Ngoc 2004.

Methods	Randomised by opening sequentially numbered envelope ‐ prepared by computer‐generated code in blocks of 10. Recruitment took place at Hung Vuong Hospital in Ho Chi Minh City, Vietnam, from January through August 2003.
Participants	200 women in Ho Chi Minh City, Vietnam, with non‐viable first trimester pregnancies (anembryonic or early fetal death) diagnosed by ultrasound; cervix closed.
Interventions	Oral misoprostol 800 mcg (n = 100) vs vaginal misoprostol 800 mcg (n = 98). Women reviewed after 48 hours; if retained products present, they were given option of surgical evacuation or further review after another 5 days (when evacuation was performed if there were still products present).
Outcomes	Primary: complete miscarriage without need for surgical evacuation. Secondary: adverse effects.
Funding	Funding by David and Lucile Packard Foundation
Declarations of interest	No information on conflicts of interest.
Notes	2 women lost to follow‐up.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Comment: the randomisation scheme was created by Population Council staff, using a computer‐generated code in blocks of 10.
Allocation concealment (selection bias)	Low risk	Quote: "The study investigator opened the next sequentially numbered randomized envelope to determine the treatment arm".
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	Quote: "Two women in the vaginal group and one in the oral group were lost to follow‐up. One woman in the vaginal group was later reached by telephone". Comment: table 2 shows side effects for 190 patients (not 200 patients), so there are some missing data. This was < 10% of total study population.
Selective reporting (reporting bias)	High risk	Comment: in table 2 side effects were presented for 95 patients per treatment arm, which is a sign of missing data. Analyses for these side effects were measured as an percentage of 95 women instead of 100 women. This influences the outcomes.
Other bias	Low risk	No other source of bias could be detected
Blinding of participants and personnel (performance bias) All outcomes	High risk	Quote: "Neither the investigator nor the woman was blinded to the treatment assignment". Comment: due to the nature of the interventions (oral vs vaginal medication) blinding would have been difficult. Nonetheless this might have influenced (perception of) outcome.
Blinding of outcome assessment (detection bias) All outcomes	High risk	Comment: no blinding of outcome assessment.