Study design and setting

Data from this analysis were collected as part of a randomised controlled trial comparing oxytocin with misoprostol for post-partum haemorrhage (NCT01866241). All study procedures were conducted at the Mbarara Regional Referral Hospital, a publically-funded teaching hospital in rural south-western Uganda serving 10 districts with a population of over 5 million people. The hospital delivers over 10,000 mothers annually and a pre-study review of 9,027 births over 10 months estimated a 33% incidence of maternal PPH.

Participants and Recruitment

Full study procedures have been described previously[19]. Trained midwife research assistants (MRAs) screened labouring mothers in early active labour on arrival to the prenatal ward. Eligibility criteria were 1) age above 18 years, 2) 38–41 weeks of amenorrhea and 3) anticipated uncomplicated vaginal delivery as assessed by hospital staff. The exclusion criteria included: 1) confirmed foetal death, 2) self-reported maternal heart disease, 3) mothers' current diagnosis of severe malaria or acute bacterial infection, 4) multiple pregnancy, 5) induced or augmented labour, 6) planned Caesarean delivery, 7) ante partum haemorrhage (APH), 8) reported hypersensitivity to prostaglandins and 9) altered cognitive status (ACS) as assessed by the MRAs. MRAs obtained written informed consent from all eligible participants after the birth was determined to likely be uncomplicated vaginal delivery. All consenting participants gave written informed consent, or for those who could not write, a thumbprint was made on the consent form.

Study Groups

Participants were randomly assigned to 600μg misoprostol sublingually or 10 IU oxytocin intramuscularly within one minute following birth. Delayed cord clamping was preferred and the placenta was delivered by controlled cord traction (CCT) or manually if not delivered 30 minutes postpartum as per Ugandan National Clinical Guidelines (UNCG). Further care was provided by the hospital clinical care team in collaboration with MRAs in accordance with national guidelines which recommends administration of a repeat dose of parenteral oxytocics along with bladder emptying, management of lacerations, and uterine massage if bleeding persists. Mothers were monitored for 24 hours postpartum before discharge.

Study Measures

Two blood samples for complete blood count (CBC) were drawn 1) immediately after admission and 2) prior to hospital discharge or before the first blood transfusion, for participants who received one. After the baby was born, the amniotic fluid was drained from the plastic sheet immediately. A clean plastic sheet specifically designed and piloted to collect blood for this trial was placed under the mother’s buttocks during and after third stage of labour. Blood collected onto the plastic sheet during delivery was drained into a standard calibrated measuring jar[20] Additionally, mothers were given pre-weighed standard sanitary mops to place in the perineum during the entire postpartum period. These pads were collected, weighed and added onto the volume of blood from the plastic sheet. To improve consistency in estimation of blood loss, standardized electronic scales were used to weigh soiled mops. Mops were weighed hourly for the first six hours, and then every six hours until 24 hours postpartum. At the conclusion of the 24-hour postpartum period, remaining pads were collected, weighed and added to the volume of blood from the plastic sheet. We estimated blood loss as 1 mL per each gram of weighed mop, subtracted from the dry mop weight, as previously described[21]. Study staff who performed laboratory tests and directly measured blood loss were blinded to treatment allocation. The CBC was performed using a Sysmex KX-21N™ Automated Haematology Analyzer (Sysmex Corporation, Kobe, Japan).

Statistical Analysis

We described demographic and clinical data for the cohort using standard summarization techniques. Our primary aim was to assess the diagnostic validity of weighted blood loss method (WBL) versus quantified blood loss. Our reference standard was hemoglobin (Hb) drop, which is the usual means of measuring blood loss accurately and indirectly, with direct clinical relevance to the risk of myocardial infarction[16]. We considered a dichotomous outcome of Hb drop of >10% as a reasonable definition of PPH as previously used[21]

We assessed the sensitivity and specificity of WBL using the following dichotomous thresholds: a) ≥250mLs, b) ≥500mLs, c) 750mLs, d) ≥ 1000mLs at each of the following time points 1, 2 and 24 hour-post partum.

To describe the performance of the WBL compared to hemoglobin change, we estimated sensitivity and specificity of each volume-time combination (e.g. ≥250mLs at 1 hour, 2 hours, and 24 hours), and fit a receiver operating curve at each time point (i.e. 1 hour, 2 hours, and 24 hours). Finally, we calculated the positive and negative predictive values of each volume-time combination assuming PPH prevalence of 5, 15, and 30% to assess the clinical utility of each measure in various settings. Data analyses were performed using STATA version 12.0 (Statacorp, College Station, Texas, USA).

