The current conflict in Sudan began on 15 April 2023. Since the start of the conflict there have been reports of conflict-related deaths and injuries. There is currently no way to generate precise estimates of conflict-related mortality in Sudan as government services, including vital registration functions, have collapsed or are under severe strain.
Study background
The Estimating conflict-attributable mortality in Sudan study builds on our mortality studies in Yemen and Tigray and aims to generate estimates of conflict-related excess mortality using remote methods, in this case capture-recapture. We will collect capture and recapture data using two methods: 1) online survey (disseminated via our dedicated respondent-driven sampling solution, and via social media), and 2) scraping of publicly available death notifications (via social media and lists obtained from Sudanese civil society organisations). We will use capture-recapture methods to estimate the population of deceased persons during the conflict, and to estimate the proportion of these that constitute ‘excess’ or ‘conflict-attributable’ deaths.
Our survey will ask participants to provide the names of people they know to have died within Sudan over the study period (we will include a pre-war period, defined as the five-year period preceding the conflict start date) as a comparator, the decedent’s name, gender, date of death (approximate), the state in which they died, their manner of death (e.g. violent injury, self-inflicted injury, disease), and their occupation. We will not ask for any information about living persons including the respondent completing the survey. We will mine as much of this information as possible from publicly available death notifications. We will collect data up to the end of the study period or until we have obtained a sufficiently large sample to power our estimates.
The study also aims to evaluate each approach to collecting data about descendants. We aim to assess: the rapidity with which we are able to collect data, the resources required, and the completeness of the data collected across different approaches (e.g. social media disseminated survey, publicly available civil society decedent lists). In addition, we aim to establish how each approach (or combination of approaches) to collecting mortality data compares to combined data from all four approaches, in terms of power and completeness of data.
Study rationale
Vital registration functions are often completely disrupted in conflict-settings: though raw counts of deaths have been reported these numbers lack precision, rigour, representativeness, and are likely to reflect proximal manner of death only (e.g. injury). Due to security concerns and travel restrictions ground surveys are not feasible in most conflict settings. Remotely carried out mortality surveys have been carried out in Yemen and Tigray and have generated large samples from which to estimate conflict-attributable mortality. There are few, if any, alternatives to collecting mortality data in Sudan that are both safe, accurate, and which account for the downstream effects of the conflict on mortality (e.g. due to infectious disease outbreaks, or malnutrition). Data on conflict-related mortality are essential for measuring the humanitarian need and galvanising the humanitarian community.
Aims and objectives
The study aims to generate conflict-attributable mortality estimates in Sudan, and to evaluate our approaches to collecting mortality data. The specific objectives are as follows:
1. Collect mortality data via online survey (distributed via two distinct approaches)
2. Collect mortality data using publicly available information (via two distinct approached)
3. Estimate conflict-attributable mortality using capture-recapture methods.
4. Evaluate different approaches to collecting mortality data in Sudan including by assessing the rapidity with which we can develop and deploy the mortality survey, determining the resources required, and assessing the completeness of data collected.
5. Evaluate whether or not any one of the four approaches (or any combination of approaches) is capable of informing a sufficiently powered estimate compared to combining data from all four approaches.
Methods
Survey
We will collect data using an online survey created in ODK. The survey will be distributed using two approaches:
Respondent-driven sampling
As part of our mortality estimation studies in Yemen and Tigray we have developed a bespoke respondent-driven sampling (RDS) solution which we have used to anonymously distribute the ODK survey, and to enable respondents to invite onward potential-respondents from within their networks and without the involvement of the study team. Survey invitations are generated and sent by the RDS solution based on templates we will co-produce with our Sudanese colleagues. The RDS solution will support dissemination within Sudan and amongst the diaspora/refugee populations and will thus be available in both English and Arabic.
Social media
The team in Sudan will also disseminate the survey (slightly adapted to remove meta-data required by the RDS platform and to include a hidden field which indicates that respondents have been recruited though the social media approach). The survey questions will be the same. We will only send out the survey using open social media accounts (i.e. no private accounts). The recruitment information will be similar to the invitation email generated by RDS. We will adapt it, insubstantially, with support from the Sudan team when they are available.
Data mining
We will use two approaches to obtaining publicly-available lists of decedents: 1) social media, and 2) civil society organisation lists.
Social media
We will review public-facing social media sites (e.g. Facebook, Twitter) to retrieve names and other available information about deceased persons. We will only access public (i.e. not private) accounts. The team in Sudan will identify social media accounts. The team in London will then use NVivo to pull data from the sites at regular intervals. We will export the data into a suitable format for review and extraction of decedent lists.
Civil society organisations
Currently, there are some publicly available lists of deaths and injuries collected by Sudanese civil society groups. The team in Sudan will identify these lists and will create a consolidated list.
Analysis
We will link records across and within the four lists based on the name and other information about the deceased. We will then analyse the overlap within and between lists to estimate the number of deaths or injuries not captured which will be added to the number present on at least one list to yield total mortality, as per capture-recapture analysis (also known as multiple systems estimation). We will fit all possible log-linear models with Bayesian model averaging to estimate the true number of deaths and injuries. We have previously tested this method with ground researchers in four settings and have also implemented it remotely to estimate mortality during demonstrations in Sudan. Note that the method requires lists to be independent, but not complete; we anticipate and account for a considerable degree of imprecision and incompleteness.
Evaluation methods
The study includes a rapid process evaluation to determine the rapidity with which we are able to collect data, the resources required, and the completeness of the data collected across different approaches. In addition, we aim to establish how each approach (or combination of approaches) to collecting mortality data compares to combined data from all four approaches, in terms of power and completeness of data. For the latter we will simply carry out subgroup analysis for each approach and combination of approaches to determine which methods (or combination of methods) produce the most suitable lists (in terms of study power), balanced against the time and resources required for each approach (or combination of approaches).
Ethics and governance
This study has been approved by the LSHTM Observational Research Ethics Committee (Approval # 29595).
Additional queries can be addressed to Dr Maysoon Dahab.