The study retrieval and selection strategy are illustrated in Figure 1. Of the 878 publications meeting initial search criteria, 65 papers were retrieved. On detailed evaluation, 53 articles were excluded leaving a total of 13, which were included in the meta-analysis.
Quality of the included articles
Basing on the Murad’s quality assessment tool, the four domains of selection, ascertainment, causality and reporting were used to assess the potential bias in the selected articles. On evaluation, the included articles addressed most of the relevant questions in the four assessment domains, as shown in Figure 2.
Pooled Case Fatality Rate of EVD outbreaks
We used 13 articles (32,300 cases and 13,727 deaths) that reported on EVD outbreaks in the last ten years in Africa. Their CFR ranged from 28-100%. We used the random-effects model and calculated the pooled CFR, which was found to be 60% (95% CI: 47-73%), with P<0.001. However, there was significant heterogeneity observed among the outbreaks with I2=99.6%, P<0.001, as shown in figure 3. In addition, sensitivity analysis by removing outbreaks that reported single case and those with CFR of 100% showed no effect on the observed heterogeneity (I2=99.7%, P<0.001), but the pooled CFR reduced to 57% (95% CI:45-69%).
Factors influencing the CFR of EVD outbreaks
Meta-regression was done to explore the possible sources of the observed heterogeneity (Table 2). The analysis was done for four pre-selected factors; start (time) of the outbreak, species involved, country of outbreak and number of cases in the outbreak. Of these factors, the country of the outbreak (P=0.046) and the number of cases (P<0.001) were the only factors identified to cause the between-outbreak variance.
Outbreaks were, first, categorised according to the start (time) of the outbreak; 1st half of the year (5 outbreaks) and 2nd half of the year (8 outbreaks). On subgroup analysis, there was no heterogeneity noted among EVD outbreaks that started in the 1st half of the year. However, substantial heterogeneity remained among outbreaks that started in the second half of the year (I2=99.7%, P<0.001) (Table 2).
Regarding the species causing the outbreak, outbreaks caused by Sudan ebolavirus (2 outbreaks) showed no heterogeneity, unlike Zaire ebolavirus outbreaks (10) which showed a significant heterogeneity (I2=99.7%, P<0.001). It was also noted that Sudan ebolavirus had the highest CFR of 80%, followed by Zaire ebolavirus (58%) and Bundibugyo ebolavirus (55%), as shown in table 2. However, Zaire ebolavirus was responsible for 10 outbreaks out of the 13 recorded outbreaks.
According to the region/country of the outbreak, outbreaks (3) in Uganda (East Africa) and DRC (Central Africa-5 outbreaks) showed no significant heterogeneity on subgroup analysis. However, outbreaks from West Africa (5) had persistent heterogeneity (99.8%, P<0.001). In the last ten years, DRC was noted to have the most number of EVD outbreaks (5 outbreaks) with a CFR of 65% (95% CI: 59-71%), followed by Uganda (3 outbreaks), with CFR=83% (95% CI: 60-99%) (Figure 4). The West African countries included: Liberia, Sierra Leone, Guinea, Mali and Nigeria, each with one EVD outbreak.
When sub-grouped according to the number of cases in each outbreak, outbreaks with less than 1000 cases (9 outbreaks) showed no significant heterogeneity, compared to those with above 1000 cases (4 outbreaks) which showed substantial heterogeneity (I2= 99.9%, P<0.001) (Figure 5). It was noted that four countries reported EVD outbreaks with above 1000 cases, which included; Liberia, Sierra Leone, Guinea and DRC.
Publication bias
Publication bias was assessed in the EVD articles using visual inspection of the funnel plot, which showed a slight asymmetry in the outbreaks, implying probable publication bias towards small EVD outbreaks (Figure 6). However, further evaluation using Egger’s test showed no significant publication bias in EVD outbreaks (P= 0.872). Nevertheless, when the trim-and-fill analysis was executed, the adjusted CFR was 55.5% (95%CI: 44.3-69.6%), after filling in 2 missing outbreaks.