Search results
The search through the Cochrane Reviews database identified 93 trials which should be screened and assessed for eligibility (Fig. 1). Of these 93 trials, 30 were eligible for inclusion for grading and meta-analysis.
There were 28 trials which reported results for the minor PPH outcome and 21 trials for major PPH. Assessment of blood loss was measured or calculated using Hb only in 22 trials and was estimated only in seven trials. One trial reported estimated blood loss ≥500 ml (minor PPH), and a measured Hb of 2 g/dL (major PPH) [9].
Of the excluded trials, 50 were excluded during screening and a further 13 were excluded as they did not meet the review eligibility criteria.
Risk of bias of included trials
Risk of bias assessment summaries are shown in Figs. 2 and 3. Individual risk of bias assessments can be found in Additional file 1.
The risk of bias was assessed independently by LH and AW for 14 trials [10,11,12,13,14,15,16,17,18,19,20,21,22,23]. Six assessments [24,25,26,27,28,29] were produced during a review by Gallos et al. [30]. The risk of bias assessments for the Althabe 2009 [31] and Deneux-Tharaux 2013 [32] trials were completed during a review by Hofmeyr, Mshweshwe and Gülmezoglu [33]. Oladapo, Okusanya and Abalos [34] completed a risk of bias assessment for the Dagdeviren 2016 [35] trial. The risk of bias for the Jackson 2001 [9] trial was completed during the review by Soltani, Hutchon and Poulose [36]. Westhoff, Cotter and Tolosa [37] carried out a risk of bias assessment for the Poeschmann 1991 [38] trial. Where appropriate, risk of bias assessments from multiple sources were used to produce a single risk of bias for the current review. Risk of bias assessments for the Stanton 2013 [39] trial were completed by Gallos et al. [30] and Pantja et al. [40]. The Rogers 1998 [41] and Prendiville 1988 [42] risk of bias assessments were completed by Begley et al. [43] and Gallos et al. [30]. The Groot 1996 [44] trial risk of bias was completed by Gallos et al. [30], Liabsuetrakul et al. [45] and Westhoff, Cotter and Tolosa [37]. The Attilakos 2010 [46] risk of bias was assessed by two previous reviews, Gallos et al. [30] and Su, Chong and Samuel [47].
Selection bias was judged to be low risk in 87% of trials for random sequence generation and for 77% of trials for allocation concealment. Four trials were considered to have an unclear risk of selection bias as the methods to describe adequate random sequence generation were not reported or inadequately described [14, 15, 28, 42]. There was an unclear risk for five trials [19, 28, 29, 35, 39] concerning allocation concealment as it was not reported appropriately. The Diop 2016 [11] and Raghavan 2015 [21] trials were cluster randomised trials meaning allocation would have been known in advance so were judged to be high risk for inadequate allocation concealment.
Performance bias was low risk in 47% of trials, unclear risk in 7% and high risk in 47%. Detection bias showed similar results with 50% of trials deemed low risk, 17% unclear and 33% high risk. The large number of trials judged to be unclear or high risk for performance and detection bias were commonly due to difficult blinding participants and personnel to interventions due to the differences in administration [11, 16, 19, 21, 22, 31, 32, 35, 39, 41, 42, 44], unclear reporting of methods [28, 29] and a likelihood that blinding could be broken [18, 20]. The Deneux-Tharaux 2013 [32] and Groot 1996 [44] trials were judged to be high risk of performance bias but low and unclear risk of detection bias, respectively, due to the use of objective measurements for the primary outcome.
Attrition bias was low risk in 93% of trials. The Orji 2008 [28] trial was considered to have an unclear risk of attrition bias as the authors did not report attrition or any incomplete data. The Diop 2016 [12] trial was also considered to have unclear risk as the attrition rates in the misoprostol and oxytocin arms, 27.5 and 22.5% respectively, were high but the authors stated that all women given an intervention were followed up and were included in the analysis. Reporting bias was low risk in 63%, unclear in 30% and high risk in 7%. For eight trials deemed to have unclear risk of reporting bias, the protocol was not published or could not be accessed [10, 18, 25, 27,28,29, 41, 42]. The Shady 2017 [22] trial was judged to have an unclear risk of reporting bias as the definition of PPH was unclear. The protocol for the Hofmeyr 2011 [13] trial stated the rates of transfusion and haemoglobin < 8 g/dl after 24 h would be reported as secondary outcomes. However, this was not reported in the published trial results and was therefore judged to be high risk of selective reporting. The Lamont 2001 [15] trial did not report all expected outcomes such as additional surgery and transfusion rates so was also considered to be high risk.
Other bias was assessed on an individual basis and included a high risk of bias for early termination of the Poeschmann 1991 [38] and Rogers 1998 [41] trials. The Prendiville 1988 [42] trial was deemed to have a high risk of bias due to a smaller sample size and a protocol change after five months. The data of women included before the protocol change was still included within the final analysis. The Diop 2016 [11] trial changed the huts receiving the misoprostol or oxytocin intervention after initiation of the trial which has been considered to have a high risk of bias.
