The trial was implemented between January 2006 and May 2007, depending on the country. Before participation, each centre was visited by the national coordinator. At the visit, staff were reminded of the EUPHRATES consensus statement on the prevention and management of postpartum haemorrhage and familiarised with the processes and the data collection instrument.

A second visit from the national coordinator took place in the intervention group after randomisation, when use of the collector bag was explained to birth attendants with the aid of standard written instructions and a training video. The bag was to be placed under the mother?s pelvis as soon as the baby was born and before delivery of the placenta. The bag was transparent and graduated, allowing continuous monitoring of blood loss. It did not require sterilisation and could be used in the dorsal, lateral, or lithotomy positions. Women delivering standing or crouching could be offered the opportunity to lie down for the third stage, allowing the bag to be placed under their pelvis. The bag was to be left in situ until the birth attendant was no longer concerned about blood loss, such as when a sanitary towel was applied to the vulva. Bags were purchased centrally and provided to each cluster in the intervention arm.

No collector bag was used in the control group, with postpartum blood loss being assessed visually. The control group was monitored during the study period to assess compliance with allocation.

The primary outcome was the incidence of severe postpartum haemorrhage after vaginal deliveries, defined as a composite of all women who experienced one or more of blood transfusion, intravenous plasma expansion, arterial embolisation, surgical procedure, admission to an intensive care unit, treatment with recombinant factor VII, and death. Secondary outcomes were each of the components of the primary outcome, manual removal of the placenta, and administration of prostaglandins after delivery.

Each participating centre was asked to collect data for four months from all women who had had a vaginal delivery: for one month before randomisation (baseline period) and for three months after randomisation in the control group (trial period). The three month period of data collection in the intervention group followed a two week training period during which the unit started using the collector bag.

Data were collected using a form filled in by the birth attendants for each vaginal delivery. Information was collected on maternal age, induction of labour, mode of delivery, number of babies, birth weight, use of prophylactic uterotonics, and outcome data. A second form was used for deliveries with severe postpartum haemorrhage, with information collected on the delivery and management of postpartum haemorrhage. This form was used to cross check criteria for the primary outcome.

The sample size calculation took into account the cluster randomised design; we estimated the intracluster correlation coefficient to be 0.01. With the assumption of an event rate for the primary outcome of 2.5% in the control group, 82 clusters (41 in each arm of the trial) were required to detect a decrease in the event rate to 1.5% (a 40% relative risk reduction) with 80% power, a two sided significance level of 5%, and an average cluster size of 300 women.²⁵

The participants and maternity units were analysed in the groups to which they were assigned regardless of the management received by individual women or deviation from the protocol.

We summarised the baseline characteristics of the maternity units and individual women using counts (percentages) for categorical variables, means (standard deviations) for normally distributed continuous variables, or medians (interquartile ranges) for other continuous variables. Comparative statistical analysis was done at both individual and cluster level and took into account the effect of clustering. All statistical tests were two sided (5% significance level) and not adjusted for multiple comparisons. Statistical analyses were done using SPSS version 17 and Stata v10.0 software.

For analysis at the individual level we compared primary and secondary outcomes between the two study groups both unadjusted and adjusted for the effect of clustering. To determine the magnitude and direction of any differences in outcomes between the two groups, we calculated crude odds ratios and 95% confidence intervals. We also used logistic regression to adjust for clustering and key prognostic factors. The cluster randomised design imparts a data structure that facilitates the calculation of a valid intracluster correlation coefficient, ?.

Cluster level analysis was carried out only on the primary outcome. Some hospitals contributed fewer events than others and some recruited fewer women. We allowed these hospitals to have less effect on the treatment estimate by weighting the analysis on the basis of the precision?that is, we calculated the weighted mean difference for the treatment comparison. We used a weighted linear regression model to test the effect of the intervention on the rate of severe postpartum haemorrhage during the trial period, adjusting for the baseline rate, expressed as the weighted mean difference (95% confidence interval).

