Introduction

Normal luteal function is an essential component for pregnancy maintenance. In natural ovulatory cycles, the corpus luteum can produce adequate progesterone after ovulation until the placental function starts at seven weeks of gestation. Ovarian stimulation (OS) techniques, either with gonadotropin-releasing hormone (GnRH) agonist or antagonist protocols, often induce endocrine defects in the luteal phase with increasing evidence suggesting that the resulting luteal-phase dysfunction may lead to lower pregnancy rates in in vitro fertilization (IVF) and/or ICSI (intracytoplasmic sperm injection) cycles1,2. To counteract these effects, luteal-phase support (LPS) is a well-known intervention for almost all stimulated assisted reproductive technology (ART) cycles3. Progesterone is amongst the most commonly, exogenously supplemented compounds employed as support of the luteal phase; however, the route of progesterone administration remains controversial4. In addition to the route of progesterone supplementation, disparities across literature are also present, regarding LPS dosage, duration and its use as monotherapy or in the context of combinatorial treatment with compounds such as oestradiol, Dehydroepiandrosterone (DHEA), gonadotropin-releasing hormone agonist (GNRH-a) and/or human chorionic gonadotropin (hCG)2,4. A plethora of previous pairwise and network meta-analyses has been published in an effort to discern the optimal LPS protocol in fresh cycles5,6,7,8,9,10. However, significant modifiable limitations were recognised. Amongst the pairwise analyses, the one-to-one comparison of specific LPS protocols, dimmed the option of a holistic picture of LPS variability and efficacy to be provided. The homogenisation of LPS protocols under a single agent umbrella did not allow for the appreciation of combinatorial protocols whilst combination of patient populations undergoing both fresh and frozen embryo transfers introduced a significant degree of data bias. Lastly, the effect of LPS selection under different ovarian stimulation protocols had not been previously addressed despite the significant impact upon clinical outcomes11,12.

Given the significance of clinical implications of appropriate LPS selection upon pregnancy outcomes, the present network meta-analysis compared mono-and multi-compound LPS regimens for women undergoing fresh cycles of IVF or ICSI in respect to core aspects of IVF/ICSI success (live birth, clinical and biochemical pregnancy rate, miscarriage, multiple pregnancy and ovarian hyperstimulation events). Additionally, the optimal LPS protocol in both agonist and antagonist OS has been explored.

Methods

Search strategy and selection criteria. The present study was prospectively registered under the PROSPERO database CRD42022358986 and conducted according to the PRISMA-NMA checklist13. Twelve databases, namely Embase (OVID), MEDLINE (R) (OVID), GlobalHealth (Archive), GlobalHealth, Health and Psychosocial Instruments, Maternity & Infant Care Database (MIDIRS), APA PsycTests, ClinicalTrials.gov, CENTRAL, Web of Science, Scopus and HMIC Health Management Information Consortium and two prospective registers, MedRxiv, Research Square were searched from inception to August 1st 2023. Search strategy was as follows and adapted per requirements of each target database (luteal and (support or supplementation or addition) and (assisted reproduction or IVF or ICSI or in vitro fertilization) and fresh). mp. [mp = ti, ab, hw, tn, ot, dm, mf, dv, kf, fx, dq, cw, ta, te, bt, nm, ox, px, rx, an, ui, sy, ux, mx]. To ensure that all previous meta-synthesised evidence have been identified and assessed, a snowball approach has also been implemented, where the search to the databases described above was also conducted with a limit to include only meta-analyses (N = 102). The original studies included in the relevant meta-analysis manuscripts and were extracted and deduplicated (N = 169). Those were compared to the manuscripts identified through the classical search (Fig. 1). All study designs were included in the initial search but only Randomised Control Trials (RCTs) met abstract selection criteria. No language or geographical restrictions were applied.

Figure 1
figure 1

PRISMA Flow chart. PRISMA 2020 flow diagram for new systematic reviews which included searches of databases, registers and other sources.

Full size image

For both systematic review and network meta-analysis (NMA), RCTs comparing pharmacological treatments administered for luteal support, either as monotherapy or combinatorial therapy, against placebo or other active agents administered either as mono- or combinatorial therapy for women undergoing fresh IVF/ICSI cycles were included. Studies reporting outcomes from oocyte donation cycles, comparing dosage or timing of same compound, or including patients that had undergone Intrauterine insemination (IUI) or Gamete intrafallopian transfer (GIFT) and zygote intrafallopian transfer (ZIFT) and studies where the route or compound of LPS was not stated or ≥ 4 embryos transferred were excluded (Table S1). Non-blind, single and double-blind studies were included in the analysis. Two independent researchers (SLK, KS) independently selected the studies, reviewed the main reports and supplementary materials, extracted the relevant information from the included trials, and assessed the risk of bias. Any discrepancies were double-checked and resolved by discussion with other members of the review team (GG, NB, DM).

