This is a secondary analysis of the data from the WHO CHAMPION trial. The primary aim of this analysis was to describe the association between the duration of TSL and postpartum blood loss among women with AMTSL with vaginal delivery, without interventions and with duration of TSL up to 60 minutes, particularly to see whether there is a critical duration above which blood loss increases. The secondary aim of the analysis was to define the association described above separately for women receiving Oxytocin and for women receiving HS carbetocin.
For the purpose of this analysis, the duration of the TSL was defined as the time interval from the birth of the baby until the delivery of the placenta.
Analysis Population: To assess the association between the duration of third stage of labour and postpartum blood loss, a subcohort of the CHAMPION modified ITT population (29,539 women with vaginal delivery and confirmed consent) was selected by excluding women with missing blood loss (69), with missing TSL duration (5) or TSL duration more than 60 minutes (106) and women with interventions (19,319) (Fig. 1). Thus, the subcohort consisted of 10,040 women.
The rationale for excluding women with interventions was as follows: TSL duration may be influenced by interventions to treat excessive bleeding. For example, excessive bleeding in the context of a retained placenta will induce treatment with uterotonics and probably a manual removal of placenta, which will determine the actual TSL duration according to the definition. To prevent this bias, the subcohort analyzed excluded women that had received any of the following interventions: episiotomy or perineal tear requiring suturing, use of additional uterotonics, manual removal of placenta, suturing cervical/ high vaginal tear, bimanual uterine compression, intrauterine balloon tamponade, exploration of the uterine cavity under anaesthesia, uterine or hypogastric artery ligation, uterine compression sutures, hysterectomy, manual/ surgical correction of uterine inversion, admission to Intensive Care Unit (ICU), or blood transfusion.
The rationale for excluding women with TSL duration > 60 minutes is explained under Statistical methods.
The uterotonic received had been assigned at random in the CHAMPION trial within centers, to Oxytocin or HS carbetocin. As both the duration of TSL and the delivery with or without intervention were variables collected after the randomization took place, the relationship between blood loss and these factors was studied to explore the association and not to establish cause-effect relationships.
A generalized linear model was used to fit blood loss as a function of duration of TSL in minutes, center and arm, with the normal distribution and the log link function. The logarithmic transformation for blood loss was used based on the results from Carvalho et al 2018  showing that postpartum blood loss has a lognormal distribution. For TSL duration, the logarithmic transformation was used because the distribution of duration was very asymmetric. In a study of outliers including all data in the whole range of TSL duration, TSL duration values above 60 minutes (106 women, 0.36% of the CHAMPION modified intention-to-treat population) were found to have high leverage  and thus justified their exclusion from the analysis. As a first step, a model was fitted including the three two-factor interactions and the three-factor interaction, for the factors center, arm and TSL duration. Then, except for the interaction arm by TSL duration, only the interactions that were significant at the 5% level were kept in the final models. The interaction arm by TSL duration was kept in the model to obtain predicted values separately for the two drugs. Analysis of residuals was used to assess the quality of the fits.
Plots of the blood loss fitted values from the above models versus TSL duration were constructed, by arm, with 95% confidence bands obtained from the models by maximum likelihood. A test of interaction of TSL duration by arm was conducted.