The DRC remains affected by crisis conditions. During the main epidemic year of 2018, 1.8 million people were displaced by violence and 12.8 million were in need of humanitarian assistance [24]. With > 100 active armed groups, the eastern DRC is a challenging environment in which to conduct epidemic responses [8]. In such a context, the National Red Cross Society, supported by the IFRC, can leverage its large volunteer network and mandate to support epidemic-affected communities. The Red Cross-led SDB service became available within the first days of the epidemic declaration and achieved a remarkable volume of activity over the 14-month period we analysed. The service’s scale-up was however somewhat delayed, potentially missing opportunities to contain transmission clusters during the first 2–3 months of the outbreak. Conversely, towards the end of the epidemic, the introduction of more specific case definitions or criteria for requesting SDB might have improved efficiency.

SDB teams almost ubiquitously responded to alerts and met timeliness targets. The proportion of successful SDB responses was below target, though low performance was mainly circumscribed to a period from February to May 2019: this period saw high-profile killings of EVD responders, attacks against treatment centres [25, 26] and increased insecurity in health zones reporting EBOV transmission, particularly in Ituri [27], as shown in Fig. 3. Direct threats against Red Cross international staff also occurred during this period, resulting in temporary evacuations from North Kivu. Performance was also lower when SDB teams responded to suspected, rather than confirmed EVD deaths. Despite these reductions in service performance, we find in a separate paper (in preparation) that SDB was associated with substantial reductions in EBOV transmission.

We identified notable risk factors for SDB failure. Plausibly, local insecurity and interruptions to the EVD response were associated with higher odds of failure. Insecurity constrained the EVD response generally and, specifically, reduced SDB access to communities [28]. Burial supported by Civil Protection teams (mobile or CEHRB), and/or by CEHRB teams supported by the Red Cross, was associated with a lower odds of failure, suggesting the benefit of empowering local authorities and localising services to affected communities themselves (note that the true proportion of CEHRB SDBs includes an unknown number supported by Civil Protection). Strong community engagement has been shown to contribute to the effectiveness of EVD response in the West African epidemic [29, 30]. On the other hand, Civil Protection teams, unlike the Red Cross, sometimes travelled with armed escorts and may have contributed to ‘militarising’ the EVD response, with consequences for trust and acceptance [31].

Generally, SDB teams encountered limited community and/or family nonacceptance, but its occurrence was associated with a very high odds of failure, as noted in the West Africa epidemic and more broadly for EVD response interventions in DRC [32,33,34,35]. In the DRC, the Red Cross has engaged with affected communities through a Community Engagement and Accountability approach [36], whereby community volunteers routinely collect information regarding local knowledge, attitudes and practices in order to systematically respond to needs and concerns and adapt EVD response programming to meet communities’ expressed needs. These volunteers accompany SDB teams, liaising with families, community and religious leaders to safely adjust burial procedures based on local customs and preferences. In West Africa, a similar model of mediation, where implemented, increased SDB acceptance [37].

We found that the presence of communication infrastructure (radio, mobile network), as well as community exposure to EVD (health zones being in the midst or past their epidemic; a nearby EVD treatment centre) were associated with higher SDB failure odds. Communication means should in theory facilitate dissemination of response information and healthy behaviours [38]. In this scenario, however, increased communication may have facilitated the spread of negative views of the EVD response, information on inappropriate practices among responders, or conspiracy theories. The presence of communication infrastructure may have had a particularly marked effect once the epidemic had reached a given HZ, explaining why pre-epidemic SDBs were more successful: however, this effect modification was not evident statistically. Moreover, the above risk factors may be proxies rather than direct measures of the infrastructure causal domain. Generally, our and other studies’ findings need to be interpreted with reference to a context of chronic disenfranchisement and inadequate provision of health and other essential services, which time-bound, outbreak-focussed community engagement activities might not sufficiently address [8].

This was a retrospective evaluation, featuring no direct observation of SDB activities and relying principally on monitoring data not originally collected for research purposes. While we observed no patterns suggesting data fabrication, some SDB teams may have felt an incentive to overstate successful activities. Instances of CEHRB burial were anecdotally subject to greater-than-average data completeness and timeliness problems, which may have resulted in under-reporting of failed or untimely alert responses, and thus an over-estimation of the programme’s performance. The SDB dataset contained some ambiguities. EVD status was not ascertained or communicated to the SDB database for a large proportion of deaths in the community: these decedents were considered suspect cases and thus included in our analysis, but some may in fact not have met the suspect case definition, potentially biasing our estimates of key performance indicators. Findings could have been strengthened through qualitative exploration of the perceptions of EVD-affected communities and SDB staff. Separately, we are analysing community feedback data collected by Red Cross volunteers to shed light on this. Lastly, our risk factor analysis was limited by the explanatory and confounder variables that we were able to source: the model would have been strengthened by other variables for which we were unable to find appropriate and complete data: hidden confounding may therefore be present in our analysis. Specifically, the extent to which community engagement and feedback during specific SDB instances were used to adapt the service locally or generally could have been an important factor behind SDB success, and would have thus confounded the other associations. Separately the limited quality of data may have introduced random or non-random error (the former is likely, and would generally bias estimates of association towards non-significance). Generally, exploratory risk factor models with multiple adjustment are subject to the well-described ‘Table 2 fallacy’ [39], and should be interpreted with caution so as, in this case, identify factors that should be evaluated further and considered carefully in future interventions.

While we were able to analyse SDB performance when the service was called upon, we were unable to quantify its coverage (proportion of EBOV+ decedents for whom a SDB was performed). The coverage numerator (EBOV+ SDB burials) is unclear due to the many SDB instances for which no EBOV serostatus was recorded in the database after sample collection (Table 3). The denominator contains an unknown number of EVD deaths that escaped case ascertainment (at least one report [40] suggests a substantial fraction of undetected cases). Because burial at the community level mostly could not be delayed while awaiting results of EBOV testing, the SDB service targeted all decedents that community members may have suspected as EVD cases, meaning about four times as many SDBs were performed as all known EVD cases in the epidemic. If a crude measure of coverage is adopted, whereby any decedent during the epidemic should have received SDB, then over the entire population of HZs affected by the epidemic over the 14 months of the analysis period, and assuming a crude annual death rate of about 10 per 1000 for DRC [41], some 66,000 deaths would have been expected, meaning SDB coverage would have been ≈ 20%: this is likely to be a gross underestimate of the true coverage, as it is plausible that true EVD cases would have been much more likely than at random to have been reached by the SDB service, since the latter was only triggered by cases meeting the EVD suspect definition.

Generally, our study evaluates only one aspect of the SDB service and omits dimensions such as its feasibility, fidelity to different components, acceptability, equity and cost-effectiveness, some of which would be illuminated by a more comprehensive implementation science approach to evaluation adopting epidemiological, social science and health economics methods [42].

This evaluation was enabled by collection of systematic, quality monitoring data on SDBs by the IFRC and partners, reinforcing the benefit of such programmatic data collection. Our evidence shows that a large-scale, timely and moderately performant SDB service was feasible despite the challenging circumstances of the EVD response in eastern DRC. Failed SDBs due to the inability to secure the body or supervise burials suggest weaknesses in community engagement. The considerable number of SDBs for which EVD status remained unclear indicates a gap in testing or reporting of results. Community and/or family nonacceptance, while infrequent, is a key barrier to SDB effectiveness: an acceptance approach emphasising cooperation and with communities and other stakeholders should be pursued [17, 43]. Critically, the apparent success of localised SDB implemented by static teams within affected communities (the CEHRB programme) suggests this model should be relied upon and documented further in future EVD epidemics, particularly where accessibility is a constraint for mobile teams. Its potential benefits, however, should be carefully weighed against risks (insufficient burial safety, infection of SDB performers and family members). Lastly, our findings suggest that the availability of communication means and access to news may not necessarily improve community sentiment about epidemic responses, and trust in control interventions or their implementing actors: as noted by others [8, 34], solutions to this are likely to involve a better understanding of how information flows, engagement with communities through consultative processes based on active social listening, and empowering people themselves to be the main bearers of helpful knowledge.