Data extraction

Events (%, N) of clinical pregnancy, live birth, biochemical pregnancy, miscarriage, multiple pregnancy and OHSS, as previously defined, and the total number of patients exposed per treatment were extracted. Patient demographics and treatment specific parameters were also collected to allow for NMA transitivity analysis and comprehensive exploration of employed treatments across studies. Crude demographic and clinical data were collected. Per study, the total percentage of fresh cycles, Day 3 ETs was calculated from the reported, individual study data (Figs. 2, 3, 4). Missing SD or IQR were calculated from p values, t values, and standard error (SE) to allow for data harmonisation. When mean and standard deviation values were recorded, Bland’s method was employed to calculate median and IQR (Wan et al., 2014). Additionally, treatment specific parameters, namely active compound (Progesterone, Estradiol, hCG, GNRH-a, DHEA), brand name, route of administration [O, IM, SC, PV, PR, Topical (Patch)] dose (Progesterone, Estradiol, DHEA and GnRH agonist in mg, hCG in IU, median day of treatment initiation and SD, median end of treatment (weeks) and SD, number of patients exposed to named compound (Table 1). Lastly, implantation and fertilisation rates (%) were extracted as reported per study, given the inclusion of ≥ 1 embryos per study and aggregate data analysed as descriptive statistics (Fig. S1).

Figure 2
figure 2

Population percentage and crude numbers exposed to each luteal support regimen and baseline demographic characteristics. Percentage and number of participants exposed to each luteal support protocol (A), Comparison of median participant age [95% CrI] (B) and median BMI [95% CrI] (C) per luteal support intervention. Reference group was considered to be VP. One way ANOVA analysis was employed as data values abided by gaussian distribution. Two decimal p values and asterisk annotation of significance where p-value < 0.05, it is flagged with one star (*), p-value < 0.01, 2 stars (**), p-value < 0.001, three stars (***). placebo (no exposure), SCP (Subcutaneous progesterone), VP (vaginal progesterone), IMP + VP (intramuscular progesterone and vaginal progesterone), VP + OE (vaginal progesterone and oral estradiol), IMP (intramuscular progesterone), VP + PatchE (vaginal progesterone and patch oestrogen), IMP + OE (intramuscular progesterone and oral estradiol), IMHCG (intramuscular hCG), SCP + VP, Intranasal GnRH-a, OP (oral progesterone), IMP + IME (intramuscular progesterone and intramuscular estradiol), IMP + VP + OE (Intramuscular progesterone, vaginal progesterone and oral estradiol), IMP + VE (Intramuscular progesterone and vaginal estradiol), VP + SCGNRH-a [(Vaginal progesterone and subcutaneous GNRH agonist (GNRH-a)], VP + OE + SCGNRH-a (Vaginal progesterone, oral estradiol and subcutaneous GNRH-a), RP (Rectal progesterone), SCHCG (subcutaneous HCG), VP + DHEA (vaginal progesterone and oral DHEA), IMP + VP + SCGNRH-a (Intramuscular progesterone, vaginal progesterone and subcutaneous GNRH-a), OP + VP (oral progesterone and vaginal progesterone).

Full size image
Figure 3
figure 3

Comparison of clinical parameters [Median, 95% CrI] across treatment groups. Duration of infertility (A), Percentage of population diagnosed with primary infertility (B) or secondary infertility (C), Basal AMH ng/ml (D), basal LH IU/L (E), FSH IU/L (F). Only comparisons that reached statistical significance are depicted. The reference group was PVP. Two decimal p values and asterisk annotation of significance where p-value < 0.05, it is flagged with one star (*), p-value < 0.01, 2 stars (**), p-value < 0.001, three stars (***), p-value < 0.0001, four stars (****). LPS (luteal support), AMH (anti-mullerian hormone), FSH (Follicle stimulating hormone), LH (luteinising hormone), placebo (no exposure), SCP (Subcutaneous progesterone), VP (vaginal progesterone), IMP + VP (intramuscular progesterone and vaginal progesterone), VP + OE (vaginal progesterone and oral estradiol), IMP (intramuscular progesterone), VP + PatchE (vaginal progesterone and patch oestrogen), IMP + OE (intramuscular progesterone and oral estradiol), IMHCG (intramuscular hCG), SCP + VP, Intranasal GnRH-a, OP (oral progesterone), IMP + IME (intramuscular progesterone and intramuscular estradiol), IMP + VP + OE (Intramuscular progesterone, vaginal progesterone and oral estradiol), IMP + VE (Intramuscular progesterone and vaginal estradiol), VP + SCGNRH-a [(Vaginal progesterone and subcutaneous GNRH agonist (GNRH-a)], VP + OE + SCGNRH-a (Vaginal progesterone, oral estradiol and subcutaneous GNRH-a), RP (Rectal progesterone), SCHCG (subcutaneous HCG), VP + DHEA (vaginal progesterone and oral DHEA), IMP + VP + SCGNRH-a (Intramuscular progesterone, vaginal progesterone and subcutaneous GNRH-a), OP + VP (oral progesterone and vaginal progesterone).

