Nonlinear relationship between gonadotropin total dose applied and live birth rates in non-PCOS patients: a retrospective cohort study

The purpose of this article is to explore the relationship between the total dose of follicle-stimulating hormone (FSH) applied during controlled ovulation stimulation and the live birth rates (LBRs) in non-PCOS population. Many studies have found no difference between the dose of FSH application and pregnancy outcomes such as clinical pregnancy rates after fresh embryo transfer. However, a recent large retrospective analysis found a negative correlation between live birth rates and increasing dose of FSH. It is still controversial about the association between FSH dose and LBRs. In addition, no studies have yet explored the nonlinear relationship between FSH and LBRs. This cohort study included a total of 11,645 patients who had accepted IVF/intracytoplasmic sperm injection (ICSI) at the second hospital of Hebei medical university between December 2014 to December 2019. PCOS was identified by Rotterdam PCOS criteria. We researched the association between FSH total dose and live birth rates (LBRs) using multivariate regression analysis. In addition, a model for nonlinear relationships based on a two-part linear regression was applied. The analysis of threshold effects indicated that LBR increased with every 1000 IU FSH when the concentration of FSH was lower than 1410 IU (OR 1.55, 95% CI [1.05, 2.28]); however, a negative association between FSH dose and LBR (OR 0.94, 95% CI [0.89, 0.99]) was found when the FSH total dose was higher than 1410 IU. It is worth noting that the relationship between LBR and FSH dose varied among patients of different ages (OR 0.92 vs 1.06, P for interaction < 0.05).

www.nature.com/scientificreports/no statistically significant difference in cumulative live birth rates between the two groups receiving different FSH doses 24 .However, few studies with adequate sample size have investigated whether FSH applied provides an independent factor affecting LBRs from fresh cycles in non-PCOS women.
Given the contradictory evidence above, there is growing interest in the influence of the dose of FSH on live births.How to determine the dose of FSH during COS is necessary to clinician's judgement.This paper aimed to research the relationship between total FSH dose and live births in the non-PCOS population and to explore whether there is a dose-response inflection point for live births and FSH applied.

Study population and design
Our study population includes all autologous IVFs between December 2014 and December 2019.Cancellation of oocytes retrieval cycles, donor oocytes cycles, freeze-thaw cycles, embryo transfer was not performed for various reasons (lack of good quality embryos, endometrial abnormalities, etc.), patients with PCOS, and missing data on pregnancy outcome were not included.The detailed flow chart is shown in Fig. 1.
The ovarian stimulation protocol has been introduced previously 25 .In brief, ovarian stimulation in fresh cycles was achieved with combination of gonadotropin-releasing hormone agonist, antagonist and recombinant follicle-stimulating hormone.Serial ultrasound examinations and hormone measurements were performed the ovarian response.All patients received recombinant FSH and/or HMG for COS.The dose of HMG was determined based on multiplying vials number of HMG given by 75 IU and adding this to the total amount of FSH given.When at least 1 follicle ≥ 18 mm in diameter or 3 follicles ≥ 17 mm in diameter were monitored in both ovaries during ovulation promotion, 10,000 units of human chorionic gonadotropin (hCG) injection was given.Oocytes were retrieved under transvaginal ultrasound guidance 36-37 h later.For patients with possible ovarian hyperstimulation syndrome, the embryos are frozen.
Basic demographic characteristics include patient age, BMI, antral follicle count, and type of infertility.Embryo indicators include the number of embryos transferred, stage of embryo culture.Outcome indicators include number of oocytes retrieved, clinical pregnancy rate, and LBRs.
When evaluating the cleavage stage embryos, the selected indicators mainly include the number of blastomeres and the fragmentation rate 26 .Morphological scoring is mainly based on Gardner and Schoolcraft scoring system 27 .Fresh embryo transfer was conducted on the 3rd to 5th day after retrieval.Progesterone gel (Crinone, Merck Serono, Watford, UK) was administrated for luteal phase support starting from the oocyte retrieval day until 14 days after ET.

Outcome
A positive blood hCG test was performed 14 days after embryo transfer and a positive result was regarded as a biochemical pregnancy.Ultrasound was performed 30 days after embryo transfer and the presence of a gestational sac was regarded as clinical pregnancy.Live birth was defined as the birth of at least one living child.

Covariates assessment
Covariates were chosen based on clinical experience and studies published in recent years, with the following covariates being included in the model.The continuous variables mainly include female age (in years), infertility duration (in years), body mass index (BMI, kg/m2), and endometrial thickness on the hCG trigger day (mm).The categorical variables mainly included infertility type (primary or secondary infertility), fertilization method (IVF, ICSI or IVF&ICSI), and type of embryo transferred (cleavage stage or blastocyst stage).