Analyses were done to test the effect of the intervention on the main components of the primary outcome (table 3). The proportion of women who had blood transfusion, surgical procedure, embolisation, or manual removal of placenta did not substantially differ between the intervention and control groups, whether after adjustment for cluster or after further adjustment for other prognostic factors. No maternal deaths occurred.

The proportion of women in receipt of intravenous plasma expanders or prostaglandins differed between intervention and control groups, but the differences were not significant after adjustment for clustering effect.

This trial provides new results on an unexplored although controversial aspect of care in the third stage of labour. Although objective measurement has been shown to increase the accuracy of assessing postpartum blood loss compared with visual estimation,^{15, 16, 17, 18, 19, 20, 21} the routine use of a collector bag was not associated with a significant decrease in severe postpartum haemorrhage. This result constitutes an important contribution to the ongoing debate on strategies to improve the care of women with postpartum haemorrhage and to decrease the incidence of severe cases. Additionally, the cluster randomised design and the large number of clusters and their diversity provide good external validity to this trial. Small deviations from the protocol did occur for data collection, but sensitivity analyses showed that none of these changed the internal validity of the trial.

Heterogeneity of baseline rates for severe postpartum haemorrhage was large between maternity units (0% to 13.4%). In theory such a variation should be an asset and reflect a broad range of levels of risk in the participating maternity units. Because these differences were strongly related to the country, however, some concern remains about the criteria in use for the management of postpartum haemorrhage in different parts of Europe. Sensitivity analysis showed that this aspect did not alter the results.

Baseline data showed some heterogeneity between the intervention and control groups. Heterogeneity in postpartum haemorrhage related practices and rates has been reported across maternity units in Europe, both between and within countries.^{4, 23} Randomisation should balance these differences between the two arms. However, given the number of units randomised, although large for a cluster randomised controlled trial, there is a slight possibility of residual imbalance. Analyses were, however, adjusted for the main determinants of postpartum haemorrhage (individual level analysis) and baseline rate of severe postpartum haemorrhage (cluster level analysis); in addition, sensitivity analysis indicated that the absence of a major effect of the intervention was similar whether the maternity units had a high or low baseline rate of severe postpartum haemorrhage. In consequence, any perceived or real imbalance in these characteristics should have little or no effect on the findings.

Different mechanisms may explain the absence of difference in the rates of severe postpartum haemorrhage between maternity units that used the bag and those that visually assessed blood loss. This may result from a lack of compliance to the intervention. This is, however, unlikely because of the persistent absence of difference between the two groups when the analysis was restricted to those maternity units that used the bag in a high proportion of deliveries.

One potential reason for the apparent ineffectiveness of the intervention might be incorrect use of the bags; in particular, that they were covered most of the time and thus could not be viewed. Such misuse is unlikely to explain the observed lack of effect as detailed oral and written instructions were provided and the training video clearly showed the care giver watching the bag and the graduations.

Participation in the study may indicate a particular interest in the management of postpartum haemorrhage so that existing management had little room for improvement. Such a selection process is, however, unlikely in these maternity units, owing to the variety of baseline rates of severe postpartum haemorrhage.

It may be hypothesised that the intervention has a double effect, in two opposite directions: increasing the rate of ascertainment through increased vigilance and decreasing the prevalence through timely management of excessive bleeding. If these two components were of the same order of magnitude, the global effect would be no effect. If this explanation was realistic, however, different effect sizes would be expected with different baseline rates or different degrees of compliance, or both. None of this occurred, making it unlikely that a benefit of the intervention in terms of decreased severe outcome was balanced by an equivalent increase in ascertainment. In fact the intervention seemed to increase the rates of postpartum haemorrhage, reflecting that the intervention was possibly more effective in improving ascertainment than in changing practice.

