Selection of studies
A total of 918 studies were retrieved from sources by pre-specified search strategies. Of these, 47 studies were duplicates and are removed. In preliminary screening, 815 studies (802 by looking at the title and 13 by abstract) were removed. A total of 56 studies were identified for checking eligibility for full text review. Of these, 21 studies were excluded with the following reasons:
A case control study from Bonga General and Mizan Tepi University Teaching Hospitals, a facility based case control study from Arbaminch general hospital, a prospective cohort study from tertiary care hospitals in Addis Ababa, a case control study from public hospitals of Tigray, a prospective cohort study from Northern Ethiopia, a cross-sectional study on neonatal mortality from Ethiopia, a case control study from Debre Tabor town, a hospital based retrospective cohort study from Somali Ethiopia, a case control study from Hawassa University hospital were excluded due to lack of relevant data for this review. A 5 years retrospective survey from Tikur Anbesa Hospital, a matched case control study from Kalu district, a retrospective chart review from Yekatit 12 hospital, Addis Ababa, a cross-sectional study from Addis Ababa hospital, a cross-sectional study from Jimma Hospital and a retrospective review on perinatal mortality audit from Jimma hospital were excluded due to lack of access to full text. A retrospective study from Mettu Karl referral hospital Ethiopia, a cross-sectional study from Amhara region hospitals, an institutional based cross-sectional study from Debremarkos referral hospital and a retrospective cohort study from Wolita Sodo referral hospital were excluded since they were conducted on high risk groups for perinatal deaths. A study from Kersa demographic and health surveillance system site was excluded due to double reporting in publications. Lastly, an interventional study from Ethiopia was excluded because of the design as this review focus on observational studies.
Finally, a total of 35 studies that fulfill the pre-specified inclusion criteria were obtained (Fig. 1).
Characteristics of included studies
A total of 35 studies included in this review. However, studies included in the meta-analysis were thirty four [1,2,3, 8, 11, 12, 18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45]. Pregnant women, women attending for labor and delivery service and newborns were study participants considered for data source in primary studies.
Of 34 studies included in the meta-analysis, 30 studies contained data relevant for perinatal mortality and trend estimation. From 30 studies, 25(83.3%) and 18(60%) studies had data relevant for pooling stillbirth and early neonatal mortality rates, respectively. One study (25) had crudely reported perinatal mortality without specifying either stillbirth or early neonatal mortality. From 34 studies, only 8(23.5%) contained data relevant to see the relationship between inter-pregnancy interval and perinatal mortality.
Perinatal mortality rate (PMR) in primary studies ranges from 13.98 (15) to 85.3 (24) per 1000 total births in community based studies. Whilst it ranges from 27.6 (25) to 173 (23) per 1000 total births in health institution based studies. Stillbirth rate (SBR) ranges from 5.9 (26) to 85.3 (24) per 1000 total births in community based studies while from 22.7 (31) to 173.3(23) per 1000 total births at health institution based studies. Early neonatal mortality rate (ENMR) ranges from 10.8(26) to 41.4(32) per 1000 live births in community based studies while it ranges from 24.8(5) to 110(30) per 1000 live births at health institution. All conditions (PMR, SBR and ENMR) were higher in health facility based studies than community based studies. Inter-pregnancy interval less than 15 months or birth interval less than 24 months was significantly associated with perinatal mortality in 5(62.5%) of the 8 studies. In two studies however (20, 26) it was not significantly associated with perinatal mortality. Categorization of inter-pregnancy interval or birth interval in these two studies was not in line with most of the literatures. In one study (34), it is unclear that, whether it was fitted in the adjusted model or not.
Of 34 studies, 26(76.4%) were conducted in both (urban and rural) settings. The others: 4(11.8%) in urban and 4(11.8%) in rural settings alone. Sixteen (47.1%) were community based while 18(52.9%) were institution based studies. Regarding regions: Amhara 8(23.6%), Oromia 6(17.7%), SNNP 5(14.7%), Addis Ababa 3(8.8%), Tigray 5(14.7%), Benishangul-Gumuz 1(2.9%), mixed (more than one region) 3(8.8%) and the others 3(8.8%) were National (Ethiopia) Demographic and Health Surveys (EDHS). Concerning the study design: cross-sectional 21(61.8%), follow ups 6(17.6%), prospective cohort 1(2.9%), case control 4(11.8%), nested case control 2(5.9%).
Regarding year of publication: 2(5.9%) published before 2000, 11(32.3%) published from 2001 to 2015 and 21(61.8%) published from 2016 to 2019.