Treatment procedures
Starting in the mid-luteal phase, the patient was given 0.1 mg/day of Decapeptyl (Ferring AG, Dübendorf, Switzerland) as part of the gonadotropin releasing hormone agonist (GnRH-a) protocol.After about 14 days, gonadotropin was given until the hCG trigger day once down-regulation standards (LH < 5 IU/L, E2 < 50 pg/L, FSH < 5 IU/L, thickness of endometrium < 5 mm, follicle diameter and no functional ovarian cyst) were satisfied by serum endocrinology and transvaginal ultrasound examinations.On the second day of the menstrual cycle, patient received 3.75 mg of triptorelin (Ipsen Pharma Biotech, Signes, France) as part of the prolonged GnRH agonist protocol.Once pituitary-ovarian suppression was established, gonadotrophin was administered 28-31 days later.Throughout therapy, the ovarian response was tracked using vaginal ultrasonography and serum hormone levels were measured.
The GnRH antagonist (GnRH-ant) protocol began on day 2 or 3 of menstruation.Daily injections of 0.25 mg Cetrotide (Baxter Oncology GmbH, Frankfurt, Germany) were administered until the hCG trigger day once the leading follicle reached 14 mm diameter.
The injection used as the hCG trigger was either 250 µg of recombinant human follitropin alfa (MerckSerono S.p.A., Geneva, Switzerland) or 8000-10,000 IU of hCG (Lishenbao, Livzon Pharmaceutical Co., Ltd.).The trigger day was the combination of > three follicles with > 17 mm diameter, or > 18 mm diameter in one dominant follicle with acceptable serum hormone levels.On the day of the hCG trigger, patients had blood testing with an empty stomach.
Under the supervision of vaginal ultrasound guidance, posterior fornix puncture and oocyte retrieval were carried out 36-37 h after the hCG triggering.
Blastomere number, degree of fragmentation, embryo symmetry, multinucleation, and degree of cell fusion were used to score the embryos after 72 h.The embryo grading criteria was used to determine which embryos were chosen for transfer 28 .A fresh ET (with no more than two embryos) might be carried out if the endometrium was likewise in good condition (thickness ≥ 8 mm; acceptable morphology) and there were no transfer-related contraindications.On the day of oocyte retrieval, luteal support was initiated with the use of oral contraceptive tablets (10 mg twice daily) and 8% progesterone sustained-release vaginal gel (Xenoto, Merck Serono, Germany), in accordance with consensus 29 .The progesterone and estrogen dosages didn't alter until 14 days after ET, when the blood β-hCG level was measured.

Statistical analysis
When analyzing the baseline characteristics, the method of mean ± SD is applied to the measurement data, while the method of counting and proportion is applied to the categorical variables.Chi-square tests (for categorical variables), analysis of covariance (normally distributed continuous variables) or a Kruskal-Wallis test (nonnormally distributed continuous variables) were used to compare the difference between groups.In this study, we aimed to explore the association between FSH total dose and LBRs based on logistic regression analysis.Crude odds ratios (OR) and adjusted OR with 95% confidence interval (CI) were calculated.Tertiary groups were applied to flexibly model and visualize the association between the FSH total dose and LBR.
The generalized estimating equation (GEE) framework was used to construct univariable and multivariable regression models for the prediction of live birth.This was done to control for the non-independence of data since some patients contributed more than one cycles in the dataset analyzed.In the process of regression analysis, the result of smooth curve fitting is used to determine whether the independent variable is divided into intervals and general addictive model (GAM) was allied using R packages: doBy, plotrix, stringi, stringr, survival, rms, nnet, car, mgcv, gdata, Hmisc, SparseM, carData, nlme, lattice, Formula, ggplot2.Under the condition of dividing into intervals, each interval is fitted based on the piecewise regression method.When comparing non-segmented and single-segment models, the log-likelihood ratio method is used to determine whether there is a threshold based on the results obtained.According to the model analysis results, the maximum likelihood of the inflection point of the connecting line segment is determined, and the two-step recursive method is applied in the regression (using R packages: doBy, plotrix, stringi, stringr, survival, rms, nnet, car, mgcv, MASS, mgcv, gdata, geepack).
Step 1: Narrow the inflection point to the 10% range of independent variables, and increase it by 5% in the range of 5% to 95%, so that 19 percentage points can be determined, and then set these points as inflection points, so as to obtain the segmented regression model, and compare the results to obtain the maximum likelihood model corresponding to a point.The inflection point is reduced to the ± 4% range of the percentile, and the inflection point determined according to this is marked as Kmin and Kmax.
Step 2: Analyze based on the recursive method to determine the exact inflection point between the two inflection points.The corresponding process is as follows: first, run the model of three inflection points within the two inflection points, and compare the results to determine the maximum likelihood model corresponding to the highest quartile.Then reduce the range of Kmin and Kmax to ± 25% of the corresponding quartile, and carry out iterative operation continuously to finally obtain the specific value of the target variable.Set this point as the inflection point, and the piecewise regression can obtain the optimal result.

