Estimating the net effect of progesterone elevation on the day of hcg on live birth rates after IVF: a cohort analysis of 3296 IVF cycles

Transcription

1 Human Reproduction, Vol.30, No.3 pp , 2015 Advanced Access publication on January 12, 2015 doi: /humrep/deu362 ORIGINAL ARTICLE Reproductive endocrinology Estimating the net effect of progesterone elevation on the day of hcg on live birth rates after IVF: a cohort analysis of 3296 IVF cycles Christos A. Venetis 1, *, Efstratios M. Kolibianakis 2, Julia K. Bosdou 2, George T. Lainas 3, Ioannis A. Sfontouris 3, Basil C. Tarlatzis 2, and Tryfon G. Lainas 3 1 Women s and Children s Health, St George Hospital, University of New South Wales, Sydney, NSW, Australia 2 Unit for Human Reproduction, 1st Department of Obstetrics and Gynaecology, Medical School, Aristotle University of Thessaloniki, Thessaloniki, Greece 3 Eugonia - IVF Unit, Athens, Greece *Correspondence address. venetis@gmail.com; c.venetis@unsw.edu.au Submitted on July 26, 2014; resubmitted on December 2, 2014; accepted on December 29, 2014 study question: What is the proper way of assessing the effect of progesterone elevation (PE) on the day of hcg on live birth in women undergoing fresh embryo transfer after in vitro fertilization (IVF) using GnRH analogues and gonadotrophins? summary answer: This study indicates that a multivariable approach, where the effect of the most important confounders is controlled for, can lead to markedly different results regarding the association between PE on the day of hcg and live birth rates after IVF when compared with the bivariate analysis that has been typically used in the relevant literature up to date. what is known already: PE on the day of hcg is associated with decreased pregnancy rates in fresh IVF cycles. Evidence for this comes from observational studies that mostly failed to control for potential confounders. study design, size, duration: This is a retrospective analysis of a cohort of fresh IVF/intracytoplasmic sperm injection cycles (n ¼ 3296) performed in a single IVF centre during the period participants/materials, setting, methods: Patients in whom ovarian stimulation was performed with gonadotrophins and GnRH analogues. Natural cycles and cycles where stimulation involved the administration of clomiphene were excluded. In order to reflect routine clinical practice, no other exclusion criteria were imposed on this dataset. The primary outcome measure for this study was live birth defined as the delivery of a live infant after 24 weeks of gestation. We compared the association between PE on the day of hcg (defined as P. 1.5 ng/ml) and live birth rates calculated by simple bivariate analyses with that derived from multivariable logistic regression. The multivariable analysis controlled for female age, number of oocytes retrieved, number of embryos transferred, developmental stage of embryos at transfer (cleavage versus blastocyst), whether at least one good-quality embryo was transferred, the woman s body mass index, the total dose of FSH administered during ovarian stimulation and the type of GnRH analogues used (agonists versus antagonists) during ovarian stimulation. In addition, an interaction analysis was performed in order to assess whether the ovarian response (,6, 6 18,.18 oocytes) has a moderating effect on the association of PE on the day of hcg with live birth rates after IVF. main results and the role of chance: Live birth rates were not significantly different between cycles with and those without PE when a bivariate analysis was performed [odds ratio (OR): 0.78, 95% confidence interval (CI): ]. However, when a multivariable analysis was performed, controlling for the effect of the aforementioned confounders, live birth rates (OR: 0.68, 95% CI: ) were significantly decreased in the group with PE on the day of hcg. The number of oocytes retrieved was the most potent confounder, causing a 29.4% reduction in the OR for live birth between the two groups compared. Furthermore, a moderating effect of ovarian response on the association between PE and live birth rates was not supported in the present analysis since no interaction was detected between PE and the type of ovarian response (,6, 6 18,.18 oocytes). & The Author Published by Oxford University Press on behalf of the European Society of Human Reproduction and Embryology. All rights reserved. For Permissions, please journals.permissions@oup.com

2 Progesterone elevation and live birth rates 685 limitations, reasons for caution: This is a retrospective analysis of data collected during a 12-year period, and although the effect of the most important confounders was controlled for in the multivariable analysis, the presence of residual bias cannot be excluded. wider implications of the findings: This analysis highlights the need for a multivariable approach when researchers or clinicians aim to evaluate the impact of PE on pregnancy rates in their own clinical setting. Failure to do so might explain why many past studies have failed to identify the detrimentaleffectofpeinfreshivfcycles. study funding/competing interest(s): None. Key words: hcg / IVF/ICSI outcome / ovarian stimulation / progesterone / pregnancy Introduction The potential association of serum progesterone