Study id, author, year of publication, region, study design, sample size, exposure variable and outcome data were extracted to describe the characteristics of the studies. The total sample size of the studies was 141,835 (136,168 to determine the magnitude of perinatal mortality and 5667 to see the relationship between inter-pregnancy interval and perinatal mortality). A sample size considered for primary studies ranges from 300 to 20,161. For the assessment of the relationship between inter-pregnancy interval and perinatal mortality, the sample size considered ranges from, 219 to 1949 (See Table 1).
Quality (risk of bias) assessment for the included studies
Of 24 cross-sectional studies: except for 4 studies (ID: 2, 24, 15, 33), measurement of exposure in a valid and reliable way was unclear for 16 (66.6%) studies. The confounder factors identified way was unclear for 9 studies (ID: 21, 5, 22, 11, 17, 18, 23, 19, 16). Strategies to deal with confounding factors were also unclear for 2 studies (ID: 17, 19). None of the case control studies used matching for potential confounders. From the follow up studies; strategies to address the incomplete follow up was not utilized in 4 studies (ID: 1, 6, 9, 14). According to Cohen’s Kappa statistics (K = 0.65, % of agreement = 97.2%), the two reviewers (BH and AE) have “substantial agreement” to include studies in the meta-analysis . One study  was not considered for data extraction since it scores below 50% for JBI quality indicators. See Table S1 in Additional file 3 for quality assessment scores.
Pooled estimate of perinatal mortality rate
Forest plot was used to identify heterogeneity between studies. The result of the forest plot indicated that there is heterogeneity among studies (I2 = 99%, p < 0.0001). The result in fixed effect suggests the need to fit the random effect model to handle heterogeneity.
In random effect model, tau2 using Dar-Simonial Liard (DL) quantified the amount of total heterogeneity to be 0.0046. Using double arcsine transformed proportion in random effect model, the weighted average from each study using the inverse of variances yield a more valid effect size.
Accordingly, the pooled perinatal mortality rate in Ethiopia was 51.3 per 1000 total births with 95% CI (40.8, 62.8) (Fig. 2). Perinatal mortality includes stillbirth and early neonatal death. The pooled stillbirth rate was 36.9 per 1000 births with 95% CI (27.3, 47.8). The pooled early neonatal mortality rate was 29.5 per 1000 live births with 95% CI (23.9, 35.6).
The result of sensitivity analysis indicated that no study is found to be removed since the estimate of each study, when removed, is within the confidence interval of the pooled perinatal mortality rate (40.8, 62.8) (Figure S1 in Additional file 4).
The results of subgroup analysis for magnitude of perinatal mortality rate was indicated below (Table 2).
The presence of small study effects/publication bias was examined using regression test (Egger test). The test result indicated that there was no any small study effects or publication bias (P = 0.076).
Trend of perinatal mortality
The cumulative meta-analysis indicated that, there is slight reduction trend with some ups and downs in between years of publication. From the Fig. 3, we can see that; perinatal mortality rate was reduced from 66 per 1000 total births in 1997 to 51.3 per 1000 total births in 2019. Significant reduction trend in perinatal mortality rate was observed from 1997 to 2014 (p = 0.014) but in the last 5 years (2015–2019) no reduction was seen. Overall, regression test indicated that there was no statistically significant reduction trend in perinatal mortality rate from 1997 to 2019 (P = 0.08). Since perinatal mortality rate was the sum of stillbirths and early neonatal mortality rates it is important to see the trends of them separately. In this regard, different kinds of trends were observed for stillbirth and early neonatal mortality rates (Fig. 4). There was no significant reduction trend in stillbirth rate rather it increased from 23.7 in 1997 to 36.9 in 2019 per 1000 births (P = 0.25). A significant reduction trend was observed in early neonatal mortality rate from 51 in 1997 to 29.5 in 2019 per 1000 live births (P = 0.01).
Relationship between inter-pregnancy interval and perinatal mortality
The relationship between inter-pregnancy interval less than 15 months and perinatal mortality was observed. Accordingly, children who were conceived within 15 months of their preceding child birth were nearly three times more likely risked to death during perinatal periods as compared to their counter parts; OR = 2.76, 95% CI: (2.1, 3.6). There is low heterogeneity among studies (I2 = 46%, P = 0.07) (Fig. 5).
Regression test for funnel plot asymmetry showed that there is no evidence of publications bias for trim and fill analysis (P = 0.313) (Fig. 6).
Key findings of the review
Based on the review questions the following findings were obtained (Table 3).