Postpartum haemorrhage (PPH) rates in randomized trials of PPH prophylactic interventions and the effect of underlying participant PPH risk: a meta-analysis

Background Postpartum haemorrhage (PPH) remains a leading cause of maternal mortality. Many trials assessing interventions to prevent PPH base their data on low risk women. It is important to consider the impact data collection methods may have on these results. This review aims to assess trials of PPH prophylaxis by grading trials according to the degree of risk status of the population enrolled in these trials and identify differences in the PPH rates of low risk and high risk populations. Methods Systematic review and meta-analysis using a random-effects model. Trials were identified through CENTRAL. Trials were assessed for eligibility then graded according to antenatal risk factors and method of birth into five grades. The main outcomes were overall trial rate of minor PPH (blood loss ≥500 ml) and major PPH (> 1000 ml) and method of determining blood loss (estimated/measured). Results There was no relationship between minor or major PPH rate and risk grade (Kruskal-Wallis: minor - T = 0.92, p = 0.82; major - T = 0.91, p = 0.92). There was no difference in minor or major PPH rates when comparing estimation or measurement methods (Mann-Whitney: minor - U = 67, p = 0.75; major - U = 35, p = 0.72). There was however a correlation between % operative births and minor PPH rate, but not major PPH (Spearman r = 0.32 v. Spearman r = 0.098). Conclusions Using data from trials using low risk women to generalise best practice guidelines might not be appropriate for all births, particularly complex births. Although complex births contribute disproportionately to PPH rates, this review showed they are often underrepresented in trials. Despite this, there was no difference in reported PPH rates between studies conducted in high and low risk groups. Method of birth was shown to be an important risk factor for minor PPH and may be a better predictor of PPH than antenatal risk factors. Women with operative births are often excluded from trials meaning a lack of data supporting interventions in these women. More focus on complex births is needed to ensure the evidence base is relevant to the target population.


Low risk
Sequence generated at the coordinating centre using computergenerated list of numbers with randomly permuted blocks of 4-6 in a 1:1 ratio

Low risk
The randomisation sequence (1:1 ratio-blocks of 10, no stratification) was generated by computer.
Allocation concealment (selection bias) Low risk The preparation of the ampoules was undertaken by DHP Ltd. (Powys, UK). All boxes and ampoules were identically labelled. The random allocation sequence was not known to the investigators until the study had finished and the analysis was started.
Blinding of participants and personnel (performance bias) All outcomes Low risk Study participants and caregivers were blinded to treatment allocations.
Blinding of outcome assessment (detection bias) All outcomes Low risk Assessors were blinded to treatment allocations.
Incomplete outcome data (attrition bias) All outcomes Low risk Data were collected completely from all randomised study participants.

Low risk
The study report matches the study protocol that was registered prospectively (EudraCT 2005-002812-94).

Low risk
Randomisation was achieved using a computer-generated random list.
Allocation concealment (selection bias) Low risk Investigators used identical study boxes. Care was taken that no difference could be seen or heard between the packages of the ergometrine/placebo tablets and the oxytocin ampoules.
Blinding of participants and personnel (performance bias) All outcomes

High risk
The study was "double-blind" with placebo to match ergometrine treatment, but "to allow comparison with a standard prophylactic regimen a third group receiving the standard intramuscular oxytocin was added, but for obvious reasons this could not be conducted in a blind manner." Blinding of outcome assessment (detection bias) All outcomes Unclear risk Assessor blinding was not reported, however, objective blood loss measurements were used.
Incomplete outcome data (attrition bias) All outcomes Low risk "4 women with exclusion criteria were entered erroneously. They are considered as non-participants."

Selective reporting (reporting bias)
Low risk It is clear that the published reports include all expected outcomes specified. Incomplete outcome data (attrition bias) All outcomes Low risk Data were collected completely from all randomised study participants.

Selective reporting (reporting bias)
Unclear risk The protocol of the study was unavailable for verification.

