Effectiveness of hormone add-on strategies in ovarian stimulation for women with poor ovarian response: a systematic review and network meta-analysis of randomized controlled trials

The following Medical Subject Heading (MeSH) terms and keywords were used to search the electronic databases MEDLINE (available through PubMed), SCOPUS, LILACS, EMBASE, Scielo.br, and PROSPERO: “ovarian stimulation”/“ovulation induction” (MeSH Unique ID: D010062), “assisted reproductive technologies” (MeSH Unique ID: D027724), “in vitro fertilization” (MeSH Unique ID: D005307), “reproductive techniques”, “POSEIDON criteria”, “Bologna criteria”, “low responders”, “poor responders”, “poor ovarian response”, “poor response”, “poor ovarian responsiveness”, “ovarian reserve” (MeSH Unique ID: D065851), “diminished ovarian reserve”, “poor ovarian response”and “poor ovarian reserve”. without any date restrictions. The search query was changed to fit the format of each database (Appendix 1). A filter for randomized controlled trials (RCTs) was used in each database.Additional searches were conducted to find more relevant papers on CINAHL, PsycINFO, and AMED in order to reduce publication bias. The Cochrane Central Register of Controlled Studies, Clinicaltrials.gov, and the World Health Organization’s International Clinical Trials Registry Platform (ICTRP) were also searched to locate additional randomized controlled trials. Additionally, the gray literature (NTIS, PsycEXTRA) was examined to find national and international conference abstracts. The included study’s references and pertinent reviews were also reviewed to locate more articles that eluded our first search.The analysis did not include any editorials, letters to the editor, comments, or second thoughts publications. A.E. and V.A. carried out the approach for the literature search.

The titles and abstracts of all the papers were separately screened by two reviewers (G.R. and J.C.C.F.) to identify which research should be evaluated further and to omit citations that were judged unnecessary by both observers.The authors, institutions, publication titles, and study findings were all concealed during this initial screening. Any ambiguity or disagreement was cleared up by talking with a third reviewer (A.F.R). Two reviewers (A.B.G. and B.M.R.) retrieved and evaluated full texts of possibly relevant articles based on the pre-established inclusion criteria. In addition, methodological validity was evaluated before being included in the review. Every doubt or disagreement was cleared up by discussion among reviewers in order to reach an agreement.Add-ons have been defined as any additional treatment used, in addition to gonadotropin releasing hormone (GnRH) analogs and gonadotropins, during the vitro fertilization (IVF)/intracytoplasmic sperm injection (ICSI) cycle, with the goal of increasing pregnancy success in women with POR. As such, hormonal add-ons directly affect the endocrine system and influence ovarian response, oocyte quality, or endometrial receptivity. In this review, we include letrozole (LTZ) and clomiphene citrate (CC) as hormonal add-ons, although they are more commonly defined as ovulation-inducing agents. We included only RCTs that allocated women with POR, according to Bologna criteria [23], undergoing ART to at least two different OS protocols with the treatment arm including at least one hormonal add-on. We chose to use the Bologna criteria to strictly limit the inclusion criteria of patients with POR in the network meta-analysis. Considering the recent introduction but the paucity of studies adopting exclusively or predominantly the POISEDON criteria, we opted to exclusively use the Bologna criteria in the present analysis to ensure greater homogeneity of the population, significantly reducing clinical and methodological variability and thus increasing the reliability and generalizability of the results thanks to their validation and consolidation in a wider amount of RCTs.

Control arms consisted of women undergoing the same OS protocol without the hormonal add-ons.

The exclusion criteria were the following: studies including different OS protocols between treatment and control groups, studies with women not matching the Bologna criteria [23] and studies evaluating non-hormonal add-ons defined as additional treatments that improve metabolic, immunologic, and supportive factors without directly modifying hormonal balance.

The extraction forms were made especially for this network meta-analysis.

