Materials and Methods
This systematic review and NMA was performed according to the PRISMA statement10. This review was registered on the PROSPERO database (CRD42020204486). The research question was “Which interventions for preventing preterm birth improve the gestational age at birth in women with twin gestation regarding cervical length?”.
We performed a Systematic Review of randomized clinical trials searching for studies on PubMed, MEDLINE, Cochrane Library, EMBASE, Web of Science, BVS, Scopus, and grey literature in May 2022. The search strategy used a combination of specific terms using Boolean connectors for each database with no exclusion criteria regarding period or language. The basic search strategy was composed by the following: “twin” OR “multiple gestation” OR “multiple pregnancy” AND “progesterone” OR “cervical cerclage” OR “pessary” OR “hydroxyprogesterone”. (Complete search strategy is available in Appendix 1).
We included randomized controlled trials comparing intervention treatment (progesterone, cerclage or pessary) with a control group or another intervention to prevent spontaneous preterm delivery in women with a twin pregnancy and a short cervix. Short cervical length was defined as a cervical length lower than 40mm11.
The primary outcome was spontaneous preterm birth (sPTB) <34 weeks. Secondary outcome was PTB<34 weeks from any cause.
We included studies reporting on natural or 17-alpha hydroxyprogesterone caproate with any administration route (oral, rectal, vaginal and intramuscular), cervical cerclage (McDonald or Shirodkar) and cervical pessary. We also included studies in which women received a combination of these interventions. The comparison group could have received a different treatment strategy, placebo or standard treatment. We excluded studies if they reported on women with preterm labor or previously threatened preterm labor, interventions for preterm premature rupture of membranes and treatment for women with cervical dilatation. No language restrictions were considered. If needed we contacted the authors to obtain additional data not presented in the papers.
Studies´ references were imported to the Endnote web reference manager, and duplicates were excluded. Two authors (TV and ABP) evaluated titles and abstracts from the selected papers. If there were conflicting decisions, a third author (RCP) makes the final decision. Finally, two authors (TV and ABP) independently accessed the full texts of potentially eligible articles, retrieved and reviewed studies for eligibility. A third author evaluated any conflicted decision. All excluded articles after full-text analysis were described according to the reason why they were excluded. The included articles were submitted to a quality assessment using RoB2 tool12 and we considered risk of bias if the study presented two or more concerns in the components. We assessed integrity using a screening checklist to assess data integrity of Randomized Clinical Trials13.
Descriptive statistical analyses were performed using Stata 17.0 and overall confidence in the results of NMA were assessed using the CINeMA tool14. We performed random-effect NMA to synthesize evidence from the entire network by integrating direct and indirect estimates for each intervention into a single summary effect, using the statistical package ‘network’ in Stata, version 17.0. League tables with summary relative effect sizes (Odds Ratios) for each possible pair of interventions were presented. We used the surface under the cumulative ranking curves (SUCRA) to rank all interventions. We assessed global inconsistency by using a design-by-treatment interaction model and local inconsistency through the side-splitting approach.