Other bias High risk
The protocol was modified after 5 months due to high blood loss in the expectant group but data from the initial 5 months were included in the analysis. Sample size was supposed to be 3900 but stopped at 1695. 30 women in the control group gave a late maternal refusal, whereas only 1 in the experimental group did so. The outcomes of these women are included in analysis.

Low risk
The 52 CHOs were randomly allocated equally to either the intervention or the control group; this allocation was stratified by both district and distance (at least 10 km, or less than 10 km) to emergency obstetric care. The randomisation sequence was determined using Stata (version 12). Allocation concealment (selection bias)

Unclear risk
Allocation concealment was not reported.
Blinding of participants and personnel (performance bias) All outcomes High risk "The random allocation was not masked." Blinding of outcome assessment (detection bias) All outcomes High risk Assessors were not blinded to treatment allocations.
Incomplete outcome data (attrition bias) All outcomes Low risk "7 and 9 enrolled women in the oxytocin and control arms, respectively, lacked a blood-loss measure." Selective reporting (reporting bias)

Low risk
The study report matches the study protocol that was registered (ClinicalTrials.gov NCT01108289). Low risk "Maternity huts with auxiliary midwives located 3-21 km from the closest referral centre were randomly assigned (1:1) by staff at Gynuity Health Projects" Allocation concealment (selection bias)

High risk
As a cluster randomised trial, allocation would be known in advance.
Blinding of participants and personnel (performance bias) Low risk "The random allocation sequence was generated by using computer-generated random numbers and was stratified by country in blocks of 6-8" Allocation concealment (selection bias)

Low risk
Study packs were selected sequentially as each woman was enrolled and "the study drug packs were prepared, by Gynuity Health Projects. The packs were identical in shape, color, weight, and feel" Blinding of participants and personnel (performance bias)

Low risk
Packs containing tablets identical in appearance were prepared at a different site, maintaining blinding of personnel and participants Blinding of outcome assessment (detection bias)

Low risk
As above Unclear risk The authors state that participants were randomly allocated to an arm but do not explain the randomisation process in full, however, we believe that randomisation was likely adequate Allocation concealment (selection bias) Low risk "randomisation slips were contained in enveloped opened by a person not involved in the trial… who resealed the envelope" Blinding of participants and personnel (performance bias)

Low risk
Blinding was achieved through preparation of the drug by an independent person not involved in further care of participants or assessments Blinding of outcome assessment (detection bias)

Low risk
As above.
Incomplete outcome data (attrition bias) Low risk "To ensure balanced randomization and to conceal treatment assignment, a computergenerated randomization list with a random block size for each hospital was used. The data coordinating center generated a randomization list stratified by hospital" Allocation concealment (selection bias) Low risk "sequentially numbered sealed opaque envelope out of a locked cabinet in the delivery room" Blinding of participants and personnel (performance bias) Low risk Envelopes were prepared by a member of the research team not involved in patient care. As misoprostol and ZB11 are administered at different times, a placebo was used alongside the allocated intervention to avoid knowledge of the intervention by the clinician or participant Blinding of outcome assessment (detection bias)

Low risk
Blinding of assessor as above, with objective measurement of the primary outcome (postpartum haemorrhage) used Incomplete outcome data (attrition bias)

Low risk
Attrition was low as data was available for 960/967 women who had been randomised to the trial (less than 1% attrition) Selective reporting (reporting bias)

Low risk
Outcomes reported as per protocol (NCT00147420)

Domain
Review authors judgement Support for judgement Random sequence generation (selection bias) Low risk "...block randomisation, stratified according to number of births (first and second and more childbirths) with block sizes of 4 and 6 and allocation ratio of 1:1." Allocation concealment (selection bias) Low risk "...opaque sequentially numbered sealed packages" Blinding of participants and personnel (performance bias) High risk A nurse not involved directly with participant prepared the drugs so "the data assessor and participants had no knowledge of the study medication." However, it is likely that blinding could have been broken. Blinding of outcome assessment (detection bias)

High risk
As above Incomplete outcome data (attrition bias)

Low risk No attrition
Selective reporting (reporting bias) Unclear risk Protocol not seen.