Trial design and study setting

This was a double-blind, placebo-controlled randomized trial that was conducted in a single center at the Multiple Pregnancy Unit of the Obstetrics Department at São Paulo University School of Medicine (São Paulo, Brazil). The study protocol was approved by the University Ethical Committee (0418/04) and was registered at ClinicalTrial.gov ( NCT01031017 ).

Participants

Pregnant women with twin pregnancies were assigned randomly according to the inclusion criteria between June 1, 2007, and Oct. 31, 2013. The following inclusion criteria were used: naturally conceived diamniotic twin pregnancies; no history of preterm delivery (<37 weeks’ gestation); gestational age of 18–21 weeks and 6 days at random assignment; absence of major fetal abnormalities (such as neural tube defects, abdominal wall defects, cardiac defects, hydrocephalus, and malformations that are associated with polyhydramnios) at the anomaly scan; no allergies to progesterone or peanuts (peanut oil is an excipient that is used in vaginal ovules); and the absence of hepatic dysfunction, porphyria, otosclerosis, malignant disease, severe depressive state, current or previous thromboembolic disease, uterine malformation, and prophylactic cerclage.

The exclusion criteria were a subsequent diagnosis of major fetal abnormalities, the presence of ovular infection, or being lost to follow up.

Intervention and random assignment

Patients who fulfilled the inclusion criteria were given oral and written information about the study. Only women who provided signed consent were included in the study. The patients were assigned randomly to 2 groups (progesterone and placebo) at a 1:1 ratio. Random assignment was performed with a computer-generated system with balanced blocks of 20 patients in each block ( www.randomization.com ). The hospital’s pharmacy department was responsible for packing and labeling the ovules (A and B); the random assignment code was kept secret until data analysis. Therefore, the patients, researchers, and clinicians who were involved in the clinical and ultrasonographic evaluations were blinded to the treatment assignment for the duration of the study.

The progesterone (200 mg of natural micronized progesterone that also contained excipients such as peanut oil, soybean lecithin, gelatin, glycerol, and titanium dioxide: Utrogestan; Besins Manufacturing Belgium, Drogenbos, Belgium) and placebo ovules (that contained excipients such as peanut oil, soybean lecithin, gelatin, glycerol, and titanium dioxide) had the same physical characteristics, and both were manipulated and provided by the supplier (Laboratory of Progesterone; Besins Healthcare, Brussels, Belgium).

Procedures

On the day of random assignment, each patient received 1 pack that contained 21 ovules and was instructed to use the vaginal ovules every night until 34 weeks and 6 days’ gestation. To monitor treatment compliance, each patient was asked to return the empty pack every 3 weeks when she received antenatal care and ultrasound examination, after which she received a new full pack.

At the first antenatal care appointment, all patients with twin pregnancies were examined according to our routine antenatal care protocol for multiple pregnancies. Additionally, evaluations for bacterial vaginosis and fungus and endocervical investigations for Mycoplasma sp, Ureaplasma sp, Chlamydia sp, and Neisseria gonorrhoeae were performed. The presence of Group B Streptococcus was investigated during the third trimester. In the cases of positive culture, treatment was provided accordingly.

Patients also underwent a detailed ultrasound examination for pregnancy dating, chorionicity determination, anomaly examination, and transvaginal measurement of cervical length. Gestational age was calculated from the first day of the last menstrual period (LMP) and was confirmed with the use of ultrasound crown-rump length measurement during the first trimester or estimated based on multiple ultrasound parameters (biparietal, head circumference, abdominal circumference, and femur length) of the largest fetus during the second trimester. When the LMP was uncertain or unknown or there was a discrepancy between the LMP-based gestational age and the ultrasound-determined dates, the gestational age was based on the earliest ultrasound findings.

Chorionicity was determined prenatally based on ultrasound examination (“lambda sign,” the presence of 2 separate placentas, or discordant fetal gender) or placental histologic examination after birth.

After random assignment, appointments for clinical and ultrasound examinations were scheduled every 3 weeks at our Multiple Pregnancy Unit. The need for admission or the management of pregnancy complications (preterm labor, corticosteroids, tocolytics, and antibiotics) was based on our clinic protocol, and all of the patients received standard care.