Statement
The studies involving human participants were reviewed and approved by the ethics committee of the Second Hospital of Hebei Medical University.The patients/participants provided their written informed consent to participate in this study.All methods were carried out in accordance with STROBE guidelines and regulations.

Results
In this study, 11,645 fresh embryo transfer cycles were analyzed.All cycles were classified into three groups according to the tertile of the FSH dose during ovulation promotion.

Characteristics of the patients
Baseline characteristics are detailed in Table 1.Women and men in the high FSH dose group were older, with higher BMI, and lower basal AFC.Regarding ovarian stimulation protocols, the prolonged GnRH-agonist protocol was applied more often in women who were given higher FSH doses than those were given lower FSH doses.www.nature.com/scientificreports/Table 2 shows the pregnancy outcomes by FSH dose tertiles divided into three groups.There was obvious difference (P < 0.001) in LBR among three groups, 44.5%, 43.5% and 39.7% respectively.The LBR was obviously lower in group with the highest applied FSH dose than in the reference group (OR 0.82, 95%CI [0.75, 0.90]).Biochemical and clinical pregnancies were also notable lower compared to that of the reference group.Pregnancy loss increased in the tertile 2 and tertile 3 groups than the reference group, but with no significance (P = 0.612).

Association between FSH dose and LBR
As shown in Table 3, multivariate regression analysis was performed to adjust for possible confounding factors.A negative association was found between FSH total dose and LBR in the crude model with every 1000 IU increase of FSH dosing (OR 0.91, 95% CI [0.87, 0.95]).According to the comparison result, the multivariable regression analyses exhibits no obvious difference after adjusting the factor for female age and BMI in model I (OR 0.97, 95% CI [0.92, 1.01]), and adjusting for female age, infertility duration, BMI, infertility type, stimulation protocol, Endometrium thickness on hCG trigger day, embryo stage and method of fertilization in model II (OR 0.96, 95% CI [0.92, 1.01]).
However, the smoothing plot demonstrated an obvious relationship between LBR and FSH total dose after adjusting for covariates, including female age, infertility duration, BMI, infertility type, stimulation protocol, Endometrium thickness on hCG trigger day, embryo stage, and method of fertilization.(Fig. 2).With the increase of the FSH dose, the risk of LB increased.When the FSH dose was about 1410 IU, LBR started to decrease (Fig. 2).
In the subsequent study, the two-stage linear regression model was selected for analysis to determine the threshold effect, and the regression results were sorted out as shown in Table 4.The results of this table can be traced, and the inflection point of all variables is 1.41(1000 IU).On the left side of the inflection point, the OR, 95% CI and P values were 1.55, 1.05-2.28and < 0.05 respectively; On the right side, the three indicators are 0.94, 0.89-0.99respectively; P < 0.05.We also performed a linear regression model with the result (OR 0.96, 95%CI [0.92-1.01],p > 0.05), indicating that the regression coefficient of FSH total dose on LBR was not significant.The log likelihood ratio check result was 0.013 < 0.05, which reflected that the two-stage linear regression model was better than the one-line linear regression model.
Moreover, as shown in Table 5 there was interaction between the female age and FSH total dose (P for interaction < 0.05).Results were different for the patients with more or less than 1 year infertility duration (OR 0.94 vs 1.06, P for interaction = 0.089), although the p-value for interaction was not significant.However, among patients with different BMI (OR 0.94 vs 0.97, P for interaction = 0.247), different AFCs (OR 0.93 vs 1.00, P  for interaction = 0.390), different endometrium thickness on hCG trigger day (OR 0.97 vs 0.88, P for interaction = 0.597), primary or secondary infertility (OR 0.96 vs 0.96, P for interaction = 0.322), different protocols (OR 0.86, 0.92 and 0.97, P for interaction = 0.094),and different method of fertilization (OR 0.93 vs 0.97, P for interaction = 0.200), the stratification factors failed to have an interaction effect with FSH total dose in LBR.

