharan Africa between 2010 and 2016, and (B) locations with mean annual incidence of more than one per 1000 people, (C) more than one per 10 000 people, or (D) more than one per 100 000 people.
>
> We estimated that 151 of 3751 districts (4·0%, 95% CrI 1·7–16·8) in our study area, home to 87·2 million people (95% CrI 60·3 million to 118·9 million), can be classified as having high incidence of cholera (figure 2 </cms/attachment/2119058242/2088621573/gr2.jpg>). 168·3 million (116·4 million to 229·4 million) OCV doses would be needed to vaccinate all eligible individuals living in these high-risk districts, which is substantially more vaccine than has been produced since the establishment of the global stockpile in 2013.7 We estimated that vaccinating all eligible individuals living in high-risk districts would directly prevent a mean of 156 536 cases (121 094–194 640) over a 3-year period. Extending vaccination to eligible individuals living in both moderate-risk and high-risk districts could directly prevent an estimated 239 518 cases (173 581–379 223), but 511·6 million (396·4 million to 635·8 million) doses of vaccine would be needed, which is more than 30 times the estimated global production in 2017.7 Indirect effects of vaccination campaigns (ie, herd protection) could protect more individuals, even with imperfect coverage, although there is conflicting evidence as to the nature and spatial scale of indirect effects.16, 17
>
> </cms/attachment/2119058242/2088621573/gr2.jpg>
> Figure 2
>
> Districts with mean annual cholera incidence above certain thresholds
>
> (A) Districts with mean annual cholera incidence of more than one case per 1000 people, (B) more than one case per 10 000 people, and (C) more than one case per 100 000 people. Districts with a mean of fewer than five cases annually are excluded. The colour scale represents the percentage of model iterations (ie, posterior draws) for which incidence exceeds the threshold, with darker shaded districts being over the threshold in a higher percentage of Markov chain Monte Carlo iterations.
>
> Substantial strides could be made towards cholera elimination with a fraction of the resources by prioritising those areas at highest risk. For example, consider the number of people who would need to be effectively targeted by an integrated cholera control programme to prevent 50% of the cholera cases in sub-Saharan Africa. If we focus interventions at the resolution of our maps (20 km × 20 km grid cells), targeting the areas in order of their number of cases from 2010 to 2016 from highest to lowest, incidence could be reduced by 50% by targeting areas containing 11·9 million people (95% CrI 5·6 million to 21·1 million), which is 1·2% of the study population (figure 3 </cms/attachment/2119058242/2088621574/gr3.jpg>). If the same strategy was used to target districts, areas containing 35·3 million people (26·3 million to 62·0 million; figure 3 </cms/attachment/2119058242/2088621574/gr3.jpg>), which consitutue 3·6% (95% CrI 2·7–6·3) of our study population, would be targeted. An alternative strategy ranking by incidence rather than the number of cases led to similar reductions (appendix). A similar approach could be used to make optimum use of existing resources. For example, if 20 million OCV doses were used in the highest incidence districts, 121 637 (29%) of 425 754 cholera cases (83 620–182 694) could be prevented over 3 years from direct effects of the vaccine.
>
> </cms/attachment/2119058242/2088621574/gr3.jpg>
> Figure 3
>
> Annual cholera cases in sub-Saharan Africa averted as a function of the number of people targeted with an ideal intervention or mix of interventions
>
> The optimum grid cell targeting curve (blue) represents a strategy targeting all 20 km × 20 km grid cells in rank order by number of cases. The optimum district targeting curve (red) represents a strategy targeting all districts in rank order by number of cases regardless of country. The green curve represents a more realistic and practical strategy that targets all high-risk districts in each country at once, with countries ranked by the number of cases prevented. Lines are the mean values and shading shows the 95% credible intervals. Strategies targeting grid cells or districts by ranked incidence instead of number of cases are presented in the appendix.