Full size image
Figure 4
figure 4

Comparison of clinical parameters [Median, 95% CrI], luteal support regimens duration and dosage across treatment groups. Progesterone levels on hCG trigger day (ng/ml) (A), Progesterone levels on embryo transfer (ET) day (ng/ml) (B), Endometrial thickness measurement on ET day (C), Ovarian stimulation protocol employed (D). Only comparisons that reached statistical significance are depicted. The reference group was PVP. Two decimal p values and asterisk annotation of significance where p-value < 0.05, it is flagged with one star (*), p-value < 0.01, 2 stars (**), p-value < 0.001, three stars (***), p-value < 0.0001, four stars (****). Abbreviations: LPS (luteal support), hCG (Human chorionic gonadotropin), hMG (human menopausal gonadotrophin), placebo (no exposure), SCP (Subcutaneous progesterone), VP (vaginal progesterone), IMP + VP (intramuscular progesterone and vaginal progesterone), VP + OE (vaginal progesterone and oral estradiol), IMP (intramuscular progesterone), VP + PatchE (vaginal progesterone and patch oestrogen), IMP + OE (intramuscular progesterone and oral estradiol), IMHCG (intramuscular hCG), SCP + VP, Intranasal GnRH-a, OP (oral progesterone), IMP + IME (intramuscular progesterone and intramuscular estradiol), IMP + VP + OE (Intramuscular progesterone, vaginal progesterone and oral estradiol), IMP + VE (Intramuscular progesterone and vaginal estradiol), VP + SCGNRH-a [(Vaginal progesterone and subcutaneous GNRH agonist (GNRH-a)], VP + OE + SCGNRH-a (Vaginal progesterone, oral estradiol and subcutaneous GNRH-a), RP (Rectal progesterone), SCHCG (subcutaneous HCG), VP + DHEA (vaginal progesterone and oral DHEA), IMP + VP + SCGNRH-a (Intramuscular progesterone, vaginal progesterone and subcutaneous GNRH-a), OP + VP (oral progesterone and vaginal progesterone).

Full size image
Table 1 Included studies.
Full size table

Outcomes

The NMA primary outcomes were clinical pregnancy, defined as the presence of a gestational sac, with or without a fetal heartbeat on ultrasonography (US) and live birth, defined as the number of deliveries that resulted in live born neonate/s. Regarding live birth, singleton and non-singleton deliveries were considered as a single event. Secondary outcomes included biochemical pregnancy, defined as positive hCG test but without US verification 2 weeks following embryo transfer (ET), miscarriage defined as the spontaneous loss of a pregnancy before the 20th week, multiple pregnancy was defined as non-singleton clinical pregnancy and OHSS events. Crude events were collected per included study, and therefore no homogenisation of extracted data was required.

Data analysis

Effect estimates were calculated as odds ratios (ORs) for all outcomes, given that all were dichotomous, with respective 95% credibility intervals (95% CrIs) using Bayesian network and pair-wise meta-analysis14 (Fig. 5, Fig. S2–S10). Of note, a credibility interval is an interval within which an unobserved parameter value falls with a particular probability in Bayesian statistics comparable to the 95% Confidence interval commonly seen in frequentist statistics15. Network meta-analysis iterations were conducted with MetaInsight visual R package16. NMA was conducted using a fixed-effects model within a Bayesian setting, as unequal heterogeneity across all comparisons was assumed. Vaginal Progesterone (VP) was used as the reference treatment given its proven superiority over placebo and the NICE guideline recommendations17. A hierarchy of treatments was calculated for each outcome, based on the p-scores and SUCRA ratings. Summary of the rank distribution of LPS treatments, interpreted as the estimated proportion of treatments worse than the treatment of reference (VP) was displayed by Litmus Rank-O-Gram graphs and Radial SUCRA18 (Fig. S5–S6). Transitivity assumption was evaluated by comparing the distribution of key study characteristics across studies grouped by comparison (age and BMI). We assessed inconsistency between direct and indirect sources of evidence using global and local approaches. We assessed global inconsistency by using a design-by-treatment test19,20. Local inconsistency was evaluated by using the back calculation and separate indirect from direct design evidence methods, comparing direct and indirect evidence for each pairwise treatment comparison and node-splitting model21 (Table S2–S3; Fig. S3–S4). Possible heterogeneity of treatment effects and the robustness of findings was explored by subgroup network meta-analyses including only trials at overall low and medium risk of bias (Table 1, Fig. S7–S8, S10). Further subgroup analysis was conducted on trials using either standard (long) GnRH agonist or standard (short) GnRH antagonist protocol for ovarian stimulation to limit data heterogeneity. If mixed populations were included in the original publication, a cut-off of ≥ 65% of patients being treated with either of the protocols, was employed to categorise studies according to subgroup (Table S4). Mixmeta package in R v4.1.2 was employed for confounder exploration in a network meta-regression model. Gelman network convergence, network deviance and ranking analysis were conducted to quantify overall network discordance (Fig. S9–S10). Intergroup differences regarding demographic and treatment parameters were quantified, where appropriate by ANOVA (for parametric distributed variables e.g., Age, BMI) or Kruskal–Wallis test (non-parametric distribution of variables, e.g., all remaining variables). Multilevel network meta‐regression for the embryological parameters (number of transferred embryos, number of retrieved and mature oocytes, peak estradiol, % of day 3 embryos transferred) was undertaken for both primary and secondary outcomes22 (Table S11–S12).