A concomitant effect in the control group may also have contributed to the absence of difference between the two arms. Contamination of the intervention to control units is unlikely since participating units were not in contact, and no use of bags was reported in all the control maternity units. Participation in a research study, independently of any specific intervention, has been reported to change behaviours of participants (Hawthorne effect).²⁶ The hypothesis that the management of postpartum haemorrhage would have improved in the control arm is, however, not supported by the absence of change in the rate of severe postpartum haemorrhage between the baseline and trial periods in this group.

The most plausible explanation of the negative result of this trial is that having a more accurate assessment of postpartum blood loss is not by itself sufficient to change behaviours of care givers and improve the management of postpartum haemorrhage. Lack of identification of women with excessive postpartum bleeding is a problem, potentially leading to higher levels of medical intervention if the bleeding progresses to severe haemorrhage. We designed a strategy to increase the awareness of care givers. The fact that this has not translated into a change in clinical outcomes probably reflects the complexity of decisions on management, which are influenced by multiple factors such as organisation of the delivery ward and how care givers perceive and cope with emergencies.

We did not find any other published study assessing the effectiveness of the collector bag. We have, however, identified other large multicentre randomised trials in the specialty of maternal and child health where a diagnostic or screening test was evaluated but without any associated instructions about the management of abnormal results.^{27, 28, 29} None of these trials showed benefit with the introduction of the test. In addition one study showed that simple information is not sufficient to have an effect on the readiness of birth attendants to change.³⁰ These various reports suggest that the effect of enhanced diagnostic methods should include an accompanying protocol of management and may be a specific behavioural intervention, which in effect becomes a ?complex intervention.?

The practical implication of these results for high income countries is that those units that are using a collector bag (cost per bag between ?1 (88p; $1.44) and ?11 in Europe) need to reconsider their practice and possibly reallocate the resources to other aspects of care. Units that are not routinely using the bag should keep the same policy. For resource poor countries, positive results with the use of the ?kanga collector? have been reported.³¹ This needs to be tested in a randomised design. In the current context of reported ongoing increase in the prevalence of postpartum haemorrhage, further research is needed to develop and test effective strategies to decrease the prevalence of severe postpartum haemorrhage through improvement of management. These will probably be multifaceted interventions, and in this context the collector bag may warrant further investigation.