Primary outcomes
Proportion analysis for minor PPH grades
Overall, there were 28 trials reporting the proportion of women with a minor PPH. Grade 1 included 11 trials and had a random effects pooled proportion of 0.10 (95% CI = 0.042 to 0.18), with high heterogeneity (p < 0.0001; I2 = 99.1%; 95% CI = 98.9 to 99.2%) (Additional file 2: Figure S1) [12, 13, 17,18,19, 21, 22, 26, 28, 38, 44]. Grade 2 included four trials and had a random effects pooled proportion of 0.15 (95% CI = 0.067 to 0.26), with high heterogeneity (p < 0.0001; I2 of 96.6%; 95% CI = 94.6 to 97.7%) (Additional file 2: Figure S2) [16, 24, 35, 41]. Grade 3 included two trials and had a random effects pooled proportion of 0.14 (95% CI = 0.095 to 0.18), with medium heterogeneity (p = 0.07; I2 = 69.6%) (Additional file 2: Figure S3) [29, 42]. Grade 4 included 11 trials and had a random effects pooled proportion of 0.10 (95% CI = 0.072 to 0.14), with high heterogeneity (p < 0.0001; I2 = 98.1%; 95% CI = 97.8 to 98.4%) (Additional file 2: Figure S4) [9, 10, 14, 15, 20, 23, 25, 27, 31, 32, 39]. There were no trials included in grade 5 for minor PPH.
The pooled proportion across all grades for minor PPH was 0.11 (95% CI = 0.086 to 0.13) and there was high heterogeneity across the grades (p < 0.0001; I2 = 96.1%; 95% CI = 93.4 to 97.4%) (Fig. 4). Kruskal-Wallis analysis showed no statistically significant difference in minor PPH rates between the different allocated grades (T = 0.92, p = 0.82), and no significant difference with an all pairwise comparison of the individual grades.
Proportion analysis for major PPH grades
Overall, there were 21 trials reporting results for the proportion of participants with major PPH. Grade 1 included six trials and had a random effects pooled proportion of 0.033 (95% CI = 0.011 to 0.066), with high heterogeneity across the trials (p < 0.0001; I2 = 93.8%; 95% CI = 89.8 to 95.8%) (Additional file 3: Figure S1) [12, 13, 18, 26, 38, 44]. Grade 2 included three trials and had a random pooled proportion of 0.033 (95% CI 0.012 to 0.063), with high heterogeneity (p = 0.004; I2 = 81.9%; 95%CI = 0 to 92.3%) (Additional file 3: Figure S2) [16, 35, 41]. Grade 3 included one trial with a proportion of 0.020 (95% CI = 0.014 to 0.027) (Additional file 3: Figure S3) [42]. Grade 4 included 10 trials and had a random effects pooled proportion of 0.035 (95% CI = 0.015 to 0.064), with high heterogeneity (p < 0.0001; I2 = 98.5%; 95% CI = 98.3 to 98.8%) (Additional file 3: Figure S4) [9,10,11, 14, 15, 20, 25, 31, 32, 39]. Grade 5 included one trial with a proportion of 0.049 (95% CI = 0.030 to 0.073) (Additional file 3: Figure S5) [46].
The pooled proportion across all grades for major PPH was 0.030 (95% CI = 0.022 to 0.041) with high heterogeneity across the grades (p < 0.0001; I2 = 90.8%; 95% CI = 81 to 94.4%) (Fig. 5). Kruskal-Wallis analysis showed that there is no statistically significant difference in major PPH rates between the different grades (T = 0.91, p = 0.92), and no significant difference with an all pairwise comparison of the individual grades.
Secondary outcomes
Methods comparison for minor PPH
Seven trials for minor PPH collected blood loss data based on an estimated volume (median rate = 0.12), whereas, 21 trials used a measurement of blood loss (median rate = 0.10) (Fig. 6a). Measurement of blood loss included gravimetric measurement, BRASS-V drape collection, haemoglobin or haematocrit measurement and the traditional measurement of weighing blood-soaked items. Overall, there was no significant difference in the rates of minor PPH between the methods (Mann-Whitney U = 67, p (two-tailed) = 0.75). All trials within grade 1 used a measurement to determine minor PPH (median rate = 0.061). There was no significant difference in the rates of minor PPH between measured (n = 3; median rate = 0.12) or estimated (n = 1; median rate = 0.12) methods for grade 2 trials (Mann-Whitney U = 2, p (two-tailed) > 0.99). Grade 3 contained two trials, one used a measurement (median rate = 0.17) and one used an estimate of blood loss (median rate = 0.12). There was no significant difference in the rates of minor PPH between measured (n = 6, median rate = 0.092) and estimated blood loss (n = 5, median rate = 0.086) for grade 4 trials (Mann-Whitney U = 12, p (two-tailed) = 0.66).
Methods comparison for major PPH
For major PPH data, five trials reported data based on estimated blood loss (median rate = 0.024) and 16 trials reported data using a measurement (median rate = 0.025). There was no significant difference between the methods of determining the rate of PPH (Mann-Whitney U = 35, p (two-tailed) = 0.72) (Fig. 6b). All six trials within grade 1 used a measurement of blood loss (n = 6; median rate = 0.030). Within grade 2, two trials used a measurement (median rate = 0.045) and one trial used an estimation (median rate = 0.022). There was one trial in grade 3 which used an estimate to determine PPH rate (median rate = 0.019). In grade 4, there was no significant difference between the measured (n = 8; median rate = 0.020) and estimated (n = 2; median rate = 0.034) groups for the rate of major PPH (Mann-Whitney U = 5, p (two tailed) = 0.53). Grade 5 contained one trial which used a method of estimation to determine the rate of PPH (median rate = 0.048).
Post-hoc analysis
Rate of PPH vs. % operative births
Based on the results of post hoc analysis, the percentage of operative births in a study and the rate of minor PPH are significantly correlated (Spearman r = 0.32) (Fig. 7a). However, there is no correlation between the percentage of operative births and the rate of major PPH (Spearman r = 0.098, p (two-tailed) = 0.34) (Fig. 7b).