Figure 5
figure 5

Luteal support Bayesian fixed effect consistency forest plot (Odds ratio, 95% CrI) for Clinical Pregnancy (A) Live Birth (B) Biochemical Pregnancy (C) Miscarriage (D) and Multiple pregnancy (E) OHSS (F) outcomes). Graph generated by MetaInsight R package. Tabular results of design-by-treatment interaction model consistency depicted in Table S1–S2 per outcome. Node splitting model per comparison (direct and indirect effects) depicted in Table S.6–7. placebo (no exposure), SCP (Subcutaneous progesterone), VP (vaginal progesterone), IMP + VP (intramuscular progesterone and vaginal progesterone), VP + OE (vaginal progesterone and oral estradiol), IMP (intramuscular progesterone), VP + PatchE (vaginal progesterone and patch oestrogen), IMP + OE (intramuscular progesterone and oral estradiol), IMHCG (intramuscular hCG), SCP + VP, Intranasal GnRH-a, OP (oral progesterone), IMP + IME (intramuscular progesterone and intramuscular estradiol), IMP + VP + OE (Intramuscular progesterone, vaginal progesterone and oral estradiol), IMP + VE (Intramuscular progesterone and vaginal estradiol), VP + SCGNRH-a [(Vaginal progesterone and subcutaneous GNRH agonist (GNRH-a)], VP + OE + SCGNRH-a (Vaginal progesterone, oral estradiol and subcutaneous GNRH-a), RP (Rectal progesterone), SCHCG (subcutaneous HCG), VP + DHEA (vaginal progesterone and oral DHEA), IMP + VP + SCGNRH-a (Intramuscular progesterone, vaginal progesterone and subcutaneous GNRH-a), OP + VP (oral progesterone and vaginal progesterone).

Full size image

Risk of bias assessment

Within-study bias was assessed with the Cochrane risk of bias tool RoB223 and the certainty of evidence using the GRADE Framework (Table 2). Overall network risk of bias was assessed with the Network Meta-Analysis framework (CINeMA)24 (Table S5–S10). Small-study effects and publication bias for each treatment pair was assessed using a contour-enhanced funnel plot.

Table 2 RoB Tool V. 2 risk of bias assessment and GRADE certainty in evidence ratings per study.
Full size table

Results

Included study design and quality of evidence assessment

From 1322 records initially retrieved, 76 RCTs, comparing 22 interventions of at least two arms comparing LPS protocols in fresh IVF/ICSI cycles, met the inclusion criteria (Fig. 1, Table 1)25,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94,95,96,97,98,100. Overall risk-of-bias judgement was deemed “low” for 24 studies “some concerns” for 29 and “high” for 23 studies (Table 2). Overall, GRADE confidence in evidence was deemed “high” for 34 studies, “moderate” for 23 studies and “low” or “very low” for 19 studies (Table 2).

Participant and treatment characteristics

A total of 26,536 participants were randomly assigned to any of the following 22 treatments; placebo (no exposure)[N = 727], SCP (Subcutaneous progesterone) [N = 877], VP (vaginal progesterone) [N = 13862], IMP + VP (intramuscular progesterone and vaginal progesterone) [N = 475], VP + OE (vaginal progesterone and oral estradiol) [N = 898], IMP (intramuscular progesterone) [N = 2136], VP + PatchE (vaginal progesterone and patch estrogen) [N = 179], IMP + OE (intramuscular progesterone and oral estradiol) [N = 387], IMHCG (intramuscular hCG) [N = 592], SCP + VP [N = 78], Intranasal GnRH-a [N = 23], OP (oral progesterone) [N = 3693], IMP + IME (intramuscular progesterone and intramuscular estradiol) [N = 55], IMP + VP + OE (Intramuscular progesterone, vaginal progesterone and oral estradiol) [N = 249], IMP + VE (Intramuscular progesterone and vaginal estradiol) [N = 174], VP + SCGNRH-a [(Vaginal progesterone and subcutaneous GNRH agonist (GNRH-a)] [N = 1008], VP + OE + SCGNRH-a (Vaginal progesterone, oral estradiol and subcutaneous GNRH-a) [N = 386], RP (Rectal progesterone) [N = 168], SCHCG (subcutaneous HCG) [N = 160], VP + DHEA (vaginal progesterone and oral DHEA)[N = 104] and IMP + VP + SCGNRH-a (Intramuscular progesterone, vaginal progesterone and subcutaneous GNRH-a) [N = 213] and OP + VP (oral progesterone and vaginal progesterone) [N = 92] (Fig. 2A).

Median participant age across all treatment groups was 32 years [IQR 31.75, 33.85] (Fig. 2B) and the median BMI was 23.94 (kg/m2) [IQR 22.45, 26.8] (Fig. 2B,C). Duration of infertility was of a median of 4.96 years [IQR 3.98, 6.10] (Fig. 3A). The population percentage diagnosed with primary infertility was 29.6% [Range: 10.9 to 42.62%] and secondary infertility was 34.5% [Range: 16.1 to 84.65%] and were not found to significantly differ across comparator groups (Fig. 3A–C). Median values of basal AMH, LH, FSH, progesterone levels on HCG trigger, progesterone levels on embryo transfer (ET) day, and endometrial thickness on ET day, per LPS were not found to be significantly different in comparison to the VP group (Figs. 3D–F, 4A–C). Regarding OS protocol, 54.69% of the participants underwent ovarian stimulation with a standard (long) GnRH agonist while 18.17% with a standard (short) GnRH antagonist protocol. A 1.05% underwent OS via clomiphene and HMG, 0.96% via a microdose flare and 0.54% by an ultrashort GnRH protocol (Fig. 4D)101. The remaining 24.59% of the participants underwent either a standard long or short OS protocol however the distribution was not noted in the original studies. Characteristics of embryo transfers were not consistently reported across arms of included studies (Table S.11). Of note, 20 of the 76 studies, failed to report upon these variables.

