{"paper_id":"01133629-6eba-4b6d-9558-92b8aa2806ab","body_text":"The World Health Organization estimates that infertility affects around 10 to 15% of\ncouples worldwide. Although there is no specific data for Portugal, several studies\nshow that in Western countries infertility affects around 14% of the population,\nleading many couples to resort to Assisted Reproductive Technology (ART) techniques\n( DGS, 2010 ).\nHowever, ART techniques do not guarantee success. In the 2015 Portuguese report on\nART activity it is shown that 5886 cycles of in vitro fertilization (IVF) /\nIntracytoplasmic Sperm Microinjection (ICSI) were started, resulting in 1546\nclinical pregnancies and 1379 live newborns ( CNPMA,\n2015 ).\nThe ovarian response to controlled ovarian stimulation is one of the crucial points\nfor the success of second-line ART techniques. Predicting ovarian response prior to\nthe start of stimulation is of great importance, as it not only allows the\nstimulation protocol to be adapted to maximise the predicted response, but also\nallows couples to be better advised.\nA poor response to controlled ovarian stimulation usually results in a reduced number\nof oocytes. However, the classification of patients as “poor responders” is not\nconsensual ( Ferraretti & Gianaroli, 2014 ).\nTo address the lack of a standardised definition, the European Society of Human\nReproduction and Embryology (ESHRE) published the Bologna criteria, which aimed to\ndefine which women should be considered poor responders ( Ferraretti  et al. , 2011 ), and the POSEIDON\n(Patient Oriented Strategies Encompassing IndividualizeD Oocyte Number) group also\nproposed a new classification ( Esteves  et\nal. , 2019 ).\nBoth classifications place great importance on the quantitative assessment of a\nwoman’s ovarian reserve. Various markers of ovarian reserve have been studied, in\nparticular the antral follicle count (AFC) by transvaginal ultrasound and serum\nanti-Müllerian Hormone (AMH) ( Ferraretti\n et al. , 2011 ;  Nelson, 2013 ;  Li  et\nal. , 2016 ;  Liu  et\nal. , 2023 ). When compared to each other, AMH and AFC have\nvery similar predictive potential, however, the substantial intraand inter-observer\nvariability and reduced reproducibility of AFC have favoured the use of AMH ( Satwik  et al. , 2012 ;  Vural  et al. , 2014 ).\nAn exaggerated ovarian response (generally defined as obtaining a high number of\noocytes) ( Arce  et al. , 2014 )\ncan also lead to a poor prognosis ( Vale-Fernandes\n et al. , 2023a ). The fertilization rate is lower due\nto a higher proportion of immature oocytes, and the live birth rate is also\nnegatively affected due to the deleterious effect of high serum oestradiol levels on\nembryo implantation after transfer ( Kok  et\nal. , 2006 ;  Sunkara\n et al. , 2011 ). An exaggerated ovarian response can\nalso lead to the development of Ovarian Hyperstimulation Syndrome (OHSS) ( Lekamge  et al. , 2007 ), a\npotentially severe condition.\nOne of the criticisms pointed to the mentioned ovarian reserve markers is the absence\nof cut-off values predictive of poor or exaggerated ovarian response ( Esteves  et al. , 2018 ).\nRegarding AMH specifically, some of the reasons are inter-laboratory and\ninter-individual variability ( Nelson & La Marca,\n2011 ).\nIndividualising the controlled ovarian stimulation protocol according to the\npredicted ovarian response seems to be the best way to maximise the prognosis of the\nART techniques and reduce the associated iatrogenic risks ( Fauser  et al. , 2008 ;  Silva  et al. , 2016 ). Therefore, the study of\novarian reserve markers and the determination of precise cut-off values, adapted to\nthe population of each reproductive medical centre, is of great importance.\nThe aim of this study is to assess the correlation between serum AMH values and the\nnumber of oocytes obtained after controlled ovarian stimulation for IVF treatments,\nand determine the predictive cut-off values for poor and exaggerated response to\nstimulation, adapted to the population of the study. A comparison will also be made\nbetween the cut-offs determined and those presented in the Bologna criteria and in\nthe POSEIDON criteria.\n\nA retrospective observational cohort study was carried to assess the correlation\nbetween serum AMH values and the number of oocytes obtained after controlled\novarian stimulation for IVF treatments, and to determine which AMH cut-off\nvalues are predictive of poor and exaggerated ovarian response. The study was\napproved by the Institutional Ethics Committee. The principles of the\nDeclaration of Helsinki were followed.