Belizan 1991.

Methods	Multicentre trial. Numbered, sealed opaque envelopes, containing randomisation codes. Of 593 (calcium) vs 601 (placebo) enrolled, 14 vs 13 were lost before starting treatment and excluded from analysis; 577 vs 588 had at least partial follow‐up. Follow‐up was incomplete for 52 vs 46, but delivery data were available in 17 vs 12 of these, giving delivery data for 544 vs 554.
Participants	Nulliparous women, < 20 weeks' pregnant; BP < 140/90 mmHg (mean of 5 measurements); no present or past disease; not taking medication; normal oral glucose tolerance tests. Recruitment between 1987 and 1989.
Interventions	2 g calcium as 500 mg calcium carbonate tablets, vs identical looking placebo tablets. Compliance was 84% (calcium) and 86% (placebo).
Outcomes	Gestational hypertension (DBP 90 or more; SBP 140 or more mmHg, on 2 occasions 6 hours apart); pre‐eclampsia (gestational hypertension + proteinuria > 0.3 g/L on 2 random urine samples 6 hours apart); BP measured with random‐zero sphygmomanometers, Korotkoff sound 5. Perinatal death. Follow‐up: BP > 95th percentile for sex, age and height for children 5‐9 years.
Notes	3 hospitals in Rosario, Argentina. Data for preterm birth given as percentages, not clear what the denominators were. Assumed to be the numbers with complete follow‐up (527 vs 542) as these were the numbers which were divisible by the percentages to give whole numbers. Unpublished placental abruption data obtained from authors. Babies born in the private hospitals followed up at 7 years. Of 614 randomised (calcium 309/placebo 305), 301/299 completed the first study, 2/6 infant deaths and 1/0 maternal deaths had occurred, leaving 298/293 eligible for follow‐up. 289/285 were contacted, 10/5 refused to participate, 22/19 lived outside the country, and 257/261 were assessed (88% of those eligible). Funded by a research grant from The International Development Research Centre, Canada. CoI: not stated.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Computer‐generated random number sequence ‐ Epistats Statistical Package
Allocation concealment (selection bias)	Low risk	Complete set of numbered sealed opaque envelopes was sent to each of 3 hospitals.
Blinding of participants and personnel (performance bias) All outcomes	Low risk	Randomisation code was held centrally such that the woman and her healthcare providers were blind to her trial group. Tablets were identical in appearance, weight, colour, taste.
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Randomisation code was held centrally such that the woman and her healthcare providers were blind to her trial group. Tablets were identical in appearance, weight, colour, taste.
Incomplete outcome data (attrition bias) All outcomes	Low risk	All or partial data available for 579/593 (Ca) and 588/601 (Pl) respectively. Delivery data were available for 544 and 554 respectively.
Selective reporting (reporting bias)	Low risk	All primary outcomes addressed
Other bias	Low risk	Balanced group sizes, baseline characteristics including dietary calcium similar in both groups