Regarding LPS protocols, schemas were segregated by active compound to explore variations of dosage (median dosage and maximum dosage), initiation day, duration of LPS (weeks) as well route of administration (Tables 1, 3). The majority of LPS protocols were initiated on the oocyte pickup day (OPU), and duration of 8 weeks (SD = 2). No significant differences were noted amongst LPS protocols regarding implantation 24.55% [IQR 18.17, 28.9] or fertilisation 63.6% [IQR 61, 78.9] median rates (Fig. S1).

Table 3 LPS protocol characteristics per compound.
Full size table

Data synthesis and network meta-analysis

VP was considered as the reference treatment as previously mentioned (NICE guidelines17. In NMA, effect size estimates suggested that all LPS protocols were consistently superior to placebo, employed as a negative control for both primary and secondary outcomes, regardless of risk of bias sensitivity analysis (Fig. 5, Fig. S2–S10, Tables S2–S4).

More specifically, regarding NMA primary outcomes:

  1. 1.

    For clinical pregnancy events, reported by 74 studies, CiNeMa NMA RoB rating was deemed “moderate” and overall network incoherence was found to be moderate, χ2 7.02, 4 degrees of freedom, p-value: 0.005) (Table S5, Fig. 5A, Fig. S2A, Fig. S3A, Fig. S5A, Fig. S6A, Fig. S9A). All LPS protocols appeared to be equivalent to VP in respect to the clinical pregnancy events, except for VP + OE + SCGNRH-a, [OR 1.57 (95% CrI 1.11 to 2.22) (SUCRA: 80%; Npatients:386, “Moderate” GRADE] and VP + SCGNRH-a [OR 1.28 (95% CrI 1.05 to 1.55) (SUCRA: 80%; Npatients:583, “High” GRADE], which were found to be superior, with high SUCRA probability (Fig. 5A, Fig. S5A, S6A). Equally, VP + PatchE was also associated with higher clinical pregnancy probability, OR 1.73 (95% CrI 1.16, 2.58) (SUCRA: 79%; Npatients:179, “Moderate” GRADE). Treatments such as IMP + IME OR 2.68 (95% CrI 1.06, 7.72) (SUCRA: 90%; Npatients:55, “Low” GRADE) and were shown to be superior in comparison to VP however the certainty in evidence was deemed low given the small number of participants included and the high risk of subsequent heterogeneity.

  2. 2.

    For the live pregnancy events, reported by 43 studies, CiNeMa NMA RoB rating was deemed “moderate” and overall network incoherence was found to be moderate, χ2 10.95 (5 degrees of freedom), p value: 0.052 (Table S6, Fig. 5B, Fig. S2B, Fig. S3B, Fig. S5B, Fig. S6B, Fig. S9B). The following interventions were found to improve live pregnancy events in comparison to the reference LPS, IMHCG [OR 9.67 (95% CrI 2.34 to 73.2) (SUCRA: 92%; Npatients:592, “Moderate” GRADE)], VP + OE [OR 4.57 (95% CrI 1.26 to 20) (SUCRA: 80%; Npatients:898, “Moderate” GRADE)], VP + OE + SCGNRH-a OR [OR 8.81 (95% CrI 2.35 to 39.1) (SUCRA: 95%; Npatients:386, “High” GRADE)], VP + SCGNRH-a [OR 1.76 (95% CrI 1.45 to 2.15) (SUCRA: 72%; Npatients:1008, “High” GRADE)] (Fig. 5B, Fig. S5B).

Regarding secondary outcomes:

  1. 3.

    For biochemical pregnancy events, reported by 29 studies, CiNeMa NMA RoB rating was deemed “Moderate” and network incoherence was found to be moderate, χ2 6.60 (2 degrees of freedom), p value: 0.037 (Table S7, Fig. 5C, Fig. S2C, Fig. S3C, Fig. S5C, Fig. S6C, Fig. S9C). For VP versus all other LPS protocols. No LPS protocol appeared to result in a significantly higher biochemical pregnancy probability.

  2. 4.

    Regarding miscarriage events, reported by 41 studies, CiNeMa NMA RoB rating was deemed “Moderate” and network incoherence was found to be moderate, χ2 11.30 (4 degrees of freedom), p value: 0.023 (Table S9, Fig. 5D, Fig. S2D, Fig. S4A, Fig. S5D, Fig. S6D, Fig. S9D). VP + SCGnRH-a was found to reduce miscarriage events in comparison to the reference LPS, [OR 0.54 (95% CrI 0.372 to 0.806), Npatients:1008, “Moderate” GRADE] with a SUCRA of 82.2% (Fig. 5D, Fig. S4A, Fig. S5D). Additionally, a similar finding was confirmed for IMP + IME [OR 0.08 (95% CrI 0.01 to 0.46), Npatients:55] however the certainty in evidence was deemed “Low”.