>
> In practice, targeting the high-incidence areas irrespective of geography, population size, accessibility, or other factors is probably not feasible because of logistical challenges. A more practical strategy might be to prioritise countries on the basis of the number of cases that could be averted if all high-incidence districts in the country were targeted, then launch country-specific programmes aimed at those districts. Using this approach, 38·4% of cholera cases could be prevented by covering 50·8 million people (95% CrI 39·7 million to 62·8 million) in five countries: Somalia, Nigeria, Democratic Republic of the Congo, Sierra Leone, and Ghana (figure 3 </cms/attachment/2119058242/2088621574/gr3.jpg>).
>
> To better characterise the prevailing cholera dynamics on this endemic–epidemic spectrum (in which endemic dynamics are associated with a low CV and epidemic dynamics are associated with a high CV), we plotted the CV in year-to-year cholera incidence for each country against their mean yearly incidence per 100 000 population over that same period (figure 4 </cms/attachment/2119058242/2088621575/gr4.jpg>). This revealed a spectrum of cholera dynamics that included both high-incidence countries with endemic dynamics (eg, Democratic Republic of the Congo) and those with epidemic dynamics (eg, Guinea-Bissau). As with incidence, consideration of subnational epidemiological patterns that might have important policy implications is crucial. For instance, although, as a whole, Nigeria had consistently low cholera incidence, the eastern part of the country had large populations with high rates of epidemic cholera (figure 4 </cms/attachment/2119058242/2088621575/gr4.jpg>). Likewise, the high incidence of endemic cholera in the Democratic Republic of the Congo was driven by districts in the east of the country, whereas other areas had a more epidemic pattern (figure 4 </cms/attachment/2119058242/2088621575/gr4.jpg>). Unfortunately, for much of Africa, we do not have sufficient temporal data of high spatial resolution to classify all districts along this spectrum. However, assuming high-incidence districts within each country follow national dynamics, we estimate that 55·9 million people (95% CrI 38·8 million to 78·9 million) in Africa live in high-incidence districts with endemic (ie, predictable) cholera (CV <1·5), whereas 31·0 million (95% CrI 20·6 to 39·8 million) live in districts with more epidemic dynamics (CV >1·5).
>
> </cms/attachment/2119058242/2088621575/gr4.jpg>
> Figure 4
>
> Cholera incidence versus the coefficient of variation of the annual incidence
>
> Mean annual reported cholera incidence per 100 000 people versus the coefficient of variation of the annual incidence from 2000 to 2015 for 50 African countries based on reports to WHO (A),3 and from 2004 to 2014 for districts in Nigeria (B; states) and from 2000 to 2016 for the Democratic Republic of the Congo (C; zone de santé) using annual aggregated data for each country. Colouring of points and map areas corresponds to the position on the scatter plot, as shown in the bottom right inset in (A), to allow easier mapping between country maps and x–y plots. The size of the points correspond to the number of years that the country reported data to WHO. Countries and districts with a mean annual incidence of zero are dark green in the sub-maps, but are not plotted in the scatter plots. Black circles and crosses in (B) and (C) represent each country's position in (A). National-level data were available for both Djibouti and Eritrea; white areas on maps correspond to areas where data were not available.
>
> By characterising the geographical distribution of cholera risk in sub-Saharan Africa, we show that cholera is pervasive throughout sub-Saharan Africa, and that more than 200 million people are living in areas (ie, 20 km × 20 km grid cells) with at least some cholera incidence. We found substantial heterogeneities within and between countries; most cholera incidence is concentrated disproportionately in a small proportion (<5%) of districts. This finding highlights the importance of hotspots of high cholera burden in driving cholera incidence and a crucial role for spatial targeting in cholera control. As global cholera control efforts are intensified, with the goal of eliminating cholera as a public health threat, this study provides important estimates of baseline cholera burden and methods for tracking our progress.
>
> The utility of cholera mapping depends on alignment of results to the geographical scales at which public health policy is made and implemented. The precise scale of interest will depend on political boundaries and the intervention being considered. Because of the scope of our analysis, we focused on broad-scale interventions (eg, district level) that might be most relevant to population-level programmes such as OCV campaigns. By contrast, WaSH interventions will often be targeted according to the scale of water and sanitation infrastructure (eg, at towns or villages), and thus finer scale maps supplemented by detailed analysis of the local situation might be more informative.
