Evaluation of a multi-component early warning system for pastoralist populations in Doolo zone, Ethiopia: mixed-methods study

Abstract

Abstract Background This study evaluated an early warning, alert and response system for a crisis-affected population in Doolo zone, Somali Region, Ethiopia, in 2019–2021. To adequately cover an area populated by semi-nomadic pastoralist population with sparse access to healthcare facilities, the surveillance system included four components: health facility indicator-based surveillance, community indicator- and event-based surveillance, and alerts from other actors in the area. This evaluation described the usefulness, acceptability, completeness, timeliness, positive predictive value, and representativeness across each of these components. Methods We carried out a mixed-methods study retrospectively analysing data from the surveillance system February 2019 - January 2021 along with key informant interviews with system implementers, and focus group discussions with local communities. Transcripts were analyzed using a mixed deductive and inductive approach, with surveillance attributes used as deductive themes. Results 1010 signals were analysed; these resulted in 168 verified events, 58 alerts, and 29 responses. Most of the alerts (46/58) and responses (22/29) were initiated through the community event-based branch of the surveillance system. In comparison, one alert and one response was initiated via the community indicator-based branch and no alerts or responses via the health-facility indicator-based branch. Positive predictive value of signals received was about 6%. About 80% of signals were verified within 24 hours of reports, and 40% were risk assessed within 48 hours. System responses included new mobile clinic sites, measles vaccination catch-ups, and water and sanitation-related interventions. Focus group discussions emphasized that responses generated were an expected return by participant communities for their role in data collection and reporting. Participant communities found the system acceptable when it led to the responses they expected. Some event types, such as those around animal health, led to the community’s response expectations not being met. Community event-based surveillance was designed to be representative of both settled and semi-nomadic communities. Conclusions Event-based surveillance can produce useful data for localized public health action for pastoralist populations. This approach may be better suited to public health surveillance of pastoralist populations than community indicator-based methods. Improvements could include greater community involvement in the system design and potentially incorporating One Health approaches.