  3. 5.

    For multiple pregnancy events, reported by 21 studies, CiNeMa NMA RoB rating was deemed “High” (Table S9). Overall network incoherence was found to be low, χ2. 0.115 (2 degrees of freedom), p value: 0.94 (Fig. 5E, Fig. S2E, Fig. S4B, Fig. S5E, Fig. S6E, Fig. S9E). All LPS protocols appeared to produce similar results to PVP, except for SCP [OR 0.09 (95% CrI 0.009 to 0.556); SUCRA 1.2%, Npatients:877, “High” GRADE] resulting to significantly lower multiple pregnancy events and IMP + VP + SCGNRH-a [OR 6.88 (95% CrI 2.42 to 30.4); SUCRA 81.2%, Npatients:213, “Low” GRADE] resulting in significantly higher multiple pregnancy events.

  4. 6.

    For OHSS events, reported by 15 studies, CiNeMa NMA RoB confidence rating was deemed “Low” (Table S10). Overall network incoherence was found to be low, χ2.: 0.26 (2 degrees of freedom), p value: 0.015 (Fig. 5F, Fig. S2F, Fig. S5F, Fig. S6F, Fig. S9F). Pairwise analysis of included studies was not feasible due to the multitude of non-events (zero events of OHSS in either of the arms of the original study). All LPS protocols appeared to be associated with similar OHSS events to the reference LPS, except for OP [OR 1.87 (95% CrI 1.15 to 3.04); Npatients:3693, SUCRA 75%, “Low” GRADE] which was found to be associated with significantly higher OHSS events. The latter is likely to be a result of bias towards an OP LPS protocol selection in patients at high risk of ovarian hyperstimulation102,103.

Subgroup analysis of low and medium risk of bias studies (Figs. S7, S8, S10, Table 2) and node-splitting (Table S2–S3) did not significantly alter cumulative effects analysis or residual deviance (Fig. S10A–F). Optimal LPS per OS, long (Gonadotropin releasing hormone agonist) vs. short (GnRH antagonist) protocol, was explored to identify further sources of heterogeneity and to delineate whether a particular LPS appears to yield improved clinical outcomes in association with specific ovarian stimulation protocols (Table 4, Table S4). In view of live birth events, the following protocols were deemed optimal for participants that underwent OS by standard GnRH agonist protocol: (a) VP + OE + SCGNRH-a [OR 9.7 (95% CrI 3.73, 13.5)] (b) VP + OE [OR 4.58 (95% CrI 1.26, 20.3)], (c) VP + SCGNRH-a [OR 2.89 (95% CrI 1.46, 3.42)], and (d) IMHCG [OR 1.57 (95% CrI 2.24, 71.9)]. Of the aforementioned, the VP + OE, VP + SCGNRH-a and IMHCG comparators had a “High” GRADE rating while the VP + OE + SCGNRH-a protocol was also associated with a higher probability of miscarriage when used in combination with a GnRH agonist OS protocol, [OR 3.93 (1.69, 10.1)]. On the contrary, optimal luteal support protocols for standard GnRH antagonist OS were (a) IMHCG [OR 3.2 (95% CrI 1.54, 334.), “low” GRADE] and (b) VP + SCGNRH [OR 2.84 (95% CrI 1.35, 6.24), “High” GRADE] presenting the optimal LPS options across short protocols. Of note, IMHCG was also associated with a higher probability of miscarriage when used in conjunction with a short OS protocol [OR 2.11 (95% CrI 0.75, 6.40), high GRADE] while the opposite held true for VP + SCGNRH, which was associated with lower probability of miscarriage in short OS [OR 0.54 (95% CrI 0.37, 0.80), high GRADE]. Network meta-regression for all outcomes, according to embryological parameters, did not significantly alter effect sizes (Table S11–S12).

Table 4 Subgroup analysis of optimal LPS in short (GnRH antagonist) vs. long (GnRH agonist) protocol.
Full size table

Overall, NMA data suggest that combinatorial treatments, with the addition of SCGNRH-a on a VP base results in improved clinical pregnancy and live birth events and reduced miscarriage events in participants undergoing OS either a standard GnRH antagonist or agonist protocol. However, participants undergoing a long GnRH protocol OS appear to benefit more from IMHCG as LPS while participants undergoing a short GnRH protocol OS appear to benefit more from VP + SCGNRH, considering the reduction of miscarriage events of these luteal support protocols in conjunction to OS.

Discussion

This study is based on 76 RCTs, including 26,536 participants randomly assigned to 22 LPS protocols including non-exposure. Given the plethora of previous data suggesting that any LPS protocol is superior to non-exposure, the most widely employed LPS, vaginal progesterone, was set as a reference treatment3,17. Overall, meta-synthesized data presented here, suggest that combinatorial treatments, those with the addition of SCGnRH on a VP base result in improved clinical pregnancy, OR 1.28 (95% CrI 1.05 to 1.55) and live birth events, OR 1.76 (95% CrI 1.45 to 2.15) with high confidence in evidence. Of note, addition of oral estradiol to a VP + SCGNRH-a LPS, resulted in further improvement of clinical pregnancy events by 29% and 44% increase of a clinical pregnancy and live birth odds respectively. Of note, participants undergoing a long GnRH protocol OS appeared to benefit more from progesterone free LPS such as IMHCG in view of increased live birth, OR 1.57 (95% CrI 2.24 to 71.9) and reduced miscarriage events, OR 1.57 (95% CrI 2.24 to 71.9). However, participants appeared to be at a higher risk of OHSS, OR 1.64 (95% CrI 0.74 to 3.73). On the other hand, participants undergoing a short GnRH OS protocol appeared to benefit more from VP + SCGNRH with a live birth OR 2.84 (95% CrI 1.35 to 6.26), however while the probability of miscarriage was significantly reduced, OR 0.55 (95% CrI 0.38 to 0.80), the probability of multiple pregnancy significantly increased, OR 8.34 (95% CrI 2.57 to 37.6).

Luteal support is a critical aspect of IVF/ICSI cycles as it aids in maintaining the endometrial lining, in turn promoting embryo implantation, and supporting early pregnancy. In fresh IVF cycles, luteal support management can pose several challenges, including timing and duration of administration, individual outcome variability and tolerability of LPSs that may impact upon the success rates of the cycle. The effectiveness of luteal support in achieving live birth and clinical pregnancy rates is dependent on the timing of its administration104,104,106. Various studies have examined the optimal timepoint to initiate LPS, with only two out of five RCTs reporting statistically significant results104. Earlier evidence had suggested that delayed administration of LPS [(24 h after ovum pick-up (OPU)] may be more advantageous than pre-OPU administration (12 h prior to OPU)106. Williams et al. found initiating LPS on day 3 post OPU to be significantly better than delaying it until day 6105. Overall, these studies suggest that the optimal time for LPS administration is from the evening of OPU up until 3 days post OPU. Present NMA evidence suggested that the majority of studies favoured LPS initiation on the day of OPU (within the 24 h timeframe following the procedure), including for LPS protocols generating superior results namely, VP + SCGNRH-a and VP + OE + SCGNRH-a. Equally important to the LPS initiation timing, is the duration of luteal support administration. A recent meta-analysis including 1297 participants, indicating that continuing progesterone for two weeks after a positive pregnancy test did not have any significant impact on miscarriage or delivery rates106. The same study suggested that it is unnecessary to continue LPS for up to 10 weeks of pregnancy with further studies reaching to the same conclusion107,107,109. However, ESHRE 2020 recommendations suggest that LPS should be administered up until, at least the day of the pregnancy test101. Aggregate evidence of the present study indicate that duration of administration is highly dependent upon the selected LPS regimen, with an overall median of 8 weeks [Range 2–12] coinciding with ultrasonographic evidence of fetal motion and the concept of the luteo-placental shift110,111.

In addition, while initiation and duration of LPS treatment may appear more standardised, the selection of optimal type and dose of luteal support is largely individualised and dependent upon participant factors such as age, BMI, and reproductive history. Regardless of clinical and demographic parameters, undoubedtly the most important parameter affecting LPS selection and duration of perscription, is fundamentaly influenced by patient preference, which is in turn heavily reliant upon LPS side effect profile and tolerability, patient compliance and cost. For example, in view of treatment acceptability, IM progesterone has been widely available prior to vaginal formulation becoming available, and has been shown to have superior absorption and achieve stable serum concentration shortly after administration109,112,113. Nonetheless, administration complications involving pain, higher risk of infection, sterile abscess formation, and even rarely eosinophilic pneumonia as well as practical impediments requiring daily visits and injections, have necessitated the exploration of alternative more convinient routes, such as the one offered by the vaginal preparation114. Currently, vaginal progesterone products are administered in various ways, including pessaries, capsules, tablets, gel, and inserts which can achieve maximum serum concentration of progesterone after 3–8 h of administration, and by daily doses of 300–600 mg may achieve adequate available plasma levels115. Evidence has also shown that a 300–600 mg of vaginal micronized progesterone daily can induce similar endometrial maturation as 100 mg intramuscular progesterone daily109. By enabling direct transport of "first uterine pass" progesterone from the vagina to the uterus, vaginal preparations achieve adequate tissue levels of progesterone with lower circulating levels, indicating acceptable bioavailability111.

Given the improved outcomes regarding clinical pregnancy and live birth, achieved by VP + SCGNRH-a and VP + OE + SCGnRH-a combinatorial treatments, shown in the present work, a mention to route and dosage of gonadotropin-releasing hormone agonist is warranted. The use of GnRH-a for LPS was suggested following accidental use of GnRH agonist during this phase which resulted in improved implantation rates116. The effect of GnRH agonist has been observed at three levels: support of the corpus luteum through pituitary LH secretion, direct effects on the embryo and implantation process, and the effect upon trophectoderm cells and endometrial GnRH receptors77,116,117. A meta-analysis showed that administering a single dose of GnRH-a increased the implantation rate in cycles with GnRH antagonist and long GnRH-a protocols, clinical pregnancy rate per transfer, and ongoing pregnancy rate118, whilst another revealed that the use of GnRH-a for LPS significantly improved live birth rate, clinical pregnancy rate, and ongoing pregnancy rate119. An additional study demonstrated that a single dose of GnRH-a had similar efficacy as three doses of hCG120. One can hypothesise that the addition of a GnRH agonist can bimodally support the corpus luteum by stimulating the release of gonadotrophins from the pituitary gland, and by directly influencing the endometrium through interaction with GnRH receptors. Furthermore, research suggests that administering a single dose of GnRH agonist during the luteal phase enhances rates of pregnancy, implantation, delivery, and birth among recipients of donated oocytes whose ovulation was suppressed and corpus luteum was absent, suggesting a potential direct impact of GnRH agonist on the embryo77,98,116,119. The present work has highlighted that a single SCGNRH administration in addition to a VP protocol, can positively impact on IVF/ICSI outcomes especially in patients undergoing GnRH antagonist OS and could be reserved for more challenging cycles to optimise results. Conversely, in view of the improved clinical pregnancy and live birth outcomes achieved by the addition of oral estradiol in the VP + SCGNRH-a protocol, exploration of the possible synergistic effects of this compound is necessitated. However, a Cochrane meta-analysis did not find evidence to support routinely administering estrogen with progesterone in IVF cycles9. In antagonist cycles, progesterone levels surge, leading to a rebound decrease in serum estradiol, which in turn has formulated the hypothesis that adding doses of 2–6 mg/day of estradiol could be beneficial111. However, contemporary systematic reviews failed to confirm the beneficial effects of oral or any route of estradiol addition to progesterone LPS upon pregnancy outcomes120,120,122. Of note, novel LPS regimens involving intranasal GnRH administration has been shown promising results regarding clinical pregnancy rates and treatment tolerability however given the scarse RCT evidence, further, adequately powered, RCTs would be required to allow recommendations regarding this LPS regimen117,123.

In addition, the present work has shown that progesterone free LPS protocols, such as intramuscular hCG, may be equal, if not more effective that progesterone-based LPS in view of live birth outcomes, especially in patients undergoing a GnRH agonist OS protocol. HCG, by mimicking LH pulsatility, was initially considered the primary choice for LPS as it stimulates the corpus luteum to produce progesterone continuously. However, this approach has drawbacks, as it can elevate the risk of OHSS, a hypothesis which was also confirmed by the present NMA, albeit lacking statistical significance, OR 1.64 [95% CrI 0.75, 3.71].

Limitations and future perspectives

The optimal protocol for luteal support is a constantly evolving field of research in artificial reproduction. In view of the plethora of available LPS protocols, NMA precision of estimates provides a more comprehensive understanding of the comparative effectiveness of different protocols. In the present work, only RCT data have been employed to reach meaningful conclusions limiting inherent bias of diverse participant populations, with add-on sensitivity analysis targeted at low and moderate risk of bias studies and ovarian stimulation protocol to further explore confounding factors and detect sources of heterogeneity. Given the anticipated diversity of measured outcomes, a bayesian meta-synthesis approach has been adopted to account for the expected heterogeneity and to incorporate modelling flexibility by allowing for posterior distributions interpreted as SUCRA probabilities with the later enabling crisper communication of the uncertainty in the treatment effects estimates.

On this note, in the present study, LPS protocols have been treated as unique comparator entities, allowing for assessment of selected outcomes on a protocol- rather than a compound-level. However, side effect and safety profile of combinatorial treatments has not been assessed and therefore a significant confounder in tolerability and in turn, compliance, especially in the context of combinatorial LPS, remains to be investigated. Additionally, cost-effectiveness analysis has not been undertaken, which needs to be factored in a joined patient and clinical decision-making. Moreover, while reported, the present study did not aim to clarify of optimal initiation/cessation timing of LPS or the optimal dosage and therefore to produce concrete recommendations regarding these LPS parameters, further studies with relevant designs should be implemented. Notably, included studies were significantly heterogeneous in terms of reporting the characteristics of embryo transfers. Abeit no statistical difference was reported for variables such as follicles retrieved, peak oestradiol levels and number of embryos transferred in individuals studies, confounding effects cannot be confidently excluded. Lastly, OHSS events were found to be considerably under-reported across RCTs with only 15 studies noting such events. Reflecting on the implications of OHSS upon both the patient clinical management as well as the success of the IVF/ICSI, it would be strongly recommended that future RCTs would thoroughly record OHSS events across study arms. Overall, luteal support management in fresh IVF cycles is a complex and dynamic process that calls for careful consideration and individualised LPS selection to achieve optimal outcomes.

Conclusion

Herein meta-synthesized data suggest that combinatorial treatments, with the addition of subcutaneous GnRH agonist, on a vaginally administered progesterone LPS base, results in improved clinical pregnancy and live birth events. However, the side-effect and tolerability profile of such combinatorial LPS protocols needs to be thoroughly investigated prior to their wide-scale adoption in clinical practice.