\nThe relevant medical data was obtained from the hospital’s records. The dosing of\nAMH values was carried using the Beckman Coulter AMH Gen II kit, which employs\nan enzyme-linked immunosorbent assay (ELISA) technique, particularly a two-site\nsandwich ELISA ( Nelson & La Marca,\n2011 ). All AMH values in this article were presented in ng/mL.\nAfter applying the inclusion and exclusion criteria, a total of 1003 controlled\novarian stimulation cycles were included in the analysis.\nThe inclusion criteria were as follows: IVF/ICSI cycles performed between\nFebruary 2017 and December 2023, at the Unidade Local de Saúde de Santo\nAntónio (ULSSA) Medically Assisted Procreation Centre.\nThe following exclusion criteria were defined: serum AMH levels obtained more\nthan 6 months prior to the start of the ovarian stimulation, the presence of a\nsingle ovary, non-Caucasian ethnicity, a controlled ovarian stimulation cycle\nperformed for the purpose of oocyte donation or fertility preservation, a\ndocumented diagnosis of endometriosis, a documented history of ovarian surgery\nand the absence of essential data for the study in the medical records (absence\nof the number of oocytes obtained or the AMH value).\nThe stimulation protocol and the dose of gonadotrophins used in controlled\novarian stimulation cycles were chosen based on the experience of the\nreproductive medicine specialists, taking into account clinical (age, previous\ntreatments), analytical (AMH) and ultrasonographic (AFC) aspects.\nThe first objective was to assess the correlation between serum AMH values and\nthe number of oocytes obtained after controlled ovarian stimulation.\nThe second objective was to determine AMH cut-off values for poor response to\nstimulation and exaggerated response. Poor response to stimulation was defined\nas ≤ 3 oocytes obtained ( Ferraretti\n et al. , 2011 ) and exaggerated response was\ndefined as >15 oocytes obtained ( Broekmans,\n2019 ).\nAll statistical analyses were carried out using IBM SPSS Statistics software\n(version 26). Descriptive statistics were used to define the demographic data of\nthe population (woman’s age, body mass index (BMI), serum AMH value, ART\ntechnique - IVF/ICSI, controlled ovarian stimulation protocol used, ovarian\nresponse obtained). The normality of the variables was determined by the\nShapiro-Wilk test ( p <0.01), and descriptive statistics for\nnon-normal distributions are presented in the form of median and range. The\ncorrelation between serum AMH values and the number of oocytes obtained was\nassessed using Spearman’s correlation coefficient, with a significance level of\n5%. ROC (Receiver Operating Characteristic) curves were used to determine AMH\ncut-off values predictive of poor and exaggerated response.\n\nThe sample’s demographic characteristics and controlled ovarian stimulation cycles\nfeatures (gonadotrophins doses and number of stimulation days) are summarised in\n Table 1 . Of the 1003 cycles analysed, 453\n(45.16%) were IVF cycles, while 550 (54.84%) were ICSI cycles. The  Table 2  summarises AMH values and number of\noocytes obtained regarding the total sample and each ovarian response subgroup.\nDemographic characteristics and controlled ovarian stimulation cycles\nfeatures regarding the total sample and each ovarian response subgroup.\nAMH values and number of oocytes obtained regarding the total sample and each\novarian response subgroup.\nThe median age of the patients was 35 years, the median AMH value was 2.01 ng/mL and\nthe median number of oocytes obtained was 8.00.\nOf the 1003 cycles, 200 resulted in poor ovarian response (19.94%) and 157 in\nexaggerated ovarian response (15.65%).\nIn the subgroup of patients who had an excessive ovarian response, the median age\n(33.0) was lower than those of the other subgroups and those of the whole sample,\nwhile the median AMH value (3.92 ng/mL) was higher than those of the other\nsubgroups.\nOn the other hand, in the subgroup of patients with a poor ovarian response, the\nmedian age (37.0) was higher than that of the other subgroups and that of the whole\nsample, while the median AMH value (0.74 ng/mL) was lower. Women whose cycle was\ncancelled (zero oocytes obtained) were also assessed separately, and their\ncharacteristics are shown in  Tables 1  and\n 2 . The median AMH values for this\nsubgroup was 0.69 ng/mL. However, due to the small number of women (n=20), it was\nnot possible to draw any relevant conclusions.\nRegarding the type of protocol used, 950 of the controlled ovarian stimulation cycles\n(94.72%) were carried using a short protocol with a gonadotropin-releasing hormone\n(GnRh) antagonist and the remaining 53 (5.28%) using a long protocol with a GnRh\nagonist.\nFigures 1 ,  2  and  3  show the distribution of\nAMH values according to the ovarian response obtained.