The nation, study design, study type, inclusion and exclusion criteria, study period, location, were extracted from each study. For each group (OS + add-ons and OS without add-ons) the following study characteristics were gathered in order to characterize the included studies: factors such as the number of women, the women’s age, body mass index (BMI), the type and duration of her infertility, the AFC, the AMH, day-3 FSH, the number of previous failed fertility treatments, the cause of infertility besides poor ovarian response, the method of diagnosis, the characteristics of IVF/ICSI, the protocol for OS, the duration of the stimulation, the total dosage of gonadotropins used during the OS, the cancelation rate, the number of retrieved and metaphase II (MII) oocytes, the rates of fertilization, the clinical pregnancy rate (CPR) and the LBR.

Each abstract was examined and categorized independently by two authors (A.E. and V.A.). In order to agree on possible application, the same two writers carefully read the texts of the selected studies and independently gathered pertinent information about the research characteristics and the noteworthy findings.

After discussing each contradiction with the other three authors (J.C.C.F., A.F.R. and B.M.R.), the reviewers came to a consensus. Unpublished data were acquired by contacting the original study authors directly when the study methods specified that additional outcome data were collected.

The risk of bias for the included RCTs was assessed using the Cochrane Risk of Bias Tool version 2 (RoB 2.0) [26]. This tool categorizes the risk of bias in research into five areas: the randomization process, deviations from intended interventions, missing outcome data, outcome evaluation, and choice of reported result. Three risk levels are evaluated for every domain: “low risk”, “some concerns” or “high risk”.

Three authors (A.E., V.A. and G.R.) independently evaluated the risk of bias assessment. The dispute was settled after concealment with J.C.C.F., A.F.R. and A.B.G.

We examined the feasibility and comparability of baseline patient characteristics, the reported results, and the registration of trials before the recruitment of the first patient (for RCTs that began less than 15 years ago) to further assess the reliability of the studies. Additionally, we examined each study’s compliance with the CONSORT declaration and any believable retractions or expressions of concern using the PubMed and RetractionWatch.org databases. Furthermore, the Confidence in Network Meta-Analysis framework (CINeMA) criteria were used to evaluate the certainty of the evidence [27]. Publication bias was assessed using funnel plots and Egger’s test for each outcome reported in at least 10 trials.

The primary outcome of the present network meta-analysis was the LBR, where a live birth was defined according to the International Glossary on Infertility and Fertility Care [28] as the complete expulsion or removal of a fetus from a woman after at least 22 weeks’ gestation, which, when separated, manifests vital signs.

Secondary outcomes were the CPR, defined as pregnancy diagnosed by ultrasonographic visualization of one or more gestational sacs or definitive clinical signs of pregnancy [28]; total dosage of gonadotropins used during the OS; duration in days of the OS; the serum estrogen levels at the day of trigger with the human chorionic gonadotropin (hCG) or the agonist of the GnRH; the number of oocytes retrieved; the number of MII oocytes retrieved; the fertilization rate, defined as the number of fertilized oocytes on the total number of oocytes; the cancellation rate defined as the number of cycles cancelled due to ultrasonographic non-response to gonadotropin administration or absence of oocytes obtained out of the number of oocyte retrievals.

All data analysis and graphical representations were done using STATA version 14.1 (StataCorp, College Station, TX).

The network assumption of overall consistency was statistically confirmed for every meaningful result using the < network meta consistency > command. The local test on loop inconsistencies was then carried out using the Separating Indirect from Direct Evidence (SIDE) procedure with the command < network sidesplit all >. When no discrepancy between the global and local testing findings was found, the consistency assumption was accepted. The study’s direct and indirect comparisons would undoubtedly yield pertinent results in this case, as the consistency model demonstrated that any differences in the results were only attributable to chance mistakes and the effect of the intervention. The summary measurements were presented as odds ratios (OR) for categorical variables and standardized mean differences (SMD) for continuous variables with 95% confidence intervals (CI) using the Der Simonian and Laird random effects model. A Higgins I² score higher than 25% suggested the possibility of heterogeneity. To evaluate the efficacy of the different hormonal add-ons and rank the therapies to determine which was statistically better, a forest plot and an interval plot together with a Surface Under the Cumulative Ranking curve Area [SUCRA] ranking plot were made for each outcome under study.

The main characteristics of the included studies are summarized in Table 1, and the baseline patients’ characteristics are presented in Table 2.