The patients’ prenatal demographics (age, body mass index, personal history, family history, and obstetric history), current pregnancy complications, and postnatal follow-up data were obtained from the patients’ and neonates’ hospital records. Delivery and neonatal details from cases that were not delivered at our hospital were obtained by telephone contact by one of the researchers who were involved in the study. The infants’ postnatal data were obtained until hospital discharge.

Outcomes and statistical analysis

The primary outcome was the difference in mean gestational age at delivery. The secondary outcomes included the rate of spontaneous delivery at <34 weeks’ gestation and the postnatal data until hospital discharge: birthweight, Apgar score <7 at 5 minutes, hypoglycemia, intraventricular hemorrhage grade ≥3, jaundice, necrotizing enterocolitis, patent ductus arteriosus, retinopathy, septicemia, admission to the neonatal intensive care unit, respiratory distress syndrome, the need for mechanical ventilation, death before hospital discharge, and composite neonatal outcome (defined as the occurrence of any of the following events: intraventricular hemorrhage, necrotizing enterocolitis, respiratory distress syndrome, sepsis, and death before hospital discharge). Subgroup analyses were performed for treatment compliance <80% and ≥80%, for cervical length ≤25 mm at random assignment or at <24 weeks’ gestation, and for chorionicity.

The estimated sample size was calculated considering that, in our Department, mean gestation at delivery in twin pregnancies is 36.1 weeks (standard deviation, 3.4) and approximately 15% of the cases deliver at <34 weeks’ gestation.

At the time we designed the study, none of the randomized studies that used progesterone in twins had been published. Therefore, we based our study on the report of singleton pregnancies carried out in our Department that showed 50% reduction in spontaneous preterm birth rate after the use of vaginal progesterone and a prolongation of pregnancy of 1.5 weeks (33.5 ± 2.4 vs 32.0 ± 0.7 weeks).

Assuming that progesterone in twin pregnancies would postpone mean gestational age by 1 week compared with the placebo group, the sample size needed to demonstrate a significant difference (significance level, 5%; power, 80%, mean difference, 1; standard deviation, 3.4) was 366 cases (183 cases in each study group) with complete follow-up data.

Outcomes were analyzed according to the intention-to-treat principle. Unknown or missing values were excluded from the statistical analysis. Numeric data are expressed as the mean values and standard deviation, and categoric data are expressed as frequencies. Comparisons of baseline characteristics between treatment groups where the calculations were performed with the use of the chi-square test or Fisher exact test for categoric variables and the nonparametric Mann-Whitney U -test for numeric data. For treatment comparisons in relation to delivery and maternal outcome for binary outcomes, the odds ratio (OR) with corresponding 95% confidence intervals (CIs) were determined.

Cox proportional hazards regression model was used to examine whether the treatment (progesterone or placebo) influences the time to delivery. The same model was used for comparisons between treatment groups in relation to the time to delivery in the following subgroups: with cervical length ≤25 mm at random assignment or at <24 weeks’ gestation, analysis according to chorionicity, and according to progesterone compliance of <80% and ≥80%. The proportional hazards assumption was examined, and no violation of this assumption was found. Residuals analysis was performed with Cox Snell and deviance residuals. No observation was an influential point, and the fitted model seemed to be appropriate. For comparison between the treatment groups, the relative risks with corresponding 95% CIs were calculated.

Neonatal outcomes were analyzed by taking into account nonindependence of newborn infants from twin gestation (cluster). Therefore, we used generalized estimating equation with binomial distribution for response variable and logit link function because all neonatal variables were considered binary. The correlation structure of the response variable within a cluster that appeared to best fit was the unstructured working correlation matrix. The quasi-likelihood under independence criterion was used to select the correlation structure, and the unstructured working correlation matrix was the one with the lowest value of quasi-likelihood under independence criterion. The groups were compared by ORs with point estimate and 95% CIs that were obtained by the adjusted model, taking into account nonindependence of newborn infants from twin gestation. The same model was applied for comparison between treatment groups in relation to neonatal variables in all other subgroup analysis.

A probability value of < .05 was considered statistically significant. The data were analyzed using SPSS software (version 17; SPSS Inc, Chicago, IL) and using R software (The R Project for Statistical Computing).