on the day of human chorionic gonadotrophin (hcg) administration with the outcome of an in vitro fertilization (IVF) cycle has been one of the major controversies in the endocrinology of ovarian stimulation (Venetis et al., 2007; Kolibianakis et al., 2012). Following the publication of a recent meta-analysis of more than IVF cycles, there is now compelling evidence supporting that elevated progesterone on the day of hcg during a stimulated cycle is associated with decreased probability of pregnancy when a fresh embryo transfer is performed. In contrast, such an effect is not present in subsequent frozen thawed cycles and in oocyte donation cycles, in which embryos originating from cycles with progesterone elevation (PE) are transferred to an endometrium that has not been exposed to elevated progesterone (Venetis et al., 2013). However, that extensive review of the literature, besides providing a definite answer on the role of PE on the probability of pregnancy, generated at the same time new important clinical questions that ought to be addressed. One of the major issues in most of the published studies examining the role of PE on the day of hcg is the fact that its effect on pregnancy rates has been estimated through simple bivariate analyses, which are not able to control for confounding effects by other important parameters (Hosmer and Lemeshow, 2004; Rothman et al., 2008). For instance, the number of oocytes is strongly correlated with the probability of pregnancy and live birth (Sunkara et al., 2011) and it has been demonstrated consistently that elevated progesterone is also associated with a higher number of oocytes (Venetis et al., 2007; Bosch et al., 2010; Kyrou et al., 2012; Xu et al., 2012; Venetis et al., 2013). By not controlling for this difference in the number of oocytes, the available studies mayactually underestimate the true effect of PE on pregnancy rates. Furthermore, some of the available studies have suggested that ovarian response may play a moderating role on the effect of PE on pregnancy and that PE does not compromise pregnancy rates in high responders (Fanchin et al., 1997; Griesinger et al., 2013). A potential theory to explain such a moderating effect is that the number of oocytes might also be a proxy indicator of the quality of the resulting embryo and thus of its ability to implant. More oocytes might indicate better embryos which are still able to implant in an endometrium that is not optimal, while less oocytes might indicate embryos that already have diminished implantation potential, which is further affected by a compromised endometrium (Fanchin et al., 1996; Fanchin et al., 1997). On the other hand, previously published data rebut this hypothesis by demonstrating a detrimental effect of high progesterone regardless of the magnitude of ovarian response (Bosch et al., 2010). Other studies have found that in high responders such an effect is present at higher progesterone levels (Xu et al., 2012). It is still not known whether the number of oocytes moderates the effect of PE on pregnancy rates and this represents a source of confusion among clinicians. The aim of this study was to estimate the net effect of PE on the day of hcg on live birth rates by quantifying the effect of the most important known confounders. Moreover, a secondary aim was the exploration of the potential moderating effect of ovarian response on the effect of PE on live birth rates. Materials and Methods Population and ovarian stimulation This is a retrospective analysis of a cohort of fresh IVF/intracytoplasmic sperm injection (ICSI) cycles in which hcg was administered for triggering final oocyte maturation. These cycles were performed in a single IVF centre (Eugonia IVF Unit, Athens, Greece) during the period in which ovarian stimulation was done with gonadotrophins and gonadotrophinreleasing hormone (GnRH) analogues. Natural cycles and cycles where stimulation involved the administration of clomiphene were excluded. Furthermore, cycles in which a GnRH agonist was used for triggering final oocyte maturation (for avoidance of OHSS) were not included in this analysis, since in these cases a fresh embryo transfer is not performed. In order to reflect routine clinical practice, no other exclusion criteria were imposed on this dataset. In general, ovarian stimulation was performed by individualizing starting dose of gonadotrophins based on age, basal follicle-stimulating hormone (FSH) levels, body mass index (BMI), the presence of polycystic ovaries or not and previous response during stimulated cycles. Adjustments of gonadotrophin doses were performed after at least 5 days of stimulation and based on ovarian response as assessed by ultrasonography and serum estradiol levels. When follicular growth was deemed satisfactory, hcg ( IU urinary (u) hcg or 250 mg recombinant hcg and 5000 IU uhcg in case of OHSS risk) was administered to trigger final oocyte maturation and oocyte retrieval was performed h later. Fertilization was performed by either conventional IVF or ICSI and embryos were cultured for 2 6 days in sequential culture media (Origio, Malov, Denmark). Embryos were placed in fresh culture media on Day 3 in the case of extended culture. Embryo assessment was based on morphological evaluation. Embryos with 2 4 cells on Day 2 and embryos with 6 8 cells on Day 3 and,20% fragmentation were regarded as good-quality embryos (Grades 1 and 2) (Baczkowski et al., 2004). In the case of blastocysts, degree of expansion, inner cell mass and trophectoderm were graded according to a scoring system previously described (Gardner and Schoolcraft, 1999; Gardner et al., 2000). Embryo transfer (ET) was performed either at the cleavage (Day 2 or 3 of embryo development) or at the blastocyst stage (Day 5 or 6 of embryo development). The number of embryos replaced (range: 1 4 embryos) depended on the age of the patient and her previous history, as well as the developmental stage of the embryo and its quality. Luteal phase support