>
> Our analysis shows how the temporal distribution of cholera incidence varies substantially from country to country, and often also from district to district. Some areas have truly endemic cholera with high numbers of cases every year (eg, eastern Democratic Republic of the Congo). At the other extreme are areas where cholera incidence is concentrated in large outbreaks separated by many years of low activity. Analysis of variation is likely to help to ensure resources are used effectively. The distribution of cholera will change, but historic trends are usually the best evidence available about future disease incidence, and our confidence in future projections will be highest when past variation is low. Likewise, the best strategy for using OCV and WaSH interventions will vary depending on an area's position on the endemic–epidemic spectrum.
>
> Mapping cholera incidence on the basis of reported cases has many limitations. Cholera is probably under-reported, particularly in rural areas. Likewise, many countries report incidence of acute watery diarrhoea as a proxy for cholera, but even in hyper-endemic areas as few as 20% of acute watery diarrhoea cases test positive for Vibrio cholerae.18 Although use of several independent data sources helped to reduce the risk of bias stemming from these issues in reporting of data, correcting for these biases remains an area of ongoing research. Improvements to cholera surveillance would also help, and are a key element of the global cholera control roadmap. Year-to-year variation in incidence does not give a complete picture of local cholera dynamics, and there are many patterns of incidence that are compatible with each position on figure 4 </cms/attachment/2119058242/2088621575/gr4.jpg> (eg, 3 years of high incidence followed by 12 years of low incidence could have the same variation as peaks in incidence every 5 years); only detailed analysis of local data can give a full picture of local dynamics. Furthermore, although we identified many regions of high cholera incidence where known risk factors such as infrastructure problems or displaced populations are present, this study was not designed to assess the extent to which local cholera incidence can be attributed to these factors. Another fundamental challenge is local variations in cholera epidemiology. In some areas, cholera incidence is likely to increase during rainy periods as a result of contamination of water supplies. In others, incidence is likely to increase during periods of drought, when people are forced to use unsafe water sources. Most challenging are outbreaks associated with disruptions to the local infrastructure; for instance, the increased incidence of cholera after interruption of the chlorinated water supply in Uvira, Democratic Republic of the Congo.19 Our inability to predict such infrastructure problems, combined with local differences in cholera epidemiology, is one reason why mapping methodologies based mainly on using models fit to relationships with environmental covariates have been less successful for cholera than for other diseases (eg, malaria and dengue). These infrastructure changes and local variations in cholera epidemiology might lead to substantial deviations from our maps in the distributions of cholera risk over time, both in terms of increased and decreased incidence. Hence, these maps should be updated over time, and are not a substitute for local investigations of cholera epidemiology when planning local control strategies.
>
> The 87 million people living in districts in sub-Saharan Africa with high incidence of cholera will be best helped by improvements to local water and sanitation infrastructure that would provide broad benefits beyond this one disease and ensure sustained cholera elimination. However, experience tells us that these improvements will need substantial investment and can take years to be fully realised, and that success will be dependent on unpredictable political or economic conditions. Until that time, OCVs remain an important tool to prevent and control the spread of cholera. Although OCV supplies have increased substantially since establishment of the global stockpile7 and are likely to continue to increase, they will probably remain insufficient for broad untargeted use of the vaccine. The water and sanitation improvements that are the ultimate solution to global cholera are likely to be most effective if they are geographically targeted. Regardless of how reductions in incidence are achieved, targeting people at high risk can have effects across much larger populations. As with sub-Saharan Africa, targeted approaches are likely to be an essential component of cholera control programmes in other high-burden regions, such as south Asia, Haiti, and the Middle East. Hence, fine-scale incidence mapping exercises such as this have an important role to play globally in maximising the benefits of scant resources, forecasting demand for vaccines and other supplies, and tracking progress in the fight against cholera.
>
>
>
> For the Cholera dynamics website see http://www.iddynamics.jhsph.edu/projects/cholera-dynamics <http://www.iddynamics.jhsph.edu/projects/cholera-dynamics>