Discussion
This paper found that the live birth rate increased and then decreased when total FSH dose increasing during COS and found an inflection point.
In relevant studies, more than two doses of gonadotropins (including hMG) were investigated.As the FSH dose increased, the number of oocytes retrieved increased and the number of available embryos increased 11,12 , explaining the positive correlation in the first part of our study results-curve.
One of the highlights of this article was the joint effect of women age and FSH dose on live birth was significant.It is known that FSH high dose in women with normal ovarian reserve showed no significant differences in fertilization rates, embryo numbers or pregnancy rates between the groups [30][31][32] .In contrast, for older or lowresponse patients, it is generally accepted that increasing the FSH dose is an appropriate strategy to improve the number of oocytes retrieved.Our study found a positive correlation between their live birth rate and the total FSH dose applied in women aged > 35 years, which was obviously different from younger patients.This is in accordance with the results of several studies 21,33,34 .However, other studies have concluded that high FSH dose do not overcome the negative effect of increasing age to improve the pregnancy outcome of ART 30,35 .In addition, for the low response population, although a positive correlation was found between LBR and total FSH dose in women in the AFC < 5 group, the results were not statistically significant.The most recent study found  www.nature.com/scientificreports/ that higher dose of FSH has no obvious influence on transferable embryos, pregnancy rates or live birth rates in low response population 10,13,14,17,33,36 .In conclusion, poor ovarian response patients may not benefit from higher FSH doses.In addition, our center did not include records of serum AMH until 2019, as too little data for AMH to provide meaningful reference.Therefore, these predictors of ovarian responsiveness 9,17 that may be used by physicians to determine dose could not be included in this analysis.We also stratified BMI, AFC, endometrium thickness on hCG trigger day, type of infertility, different protocols, and fertilization method and found that FSH dose was inversely related to LBRs and the results were consistent across stratification.
In the latter part of the curve, the LBRs decreases meaningfully with higher FSH doses, by less than 10%.Our study demonstrated that total FSH doses higher than 1410 IU may be excessive and detrimental to the LBRs.This is in accordance with the growing retrospective correlation studies that also show that high FSH doses are detrimental to IVF success and live birth rates 23,[37][38][39] .Some animal studies showed that high FSH doses may impair oocyte quality and maturation [18][19][20] .In addition, the adverse effects of exogenous gonadotropins on embryonic development are best demonstrated in rodent models.In vivo and in vitro researches suggested that ovarian stimulation may delay the process of single-or two-celled mouse embryos into blastocysts [40][41][42] .It has been shown that FSH is important for oocyte-granulosa cell interactions during follicular development 18 .In FSH-deficient female mice, oocyte growth was inefficient, although normal mitochondrial DNA accumulation and Kit ligand-encoding messenger RNA splice forms (designed Kitl1 and Kitl2) were found 42 .FSH regulates connexins, the principal gap junction proteins, and cadherins, the cell-cell attachment proteins in both granulosa cells and oocytes 43 .
However, a comparison between natural cycles and GnRH agonist protocols in infertile women had no differences in oocyte cleavage rates, developmental capacity or embryo fragmentation rates 44 .With regard to embryonic chromosomes, it has been shown that in vitro mature mouse oocytes with high levels of FSH exhibit accelerated nuclear maturation and increased aneuploidy 19 .For human meiotic and mitotic chromosome 21 division errors were also associated with higher FSH doses 45 .Exogenous FSH may increase the aneuploidy rate of embryos 46 .Therefore, we consider that mild stimulation appears to have less impact on the process of www.nature.com/scientificreports/nuclear maturation and chromosome segregation, and no adverse effects on pregnancy outcome have been detected 47 .Regarding oocyte development, high doses of exogenous FSH may result in morphological abnormalities in oocytes, such as abnormalities in zona pellucida (ZP), cytoplasm and polar bodies of brown granulosa cells.In contrast, pregnancy rates were lower in brown oocytes, suggesting that brown oocytes affect embryo development 48 .FSH treatment causes transzonal projections (TZPs) to subside, altering oocyte transcriptional activity and meiotic capacity, which may lead to abnormal oocyte development 18 .Moreover, premature luteinization was recorded after patients were given high doses of FSH during ovarian stimulation, leading to significant fluctuations in the follicular microenvironment 49,50 .Therefore, this study found a negative correlation in pregnancy outcome when the total FSH dose exceeded a certain dose.It was also found that frozen embryo transfer reversed the adverse influence of high FSH on LBR in fresh cycles, indirectly suggesting that it may be the endometrium receptivity that is adversely affected by the FSH dose 51,52 .Nevertheless, the mechanisms by which high FSH doses alter follicular development and endometrial tolerance need to be further explored.
It is worth noting that there are three main protocols during ART with varying influence on endogenous gonadotropins.In the prolonged GnRH-agonist protocol, endogenous FSH and LH production is inhibited.Conversely, at the start of the GnRH-agonist protocol, endogenous FSH and LH production is increased.But in the GnRH-antagonist protocol, endogenous FSH and LH is at basal levels.The choice of protocol varies from person to person due to the different response to exogenous gonadotropins and different body weight.However, we did not find a significant difference in the relationship between LBRs and total FSH dose in our stratified analysis and interaction test.
The proper explanation for the relationship between FSH dose applied during COS and LBRs seen in nonrandomized researches may associated with patient features such as lower response level to FSH 53 , which may impact LBRs and the dosing of FSH.We were unable to judge whether the dose decision was caused by a previous response to FSH.
This study has some limitations.Firstly, the difference of rFSH and uFSH was not discussed.The relevant meta-analysis contained 14 trials with 1726 women found no difference in pregnancy and LBRs between rFSH and uFSH, nor between the hMG preparation group and the urinary purified FSH group 54 .Given that purified uFSH has partial LH activity but rFSH does not, the confounding factors involved need to be further investigated.Secondly, our study excluded patients with PCOS, a population whose hypersensitive response to exogenous gonadotropins would lead clinicians to individualize treatment 55,56 .Thirdly, this study was retrospective and there were some differences in baseline indicators.Additionally, although some baseline variables were included in the adjustment, the possibility of residual confounding privacy, such as AMH, cannot be ruled out.Hence, it is necessary to obtain high quality randomized controlled trials to explore the associate between total FSH dose applied in COS and LBR.Moreover, this study is a single-center study and more studies from other centers are needed to enhance its credibility and replicability.

