Bullough 1989.

Methods	Cluster‐randomised controlled trial with 2 arms, Malawi.
Participants	TBAs were randomised in the trial and not pregnant women. 23 TBAs (2184 deliveries) were allocated to the early suckling group and 26 TBAs (2201 deliveries) to the control group. The main analysis was carried out on 4227 singleton vaginal deliveries (2123 control and 2104 early suckling).
Interventions	Arm 1: early suckling (breastfeeding immediately after birth) Arm 2: standard care of pregnant women. In Malawi culture, standard care involves relatives looking after the newborn immediately, allowing the mother to rest. All TBAs were given training before the trial but none were made aware of the trial. Only TBAs in the intervention arm were educated about the importance of early suckling and if initiating this practice with women in their care.
Outcomes	The frequency of PPH defined by authors as blood loss > 500 mL. The mean blood loss. The frequency of retained placenta.
Notes	The unit of randomisation in this trial was the provider rather than pregnant women; this fact will produce a clustering‐effect. No adjustments were made to the data to account for possible similarities among women receiving care from the same birth attendant, either in the original trial report or in this review's analysis. Therefore, it is reasonable to assume that the confidence intervals associated with treatment effects have been underestimated. Women with vaginal tears were not excluded, but twin pregnancies were. Neonatal deaths were not excluded from the main analysis, but stillbirths were. We have added stillbirths and neonatal deaths from Table 2 to create the perinatal deaths outcome. For PPH, twin births and stillbirths are analysed separately in the trial report; we have not included these data in our review's analysis.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"Randomisation of TBAs to suckling or control groups was carried out by member of the Malawi Ministry of Health within districts and after stratification for the number of deliveries attended (more or less than 30 per month) and for distance from a telephone or health centre (less than 5 km, or 5 km and further)." Random sequence generation not described.
Allocation concealment (selection bias)	Low risk	Randomisation was done centrally, as above. All women attending a single TBA would have had the same intervention (either early suckling or standard care).
Blinding of participants and personnel (performance bias) All outcomes	Low risk	Participants: "Randomisation into early suckling and control groups was by TBAs rather than by women to reduce likelihood of contamination of the treatment group by knowledge of the intervention reaching women awaiting delivery”. Personnel: “The TBAs were not informed that they were taking part in the trials”. Comment: the trial participants and the TBAs were not aware that they were involved in a randomised study.
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Quotes: “The TBAs were not informed that they were taking part in the trial”. “TBAs were taught that the reason for this practice (early suckling) was to keep baby warm or promotes bonding.” “a record collection form was designed with drawings to represent the information required” (record keeping – quality control). Comments: TBA as outcome assessors of end points were not aware of the trial; however, because most TBAs had minimal education, trialists used methods to ensure quality control.
Incomplete outcome data (attrition bias) All outcomes	High risk	10 out of 33 (30.3%) of the TBAs in the suckling group and 9 out of 35 (25.7%) in the control group were excluded from the analysis due to problems with birth attendant training and with quality control (for example, some TBAs measured amniotic fluid along with blood for the 'blood loss' outcome). Of 4385 deliveries from the remaining TBAs, blood loss measurements were missing for 114 (59 in control group and 55 in suckling group). There are discrepancies in the numbers for frequency of PPH between the table on p. 524 and the paragraph on p. 524. We have used the larger numbers from the paragraph. The relative risk is similar with both sets of numbers.
Selective reporting (reporting bias)	Low risk	All outcomes are reported.
Other bias	High risk	Observer bias: TBAs had little formal education. Trialists excluded data from several TBAs due to concerns about the accuracy of their data collection. TBAs were only partly successful at achieving early suckling, with suckling before the delivery of the placenta in just 1464/2104 women. Analysis bias: methods used in the trial analysis were not appropriate for cluster‐randomised trial design.