Description of the EWARS

Figure 1

Figure 1. Organization of the Early Warning, Alert and Response System as used in the Democratic Republic of the Congo, August 2018–June 2020. Asterisk (*) indicates 1 negative result for a deceased...

The Alert Unit was the core functional unit around which the EWARS was organized (Figure 1); it was composed of an overall operational leader who coordinated activities, a database and information administrator, a case management leader, a Safe and Dignified Burial (SDB) leader, 3 telephone operators, 1 alert monitoring officer, 1 database manager, 1 data clerk, and 1 archivist. The main role of the Alert Unit was to gather and scan alerts from various sources, coordinate the field investigations with the rapid intervention teams, and, if relevant, organize the referral and ambulance transfer or safe burial in collaboration with the case management or SDB team. All alerts and their outcomes were entered and archived into paper-based alert and investigation forms and a Microsoft Excel database (https://www.microsoft.com). There were 4 main sources of alerts: community, in which community health workers, community members, and political and administrative authorities raised alerts; active case finding conducted in health facilities and other structures (pharmacies, churches, traditional practitioners); surveillance sites, including contact tracing teams, vaccination sites, and points of entry/points of control (PoE/PoC); and finally, public and private health facilities that ensured passive reporting.

Two toll-free numbers were activated on August 26, 2018, to enable rapid and easy alert reporting from all the sources. Calls were directed to telephone operators in the Alert Unit, which was operational 24 hours per day, 7 days per week; a smaller team for night shifts comprised the operations leader and phone managers only. In active case finding, passive reporting, or PoE/PoC, the alert notifier completed an alert form, or for community and contact tracing alerts, the telephone operator or alert monitoring officer completed the form. When telephone operators received the alerts, they checked for duplication and conducted preliminary triage to prioritize them by epidemiologic and clinical factors. Rapid intervention teams were then notified to investigate the alert onsite.

Rapid intervention teams were made up of a field epidemiologist, an infection prevention and control (IPC) officer, a communication officer, and a psychosocial worker. All these response pillars were positioned in each health area covered by the alert system, from which a senior epidemiologist would organize rapid intervention teams. The investigation consisted of a detailed history, assessment of the epidemiologic link, clinical symptoms for validation against the suspected case definition (17), and initial listing of contacts. Investigation forms were stored in the Alert Unit, and copies were sent to Ebola treatment centers (ETC) for patients requiring admission. The rapid intervention team validated or invalidated the alert on the basis of the investigation findings and provided immediate feedback to the Alert Unit contact persons, who coordinated the next steps.

For invalidated alerts, the family can proceed with ordinary burial of deceased patients, whereas living patients were referred to public healthcare facilities for free healthcare. Living patients with validated alerts were immediately transferred to a transit center, isolation center, or ETC, depending on the patient’s condition and location. There was no additional validation at triage in ETC. To reduce the risk that a transfer would refuse a patient, the intervention team would propose 2 options according to patient condition and preference: transfer the patient by ambulance or by motorcycles driven by Ebola survivors. After admission to the isolation center, patients followed the suspected case management algorithm: blood samples were taken and tested by using GeneXpert (Cepheid, https://www.cepheid.com) within 3 hours after admission. Cases confirmed by PCR were immediately admitted to an ETC for treatment. Those patients with an initial negative test were discharged pending a second negative result 72 hours later.

The SDB team were notified of validated alerts of deceased patients, then joined the rapid intervention team onsite to engage with the family. The body was secured and a swab sample taken and sent to the laboratory for testing. With family consent, SDB proceeded immediately. However, if the family refused, the body was kept at the mortuary until the laboratory result was available. If the result was negative, the body was returned to the family to proceed with ordinary burial; if the results was positive, SDB was mandatory and enforced by authorities.

Evaluation Approach and Data Sources and Indicators

We conducted a quantitative evaluation according to guidelines published by WHO (18) and the US Centers for Disease Control and Prevention (CDC) (19). We used the anonymized Alert Unit database, covering the health zones of Beni, Mutwanga, and Oicha, during August 5, 2018–June 30, 2020, to assess EWARS using the EVD suspected case definition as the standard. An alert was considered validated if it met the definition of an alert case by community-based surveillance or the definition of a suspected case by mobile teams or health stations or centers (12). An investigator would validate a suspected case on the evidence of clinical signs in the patient (Appendix Figure).

To assess the true sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) would require laboratory testing for all patients, which would not have been feasible. We calculated sensitivity as the proportion of alerts validated among all alerts meeting the suspected case definition, specificity as the proportion of invalidated alerts among all alerts not meeting the suspected case definition, PPV as the proportion of alerts that met the suspected case definition among all validated alerts, and NPV as the proportion of alerts that did not meet the suspected case definition among all invalidated alerts. We assessed timeliness as the median, range, and interquartile range (IQR) of the delay between the transmission of alert to the Alert Unit and the start of the onsite investigation. We evaluated representativeness through the geographic and demographic coverage of the alerts by comparing alert incidence by sex, age group, and health zone. We appraised usefulness by considering the number of confirmed and probable cases that were detected through the alert system. Finally, we assessed stability by considering how the system was operating over time, disruptions, and sustainability of functioning beyond the emergency response phase, notably in relation with costs and human resources. We conducted all analyses using R statistical software version 4.0.3 (20).

Outcomes of EWARS

Figure 2

Figure 2. Flow diagram of alerts in the Early Warning, Alert and Response System and their outcomes in 3 health zones, Democratic Republic of the Congo, August 2018–June 2020.

During the study period, 195,601 alerts were received; 194,768 (99.6%) from the health zones of Beni, Mutwanga, and Oicha, and 833 (0.4%) from other health zones (Figure 2). A small number (52,240, 2.7%) were reports of community deaths.