3 686 Venetis et al. was performed by administration of 600 mg vaginal micronized progesterone (Utrogestan, Laboratoires Besins International SA, France) from Day 2 to 3 post-oocyte retrieval until the 10th week of gestation, if pregnancy occurred. Serum progesterone values on the day of hcg did not affect the management of these cycles (e.g. no embryo transfers were cancelled due to elevated progesterone on the day of hcg). Outcome measures The primary outcome measure for this study was live birth defined as the delivery of a live infant after 24 weeks of gestation. Progesterone measurements Serum progesterone concentrations were determined routinely prior to initiation (Day 1 3 of the cycle) and during ovarian stimulation and on the morning of the day of hcg administration. During the study period samples were measured using an Immulite analyser and commercially available kits (DPC, Los Angeles, CA, USA). Analytical sensitivity was 0.2 ng/ml. Intra- and inter-assay precision at the concentrations, expressed as coefficients of variation were 7.9 and 10.0%, respectively. Ultrasound and other laboratory assays All ultrasound measurements were performed using a 7.5 or 6 or 5 MHz vaginal probe (Sonoline Adara, Siemens). FSH, LH, estradiol (E2) levels were measured using an Immulite analyser and commercially available kits (DPC). FSH, LH, E2 levels were measured prior to the initiation of stimulation (Day 1 3 of the cycle), while E2 and LH were also measured during later stages of ovarian stimulation according to the treating physician s preference. Analytical sensitivity was 0.1 miu/ml for FSH, 0.1 miu/ml for LH, 15 pg/ml for E 2. Intra- and inter-assay precision at the concentrations of most relevance to the current study (expressed as coefficients of variation) were 2.6 and 5.8% for FSH, 5.9 and 8.1% for LH, 6.3 and 6.4% for E2. Statistical analysis The distribution of continuous variables is described with the use of the mean and the standard deviation (SD). Categorical variables are presented as proportions and percentages of the total. Comparison of continuous variables among groups was performed with the use of the Student s t-test or Mann Whitney U-test depending on the normality of the distribution, while the Fisher s Exact test was used to compare categorical variables. Logistic regression analysis was used to evaluate the association between PE and the probability of live birth. Multivariable logistic regression analyses were performed in order to estimate the association of PE with the probability of live birth, while controlling for important confounders (Harrell, 2001; Hosmer and Lemeshow, 2004; Rothman et al., 2008). A quantitative assessment of the effect of each of the confounder on the association of PE with live birth rates was also performed. In order to check for a moderating effect of ovarian response on the association of PE with the probability of live birth while controlling for potential confounders (Jaccard and Turrisi, 2003; Rothman et al., 2008) an interaction term between the type of ovarian response [low responders (0 5 oocytes), normal responders (6 18 oocytes) and high responders (.18 oocytes)] (Griesinger et al., 2013) and PE was also introduced and evaluated into the logistic regression analyses. PE was defined as.1.5 ng/ml which is the most, frequently used cut-off used in the literature (Venetis et al., 2013). However, in order to check the robustness of the conclusions these analyses were repeated for the following cut-offs: 0.9, 1.2, 1.75, 2.0 and 2.25 ng/ml, which have also been used to define PE (Edelstein et al., 1990; Check et al., 1993; Miller et al., 1996; Xu et al., 2012). The confounders that were included in the multivariable logistic model were: (i) female age and the number of oocytes retrieved (Templeton et al., 1996; Sunkara et al., 2011), which have been consistently shown to be associated with the probability of live birth, (ii) the number of embryos transferred (Pandian et al., 2013) and the developmental stage of embryos (cleavage versus blastocyst) (Papanikolaou et al., 2008; Glujovsky et al., 2012), (iii) the presence of at least one good-quality embryo transferred, (iv) the woman s BMI, (v) the total dose of FSH administered during ovarian stimulation and (vi) the type of GnRH analogues used (agonists versus antagonists) during ovarian stimulation. The effect of these confounders was quantified by calculating the changes in the effect size of PE for live birth rates (Wang, 2007). More specifically, a