Conclusions
Our results indicated a non-linear association between the total dose of FSH application and LBR.In addition, there was an interaction between women's age and the total dose of FSH application.Therefore, more studies are needed to validate our results and hypotheses and to further investigate the underlying mechanisms.

Figure 1 .
Figure 1.Flow chart for selection of patients from December 2014 to December 2019.

Figure 2 .
Figure 2. Association between live birth and FSH total dose applied in controlled ovarian stimulation.A threshold, nonlinear association between live birth and FSH total dose was found in a generalized addictive model (GAM).Solid red line represents the smooth curve fit between these two variables.Blue lines represent the 95% of confidence interval from the fit.

Table 1 .
Demographic characteristics of the participants.Normal distribution of data was presented as mean ± standard deviation; nonnormal distribution of data was presented as median (interquartile range) and categorical data using number (percentage).BMI body mass index, FSH follicle stimulating hormone, P progesterone, E2 estradiol, LH luteinizing hormone, AFC antral follicular count, hCG human chorionic gonadotropin, GnRH gonadotropin releasing hormone, IVF in vitro fertilization, ICSI intracytoplasmic single sperm injection.

Table 2 .
Demographic characteristics of the participants stratified by outcome.

Table 3 .
Multivariable regression analysis models examining the association between FSH total dose and LBR.Model I: adjusted for female age and BMI.The generalized estimating equation model was applied as the adjusted Model I. Model II: following factors were adjusted: female age, infertility duration, BMI, infertility type, stimulation protocol, endometrium thickness on hCG trigger day, embryo stage, and method of fertilization.The generalized estimating equation model was applied as the adjusted Model II.

Table 4 .
The result of two-piecewise linear regression model.Model I, linear analysis; Model II, nonlinear analysis.LRT test, logarithmic likelihood ratio test (p value < 0.05 indicates that Model II is significantly differrent from Model I, which indicates a nonlinear relationship); adjustment variables: female age, BMI, infertility duration, infertility type, endometrium thickness on hCG trigger day, embryo stage, and method of fertilization.BMI, body mass index; hCG, human chorionic gonadotropin.

Table 5 .
Stratification analysis in different subgroups.The following factors, except the stratification factor itself, were adjusted in the multivariable analysis: female age, BMI, infertility duration, infertility type, endometrium thickness on hCG trigger day, embryo stage, and method of fertilization.BMI body mass index, AFC antral follicular count, Gn gonadotropin, P progesterone, GnRH gonadotropin releasing hormone, IVF in vitro fertilization, ICSI intracytoplasmic single sperm injection.