Roberts 1995a.

Methods	Quasi‐randomised trial (breasts rather than women were the unit of analysis).
Participants	34 lactating women located on postnatal wards in 2 Australian hospitals. Inclusion criteria: non‐Aboriginal, lactating, suffering from breast engorgement (hard, warm, painful breasts, with difficulty feeding), according to the professional judgement of the midwives caring for them.
Interventions	Group 1 (even hospital registration numbers): chilled gel pack on the right breast and chilled cabbage leaves on the left breast. Group 2 (odd hospital registration numbers): opposite to above. Leaves from common green cabbages were prepared by stripping out the large vein, cutting a hole for the nipple, rinsing and chilling. Breast‐shaped gel pack in small, medium, and large sizes were designed by the researcher to fit under the brassiere, covering the breast except for the nipple area. Treatment left on breasts for up to 8 hours, with mothers renewing the cabbage leaves and gel packs ad lib, usually every 2 to 4 hours.
Outcomes	Pre and post‐test pain rating for each breast rated on a "pain ruler" (a visual analogue scale with numbers from 0‐10, labelled with descriptions 0 = no pain, 5 = moderate pain, and 10 = excruciating pain). Descriptive data about engorgement were also collected.
Notes	Analysis was at the breast level and results were at high risk of bias and difficult to interpret. We have not been able to included data in the data tables.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	High risk	Quasi‐randomisation (by hospital number).
Allocation concealment (selection bias)	Unclear risk	The authors did not mention allocation concealment.
Blinding (performance bias and detection bias) Women	High risk	The nature of the study did not allow blinding of women.
Blinding (performance bias and detection bias) Clinical staff	High risk	The nature of the study did not allow blinding of clinicians.
Blinding of outcome assessment (detection bias)	High risk	Participants rated outcomes in a self‐administered questionnaire. Given that visibly different interventions were placed on each breast blinding of participants/outcome assessors was not possible.
Incomplete outcome data (attrition bias) All outcomes	Low risk	The authors reported on all outcomes.
Selective reporting (reporting bias)	Low risk	No other identified bias.
Other bias	High risk	The data were analysed at the breast level with no adjustment for the non‐independence of breasts. As mentioned in McLachlan 1991, when the visual analogue scale was used, it was not always easy for women to make a clear distinction between the left and right breast.