sequential approach was followed in order to quantify the effect of each confounder while controlling for the bias introduced by other more potent confounders. This sequential approach is implemented by the statistical routine described by Wang (2007). When using regression analysis clustered robust standard errors were calculated in order to correct for the non-independence of data (Rogers, 1993; Williams, 2000). All statistical analyses were performed with the use of STATA (Version 13.1, College Station, TX: StataCorp LP) and the significance level for all statistical tests was P No adjustment for multiple statistical testing was performed and hence results of secondary analyses were interpreted in a hypothesis-generating context. Results The characteristics of the sample analysed in this study are presented in Table I. In brief, 2316 patients underwent 3296 fresh IVF/ICSI cycles, while in 3135 of these cycles embryo transfer were performed [95.1, 95% confidence interval (CI): ]. Live birth was the outcome in 793 cycles (25.3% per embryo transfer, 95% CI: ; Table II). Inhibition of premature LH surge was performed with either GnRH agonists (n ¼ 1303) or GnRH antagonists (n ¼ 1993). The majority of patients (n ¼ 2936; 89.1%) were stimulated with recombinant FSH (follitropin-a or follitropin-b), while 315 patients (n ¼ 9.6%) were stimulated with urinary FSH. Only a small fraction of this population received gonadotrophins containing LH activity (n ¼ 45, 1.3%). PE (.1.5 ng/ml) (PE) on the day of hcg was observed in 243 cycles (7.4, 95% CI: ) and no significant difference in the incidence of PE was observed between GnRH agonist and antagonist cycles (8.3 versus 6.8%, respectively; P ¼ 0.117). Significant differences between cycles with and those without PE were present regarding a number of variables (Table I). Cycles with PE were characterized by lower female age, slightly increased BMI, lower basal FSH and higher basal progesterone when compared with cycles with non-elevated progesterone. Duration of ovarian stimulation, total dose of FSH, E2 levels on the day of hcg and the number of follicles 11 mm on the day of hcg were all significantly higher in the cycles with PE (Table I). More oocytes were retrieved in cycles with PE than in cycles without PE, which produced more two-pronuclei (2PN) oocytes. Although a similar number of cycles with or without PE resulted in embryo transfer, significantly more embryos were transferred when PE was present (Table II). Association of PE with live birth rates Live birth rates were not significantly different between cycles with and those without PE when a bivariate analysis was performed (Table III), regardless of the threshold used to classify cycles as those with PE. However, when a multivariable analysis was performed (OR: 0.68, 95% CI: ) live birth rates were significantly decreased in the

4 Progesterone elevation and live birth rates 687 Table I Baseline characteristics and ovarian stimulation parameters. Parameter a Serum progesterone on day of hcg... P b Total 1.5 ng/ml (n ) >1.5 ng/ml (n 5 243) Age 35.8 (5.3) 34.9 (4.5) (5.2) BMI (kg/m 2 ) 23.7 (4.3) 22.6 (3.8), (4.3) Duration of infertility (years) 3.99 (3.5) 4.3 (3.4) (3.5) Previous assisted reproduction attempts 2.1 (2.9) 2.5 (3.3) (2.9) Basal FSH (miu/ml) 9.5 (5.5) 7.7 (4.3), (5.4) Basal LH (miu/ml) 5.5 (3.2) 5.4 (2.8) (3.2) Basal estradiol (pg/ml) 32.8 (15.4) 33.6 (16.7) (15.5) Basal progesterone (ng/ml) 0.59 (0.5) 0.68 (0.4) (0.5) GnRH analogue agonist/antagonist 1195/1858. (39.1/60.9%) 108/135 (44.4/55.6%) /1993 (39.5/60.5%) Duration of stimulation (days) 10.2 (2.0) 11.0 (1.5), (2.0) Total dose of FSH (IU) 2970 (1385) 3167 (1418) (1388) Estradiol on the day of hcg (pg/ml) 1522 (1037) 2639 (1414.7), (1108) Progesterone on the day of hcg (ng/ml) 0.81 (0.3) 1.97 (0.5), (0.5) Follicles 11 mm on the day of hcg 14.0 (9.4) 20.7 (9.9), (9.6) a All values are presented as mean (SD). b Student s t-test or Fisher s Exact test for differences between normal and elevated progesterone groups. Table II Outcome of ovarian stimulation, fertilization and embryo transfer. Parameter a Serum progesterone on day of hcg... P b Total 1.5 ng/ml (n ) >1.5 ng/ml (n 5 243) Oocytes retrieved 12.1 (8.7) 19.3 (10.1), (9.1) Type of fertilization IVF/ICSI n (%) 1246/1768 (41.3/58.7) 99/141 (41.3/58.7) /1909 (41.3/58.7) Fertilization rate 70.9 (25.6) 68.9 (24.5) (25.5) 2PN oocytes 6.6 (5.4) 9.9 (6.3), (5.6) Embryo transfer (ET) n (%) 2905 (95.1) 230 (94.7) (95.1) Number of embryos transferred 2.7 (0.9) 3.2 (0.8), (0.9) Developmental stage at ET cleavage/blastocyst, n (%) 2436/617 (79.8/20.2) 182/61 (74.9/25.1) /678 (79.4/20.6) At least one good-quality embryo transferred, n (%) 1849 (60.6) 122 (50.2) (59.8) Live birth, n (%) 744 (24.4) 49 (20.2) (24.1) a All values are presented as mean (SD) or count (%). b Student s t-test or Fisher s Exact test for differences between normal and elevated progesterone groups. PE (.1.5 ng/ml) group (Table III). The same was true for live birth rates in the 0.9 and 1.2 ng/ml thresholds, whereas live birth rates in the 1.75, 2.0 and 2.25 ng/ml thresholds were not significantly different between the two groups compared (Table III). The effect sizes (odds ratios, ORs) of PE on live birth