We thank Allan Donner (Canada) and Pierre Buekens (USA) for their scientific advice and St?phane Freze and Myriam Loubriat for their contribution to collecting and cleaning the data. The following are members of EUPHRATES (EUropean Project on obstetric Haemorrhage, Reduction, Attitudes, Trial and Early warning System): Sophie Alexander (project leader, Belgium), Diogo Ayres-de-Campos (Portugal), Istvan Berbik (Hungary), Marie-H?l?ne Bouvier-Colle (France), G?rard Br?art (France), Peter Brocklehurst (UK), Vicen? Cararach (Spain), Anna Maria Marconi (Italy), Catherine Deneux-Tharaux (France), Risto Erkkola (Finland), Mathias Klein (Austria), Jens Langhoff-Roos (Denmark), Alison Macfarlane (UK), Walter Prendiville (Republic of Ireland), Jos van Roosmalen (Netherlands), Babill Stray-Pedersen (Norway), Carolyn Troeger (Switzerland), Clare Winter (UK), and Wei-Hong Zhang (Belgium). Also see web extra for a list of people who helped in each country. The following people helped run the trial in their own country: Austria: coordination?Sylvia Artner-Matuschek, Adolf Beck, Daniela Meger David; KH Ried?Penzinger Monika; LKH Bad Ischl?Carola Fuschlberger-Traxler; LKH Heinz Klagenfurt?Leipold; Hanusch KH?Daniela Meger David; AKH University?Hans Helmer, Katharina Klein; LKH Kufstein?Andrea Ehm. Belgium: Ambroise Par? (Mons)?Gilles Ceysens, Annick Nouls, Yaacoub Salame, Linda Van Lierde; Ath?Fran?oise Clerquin, Pierre Delvoye, Jean Piret; UZ Gent?Paul Defoort, Marleen Temmerman; AZ VUB (Brussels)?Maria Breugelmans, Lieve Devalckeneer, Monika Laubach; Baudour?Jo?l Annet, Renaud Paquay, Val?rie Vandenbosch; Brugmann (Brussels)?Thomas Pezin, Alain Vokaer; Erasme (Brussels)?Christine Kirkpatrick, Anne Maas; Hopital St Pierre (Brussels)?Patricia Barlow, Julie Belhomme, Nordine Ben Ali, Daniel Murillo; Ieper?Colette Berten, Geert Page; Ixelles (Brussels)?V?ronique Ziereisen; KUL (Leuven)?Bernadette Bijnens, Bernard Spitz, Joske Timmermans; St Jean (Brussels)?Xavier de Muylder, Christine Stoop; Ste Anne & St Remi (Brussels)?Paul Befahy, Anne Fostier; Tivoli?Maria Fabbricattore, Jacques Francotte, Sylvie Hollemaert; Tournai IMC?Viviane Gadenne; V?sale (Charleroi)?Patrick De Nayer, Didier Oberweiss. Denmark: coordination?Ane Rom, Birgitte Rode Diness; Gentofte Hospital?Anne Barfoed; Glostrup?Ambika Ravn; Hiller?d- Gitte Ulriksen; Slagelse?Karen Marie Wigh Felsen; Holb?k?Marianne Brandstrup Larsen; Frederiksberg?Ane Rom. Finland: Helsinki University Central Hospital?Vedran Stefanovic; Midwifery Institute of Helsinki?Veli-Matti Ulander; University central hospital of Turku?Risto Erkkola; University Hospital of Tampere-Jukka Uotila. France: Moulin Hospital?Michel Beytout, Catherine Damouret; University Hospital (Nancy)?Brigitte Guillemain; Antoine B?cl?re; University Hospital (Clamart)?Aur?lia Chauveaud; Tenon University Hospital (Paris)?Nadia Berkane, Marie-Christine Chaux; University Hospital (Rouen)?Loic Marpeau, Sabine Sionville; Villleneuve St Georges Hospital?Patricia Tran van. Hungary: coordination?Istv?n Szab?; Baranya County Hospital (P?cs)?J?zsef B?dis; City Hospital (Mosonmagyar?v?r)?Istv?n Barcza; Erzs?bet Hospital (Sopron)? K?roly P?ter Cs?csei; Petz Alad?r Teaching Hospital (Gy?r)?S?ndor Gard?; Selye J?nos Hospital (Kom?rom)?L?szl? Rokay; Szent Borb?la Hospital (Tatab?nya)?Mih?ly Moln?r; University Hospital (Sci. Univ. P?cs)?Istv?n Szab?, Tam?s Csermely; Vaszary Kolos Teaching Hospital (Esztergom)?Istv?n Berbik. Ireland: coordination?Reem Akkawi, Fidelma Cavanagh; Coombe Women?s Hospital (Dublin)?Suzanne Kelly; Our Lady of Lourdes (Drogheda)?Dalia Sikafi, Ann Keating; Cavan General Hospital?Iram Basit, Marie McCusker; Midland Regional Hospital Mullingar?Mary Corbet. Italy: Luigi Chiechi at start of project (Bari), then Valentina Mariotti (Milano). Az SS Antonio e Biagio e C Arrigo, Alexenderia?Enrico Rovetta; Osp di Bassano del Grappa(VI)?Yoram Meir; Osp Civile San Paolo, Civitavecchia?Antonio Castellano; Osp Civile S Liberatore, Atri (TE)?Claudio Angeloni; Osp San Massimo, Penne (PE)?Quirino Di Nisio; Pres Osp Di Piove Di Sacco (PD)?Antonino Oro. Netherlands: coordination?Marlies Rijnders, Esteriek de Miranda; St Lucas Hospital; Bronovo Hospital. Norway: coordination?Bente R?nnes, Rikshospitalet, Oslo University Hospital;Sykehuset Innlandet; Gjovik?Anne Kari Gjestvang, Elham Mahjoob; Sykehuset Innlandet Elverum?Agneta Stramrud. Portugal: Coordination?Maria Fatima Oliveira, Cristina Ferreirinha; Maternidade Bissaya Barreto (Coimbra)?Ascen??o Ba?a ; Maternidade Daniel de Matos (Coimbra)?Jos? Portugal; H. S. Marcos (Braga)?Luc?lia Guerra Sousa: H. S. Joao (Porto)?Cristina Ferreirinha; Senhora da Oliveira (Guimaraes)?Alice Santos. Spain: coordination?Sonia Pisa, Sara Herrero; H. Cl?nic?Enrique Barrau, Jordi Bellart, Isabel Salgado; H. Vall d?Hebr??Anna Suy; H. Sabadell?Jordi Costa, Maria Grimau; H. Joan XXIII?Ram?n M? Miralles; H. del Mar?Antoni Pay?; H. San Joan de Deu?Sergi Cabr?; H. Sant Pau?Marta Sim?; H. Germans Trias?Jos? Lecumberri. Switzerland: coordination ?Irene H?sli, Gideon Sartorius; Aarau-Monya Todesco; Basel?Gideon Sartorius; Frauenfeld?Verena Geissb?hler; Fribourg?David Stucki, Heidrun Sch?nberger; Solothurn?Suzanne Zakher; St Gallen?Gero Drack, Anika Hey-Moonen.

