The optimal duration of progesterone supplementation in pregnant women after IVF/ICSI: a meta-analysis

The inclusion criteria for eligible studies were defined a priori during the design phase of this systematic review. Randomised controlled trials investigating the duration of P supplementation for luteal phase support in IVF/ICSI cycles were included. Trials using donor oocyte cycles or frozen transfers were excluded. No limitations were placed on language, date, or publication status.

Women undergoing IVF/ICSI who were evaluated for the effects of P supplementation duration on pregnancy outcomes were included.

The interventions evaluated were early P cessation versus P continuation during the first trimester in pregnant women after IVF/ICSI.

The primary outcome chosen for the meta-analysis was live birth rate (LBR, i.e.,a baby born alive after 24 weeks gestation). Secondary outcomes included ongoing pregnancy rate (OPR, pregnancy beyond 12 weeks of gestation, as confirmed by foetal heart activity on an ultrasound), and miscarriage rate (MR, the failure to achieve live birth after a positive β-hCG test).

We performed an exhaustive electronic search in the following databases: MEDLINE (1946 to July 2012), EMBASE (1974 to July 2012), the Cochrane Central Register of Controlled Trials (CENTRAL), the Chinese biomedicine (CBM) literature database (1978 to July 2012), and the Wanfang database (1998 to July 2012). The search combined terms and descriptors related to IVF, ICSI, luteal phase support, and progesterone. To fit with the syntax used in each consulted database, the search strategy was modified with a series of terms suggestive of RCTs as set out by the Cochrane Handbook for Systematic Review of Intervention [32](Additional file 1). No limit was placed on language. We also carefully browsed the references of relevant publications and added the related publications to the search. When questions related to the design or outcomes of the trials arose, we contacted the corresponding authors to confirm the information we extracted from their trials or to clarify any ambiguities.

The risk of bias in the included studies was assessed independently by two reviewers (Xi-Ru Liu and Hua-Qiao Mu) according to the guidelines recommended in the Cochrane Handbook for Systematic Review of Intervention [32]. For each study, we assessed the risk of bias related to sequence generation, allocation, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting and other sources of bias. A judgment of ‘Yes’ meant a low risk of bias, a judgment of ‘No’ meant a high risk of bias, and ‘Unclear’ indicated an unclear risk of bias. Disagreements were discussed and resolved by consensus with a third reviewer (Qi Shi).

Data extraction was performed independently by two reviewers (Xi-Ru Liu and Hua-Qiao Mu). Discrepancies were resolved by discussion with a third reviewer (Qi Shi).We extracted the following information from each eligible study: first author, year of publication, country of origin, sample size, and a number of patient characteristics, including the IVF protocol used, the exact dose of P, the route of administration, the timing of initiation and duration of luteal phase support with P, and IVF/ICSI outcomes.

Raw data were extracted from the eligible studies for each defined outcome and pooled using Review Manager 5.1 software. Dichotomous results from each study were expressed as relative risk (RR) with 95% confidence intervals (CI). These results were combined for the meta-analysis using a fixed-effects model in the absence of statistically significant heterogeneity or a random-effects model in the presence of statistically significant heterogeneity. The inter-study heterogeneity was evaluated using the x² (Cochran's Q) statistic and the I² value. Sensitivity analyses were performed for those studies that answered the research question of interest but used a quasi-randomised approach for patient allocation. Subgroup analyses were planned a priori based on: (1) the timing of randomisation, (2) the timing of initiation of P, (3) the GnRH analogue used for LH surge inhibition, and (4) the type and dose of P administration.

A total of 1,185 trials were retrieved in the initial electronic search, 351 of which were duplicate records that were subsequently removed. An additional 821 were excluded upon title/abstract screening. The 13 remaining trials were selected for further full-text analysis. Seven of these trials were excluded. Two were retrospective cohort studies [10, 22], and one failed to implement randomisation [33]; one was the same study reported as an abstract at an earlier meeting [34]; one trial did not explicitly describe the sequence generation or allocation concealment [35]; and two trials did not meet other inclusion criteria [18, 36]. The remaining six RCTs, totalling 1,201 participants, were included in this meta-analysis. Detailed search procedures are summarised in the flow diagram (Figure 1).

Four of the six trials provided an adequate randomisation model [11, 13, 14, 23], and four an adequate mode for allocation concealment [12‐14, 23], all of which made use of sealed and opaque envelopes. One trial [17] used odd and even patient birth years for allocation and was classified as a quasi-randomised trial. Only one of the studies was a dual centre study [11]; the other five were unicentric. None of the studies blinded their personnel, participants or outcome assessors or at least did not mention blinding. Owing to the small number of included studies, it was impossible to conduct a meaningful assessment of publication bias using a funnel plot. See the risk of bias graph (Figure 2) and risk of bias summary (Figure 3) for an overview.

The six selected studies [11‐14, 17, 23] were performed in Spain, Belgium, Minnesota (USA), Egypt, Denmark, and Germany, respectively, and involved 1,201 participants who were originally studied between 1989 and 2010. In two studies, patients with a clinical pregnancy (at 5–7 weeks of gestation) were included [13, 14], and in three studies, patients with a positive β-hCG test (on the 11-16^th day post-embryo transfer (ET)) were included [11, 17, 23]. In the final study, patients were enrolled at the beginning of an IVF cycle [12]. The type and dose of P supplementation, the timing of administration and the duration of P supplementation varied among the studies. In addition, controlled ovarian hyperstimulation (COH) protocols and the basal clinical characteristics of the patients differed between studies. These data are presented in Table 1.

Two eligible studies presented data on live birth rates [11, 12]. In the study reported by Goudge et al. [12], the number of patients recruited at the beginning of an IVF cycle was converted to the number of patients with a positive β-hCG test according to the reported biochemical pregnancy rate. There were 143 events in the early P cessation group (in which P was stopped on the 11^th or 14^th day post-ET) and 150 in the P continuation group (in which P was continued until the 6^th or 7^th week of gestation). There were a total of 293 patients who gave birth to live babies out of 369 participants. The probability of live birth did not differ between the early P cessation group (77.3%, 143/185) and the P continuation group (81.5%, 150/184) (P = 0.33; RR: 0.95, 95% CI: 0.86–1.05). There was no statistical heterogeneity in this comparison (χ² = 0.05, df = 1, P = 0.82; I² = 0%) (Figure 4).

MR data were available from six studies, with 136 events out of 1166 participants; after data conversion, this figure corresponded to 69/585 in the early P cessation group and 67/581 in the P continuation group [11‐14, 17, 23]. No statistical heterogeneity was observed between the studies (χ² = 2.96, df = 5, P = 0.71; I² = 0%). There were no significant differences in the number of miscarriages between patients who received early P cessation and those who received P continuation (P = 0.96; RR: 1.01, 95% CI: 0.74–1.38) (Figure 5).

OPR data were available from six studies, with 1017 events among 1166 participants (503/585 in the early P cessation group and 514/581 in the P continuation group) [11‐14, 17, 23]. A meta-analysis of all six trials yielded an RR of 0.97 (P = 0.49; 95% CI: 0.90–1.05), indicating that there was no statistically significant difference between the early P cessation and P continuation groups (Figure 6).

Because the OPR was heterogeneous (χ² = 18.75, df = 5, P = 0.002; I² = 73%) among the included studies, a random-effects model was used. No differences were observed between the results obtained using a fixed-effects model (P = 0.28; RR: 0.98; 95% CI: 0.94–1.02) and those obtained from the random-effects model (P = 0.49; RR: 0.97; 95% CI: 0.90–1.05), although the weight of each study was altered. Next, according to the timing of randomisation, we performed a subgroup analysis and separately pooled four studies [11, 12, 17, 23] in which P was withdrawn on the day of a positive β-hCG test and two studies [13, 14] in which P was withdrawn on the day that clinical pregnancy was confirmed (5^th–7^th weeks of gestation). This stratified analysis revealed no significant differences between the groups in which P was stopped on the day of a positive β-hCG test (P = 0.29; RR: 0.91; 95% CI: 0.76–1.09) or on the day that clinical pregnancy was verified (P = 0.55; RR: 1.01; 95% CI: 0.97–1.06). Heterogeneity was detected in the subgroup of studies that randomised patients on the day of a positive β-hCG test (χ² = 15.15, df = 3, P = 0.002; I² = 80%). The study reported by Prietl [17] might be the source of heterogeneity, as it used a different luteal phase support protocol (17α-hydroxyprogesterone caproate (100mg) and oestradiol valerate (10mg) twice a week) and exhibited a high risk of bias based on sequence generation and patient allocation methods. In a sensitivity analysis, we recalculated the combined results while excluding this study. However, the results before (P = 0.49; RR: 0.97; 95% CI: 0.90–1.05) and after the sensitivity analysis (P = 0.74; RR: 1.01; 95% CI: 0.96–1.06) were not significantly different.