\nFigure 1 Distribution of AMH values in the subgroup of women with poor ovarian\nresponse.\nDistribution of AMH values in the subgroup of women with poor ovarian\nresponse.\nFigure 2 Distribution of AMH values in the subgroup of women with exaggerated\novarian response.\nDistribution of AMH values in the subgroup of women with exaggerated\novarian response.\nFigure 3 Distribution of AMH values in the subgroup of women with normal ovarian\nresponse.\nDistribution of AMH values in the subgroup of women with normal ovarian\nresponse.\nAMH showed a statistically significant correlation with the number of oocytes\nobtained. The Spearman’s correlation coefficient (p) calculated was 0.60\n( p <0.01).\nROC curves were used to calculate the cut-off values, and the predictive values\nfor poor response and exaggerated response were determined separately.\nThe ROC curve for predictive values of poor response is shown in  Figure 4 . The area under the curve (AUC) is\n0.805. Given the parameters of the curve, three different cut-off values were\ncalculated, with different sensitivities and specificities, which are summarised\nin  Table 3 .\nAMH cut-off values to differenciate poor ovarian response.\nFigure 4 ROC curve differentiating poor ovarian response according to the AMH\nvalue [area under the curve (AUC)=0.805; the marked points\ncorrespond to the cut-off values shown in  Table 3 ].\nROC curve differentiating poor ovarian response according to the AMH\nvalue [area under the curve (AUC)=0.805; the marked points\ncorrespond to the cut-off values shown in  Table 3 ].\nThe cut-off values presented in the Bologna criteria and POSEIDON criteria were\napplied to the ROC curve obtained. The POSEIDON criteria (cut-off value 1.2\nng/mL) presented a sensitivity of 57.89%, specificity of 86.06%, with positive\nand negative predictive values of 59.92% and 85.03%, respectively, regarding the\nstudy population. The Bologna criteria (cut-off values 0.5-1.1 ng/mL) presented\nsensitivity values between 26.32-54.89%, specificity between 88.23-97.97%, with\npositive and negative predictive values between 62.66-82.35% and 78.70-84.46%,\nrespectively.\nThe ROC curve for predictive values of exaggerated response is shown in  Figure 5 . The area under the curve (AUC) is\n0.807. Similarly, three different cut-off values were calculated, which are\nsummarised in  Table 4 .\nAMH cut-off values to differentiate exaggerated ovarian response.\nFigure 5 ROC curve differentiating exaggerated ovarian response according to\nthe AMH value [area under the curve (AUC)=0.807; the marked points\ncorrespond to the cut-off values shown in  Table 4 ].\nROC curve differentiating exaggerated ovarian response according to\nthe AMH value [area under the curve (AUC)=0.807; the marked points\ncorrespond to the cut-off values shown in  Table 4 ].\n\nThe results show that AMH is positively correlated to the number of oocytes\nobtained, lower AMH values predicting lower number of oocytes, and higher values\npredicting a higher number of oocytes. The correlation obtained is in line with\nthose presented by previous studies ( La Marca\n et al. , 2010 ;  Akoglu, 2018 ;  Oliveira  et\nal. , 2023 ;  Vale-Fernandes  et al. , 2023a ,  2023b ).\nIn determining cut-off values, it is important to note that they are expected to\nbe able to demarcate poor responders and over-responders with high precision,\nwithout classifying women with the potential for a good ovarian response as poor\nresponders/hyper-responders, since this could lead to the ART technique being\nabandoned. Therefore, clinicians should be aware that extreme cut-off values are\nfavoured, as they are associated with high specificity (low false-positive\nrate), even if this implies reduced sensitivity ( Ferraretti  et al. , 2011 ).\nAnalysing  Table 3 , one can see that the\nvalue of 0.18 ng/mL as a cut-off for poor ovarian response offers maximum\nspecificity, however its sensitivity is low, losing its clinical usefulness. On\nthe other hand, the value of 0.72 ng/mL, while maintaining a very high\nspecificity (95.13%), also has a higher sensitivity (43.23%).\nFor values above 0.72 ng/mL, there is a sharp drop in specificity, which\ntranslates into an increase in the number of false positives. Therefore, for the\nstudy data, 0.72 ng/mL is the ideal cut-off for differentiating women who will\ndevelop a poor ovarian response.\nThe existent studies on AMH as a predictor for poor ovarian response report\ncut-off values between 0.099 ng/mL and 1.96 ng/mL. Sensitivity and specificity\nalso vary between 44-97% and 41-100%, respectively ( La Marca  et al. , 2010 ;  Liu  et al. , 2023 ). Such\ndifferences can be attributed to the following causes:\nThe definition of poor response is not consensual in literature, so the\nnumber of oocytes that characterise a poor response is variable;\nThe sensitivity and specificity required to define a good cut-off point\nvary from author to author;\nThe use of different laboratory kits to measure AMH can also have an\nimpact on the differences observed ( Nelson & La Marca, 2011 ;  Nelson, 2013 ;  Iliodromiti\n et al. , 2015 ).\nThe variability and lack of standardisation found in the literature reinforces\nthe idea that cut-off values should be specific to each medical centre and its\npopulation, since each will have its own definition of poor response and will\nperform AMH dosages with the kit of its choice.\nThe POSEIDON criteria cut-off value (1.2 ng/mL), when applied to the study\npopulation, presented higher sensitivity but compromised specificity.\nIn the Bologna criteria, a range of cut-off values for low ovarian reserve\n(<0.5-1.1 ng/mL) is presented, due to the great variability in the literature\n( Ferraretti & Gianaroli, 2014 ).\nSimilarly to the POSEIDON criteria, some of the values contained in the range\nhave higher sensitivity but lower specificity than the cut-off calculated in\nthis study; the positive predictive value is also lower and the negative\npredictive value is very similar. However, it should be noted that the aim of\nthe Bologna criteria was not to distinguish which women are likely to develop\npoor ovarian response, but rather to reach a consensus on the definition of poor\novarian response in terms of clinical trials ( Ferraretti  et al. , 2011 ).\nThat said, the cut-off of 0.72 ng/mL is preferable for the study population\ncompared to those proposed by the published criteria, as it allows\ndifferentiation between women who will develop poor ovarian response with\ngreater specificity and higher positive predictive value.\nAnalysing  Table 4 , we can see that the\nvalue of 11.82 ng/mL as cut-off for exaggerated ovarian response has maximum\nspecificity. However, once again, it is of limited clinical usefulness as its\nsensitivity is low. On the other hand, a cut-off of 4.77 ng/mL increases\nsensitivity without compromising specificity too much. The cut-off of 3.19\nng/mL, despite maximising specificity and sensitivity in combination, has a\nspecificity of 77.95%, which is undesirable as it is a relatively low value.\nThat said, the cut-off of 4.77 ng/mL is preferable for the study population.\nTo date, few studies have been published on predictive cut-off values for\nexaggerated ovarian response. The reported values vary between 3.36-4.90 ng/mL,\nwith sensitivities and specificities between 53-91% and 70-95%, respectively\n( La Marca  et al. ,\n2010 ).\nThe main limitation of this study stems from its retrospective design. For this\nreason, certain characteristics of the sample with a possible influence on the\nresults, such as the dose of exogenous gonadotropins administered or the\nallocation of patients to the different controlled ovarian stimulation\nprotocols, were not randomised/controlled.\nTherefore, future research with a prospective and randomised design could\nminimise the potential limitations of this study and reduce some of the\nbias.\n\nThe serum AMH value proved to be a good predictor of ovarian response to controlled\novarian stimulation for IVF treatments and is very useful in supporting clinical\ndecision-making. According to the available data and for this population, 0.72 ng/mL\nwas the chosen cut-off value for differentiating poor ovarian response (specificity\nof 95.13% and sensitivity of 43.23%), and 4.77 ng/mL the chosen cut-off value for\ndifferentiating exaggerated ovarian response (specificity of 89.86% and sensitivity\nof 38.22%). However, these cut-offs should not be used as absolute discriminators,\nbut only to support decision-making. Due to the overlapping of different ovarian\nresponses for the same AMH values, the sensitivity and specificity values obtained\nfor the calculated cut-offs are not excellent and therefore should not be used as\nsole predictors of response to controlled ovarian stimulation, confirming that AMH\nalone is not able, for instance, to predict poor ovarian response, unless extremely\nlow values are considered.\nThe results obtained also highlight the importance of adapting the cut-off values to\neach medical centre. Medically assisted procreation centers will only be able to\nprovide truly effective advice to their users/patients if they use their own\ncharacteristics to predict response/success, allowing a uniform clinical decision,\nreducing risks, and not excluding any candidate with potential or including them\nwith unrealistic expectations.","source_license":"public-domain-us","license_restricted":false}