Contributors: W-HZ designed data collection tools, monitored data collection for the whole trial, wrote the statistical analysis plan, cleaned and analysed the data, and drafted and revised the paper. She is guarantor. CD-T implemented the trial in France, analysed the data, and drafted and revised the paper. PB analysed the data and drafted and revised the paper. EJ wrote the statistical analysis plan, monitored data collection for the whole trial, and revised the draft paper. MJ designed data collection tools,, monitored data collection for the whole trial, and revised the draft paper. SA initiated the collaborative project, designed data collection tools, implemented the trial for the all countries, monitored data collection for the whole trial, analysed the data, and drafted and revised the paper. All members of EUPHRATES designed the trial. Diogo Ayres-de-Campos, Istvan Berbik, Marie-H?l?ne Bouvier-Colle, Vicen? Cararach, Risto Erkkola, Mathias Klein, Walter Prendiville, Jos van Roosmalen, Babill Stray-Pedersen, and Carolyn Troeger implemented the trial in, respectively, Portugal, Hungary, France, Spain, Finland, Austria, Republic of Ireland, Netherlands, Norway, and Switzerland, and revised the draft paper. G?rard Br?art analysed the data and revised the draft paper. Alison Macfarlane and Clare Winter revised the draft paper.

Funding: W-HZ received a grant from Fondation Philippe Wiener-Maurice Anspach for the final analysis of the data. The project was funded by the European Union under framework 5 (contract QLG4-CT-2001-01352). The EU had no role in the design, management, data collection, analyses, or interpretation of the data or in the writing of the manuscript or the decision to submit for publication.

Competing interests: None declared.

Ethical approval: Ethical approval was obtained in each country from relevant local or national research ethics committees. Consent to participate was obtained from the maternity units. Because the procedure being tested was not invasive or different from current clinical practice, and because outcome data were routinely collected at maternity units and anonymously transmitted, no individual consent was sought.

Data sharing: No additional data are available.

*Baseline data were unavailable in one maternity unit in intervention group.

*Baseline data unavailable in one maternity unit.

?One of following: maternal death, transfusion, plasma expansion, surgery or embolisation, admission to intensive care unit, or treatment with recombinant factor VII.

ICC=intracluster correlation coefficient.

*Defined by one of following: maternal death, transfusion, plasma expansion, surgery or embolisation, admission to intensive care unit, or treatment with recombinant factor VII.

?Adjusted for clustering (maternity unit).

?Adjusted for clustering (maternity unit), age of mother, prophylactic uterotonics used in third stage, mode of delivery, and birth weight.