key: cord-0326031-fnq85qvd authors: Charnley, G. E. C.; Jean, K.; Kelman, I.; Gaythorpe, K. A. M.; Murray, K. A. title: Using self-controlled case series to understand the relationship between conflict and cholera in Nigeria and the Democratic Republic of Congo date: 2021-10-24 journal: nan DOI: 10.1101/2021.10.19.21265191 sha: dc46ffee35b75c16ac130398ffa23ba75f8f8c88 doc_id: 326031 cord_uid: fnq85qvd Abstract Background Cholera outbreaks contribute significantly to diarrhoeal disease mortality, especially in low-income countries. Cholera outbreaks have several social and environmental risk factors and extreme conditions can act as catalysts for outbreaks. A social extreme with known links to infectious disease outbreaks is conflict, causing disruption to services, loss of income and displacement. Methods Here, we explored this relationship in Nigeria and the Democratic Republic of Congo (DRC), by fitting publicly available cholera and conflict data to conditional logistic regression models. We used the self-controlled case series method in a novel application, to understand if an exposure period of excess risk (conflict), increased the relative incidence of cholera. We also used a sensitivity analysis to understand potential lag effects. Results We found that conflict and cholera had a strong positive relationship, especially in the first week after the event, at a national and sub-national level. Conflict increased the risk of cholera in Nigeria by 3.6 times and 2.6 times for the DRC. Conflict was attributed to 19.7% and 12.3% of cholera outbreaks in Nigeria and the DRC, respectively. This was higher for some states/provinces, with a maximum increased risk of 7.5 times. Conclusion The results found that several states/provinces with the strongest positive relationship were also areas of high reported conflict or were neighbouring states/provinces, suggesting a possible spill-over effect. Our results help highlight the importance of rapid and sufficient assistance during social extremes and the need for pre-existing vulnerabilities such as poverty and access to healthcare to be addressed. In fragile states, conflict resolution should be a top priority to avoid excess risk for both cholera and other health and social implications. Funding Natural Environmental Research Council, UK Medical Research Council, and the Department for International Development. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 24, 2021. The spatial granularity of the analysis was to administrative level 1 (states for Nigeria and 117 provinces for the DRC) and all data points that were reported on a finer spatial scale were 118 attributed to the upper level. To be included in the analysis, the state/province had to report both 119 outbreaks and conflicts during the study period. As such, 22 provinces were included for the DRC 120 and 36 states for Nigeria (states and provinces excluded are shown in S2 Information). The 121 number of conflicts and outbreaks for each province during the study period is shown below (Table 122 1 and Fig 2) . The study period was specified as Jan 1997 to May 2020, as these were the first and last reports in 125 the conflict data. The temporal scale was set to weekly, with continuous weeks from 126 epidemiological week 1 in 1997 to epidemiological week 20 in 2020 (1-1,220). Continuous weeks 127 was chosen for compatibility with the model and to include periods of conflict that endured from 128 one year into the next. Weeks was chosen, rather than days, to account for reporting lags, as 129 previous work has reported issues in the granularity of data and timeliness of reporting, especially 130 . CC-BY 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint Both the event and exposure were set as binary outcomes, either being present (1) or not (0). The observation period was the full study period (1-1,220). The exposure period was the first week 143 after conflict onset and was reported as multiple onsets, not one long exposure period (Fig 3) . The 144 event was defined by the week the cholera outbreaks was reported. Each event and exposure that occurred in the same state/province were designated an 147 identification number and a pre-exposure, exposure, and post exposure period (see S1 and S2 148 Tables). The data was fit to conditional logistic regression models (function clogit(), R package 149 "survival") [38] . The model coefficient values were used to calculate incidence rate ratio (IRR) and 150 the percentage attributable fraction (PAF), which is an estimate of the percentage of outbreaks that 151 could be attributed to conflict (full equations in S4 Information). The datasets for each country were then split by state/province and the analysis repeated for each, 154 to understand if the significance of conflict on cholera outbreaks varied by sub-national location 155 and if conflict was more important in some states/provinces compared to others. All statistical 156 analyses were carried out in R version 3.6.2 and the threshold for significance was p=<0.05. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint week after the onset of exposure, lag 1) and were named lag periods due to the potential lag effect is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 24, 2021. Lag 1 had the strongest effect on cholera outbreaks at a national level for both Nigeria and the 198 DRC, which then decreased through the weeks up to week 10. For both countries magnitude of 199 change in effect size was greatest between week 1 and week 2 and by week 6 the change was 200 minimal and plateaued (Fig 5) . From week 1 to week 10 the risk decreased from 3.6 to 2.08 for is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 24, 2021. ; https://doi.org/10.1101/2021.10.19.21265191 doi: medRxiv preprint soon after the event but remains a detectable association albeit at a lower level for potentially a 203 long period of time after the event. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 24, 2021. ; https://doi.org/10.1101/2021.10.19.21265191 doi: medRxiv preprint states/provinces found increased values. Thirty Nigerian states were found to be significant for at 207 least one of the lag periods and the most significant states predominately followed the trends of the 208 national analysis. Values ranged from Kebbi at 6.9 to 4.0 times increased risk of cholera, to Gombe 209 at 2.4 to 1.5 (Fig 6a) . is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 24, 2021. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 24, 2021. ; https://doi.org/10.1101/2021.10.19.21265191 doi: medRxiv preprint and/or NGOs [33] . Reporting delay is another potential problem and some national reporting 296 delays, have been found to range from 12 days for meningococcal disease to 40 days for pertussis 297 [33] . To help account for these issues, both the event and the exposure was set to a binary 298 outcome and weekly data was used, instead of daily. A lack of age and sex-disaggregation in the 299 data also means that demographic risks and changes were missed and more work and data is 300 needed to address this. In summary, our analysis shows a clear relationship between cholera and conflict in both Nigeria is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 24, 2021. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 24, 2021. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 24, 2021. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The study period was selected as Jan 1997 to May 2020, as these were the first and last reports in 558 the conflict data. The spatial granularity of the analysis was to administrative level 1 (states for 559 Nigeria and provinces for the DRC) and all data points that were reported on a finer spatial scale 560 were attributed to the upper level. To be included in the analysis, the state/province had to report 561 both outbreaks and conflicts during the study period, therefore 22 provinces were included for the is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 24, 2021. ; https://doi.org/10.1101/2021.10.19.21265191 doi: medRxiv preprint thus ! "# = 0), ! is the rate of outbreak occurrence in a Poisson process in the absence of conflict, 573 and IRR is the incidence rate ratio associated with exposure to conflict. With ! "$ being the 574 number of outbreaks observed in the province during the un-exposed period and being the total 575 period of observation, an estimator of ! is , which leads to: is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 24, 2021. ; https://doi.org/10.1101/2021.10.19.21265191 doi: medRxiv preprint S2 Table. The pseudo-dataset created from the data in S1 is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint . CC-BY 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 24, 2021. ; https://doi.org/10.1101/2021.10.19.21265191 doi: medRxiv preprint The top 10 causes of death Dynamics of 333 cholera outbreaks in Great Lakes region of Africa The authors declare no competing interests.