rates observed in the multivariable analyses were in all cases smaller when compared with the bivariate analyses (Fig. 1). The confounding effect of each of the aforementioned confounding variables on the association of PE with live birth rates was quantified and is presented in Fig. 2. In the multivariable model, the number of oocytes retrieved was the mostpotent confounder, causing a 29.4% reduction in the OR of PE. Subsequently, the addition of female age resulted in an 11.4% increase in the OR of PE. The addition of the variable at least one good-quality embryo transferred also increased the OR of PE by 7.1%. The addition of the number of embryos transferred (23.6%), the developmental stage of the embryos (+1.8%), the type of GnRH analogue used (+1.6%), the BMI (21.3%) and the total dose of FSH (20.3%) caused a more subtle change in the OR of PE for live birth (Fig. 2). Is ovarian response a moderator of the effect of PE? Dividing the total sample of this analysis in subgroups based on the number of oocytes retrieved yielded three groups: (a),6 oocytes, n ¼ 796 cycles, (b) 6 18 oocytes, n ¼ 1770 cycles and (c).18 oocytes, 730 cycles. The incidence of PE (.1.5 ng/ml) in these subgroups

5 688 Venetis et al. Table III Association of PE with live birth rates using a bivariate and multivariable method of analysis (per different thresholds). Outcome PE thresholds (ng/ml) OR (95% CI)... Threshold (ng/ml) Bivariate Live birth (per OPU) 1.03 ( ) 0.85 ( ) 0.78 ( ) 0.87 ( ) 0.60 ( ) 0.65 ( ) Live birth (per ET) 1.02 ( ) 0.85 ( ) 0.78 ( ) 0.88 ( ) 0.64 ( ) 0.71 ( ) Multivariable Model Live birth (per ET) a 0.77 ( ) 0.62 ( ) 0.68 ( ) 0.77 ( ) 0.55 ( ) 0.69 ( ) Adjustment for non-independence of data has been performed for both the bivariate and the multivariable model. a Since the multivariable model includes the variable number of embryostransferred, at least one embryo of good-quality transferred and developmental stage of embryo at transfer only cycles in which ET was performed were analysed. Figure 1 OR for live birth between cycles with embryo transfer with or without PE (per various thresholds) on the day of hcg using a bivariate or a multivariable analysis of the same dataset controlling for female age, number of oocytes, number and developmental stage of embryos transferred, the transfer of at least one good-quality embryo, BMI, total FSH dose and the type of GnRH analogue used (vertical error bars represent 95% CIs). was 1.4% (95% CI: ), 6.3% (95% CI: ) and 16.5% (95% CI: ), respectively (Table IV). The association of PE with the probability of live birth according to ovarian response group, when the effect of important confounders was controlled for through multivariable logistic regression analyses, is depicted in Fig. 3. No statistically significant effect was detected in cycles with low (,6 oocytes, n ¼ 796 cycles) and high (.18 oocytes, n ¼ 730 cycles) ovarian response. A statistically significant detrimental effect of PE on live birth rates was detected only in cycles with intermediate response (6 18 oocytes, n ¼ 1770 cycles) when 0.9, 1.2 and 1.5 ng/ml were used as thresholds for PE (Fig. 3). In order to test for a moderating effect of the ovarian response category on the effect of PE on live birth rates, while controlling for important confounders, an interaction term (Response Category PE) was introduced in the multivariable logistic regression models (Table V). Figure 2 Graphical representation of the change in the estimate of the effect of PE (.1.5 ng/ml) on live birth rates with each variable added to the multivariable logistic regression model (in order of importance). Although, the ORs for live birth were increased in cycles with.18 oocytes retrieved when compared with cycles with 6 18 oocytes (Fig. 3), this difference was not found to be significantly different in none of the PE thresholds when an interaction analysis was performed (Table V). Similarly, cycles with,6 oocytes did not present significant differences when compared with cycles in which 6 18 oocytes were retrieved. Discussion This study has shown that in this large cohort of cycles although the association of PE (.1.5 ng/ml) with live birth rates was not significant when a bivariate analysis was performed, the same association was statistically significant (and of a largest magnitude) when multivariable analysis was performed. This finding strongly suggests that the effect of confounders on the association of PE with live birth rates is important and for this reason a multivariable approach should be employed when