Figure 3

Figure 3. Trend in daily number of alerts from the Early Warning, Alert and Response System by final validation status in 3 health zones, Democratic Republic of the Congo, August 2018–June 2020....

On average, there were 280 alerts/day (range 2–955, median 127 alerts/day), although this value greatly varied over time. The number of daily alerts increased progressively, from 6 at the outset in August 2018 to a peak of 922 at the beginning of March 2020. We observed multiple sudden, short-lived decreases in the daily number of alerts, particularly in mid-November 2019 and early April 2020, coinciding with security incidents (see Stability) (Figure 3).

A total of 30,728 (15.8%) alerts were validated as suspected cases. Among those, 801 (2.6%) were finally classified as confirmed cases and 3 (<0.1%) as probable cases. No invalidated alerts became confirmed cases; the information recorded the first time remains in the database, and a new alert with the same information could be quickly detected.

Most (62.6%) alerts were raised by active case finding teams, followed by passive reporting from health facilities (19.0%), and community alerts (15.0%). The remainder (3.6%) originated from other sources (Table 1).

Sensitivity, Specificity, Positive Predictive Value, Negative Predictive Value

We excluded 434 alerts (0.2%) that were not investigated and 201 alerts (0.1%; 197 invalidated and 4 validated) that could not be classified according to the case definition because of missing data. A total of 17,927 (9.2%) alerts met the EVD suspected case definition. Sensitivity was 84.6% (95% CI 84.1%–85.1%) and specificity 91.2% (95% CI 91.0%–91.3%). PPV was 49.4% (95% CI 48.8%–49.9%) and NPV 98.3% (95% CI 98.2%–98.4%) (Table 2).

Indicators varied with time, health zone, and source of notification (Table 3). Overall, sensitivity increased over time, and specificity remained high throughout the study period. PPV decreased while NPV increased, which is consistent with the outbreak dynamics and the decrease in incidence toward the end of the epidemic (Appendix Figure).

Sensitivity was higher for alerts arising from surveillance sites (98.0%, 95% CI 97.4%–98.7%), community alerts (91.4%, 95% CI 90.1%–92.7%), and active case finding (87.5%, 95% CI 86.9%–88.1%) and lower for those arising from passive reporting from health facilities (65.4%, 95% CI 63.8%–67.0%). Conversely, specificity was highest in health facilities (96.2%, 95% CI 96.0%–96.4%), and was high (>90%) for all other sources except surveillance sites. Sensitivity was higher in Beni (94.8%, 95% CI 94.4%–95.2%) than in Mutwanga (54.9%, 95% CI 52.4%–57.3%) and Oicha (64.3%, 95% CI 62.8%–65.8%), but specificity was higher in Mutwanga (96.4%, 95% CI 96%–96.7%) and Oicha (93.3%, 95% CI 92.8–93.8).

Timeliness

An investigation was initiated within 2 hours from the time of alert for 188,184 (96.6%) alerts. The median time from alert transmission to the arrival of the investigation team on site was 11 minutes (IQR 10–15 minutes). Information about the time of investigation was not available for 3,475 (1.8%) alerts.

Figure 4

Figure 4. Timeliness over time of alerts from the Early Warning, Alert and Response System, Democratic Republic of the Congo, August 2018–June 2020. Timeliness is defined as weekly median time (in minutes)...

Timeliness of responses varied over time with substantial delays were observed at the outset of the system implementation, with greatest delays in Mutwanga (Figure 4). We saw no marked difference in timeliness by source of notification.

Representativeness

Figure 5

Figure 5. Trend in daily number of alerts in the Early Warning, Alert and Response System in 3 health zones in the Democratic Republic of the Congo, August 2018–June 2020.

We observed substantial variations in the alert incidence between the health zones. On average, there were 241 (range 2–789) alerts/day in Beni, 42.3 (range 1–181) alerts/day in Mutwanga, and 25.4 (range 1–138) alerts/day in Oicha. The alert incidence in the population followed a similar pattern: an average of 36 alerts/1,000 inhabitants/week in Beni, 2.5 alerts/1,000 inhabitants/week in Oicha, and 2.4 alerts/1,000 inhabitants/week in Mutwanga. In Beni, the incidence of alerts increased progressively from the outset (Figure 5). However, in Mutwanga and Oicha, incidence remained low until the beginning of November 2019, when it rapidly increased following community transmission.

We observed more alerts among female (56.3%) than male (43.4%) patients. Children <5 years of age were the most represented (23.6%), followed by patients 20–29 years of age (18.9%) and 10–19 years of age (18.9%); these percentages approximate the age and sex breakdown of the local population, with the exception of children 5–9 years of age, who were underrepresented (11%).

Usefulness and Cost

The EWARS system led to the detection of 801 confirmed and 3 probable cases, which equates to 242 alerts notified and 38 alerts validated for each case detected by the system. The total direct and indirect costs associated with EWARS implementation and maintenance was US $353,525 over the 2-year period of operation (Table 4), yielding a minimum value of US $1.8/alert and US $438/case detected.

Stability

The alert system operated 24 hours per day, 7 days per week, including a minimal night team to ensure continuity. Continuous communication and reporting of alerts was possible by a comprehensive and stable mobile phone coverage covering all health areas. As such, alerts were collected and analyzed on a continuous basis, and reports were produced and distributed daily. However, despite the continuous availability of human resources and communication networks, the system was severely disrupted by security incidents. Security incidents coincided with decreases in the number of alerts, affecting both the reporting and investigation of alerts (Figure 3). The Alert Unit ceased operations following the standard 90-day period of heightened surveillance after the declaration of the end of the outbreak, as determined by WHO (21).