Ethical Approval

This study was approved by the Institutional Review Council of Mbarara University of Science and Technology and Uganda National Council of Science and Technology, and registered with clinicaltrials.gov (NCT01866241). All consenting participants gave written informed consent, or for those who could not write, a thumbprint was made on the consent form as approved by the ethics committees. All authors report no conflicts of interest.

Main findings

We conducted a post-hoc analysis of a randomized controlled trial of women undergoing uncomplicated vaginal deliveries at a publically-funded regional referral hospital in south-western Uganda to assess the diagnostic accuracy of WBL, compared to change in peri-partum haemoglobin. There are three principle findings from this study. First, the WBL generally has a poor sensitivity for detection of PPH when compared to a peri-partum haemoglobin drop >10%. This is evidenced by the fact that the sensitivity of weighed blood loss failed to detect 25% or more cases of PPH in all women aside from those with ≥ 250mL at 24 hours postpartum, at which point the corresponding specificity was only 5%. As such, our results suggested that WBL will miss an important minority of PPH cases and that no single time-volume threshold can be used to ensure an accurate diagnosis.

Second, in contrast, the specificity of the WBL was generally high, such that multiple volume-time combinations had specificities over 99%. For example, less than 1% of women with WBL ≥750mL at 1 or 2 hours, or with WBL ≥1000mL at any time point was free from PPH, as measured by a haemoglobin drop of >10%. The low rate of false positives at these blood-volume-time combinations makes WBL a promising diagnostic tool for detecting PPH when blood loss exceeds750 mL at less than 2 hours or if it exceeds 1,000 mL at any time.

Lastly, we found the highest utility of the WBL method was observed in high PPH prevalence settings. In low prevalence settings, the combination of a rare PPH outcome and poor sensitivity of the method gives it an overall low utility. For example in areas with a 5% prevalence of PPH, the NPV of weighed blood loss is always approximately 95% (which it would be without any diagnostic test because 5% of women would be expected to have the outcome) and the PPV never exceeds 70%. In contrast, the high specificity and PPV (>90%) of the WBL method for blood loss thresholds of ≥750 at 1 hour postpartum and for ≥1,000 mL at any time point make it a reasonable option as a means of identifying PPH in certain settings where blood cell indices cannot be promptly accessed, such as in the case of rural or home deliveries. Taken together, our results suggest that the WBL method could serve as a valuable “rule-in” test in resource-limited settings where prevalence of PPH is high, but should not be used alone to reassure clinical staff that women have not had a significant bleeding event to cause a serious complication.

Prior studies have had mixed results in comparing the accuracy of measured blood loss to quantified methods of blood loss. While Patel and colleagues found a positive correlation between measured blood loss and other laboratory method of measuring blood loss by photo spectrometry [12], their study was limited by a relatively small sample size (n = 10). Other studies have reported promising correlations between measured blood loss and laboratory methods that quantify blood loss using absorbent paper or sanitary pads and photometric measurement of alkaline hematin [22]. A notable exception is a study by Tourne and colleagues, who compared red cell indices to direct weight of a transparent collector plastic bag placed under the pelvis of mothers just after delivery[14]. Their results were consistent with our study, in that they reported a low sensitivity (38.8%) and high specificity (96%) for weighed blood loss of ≥500mL at approximately 1 hour postpartum. However, that study used a strict criterion of determining PPH, defined as a fall of hematocrit more than 10 points or by a fall of haemoglobin by more than 3g/dL, measured 3 days postpartum as a reference standard.

Strengths and Limitations

Our study had a number of strengths. We used a quantified reference standard that objectively considers all routes of blood loss, including haemolysis and internal bleeding. We also used standardized techniques to measure blood loss in all study participants with calibrated scales and sanitary mops. The mops, sheets, and scales used for the study were locally available and thus potentially reproducible in similar settings. Lastly, we had a large enough sample size (>1,000 participants) to thoroughly evaluate the WBL method with relatively narrow precision. There were also limitations to this work. Most importantly, we found clumping in blood loss estimates at both 400 and 500mL, likely due to rounding by study staff, which might misrepresent true blood loss in a sub-set of women. However, this form of clumping should only affect estimates at the 500mL threshold. Secondly, accurate estimation of blood loss through weighing is dependent on diversion of other fluid from the mops and collector bags. Although failure to do so might lead to mis-estimation of true blood loss[16], this limitation is likely to be present in non-research settings, so might be an accurate assessment of the technique in actual clinical conditions.