6 Progesterone elevation and live birth rates 689 Table IV Distribution of cycles with or without PE according to the threshold used. Threshold used (ng/ml) <6 oocytes (n 5 796) 6 18 oocytes (n ) >18 oocytes (n 5 730) Total (n ) With PE, n (%) With PE, n (%) With PE, n (%) With PE, n (%) (19.9) 849 (48.0) 470 (64.4) 1477 (44.8) (4.9) 337 (19.0) 254 (34.8) 630 (19.1) (1.4) 112 (6.3) 120 (16.4) 243 (7.4) (0.9) 63 (3.6) 73 (10.0) 143 (4.3) (0.4) 30 (1.7) 29 (4.0) 62 (1.9) (0.3) 20 (1.1) 19 (2.6) 41 (1.2) Figure 3 OR for live birth after multivariable analysis according to ovarian response group (per various PE thresholds) corrected for female age, number of oocytes, number and developmental stage of embryos transferred, the transfer of at least one good-quality embryo, BMI, total FSH dose and the type of GnRH analogue used (vertical error bars represent 95% CIs). one wishes to more accurately estimate the net effect of elevated progesterone on the day of hcg on the live birth rate of patients after IVF. Based on the results of the present analysis, the bias seems to originate from the imbalance between the two groups compared in a number of parameters. Most importantly, PE has been shown to be associated with a higher number of oocytes retrieved not only in this analysis, but in most of the previously published studies (Bosch et al., 2010; Yding Andersen et al., 2011; Kyrou et al., 2012; Ochsenkühn et al., 2012; Huang et al., 2012b; Griesinger et al., 2013; Venetis et al., 2013). The number of oocytes retrieved is a known predictor of pregnancy (Sunkara et al., 2011). By controlling for this imbalance through multivariable regression analyses, the OR for live birth rate in PE versus non-pe cycles was decreased substantially. Controlling for the bias of female age, the transfer of at least one good-quality embryo and the number of embryos transferred resulted also in a significant adjustment on the effect size of the association between PE and live birth rates. The remaining covariates that were evaluated (developmental stage of embryos, BMI, total dose of FSH, type of GnRH analogue) had a smaller effect. These findings highlight the risk of reaching spurious conclusions when data derived from observational studies are not properly analysed. The observational nature of data renders this type of studies particularly vulnerable in numerous kinds of biases. Identifying the true effect an exposure might have on an outcome entails the proper design and analysis of such studies. Multivariable regression analysis aims at mathematically removing the effect of confounders or other forms of bias and thus providing a much better approximation of the true effect of the independent on the dependent variable (Grimes and Schulz, 2002; Hosmer and Lemeshow, 2004; Rothman and Greenland, 2005). This approach is particularly useful when examining associations such as the one of PE on the day of hcg with live birth rates, which can be assessed only through observational studies. Interestingly, a multivariable approach in order to identify the true effect of PE on pregnancy rates has been attempted only in a small fraction of the studies that have been published so far regarding this issue (Huang et al., 2012a,b; Griesinger et al., 2013). In the study by Bosch et al., (2010), the researchers opted to control for various confounders by stratification (Bosch et al., 2010). Thus, most of the studies that have evaluated this association have relied on bivariate analyses, which, as has been demonstrated in this study, can lead to erroneous conclusions. This might be a reason why many researchers have failed to confirm the detrimental effect of PE on pregnancy rates, leading to a long-lasting discrepancy in the relevant literature (Miller et al., 1996; Yding Andersen et al., 2011). It has also been suggested that ovarian response may play a moderating role on the effect that PE has on pregnancy rates. Recently, a pooled analysis of data from six trials (n ¼ 1890 cycles) by Griesinger et al. (2013) supported that PE (.1.5 ng/ml) does not seem to compromise ongoing pregnancy rates in high responders (.18 oocytes) (Griesinger et al., 2013). On the other hand, Bosch et al. (2010) showed that a detrimental effect of PE on ongoing pregnancy rates is present regardless of the magnitude of ovarian response (Bosch et al., 2010). Similarly, a study analysing the effect of PE according to the different ovarian responses supported that a negative effect of PE on pregnancy rates is present regardless of ovarian response, although a higher PE threshold should be used in high responders (Xu et al., 2012). In the present study, it was shown that an interaction between the category of ovarian response (,6, 6 18 and.18 oocytes) and PE could not be detected. Hence, the present analysis does not provide evidence to support a moderating effect of ovarian response on the association of PE with live birth rates. The subgroup analysis performed according to the ovarian response did not detect a significant difference in live birth rates between cycles with and without PE in the.18 oocytes group. Nevertheless, it should be noted that the fact that the ORs were uniformly,1 (Fig. 3)