Among the 22 included studies, all were RCTs [12, 13, 16, 17, 89‐106]. Of these, 5 studies [13, 92, 95, 103, 105] where from China, 3 studies [89, 97, 99] were from Iran, 2 studies [16, 102] were from Turkey, 2 studies were from Greece [91, 100], 2 studies were from Egypt [90, 96], 1 study was from Denmark [94], 1 study was from Italy [98], 1 study was from Spain [101], 1 study was from Switzerland [17], 1 study was from Australia [12], 1 study was from Hong Kong [106], 1 study was from Israel [104], and 1 study was from Vietnam [93]. A total of 4131 infertile patients with a diagnosis of POR according to Bologna criteria [23] were included in this analysis; of these, 2069 patients underwent OS without hormonal add-ons and 2062 patients received hormonal additions in the context of OS. Two hundred ten patients were given LTZ as a hormonal add-on during the OS [16, 95, 97], 214 patients were given CC during the OS [17, 98, 100], 509 patients were given rLH during the OS [94, 102, 104], 276 patients were pre-treated with estrogen before the OS [13], 493 patients were pre-treated with DHEA before the OS [92, 103, 105, 106], 167 patients were pre-treated with testosterone before the OS [89, 91, 93, 99, 101], and 200 patients were given GH during the OS [12, 90, 96].

Figure S1a shows the quality of the methodology used in the 22 included RCTs, and Figure S1b shows a percentage-based assessment of the methodological quality of these trials. Overall, one RCT [90] was deemed at high risk of bias, six RCTs [89, 93, 98‐101] were judged as having “some concerns” and 15 RCTs [12, 13, 16, 17, 92, 94‐97, 102‐106] were considered at low risk, respectively.

In detail, nine trials [90‐92, 95, 98‐100, 105, 106] had some or severe concerns for biases arising from the randomization process (Figure S1a-b). No study was deemed at high risk for bias due to deviations from intended interventions or missing data. On the contrary, there was a high risk of bias in the measurement of the outcomes in three studies [89, 93, 101]. Conversely, the vast majority of RCTs were at low risk for selection biases (Figure S1a-b).

All the included RCTs [12, 13, 16, 17, 70, 89‐105] were registered in valid prospective repositories prior to the enrollment of the first woman.

Certainty of evidence according to CINeMA criteria ranged from very low to low, principally due to major concerns in the imprecision and incoherence domains, which were related to the low number of direct comparisons available (Supplementary Table 2). Potential publication bias was visually displayed in Figure S2a-g.

The aim of this systematic review and network meta-analysis of RCTs was to evaluate the reproductive outcomes of hormonal add-ons in women with POR undergoing ART. For the LBR, which was our primary outcome, no significant differences were observed between the evaluated interventions; however, testosterone showed the highest probability of improving LBR according to SUCRA analysis. Meanwhile, GH showed the highest SUCRA score for CPR. Moreover, GH also demonstrated better results in terms of total number of retrieved oocytes and mean number of MII oocytes. Finally, LTZ demonstrated benefits in reducing gonadotropin usage and stimulation duration.

Several add-ons to OS have been explored in POR women. However, available studies did not ensure their effectiveness. A recent meta-analysis of RCTs concluded that pretreatment with testosterone appears to be related to a better LBR in women with POR [21]. The authors of the study, however, compared not only hormonal add-ons with controls, but also variations in OS protocols including GnRH agonists or antagonists, different types of gonadotropins, and dual-stimulation regimen. This heterogeneity, combined with the limitations derived from a classic meta-analysis approach, makes it difficult to interpret how their conclusions were derived. In fact, a conventional pairwise meta-analysis, which only compares treatments directly from primary studies, does not allow treatments to be ranked according to their efficacy or safety, offering a more limited basis for clinical decision-making. Because the ideal goal should be to compare available hormonal add-on therapies within each other to achieve a broader and more quantitative view of their comparative effectiveness of, a network meta-analytic approach should be preferred, as performed by Zhang et al. in 2020 [22]. However, this study is not without strong limitations. First, the reliability of the included studies has not been evaluated and is not adequate to the actual trustworthiness standards [107]. In addition, although the authors claimed to include only patients defined as POR according to the March 2010 Bologna criteria of the European Society of Human Reproduction and Embryology (ESHRE) [23], multiple studies published before this date were included, some of which enrolled patients who could not be defined as POR according to them [23]. This lower homogeneity of the included population consequently significantly reduces the clinical and methodological variability and thus also the reliability and generalizability of the results. Moreover, a crucial strength of our analysis compared to that of Zhang et al. [22] concerns the primary outcome analyzed. The choice of LBR as the primary outcome, compared to the CPR used by the authors of the previous network meta-analysis, represents a substantial advantage. In fact, the ESHRE explicitly recommends the use of LBR as the primary outcome in clinical trials and systematic reviews conducted with the aim to evaluate the clinical effectiveness of using add-ons in ART treatments, as LBR is considered the most clinically relevant parameter, being directly related to the final success of treatment and reflecting the ultimate goal of couples undertaking reproductive medicine treatment [108]. Therefore, the use of LBR as the primary endpoint in this network meta-analysis significantly increases its clinical relevance and relevance to the information needs of physicians and patients, clearly distinguishing it from the previous meta-analysis by Zhang et al. [22].

It is crucial to underline that SUCRA rankings reflect the probability that an intervention is among the best options based on the entire network of evidence. However, a higher SUCRA score does not imply that the intervention is statistically superior to others in direct or indirect comparisons. In our analysis, no statistically significant differences were observed among most pairwise comparisons, which suggests limited certainty in the relative superiority of the treatments. Nevertheless, the consistency of higher SUCRA values for GH and testosterone across multiple outcomes suggests a favorable trend that warrants further investigation. Therefore, SUCRA rankings should be interpreted as hypothesis-generating tools rather than definitive clinical guidance, particularly in this scenario where statistical significance is lacking. Our findings highlight GH and testosterone as the agents with the highest probability of improving reproductive outcomes in women with POR undergoing ART.

The exact mechanism by which GH might improve the outcomes of OS remains unclear. However, it is known that GH enhances insulin-like growth factor 1 (IGF-1) activity in the follicle, which plays a critical role in granulosa cell function and follicular maturation [109]. The studies included in our meta-analysis [12, 18, 90] reported a higher number of oocytes retrieved after the administration of GH in POR women. Nevertheless, the benefit for the LBR could not be clearly demonstrated. A review by Kolibianakis et al. in 2009 [110] had already suggested a significant increase in the number of retrieved oocytes and an improvement in LBR when GH was added to OS in poor responders. However, with the implementation of Bologna criteria, recent reviews concluded that the effect of GH administration on ovarian response or pregnancy rate is still contradictory and remains unclear [21, 111]. A recent Cochrane on this topic also reported similar results [112]. Conversely, the use of adjuvant GH in OS protocols has an uncertain effect on LBR and the average number of oocytes retrieved in patients expected to be good responders. Therefore, although promising, the use of GH as an add-on in women with POR requires further high-quality studies to confirm its efficacy, given the paucity of trials and their inherent biases and imprecisions.

Androgens play a crucial role in folliculogenesis. Specifically, they influence the proliferation of granulosa cells, the number of growing follicles as well as the expression of FSH receptors in the ovary [113]. Different agents have been explored to increase intraovarian androgen concentration: transdermal testosterone, DHEA, and androgen modulators such as LTZ. Our findings suggest that transdermal testosterone could have a positive impact on the LBR and CPR. However, the available evidence regarding the clinical utility of transdermal testosterone in poor responders shows contradictory results. Some authors reported an increase in pregnancy rates [93, 99] and ovarian response [89, 99] after testosterone administration. In contrast, well-designed clinical trials did not show differences in their primary outcomes (number of oocytes and pregnancy rate) using testosterone [91, 101]. Recently, a conference abstract of a multicenter multinational RCT concluded that testosterone pre-treatment prior to IVF/ICSI does not increase clinical pregnancy rates in poor ovarian responders [114]. Nevertheless, a recent meta-analysis [115] concluded that the probability of obtaining more oocytes and clinical pregnancies was significantly increased in women pretreated with transdermal testosterone compared with those who were not, with a lower cancellation rate in the testosterone group. However, available studies lack homogeneity regarding type, dose, and duration of testosterone supplementation. Thus, although it is suggested that the use of testosterone in POR women may improve reproductive outcomes, a consensus on the supplementation protocol needs to be reached.

In the current meta-analysis, the use of DHEA has shown no benefit in improving reproductive outcomes, except for a decrease in the cancellation rate. This contrasts with other reviews and meta-analyses suggesting that DHEA could improve oocyte response and pregnancy rates in patients with poor response [21, 22].

Another add-on that showed some results in our network meta-analysis was LTZ. LTZ is an aromatase inhibitor that increases intraovarian androgens by inhibiting the conversion of androgens into estrogens. In concordance with other authors [21], we did not observe a benefit with LTZ in the number of oocytes retrieved or pregnancy rate although we observed a reduction of gonadotropin consumption and duration of stimulation with the use of LTZ. Only a few studies have evaluated the use of this add-on in patients with POR [8, 16, 97]. Authors who have evaluated the use of LTZ in mild stimulation compared to conventional stimulation without LTZ have observed no differences in reproductive outcomes [8, 16]. However, Moini et al. [97] showed that patients treated with conventional doses of gonadotrophins plus LTZ had an increase in the number of oocytes retrieved along with a decrease in gonadotropin consumption and duration of stimulation compared to the women without LTZ addition. LTZ is therefore a promising treatment that requires further exploration. According to our results, neither the use of CC, estrogen priming nor the addition of rLH would gain an additional benefit in patients with POR.

While this network meta-analysis was limited to hormonal add-ons, other interventions such as coenzyme Q10, low-dose hCG, and corifollitropin alfa showed promising potential in improving ovarian response. Coenzyme Q10, through its antioxidant and mitochondrial-supportive properties, may enhance oocyte quality in women with POR [19]. Low-dose hCG, when added during OS, may mimic LH activity and support follicular maturation, although high-quality RCTs in Bologna-defined POR are lacking [116]. Corifollitropin alfa offers the advantage of simplified stimulation protocols with a single injection replacing daily FSH, and may be useful in selected POR [117]. However, the current body of evidence is either methodologically limited or inconsistent with our inclusion criteria. Future high-quality RCTs are needed to assess the role of such agents and determine their integration into individualized OS strategies.

Several limitations should be considered while interpreting our findings. The certainty of evidence was primarily low due to imprecision and incoherence, mainly stemming from the limited number of direct comparisons available and variations in study methodologies.

The dosages, the beginning time, the length of ovulation stimulation, and the GnRH analog protocols employed in the suggested interventions and OS protocols also showed notable differences. Therefore, some small heterogeneity among included trials may have influenced the findings. Due to the small sample sizes of the eligible trials, it was typically impossible to properly investigate this clinical heterogeneity. Consequently, the current systematic review and network meta-analysis permitted only broad comparisons between various hormonal adjuvant treatments and no treatment, lowering the certainty of evidence for imprecision due to the lack of significant differences in several comparisons. Future research will allow for the assessment of a possible moderating influence of these parameters on the effectiveness of the interventions under investigation.

Conversely, to ensure the robustness and validity of such findings, we included only studies in which the POR followed the Bologna criteria [23], excluding many trials that might increase inherent heterogeneity and thus be a strength of this network meta-analysis.

Moreover, all the most critical ART outcomes, including also the LBR, were analyzed, emphasizing the importance of our findings in clinical practice. Additionally, the consistency of the reported findings, showing agreement in the top-ranked add-ons (GH and testosterone) for most study outcomes, provides useful indications for routine practice in practitioners dealing with women with POR undergoing ART.

In conclusion, the results of the present systematic review and network meta-analysis of RCTs suggest that the addition of testosterone and GH may offer benefits to POR patients undergoing OS. The addition of LTZ also showed some sort of benefit, showing the improvement of some OS parameters. In contrast, the addition of DHEA, CC, estrogen, or rLH did not show significant benefit. However, the current certainty of evidence is still limited to low and very low; therefore, it is crucial to further investigate the real clinical value of these additional hormone treatments with further rigorous studies.