7 690 Venetis et al. Table V Interaction between ovarian response category and the effect of PE on live birth rates (derived from a multivariable logistic regression model with live birth as a dependent variable and PE, ovarian response category, the interaction term: PE 3 ovarian response category, female age, number of oocytes, number and developmental stage of embryos as independent variables). Outcome PE thresholds (ng/ml) Ratio of OR for live birth (95% CI)... Threshold (ng/ml) ,6 versus 6 18 oocytes 0.67 ( ) 2.51 ( ) Not estimable Not estimable Not estimable Not estimable.18 versus 6 18 oocytes 0.99 ( ) 4.08 ( ) 1.88 ( ) 1.94 ( ) 1.95 ( ) 2.52 ( ) All analyses have been performed while adjusting for the non-independence of data. and that other two groups were substantially smaller in size than the 6 18 group where a significant difference was detected, might suggest that a detrimental effect could be indeed present in the population and the finding of the present analysis constitutes a type II error. Under the same prism, the fact that the ORs in the.18 oocytes group were consistently higher when compared with the 6 18 oocytes group (Fig. 3), might indicate that this group of patients is less affected by PE than the 6 18 oocytes group. However, these are hypotheses that should be tested in future sufficiently powered studies. An issue of the current study that needs to be discussed is that not all potential confounders were included in the multivariable model and thus residual confounding may be present. The number of oocytes remained the most important confounder and this is in accordance with prior knowledge on the independent association of the number of oocytes with the probability of live birth and PE. Nevertheless, it should be emphasized that, despite the high number of potential covariates used in the multivariable analyses, residual or unmeasured confounding might still be present (e.g. the data derive from over a decade of clinical practice). The results of the current study indicate that researchers and clinicians trying to assess the effect of PE on live birth rates in their population and/ or identify the optimal cut-off in their own setting, should utilize a multivariable approach, so that the true estimate of the underlying association is revealed, devoid of any major confounding effects (such as the difference in the number of oocytes retrieved). In this way, they will be able to more accurately predict the prognosis of the cycle and decide on whether embryo transfer should be performed immediately or deferred to a frozen thawed cycle where the endometrial receptivity has not been compromised by elevated progesterone during the stimulated cycle (Shapiro et al., 2010, 2011; Venetis et al., 2013). In conclusion, this study highlights the magnitude of the potential confounding bias that can be present when assessing the effect of PE on the day of hcg on live birth rates and the importance of adopting a methodology able to control for confounding effects from multiple sources. The use of a multivariable approach will allow clinicians to better estimate the net effectofpe onthe probabilityof live birthrateafter IVF and hence make properly informed choices regarding the management of their patients. Authors roles C.A.V.: Conceived the idea of the study, contributed to the construction of the protocol, performed the analyses and interpretation of the data and drafted the manuscript. E.M.K.: Reviewed the protocol, contributed to data extraction, contributed to the analyses and the interpretation of the data, revised the manuscript for important intellectual content. J.K.B.: Contributed to data extraction and the interpretation of the data, revised the manuscript for important intellectual content. G.T.L. and I.A.S.: Contributed to data extraction and the interpretation of the data, revised the manuscript for important intellectual content. B.C.T.: Reviewed the protocol of the study, contributed to the interpretation of the data and revised the manuscript for important intellectual content. T.G.L.: Contributed to the data extraction, interpretation of the data and revised the manuscript for important intellectual content and had overall supervision of the study. All authors approved the final version of the manuscript. Funding No external funding was used for this study. Conflict of interest None of the authors has any conflict of interest to declare. References Baczkowski T, Kurzawa R, Glabowski W. Methods of embryo scoring in in vitro fertilization. Reprod Biol 2004;4:5 22. Bosch E, Labarta E, Crespo J, Simon C, Remohi J, Jenkins J, Pellicer A. Circulating progesterone levels and ongoing pregnancy rates in controlled ovarian stimulation cycles for in vitro fertilization: analysis of over 4000 cycles. Hum Reprod 2010;25: Check JH, Lurie D, Askari HA, Hoover L, Lauer C. The range of subtle rise in serum progesterone levels following controlled ovarian hyperstimulation associated with lower in vitro fertilization pregnancy rates is determined by the source of manufacturer. Eur J Obstet Gynecol Reprod Biol 1993; 52: Edelstein MC, Seltman HJ, Cox BJ, Robinson SM, Shaw RA, Muasher SJ. Progesterone levels on the day of human chorionic gonadotropin administration in cycles with gonadotropin-releasing hormone agonist suppression are not predictive of pregnancy outcome. Fertil Steril 1990; 54: Fanchin R, Righini C, Olivennes F, de Ziegler D, Selva J, Frydman R. Premature progesterone elevation does not alter oocyte quality in in vitro fertilization. Fertil Steril 1996;65:

8 Progesterone elevation and live birth rates 691 Fanchin R, Righini C, Olivennes F, Ferreira AL, de Ziegler D, Frydman R. Consequences of premature progesterone elevation on the outcome of in vitro fertilization: insights into a controversy. Fertil Steril 1997; 68: Gardner DK, Schoolcraft WB. Culture and transfer of human blastocysts. Curr Opin Obstet Gynecol 1999;11: Gardner DK, Lane M, Stevens J, Schlenker T, Schoolcraft WB. Blastocyst score affects implantation and pregnancy outcome: towards a single blastocyst transfer. Fertil Steril 2000;73: Glujovsky D, Blake D, Farquhar C, Bardach A. Cleavage stage versus blastocyst stage embryo transfer in assisted reproductive technology. Cochrane Database Syst Rev 2012;7:CD Griesinger G, Mannaerts B, Andersen CY, Witjes H, Kolibianakis EM, Gordon K. Progesterone elevation does not compromise pregnancy rates in high responders: a pooled analysis of in vitro fertilization patients treated with recombinant follicle-stimulating hormone/gonadotropinreleasing hormone antagonist in six trials. Fertil Steril 2013;100: e Grimes DA, Schulz KF. Bias and causal associations in observational research. Lancet 2002;359: Harrell FE. Regression Modeling Strategies: with Applications to Linear Models, Logistic Regression, and Survival Analysis. Berlin: Springer, Hosmer DW Jr, Lemeshow S. Applied Logistic Regression. New Jersey, NJ: John Wiley & Sons, Huang CC, Lien YR, Chen HF, Chen MJ, Shieh CJ, Yao YL, Chang CH, Chen SU, Yang YS. The duration of pre-ovulatory serum progesterone elevation before hcg administration affects the outcome of IVF/ICSI cycles. Hum Reprod 2012a;27: Huang R, Fang C, Xu S, Yi Y, Liang X. Premature progesterone rise negatively correlated with live birth rate in IVF cycles with GnRH agonist: an analysis of 2566 cycles. Fertil Steril 2012b;98: Jaccard J, Turrisi R. Interaction Effects in Multiple Regression, 2nd edn, Thousand Oaks, CA: Sage, Kolibianakis EM, Venetis CA, Bontis J, Tarlatzis BC. Significantly lower pregnancy rates in the presence of progesterone elevation in patients treated with GnRH antagonists and gonadotrophins: a systematic review and meta-analysis. Curr Pharm Biotechnol 2012;13: KyrouD,Al-AzemiM,PapanikolaouEG, Donoso P, Tziomalos K, Devroey P, Fatemi HM. The relationship of premature progesterone rise with serum estradiol levels and number of follicles in GnRH antagonist/recombinant FSH-stimulated cycles. Eur J Obstet Gynecol Reprod Biol 2012;162: Miller KF, Behnke EJ, Arciaga RL, Goldberg JM, Chin NW, Awadalla SG. The significance of elevated progesterone at the time of administration of human chorionic gonadotropin may be related to luteal support. J Assist Reprod Genet 1996;13: Ochsenkühn R, Arzberger A, von Schönfeldt V, Gallwas J, Rogenhofer N, Crispin A, Thaler CJ, Noss U. Subtle progesterone rise on the day of human chorionic gonadotropin administration is associated with lower live birth rates in women undergoing assisted reproductive technology: a retrospective study with 2555 fresh embryo transfers. Fertil Steril 2012; 98: Pandian Z, Marjoribanks J, Ozturk O, Serour G, Bhattacharya S. Number of embryos for transfer following in vitro fertilisation or intra-cytoplasmic sperm injection. Cochrane Database Syst Rev 2013;7:CD Papanikolaou EG, Kolibianakis EM, Tournaye H, Venetis CA, Fatemi H, Tarlatzis B, Devroey P. Live birth rates after transfer of equal number of blastocysts or cleavage-stage embryos in IVF. A systematic review and meta-analysis. Hum Reprod 2008;23: Rogers W. Regression standard errors in clustered samples. Stata Technical Bull 1993;13: Rothman KJ, Greenland S. Causation and causal inference in epidemiology. Am J Public Health 2005;95(Suppl 1):S144 S150. Rothman KJ, Greenland S, Lash TL. Modern Epidemiology. Philadelphia, PA: Lippincott Williams & Wilkins, Shapiro BS, Daneshmand ST, Garner FC, Aguirre M, Hudson C, Thomas S. Embryo cryopreservation rescues cycles with premature luteinization. Fertil Steril 2010;93: Shapiro BS, Daneshmand ST, Garner FC, Aguirre M, Hudson C, Thomas S. Evidence of impaired endometrial receptivity after ovarian stimulation for in vitro fertilization: a prospective randomized trial comparing fresh and frozen thawed embryo transfer in normal responders. Fertil Steril 2011;96: Sunkara SK, Rittenberg V, Raine-Fenning N, Bhattacharya S, Zamora J, Coomarasamy A. Association between the number of eggs and live birth in IVF treatment: an analysis of treatment cycles. Hum Reprod 2011;26: Templeton A, Morris JK, Parslow W. Factors that affect outcome of in-vitro fertilisation treatment. Lancet 1996;348: Venetis CA, Kolibianakis EM, Papanikolaou E, Bontis J, Devroey P, Tarlatzis BC. Is progesterone elevation on the day of human chorionic gonadotrophin administration associated with the probability of pregnancy in in vitro fertilization? A systematic review and meta-analysis. Hum Reprod Update 2007;13: Venetis CA, Kolibianakis EM, Bosdou JK, Tarlatzis BC. Progesterone elevation and probability of pregnancy after IVF: a systematic review and meta-analysis of over cycles. Hum Reprod Update 2013;19: Wang Z. Two post-estimation commands for assessing confounding effects in epidemiological studies. Stata J 2007;7: Williams RL. A note on robust variance estimation for cluster-correlated data. Biometrics 2000;56: Xu B, Li Z, Zhang H, Jin L, Li Y, Ai J, Zhu G. Serum progesterone level effects on the outcome of in vitro fertilization in patients with different ovarian response: an analysis of more than cycles. Fertil Steril 2012; 97: e1324. Yding Andersen C, Bungum L, Nyboe Andersen A, Humaidan P. Preovulatory progesterone concentration associates significantly to follicle number and LH concentration but not to pregnancy rate. Reprod Biomed Online 2011; 23: