Predictors of poor ovarian response in in vitro fertilization: a prospective study comparing basal markers of ovarian reserve

Transcription

1 FERTILITY AND STERILITY VOL. 77, NO. 2, FEBRUARY 2002 Copyright 2002 American Society for Reproductive Medicine Published by Elsevier Science Inc. Printed on acid-free paper in U.S.A. Predictors of poor ovarian response in in vitro fertilization: a prospective study comparing basal markers of ovarian reserve László F. J. M. M. Bancsi, Ph.D., a Frank J. M. Broekmans, Ph.D., a Marinus J. C. Eijkemans, M.Sc., b Frank H. de Jong, Ph.D., c J. Dik F. Habbema, Ph.D., b and Egbert R. te Velde, Ph.D. a University Medical Center Utrecht, Utrecht, and Erasmus University Rotterdam, Rotterdam, The Netherlands Received December 29, 1999; revised and accepted August 28, Presented in part at the 10th Reinier de Graaf Symposium, Zeist, The Netherlands, September 9 11, Reprint requests: László F. J. M. M. Bancsi, Ph.D., Department of Reproductive Medicine, Division of Obstetrics, Neonatology and Gynecology, University Medical Center Utrecht, Heidelberglaan 100, 3584 CX Utrecht, The Netherlands ( lbancsi@azu.nl). a Department of Reproductive Medicine, Division of Obstetrics, Neonatology and Gynecology, University Medical Center Utrecht. b Department of Public Health, Faculty of Medicine and Health Sciences, Erasmus University Rotterdam. c Department of Internal Medicine III, Erasmus University Medical Center, Rotterdam /02/$22.00 PII S (01) Objective: To identify and quantify predictors of poor ovarian response in in vitro fertilization (IVF). Design: Prospective study. Setting: Tertiary fertility center. Patient(s): One hundred twenty women undergoing their first IVF cycle. Intervention(s): Measurement of the number of antral follicles and the total ovarian volume by ultrasound, and of basal levels of FSH, E 2, and inhibin B on cycle day 3. Main Outcome Measure(s): Ovarian response, and clinical and ongoing pregnancy rates. Result(s): The antral follicle count was the best single predictor for poor ovarian response: area under the receiver operating characteristic curve Addition of basal FSH and inhibin B levels to a logistic model with the antral follicle count significantly improved the prediction of poor response; the addition of basal E 2 levels and total ovarian volume did not improve the prediction. To express the discriminative performance of this model toward poor response, a maximum area under the receiver operating characteristic curve of 0.92 was calculated. Poor responders had significantly lower clinical and ongoing pregnancy rates than did normal responders. Conclusion(s): Our data demonstrate that the antral follicle count provides better prognostic information on the occurrence of poor response during hormone stimulation for IVF than does the patient s chronological age and the currently used endocrine markers. However, endocrine tests remain informative. Multivariate models can achieve more accurate predictions of outcomes of complex events like ovarian response in IVF. (Fertil Steril 2002;77: by American Society for Reproductive Medicine.) Key Words: Antral follicle count, FSH, estradiol, inhibin B, ovarian reserve, poor response, IVF, logistic model, prognosis, ovarian aging, ultrasound The concept of diminished ovarian reserve has gained general acceptance in infertility medicine. In in vitro fertilization (IVF), the association of poor ovarian response due to diminished ovarian reserve with cycle cancellation and a significant decline in success rates is well known (1, 2). Correct identification of patients who are at risk for poor response can help physicians to individualize counseling and permit the patients to decide whether to undergo a demanding infertility treatment. Accurate assessment of ovarian response potential before the patient enters an IVF program is, therefore, of pivotal importance. It is generally acknowledged that reproductive aging is related to both a quantitative and a qualitative reduction of the primordial follicle pool. As women age, their ovarian reserve diminishes, and the rates of both spontaneous and treatment-induced pregnancies decline. However, for individual predictions of ovarian responsiveness and IVF success rates, chronological age alone is of limited value. Basal FSH was the first widely used endocrine marker of ovarian reserve that had better potential than age for predicting decreased ovarian function and lowered success rates after IVF (3). Although related to the same phenomenon, FSH 328

2 and age were shown to be independent prognosticators of assisted reproduction outcome (4). At present, several other markers of ovarian reserve are advocated. Elevated early follicular E 2 levels have been demonstrated to correlate with reduced ovarian responsiveness (5 7). This was an important finding, as elevated E 2 levels may be able to suppress FSH into the normal range in women who have substantially diminished ovarian reserve and thus may cause false-negative FSH test results. Also basal inhibin B has been proposed as an endocrine prognosticator for assisted reproduction outcome, although reports are conflicting (8 10). Various reports have been published recently on the usefulness of ovarian ultrasonography characteristics in predicting impaired ovarian potential during hormone stimulation. The antral follicle count as well as ovarian volume appeared to be indicative of poor response in assisted reproduction (11 13). This prospective study was designed to compare several basal ovarian reserve markers that are currently used. Our aim was to determine which markers significantly contribute to the prediction of poor response in IVF, to identify the best single predictor, and to evaluate the additional predictive value of the remaining markers in bolstering the best single predictor. MATERIALS AND METHODS Patients Between January 1997 and April 1998, we prospectively included in our study 130 women who were referred to our IVF center for their first IVF cycle in this pregnancy attempt. Five women already had had a child after IVF. Our inclusion criteria were [1] a regular spontaneous menstrual cycle (25 to 35 days); [2] presence of both ovaries; [3] no evidence of endocrine disorders (normal levels of thyroid-stimulating hormone, testosterone, androstenedione, and prolactin), and [4] written informed consent. The study was approved by the institutional review board. All of the patients were assigned to one of three groups, based on the main cause of infertility: tubal factor, male factor, or unexplained infertility. Conventional IVF was planned for the majority of patients (n 112); the remaining 18 patients were scheduled for intracytoplasmic sperm injection (ICSI). According to national guidelines, patients over 41 years of age should not enter routinely into an IVF program in the Netherlands. Additionally, most Dutch IVF centers apply a basal FSH threshold level to select patients for treatment. In our center, routine IVF treatment is not offered to patients if their basal FSH level exceeds 15 IU/L because in a previous study we found that pregnancy rates are very poor with such high FSH levels (14). Patients up to 45 years of age were accepted in this study after appropriate counseling, irrespective of their basal FSH level. Basal Ovarian Reserve Screening On day 3 of a spontaneous cycle, venous blood samples were drawn from each patient. Serum samples were allowed to clot at room temperature for at least 1 hour. All samples (plasma and serum) were centrifuged within 2 hours after withdrawal and stored at 20 C until assayed. The endocrine screening program comprised a serum inhibin B measurement and a plasma assay of FSH and E 2. Immediately after blood samples had been taken, ovarian ultrasonography was performed using a Toshiba Capasee SSA-220A (Toshiba Medical Systems Europe BV, Zoetermeer, The Netherlands) with a 7.5 MHz transvaginal probe. Round or oval echo-free structures were regarded as follicles and all follicles up to 5 mm were included in the data analysis. Ovarian volume was subsequently computed using the prolate ellipsoid formula: Ovarian volume D 1 D 2 D 3 /6, where D 1,D 2, and D 3 are maximal perpendicular diameters of each ovary. The volumes of both ovaries were added, resulting in the total ovarian volume. One investigator (LB) performed the whole screening program. Hormone Assays Estradiol and FSH were measured in plasma specimens with the AxSYM immunoanalyzer (Abbott Laboratories, Abbott Park, IL) according to the instructions of the manufacturer. The assays are based on the microparticle enzyme immunoassay technology (MEIA). The standard of the FSH assay is calibrated against the WHO Second International Reference Preparation for human FSH (78/549). For FSH, the interassay variation was 6.0%, 6.6%, and 8.0% at levels of 5, 25, and 75 IU/L, respectively (n 46). The E 2 assay is standardized to gas chromatography/mass spectrometry. The between-run variation of the E 2 assay at 300, 1105, and 2,626 pmol/l was 12.5%, 7.5%, and 4.9%, respectively (n 29). Serum inhibin B levels were determined using an immunoenzymometric assay (Serotec, Oxford, United Kingdom) as described by Groome et al. (15). Intra-assay and interassay coefficients of variation were 14.6% and 14.0%, respectively. Treatment Protocol The IVF treatment followed within 3 months of the basal ovarian reserve screening. A detailed description of the protocol has been published previously (16). Briefly, pituitary desensitization by leuprolide acetate (Lucrin; Abbott, Hoofddorp, The Netherlands) administration was started in all women in the midluteal phase. After menstruation, ovarian stimulation was started in a fixed daily dose of 150 IU follitropin alpha (Gonal-F; Serono Benelux BV, s Gravenhage, The Netherlands) and was monitored by transvaginal ultrasonography and E 2 measurements. After 7 days the gonadotropin dose was adjusted individually. FERTILITY & STERILITY 329

3 When at least three leading follicles had developed, hcg (Profasi, Serono Benelux BV) was administered; transvaginal follicular aspiration took place 36 hours later. A maximum of two embryos was replaced in women under 38 years of age. In older women the maximum number of embryos at transfer was three. The luteal phase was supported either by hcg (Profasi, Serono Benelux BV) or micronized progesterone (Progestan, Nourypharma BV, Oss, The Netherlands). Ovarian Response Currently, no uniform definition for poor response is available (17). From a practical point of view, given a mean fertilization rate of 50% to 60% in IVF, four oocytes are needed to reach an average of two embryos available for transfer. Hence, we defined poor response as: [1] collection of fewer than four oocytes at retrieval or [2] cycle cancellation because of impaired follicular reaction ( 3 follicles) in response to exogenous gonadotropins. High response was defined as the collection of 20 oocytes at retrieval. Patients whose cycles were canceled because they were considered to be at risk of ovarian hyperstimulation syndrome (OHSS) due to exaggerated follicle growth ( 30 follicles and/or peak E 2 15,000 pmol/l) were also defined as high responders. In all other cases ovarian response was classified as normal. For the purpose of analysis, normal and high responses were considered as one group. Pregnancy Clinical and ongoing pregnancies were defined as the presence of fetal cardiac activity beyond 6 and 12 weeks of gestation, respectively. For this study, a multiple pregnancy was regarded as one pregnancy. Eligibility for Analysis Only data from the first IVF cycle of patients who had completed the basal ovarian reserve screening program were analyzed. Data from patients whose cycles were canceled because of the risk of ovarian hyperstimulation syndrome or because of poor response ( 3 follicles) to hormone stimulation were included in ovarian response analyses but not in pregnancy rate calculations, because it cannot be excluded that such patients would have become pregnant if IVF were performed. However, patients with complete absence of follicle growth and levels of E pmol/l were considered to have a zero chance of pregnancy; therefore, data from their cycles were included in pregnancy rate calculations. Statistical Analysis Data were analyzed with the SPSS (SPSS Inc., Chicago, IL) and the SAS (SAS Institute, Cary, NC) programs. We used the Mann Whitney U test, the chi-square test, and Fisher s exact test for comparing screening and treatment variables between poor responders and normal responders and for comparing screening and IVF outcome parameters between two groups with different age and FSH characteristics. A two-sided P.05 was considered to indicate statistical significance. Furthermore, we applied Pearson s correlation where appropriate. Univariate and multivariate logistic regression analyses were used to study the association between six potential prognostic variables (age, FSH, E 2, inhibin B, the antral follicle count, and total ovarian volume) and ovarian response. Forward as well as backward selection of parameters was applied, using P.05 and P.10 for entry or deletion, respectively. The area under the receiver operating characteristic curve (ROC AUC) was computed to assess the predictive accuracy of the logistic models, yielding values from 0.5 (no predictive power) to 1.0 (perfect prediction). Selection of relevant predictors and estimation of parameters were performed on a dataset with a limited number of events (i.e., poor response: n 36; also see the results section). Therefore, to exclude overoptimism when applying the final logistic model and to assess the stability of the selection of variables, a correction was performed (18). In this so-called bootstrap sampling procedure, the selection and estimation process was repeated 200 times after creating 200 new datasets of 120 cases (bootstrap samples) by drawing cases with replacement from the original data. The original statistical analysis was performed on each bootstrap sample, yielding 200 sets of predictors and parameter estimates. The model estimates of each bootstrap sample were evaluated on the original data and a new mean ROC AUC was calculated, now corrected for statistical overoptimism and thus yielding a more accurate estimate of future performance. The Hosmer-Lemeshow goodness of fit test was used to check for lack of fit ofthefinal logistic model. RESULTS Of the 130 included patients, 120 became eligible for analysis (102 patients undergoing conventional IVF and 18 patients undergoing ICSI). Six patients conceived spontaneously while on the waiting list for IVF, two patients dropped out due to intercurrent disease, and two patients withdrew their consent. Thirty-six patients were diagnosed as poor responders. In 20 patients, follicular aspiration yielded less than four oocytes. Sixteen cycles were canceled: six cycles in patients with one or two follicles during hormone stimulation (not included in pregnancy calculations), and 10 cycles in patients in whom there was no ovarian response at all (included in pregnancy calculations). Ten patients were diagnosed as high responders: seven cycles were canceled due to exaggerated ovarian response (not included in pregnancy calculations), and in three patients more than 20 oocytes were retrieved (included in pregnancy calculations). Thus, of the 120 patients with data for ovarian response analysis, 107 were included into pregnancy calculations. Table 1 presents clinical, endocrine, and ultrasound data 330 Bancsi et al. Predictors of poor ovarian response in IVF Vol. 77, No. 2, February 2002

4 TABLE 1 Screening, treatment, and outcome variables. Variable Overall group (n 120) Normal responders (n 84) Poor responders (n 36) P Screening Mean ( SD) age (y) No. (%) of patients with primary infertility 66 (55%) 48 (57%) 18 (50%).65 No. (%) of patients with indicated infertility diagnosis Tubal factor 23 (19%) 17 (20%) 6 (17%) Male factor 59 (49%) 50 (60%) 9 (25%).001 Unexplained 38 (32%) 17 (20%) 21 (58%) Mean ( SD) duration of infertility (mo) Mean ( SD) body mass index (kg/m 2 ) Mean ( SD) day 3 FSH level (IU/L) Mean ( SD) day 3 E 2 level (pmol/l) Mean ( SD) day 3 inhibin B level (pg/ml) Mean ( SD) day 3 antral follicle count (n) Mean ( SD) day 3 total ovarian volume (ml) Treatment Mean ( SD) no. of ampules Mean ( SD) duration of stimulation (d) Rate of gonadotropin dose increase (%) 25 (30/120) 18 (15/84) 42 (15/36).006 Mean ( SD) peak E 2 level (pmol/l) 5,825 4,259 7,594 3,779 1,696 1, Outcome Mean ( SD) no. of oocytes a Mean ( SD) fertilization rate (% of oocytes) Mean ( SD) no. of embryos transferred b Implantation rate per embryo (%) 26 (44/167) 30 (41/137) 10 (3/30).024 Clinical pregnancy rate (%) c 32 (34/107) 40 (31/77) 10 (3/30).003 Ongoing pregnancy rate (%) c 25 (27/107) 31 (24/77) 10 (3/30).024 Note: P values for comparison between normal and poor responders. a Data for oocyte retrieval cycles: overall n 97, normal responders n 77, and poor responders n 20. b Data for embryo transfer cycles: overall n 83, normal responders n 65, and poor responders n 18. c Data for oocyte retrieval cycles or cycle cancellation due to complete absence of follicular response: overall n 107, normal responders n 77, and poor responders n 30. of both the basal ovarian reserve screening program and the treatment cycle for the overall group (n 120), and for poor responders (n 36) and normal responders (n 84) separately. In poor responders we found substantially reduced clinical and ongoing pregnancy rates when compared to normal responders. Poor responders had slightly fewer embryos transferred and showed a markedly lowered implantation rate per embryo. The mean embryo quality was similar in both groups (P.28). Note the coincidental equality of implantation rate per embryo and of clinical and ongoing pregnancy rates. Table 2 depicts odds ratios, statistical significance, and ROC AUCs for six basal ovarian reserve markers. As a single predictor, the antral follicle count appeared to have the best discriminative potential for poor response, expressed by the largest ROC AUC of 0.87; this was not, however, a statistically significant difference from the second best ROC AUC of 0.84 for basal FSH. The antral follicle count also appeared to be the best predictor in the subsequent multivariate analysis, as it was selected first. We therefore considered this variable to be the most powerful prognosticator of poor response. Multivariate analysis resulted in a predictive logistic model with three variables. The antral follicle count was selected in the first step, followed by inhibin B in step two, and finally by FSH in step three. The ROC AUC stepwise increased from 0.87 to Backward stepwise elimination yielded the same selected variables and parameter estimates. After correction for overfitting by the bootstrapping procedure, the ROC AUC for the three-variable model was In this procedure the antral follicle count, inhibin B, and FSH were selected in the majority of samples (93%, 76%, and 83%, respectively), whereas all other variables were selected in a minority of samples (E 2 1.5%, age 5%, and ovarian volume 25%), thus confirming the stability of variable selection. The Hosmer-Lemeshow goodness of fit test of the full model (antral follicle count, inhibin B, and FSH) showed no lack of fit (P.36). To illustrate the possible clinical application of a multi- FERTILITY & STERILITY 331

5 TABLE 2 Univariate and multivariate logistic regression analysis with odds ratios and area under the receiver operating characteristic curve (ROC AUC) of basal ovarian reserve markers for the prediction of poor response in IVF. Odds ratio (95% CI) P ROC AUC Univariate analysis Age (per year) 1.08 ( ) FSH (per IU/L) 1.41 ( ) E 2 (per pmol/l) ( ) Inhibin B (per pg/ml) 0.98 ( ) Antral follicle count (per follicle) 0.70 ( ) Total ovarian volume (per ml) 0.94 ( ) Multivariate analysis (all six variables) Step 1: antral follicle count (per follicle) 0.70 ( ) Step 2: inhibin B (per pg/ml) 0.98 ( ) Step 3: FSH (per IU/L) a 1.27 ( ) Age not selected.66 NA E 2 not selected.63 NA Total ovarian volume not selected.13 NA Restricted multivariate analyses (two variables) FSH inhibin B Step 1: FSH (per IU/L) 1.41 ( ) Step 2: inhibin B (per pg/ml) 0.98 ( ) Antral follicle count FSH Step 1: antral follicle count (per follicle) 0.70 ( ) Step 2: FSH (per IU/L) 1.27 ( ) Note: NA not applicable. a Logistic regression model: probability of poor response 1 e ( number of antral follicles inhibin B FSH) number of antral follicles inhibin B FSH) e( variate predictive model to estimate the chance of developing a poor ovarian response in IVF, we constructed a formula that can be easily applied in daily practice. With this formula a so-called poor response score (PRSc) can be calculated as follows: PRSc (12 FSH [IU/L]) (14 antral follicle count [n]) (1 inhibin B [pg/ml]). The poor response score can subsequently be transformed into a probability of poor response using the curve in Figure 1. For example, if a patient has an FSH level of 18 IU/L (score: 216), an antral follicle count of 3 (score: 42), and an inhibin B level of 50 pg/ml (score: 50), the poor response score equals 124. Such a poor response score would correspond with a probability of poor response of about 90% (see Fig. 1). Although poor responders were more often diagnosed with unexplained infertility, infertility diagnosis as such was not associated to poor response in a multivariate analysis (P.41). Moreover, a forced start with age and infertility diagnosis again yielded a logistic model with three selected variables (number of antral follicles, inhibin B, and FSH) with equal performance (area under the receiver operating characteristic curve 0.92). We also assessed the predictive accuracy of logistic models based on only two variables. Starting off with the antral follicle count, the ROC AUCs significantly increased to 0.90 after the addition of either inhibin B ( step 2 of the multivariate analysis) or FSH. The model solely based on endocrine variables yielded an ROC AUC of Figure 2 shows the distribution of the predicted probabilities of poor response for normal and poor responders, according to the three-variable logistic model. For example, of the 54 patients who were predicted to have the lowest probability of poor response (0 to 0.10), 51 had a normal response and 3 had a poor response. Of the 10 patients who were predicted to have the highest probability of poor response (0.90 to 1.00), 9 were predicted correctly, but one patient had a normal response. From the combined information provided by both histograms, sensitivity, specificity, and predictive values of the logistic model can be calculated at different cut-off points. Thus, Figure 2 is an alternative, but clinically more understandable, presentation of a receiver operating characteristic curve. If a cut-off probability of 0.50 for poor response would be adopted arbitrarily, the performance of the logistic model is as follows: sensitivity 0.75, specificity 0.95, positive predictive value 0.87, negative predictive value 0.90, and in 89% of all patients response was predicted correctly. To compare the performances of several logistic models, 332 Bancsi et al. Predictors of poor ovarian response in IVF Vol. 77, No. 2, February 2002

6 FIGURE 1 Poor response score based on the three-variable logistic model (antral follicle count, basal FSH, and inhibin B) and predicted probability of poor response. we calculated the accuracy of different combinations of the three relevant variables, selected by the multivariate analysis, to predict poor response. Table 3 demonstrates the impact of adding a second and third variable to a single variable model, again at a cut-off probability of 0.50 for poor response. With the three-variable model, the proportion of patients that is misdiagnosed (11%) is almost halved when compared to the proportion of misdiagnoses by the best single variable model of the antral follicle count (20%). The difference between the full model and the model with the antral follicle count and FSH is only minimal at this cut-off. The increase in correct predictions with the three-variable model as compared to the model with FSH and inhibin B is still considerable. From the total group of 120 women, 18 (15%) were over 41 years of age (median 42.2, range: 41.3 to 44.0), and 10 (8%) of the women presented with an elevated ( 15 IU/L) basal FSH level (median 18.9, range: 16.2 to 55.4). In Table 4 basal ovarian reserve screening variables and treatment outcome are presented for normal and poor responders in two groups of patients. To this end, patients were split up by age and basal FSH level: [1] age 41 years and FSH 15 IU/L (n 92), and [2] age 41 years and/or FSH 15 IU/L (n 28). Group 2 represents patients to whom IVF would not have been offered routinely in our clinic. The poor response rate was significantly higher in Group 2 (57% vs. 22%; P.001). However, the poor response rate in Group 1, despite selection by age and FSH level, still appeared to be considerable. The good result obtained in normal responders in Group 2 is striking, whereas in poor responders no pregnancy occurred. As in the whole population, the predictive logistic model performed well in both groups at a cut-off of 0.50 for the predicted probability of poor response. In Group 1 a positive predictive value of 0.93 and a negative predictive value of 0.91 were calculated. In Group 2 the positive and negative predictive values were found to be 0.93 and 0.85, respectively. DISCUSSION Gougeon et al. (19) demonstrated a clear association between the number of growing follicles and the number of resting follicles in the follicle pool. Moreover, a study by Pellicer et al. (20) revealed a high correlation between the FERTILITY & STERILITY 333

7 FIGURE 2 Distribution of predicted probabilities of poor ovarian response for normal responders (top panel) and poor responders (bottom panel) according to the three-variable logistic model (antral follicle count, basal FSH, and inhibin B). number of selectable follicles (2 to 5 mm) measured by transvaginal three-dimensional ultrasonography and the number of selectable follicles in histological slices. Therefore, it is plausible that the number of antral follicles, originating from the cohort of growing follicles, reflects the size of the pool of resting follicles, and thus ovarian reserve. Our data indicate that, on cycle day 3, the number of antral follicles provides better prognostic information on the occurrence of poor ovarian response during hormone stimulation for IVF than the currently used endocrine markers and chronological age. As expected, clinical and ongoing pregnancy rates were significantly reduced in poor responders. Although poor responders had fewer embryos transferred, the important reduction in pregnancy rates is mainly due to the decrease in implantation rates. This supports the concept of qualitative and quantitative ovarian reserve being closely related. Nasseri et al. (21) reported on a significant higher incidence of fetal aneuploidy, originating mainly from the oocyte, in abortuses in women with elevated basal FSH and/or E 2 levels. It is possible that the reduction of implantation potential of poor responder embryos is also related to an increase in chromosomal abnormalities in the oocyte. Like Chang et al. (11) and Tomás et al. (13), we analyzed the number of antral follicles measuring up to 5 mm. However, in their studies antral follicle counts were mainly performed after complete pituitary suppression, just before gonadotropin stimulation. In our study, antral follicles counts were obtained on the same day as the endocrine screening, in a spontaneous cycle. Because follicles can vary up to 10 mm before the dominant follicle is identified, we also analyzed the number of follicles up to 10 mm as a predictor for poor response, as this might represent the antral follicle cohort to its maximum extent (22). The number of follicles up to 10 mm correlated highly with the antral follicle count up to 5 mm (r 0.95, P.001). This explains why, although it is significant in a univariate analysis (odds ratio 0.77; 95% CI ; P.001), the antral follicle count up to 10 mm was not selected into a multivariate model after the number of follicles up to 5 mm was taken into account. In our study, all ultrasound scans were performed by a single investigator. The clinical applicability of the antral follicle count not only depends on predictive accuracy but also on intraobserver and interobserver reproducibility. We are not aware of any studies concerning interobserver variation in the assessment of the number of small antral follicles by transvaginal ultrasonography. Intraobserver variation only has been studied in a small group of volunteers (22). Data from a prospective study that was performed recently by our group indicated adequate agreement, both within and between observers, with respect to the measurement of the number of small antral follicles by transvaginal ultrasonography (23). We were able to construct a logistic three-variable model with good predictive performance for ovarian response. With this model the number of correct predictions is substantially higher than with the models with one or two variables. For example, when compared to the single-variable model with the antral follicle count, the three-variable model improved both the recognition of poor responders ( 5) and the identification of normal responders ( 6). When the full model is compared to the model with FSH and inhibin B, the gain in correct predictions is almost solely caused by better identification of poor responders ( 6). 334 Bancsi et al. Predictors of poor ovarian response in IVF Vol. 77, No. 2, February 2002

8 TABLE 3 Performances of several logistic models for the prediction of poor response at a cut-off point of 0.50 for the probability of poor response. Predictive model Sens Spec PPV NPV ROC AUC Correct predictions (%) Inhibin B (77%) FSH (78%) Antral follicle count (80%) FSH inhibin B (83%) Antral follicle count inhibin B (83%) Antral follicle count FSH (87%) Antral follicle count FSH inhibin B (89%) Note: PPV positive predictive value; NPV negative predictive value; ROC AUC area under the receiver operating characteristic curve. The results of analyses for pregnancy were far less promising. Only for basal inhibin B did we find a statistically significant, though weak, relation with clinical pregnancy (P.033). The relation of inhibin B with ongoing pregnancy was not statistically significant (P.14). Despite the fact that also basal FSH level and number of antral follicles were significantly related to poor response, we failed to find an effect of these variables on pregnancy rates. Obviously, the occurrence of pregnancy is influenced by many more factors than ovarian reserve. In daily practice, it is our policy to withhold IVF treatment from patients who have a suspected poor prognosis because of diminished ovarian reserve. Hence, patients over 41 years of age (according to national guidelines) or with elevated FSH levels ( 15 IU/L, our institutional guideline) are not offered the option of proceeding with IVF. In this study IVF was offered to all patients, regardless of their FSH level and up to a maximum age of 45 years if all of the study s inclusion criteria were met. Thus, a total of 28 patients who fulfilled at least one of the aforementioned criteria were allowed to enter the IVF program (see Table 4). Interestingly, 12 of these poor prognosis patients had a normal ovarian response, 11 of whom had a probability of poor response that was substantially less than 50% according to the three-variable model. Five of these 12 patients did conceive: four women were over 41 years (42, 42, 42, and 44 years) but had a normal basal FSH level (5.1, 5.5, 7.1, and 7.2 IU/L, respectively); one patient had an elevated FSH level (17.1 IU/L, age 39 years). In spite of our small number of patients, these data underline the importance of looking beyond age and FSH. Therefore, in patients of advanced age or with elevated FSH levels it seems worthwhile to select TABLE 4 Basal ovarian reserve screening variables and treatment outcome in normal responders (NR) and poor responders (PR), subdivided by age and basal FSH level: Group 1 age 41 years and FSH 15 IU/L, Group 2 age 41 years and/or FSH 15 IU/L. Group 1 Group 2 Variable NR (n 72) PR (n 20) P NR (n 12) PR (n 16) P Screening Mean ( SD) age (y) Mean ( SD) day 3 FSH level (IU/L) Mean ( SD) day 3 E 2 level (pmol/l) Mean ( SD) day 3 inhibin B level (pg/ml) Mean ( SD) day 3 antral follicle count (n) Mean ( SD) day 3 total ovarian volume (ml) Treatment outcome Clinical pregnancy rate (%) a 40 (26/65) 20 (3/15) (5/12) 0 (0/15).010 Ongoing pregnancy rate (%) a 32 (21/65) 20 (3/15) (3/12) 0 (0/15).08 Note: Values in parentheses are proportions. a Data for oocyte retrieval cycles or cycle cancellation because of complete absence of follicular response (overall: n 107; Group 1: normal responders n 65, poor responders n 15; Group 2: normal responders n 12, poor responders n 15). FERTILITY & STERILITY 335

9 those who are likely to respond well to gonadotropins (24). The predictive model, based on the antral follicle count, FSH, and inhibin B might be a useful tool to identify such patients. Although the bootstrapping procedure confirmed the statistical stability of the predictive model, prospective validation is mandatory. Also, the limited role of E 2 and ovarian volume should be confirmed. It is still unclear whether repeated measurements in different cycles of the predictors described in this report, or more invasive dynamic ovarian reserve tests (clomiphene citrate challenge test, GnRH agonist stimulation test, and exogenous FSH ovarian reserve test) can be of additional predictive value. These tests have been described to have good predictive power for IVF outcome (25 27), but the question of whether the predictions are better than those based upon basal FSH levels has yet to be addressed in detail. In conclusion, the results of our study indicate that the number of antral follicles is the best currently available basal marker of ovarian reserve in terms of predicting poor response in IVF. In itself, the performance of the antral follicle count is only fair. Because the availability of transvaginal ultrasonography is a prerequisite for IVF, counting the number of antral follicles could be considered as a standard ovarian reserve screening test before IVF is applied. If maximum accuracy in counseling is warranted, endocrine testing, especially of FSH and inhibin B, should not be abandoned, as it will provide additional predictive information in addition to the antral follicle count. Acknowledgments: The authors thank Ben W. J. Mol, Ph.D., for his critical review of the manuscript. References 1. Pellicer A, Lightman A, Diamond MP, Russell JB, DeCherney AH. Outcome of in vitro fertilization in women with low response to ovarian stimulation. Fertil Steril 1987;47: Jenkins JM, Davies DW, Devonport H, et al. Comparison of poor responders with good responders using a standard buserelin/human menopausal gonadotrophin regime for in-vitro fertilization. Hum Reprod 1991;6: Scott RT, Toner JP, Muasher SJ, Oehninger S, Robinson S, Rosenwaks Z. Follicle-stimulating hormone levels on cycle day 3 are predictive of in vitro fertilization outcome. Fertil Steril 1989;51: Toner JP, Philput CB, Jones GS, Muasher SJ. Basal follicle-stimulating hormone level is a better predictor of in vitro fertilization performance than age. Fertil Steril 1991;55: Evers JL, Slaats P, Land JA, Dumoulin JC, Dunselman GA. Elevated levels of basal estradiol-17 predict poor response in patients with normal basal levels of follicle-stimulating hormone undergoing in vitro fertilization. Fertil Steril 1998;69: Licciardi FL, Liu HC, Rosenwaks Z. Day 3 estradiol serum concentrations as prognosticators of ovarian stimulation response and pregnancy outcome in patients undergoing in vitro fertilization. Fertil Steril 1995; 64: Smotrich DB, Widra EA, Gindoff PR, Levy MJ, Hall JL, Stillman RJ. Prognostic value of day 3 estradiol on in vitro fertilization outcome. Fertil Steril 1995;64: Balasch J, Creus M, Fabregues F, Carmona F, Casamitjana R, Ascaso C, Vanrell JA. Inhibin, follicle-stimulating hormone, and age as predictors of ovarian response in in vitro fertilization cycles stimulated with gonadotropin-releasing hormone agonist-gonadotropin treatment. Am J Obstet Gynecol 1996;175: Hall JE, Welt CK, Cramer DW. Inhibin A and inhibin B reflect ovarian function in assisted reproduction but are less useful at predicting outcome. Hum Reprod 1999;14: Seifer DB, Lambert-Messerlian G, Hogan JW, Gardiner AC, Blazar AS, Berk CA. Day 3 serum inhibin-b is predictive of assisted reproductive technologies outcome. Fertil Steril 1997;67: Chang MY, Chiang CH, Hsieh TT, Soong YK, Hsu KH. Use of the antral follicle count to predict the outcome of assisted reproductive technologies. Fertil Steril 1998;69: Lass A, Skull J, McVeigh E, Margara R, Winston RM. Measurement of ovarian volume by transvaginal sonography before ovulation induction with human menopausal gonadotrophin for in-vitro fertilization can predict poor response. Hum Reprod 1997;12: Tomás C, Nuojua-Huttunen S, Martikainen H. Pretreatment transvaginal ultrasound examination predicts ovarian responsiveness to gonadotrophins in in-vitro fertilization. Hum Reprod 1997;12: Bancsi LF, Huijs AM, den Ouden CT, Broekmans FJ, Looman CW, Blankenstein MA, te Velde ER. Basal follicle-stimulating hormone levels are of limited value in predicting ongoing pregnancy rates after in vitro fertilization. Fertil Steril 2000;73: Groome NP, Illingworth PJ, O Brien M, Pai R, Rodger FE, Mather JP, McNeilly AS. Measurement of dimeric inhibin B throughout the human menstrual cycle. J Clin Endocrinol Metab 1996;81: van Kooij RJ, Looman CW, Habbema JD, Dorland M, te Velde ER. Age-dependent decrease in embryo implantation rate after in vitro fertilization. Fertil Steril 1996;66: Lashen H, Ledger W, Lopez-Bernal A, Barlow D. Poor responders to ovulation induction: is proceeding to in-vitro fertilization worthwhile? Hum Reprod 1999;14: Harrell FE Jr, Lee KL, Mark DB. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med 1996;15: Gougeon A. Caracteres qualitatifs et quantitatifs de la population folliculaire dans l ovaire humain adulte. Contracept Fertil Sex 1984;12: Pellicer A, Ardiles G, Neuspiller F, Remohi J, Simon C, Bonilla- Musoles F. Evaluation of the ovarian reserve in young low responders with normal basal levels of follicle-stimulating hormone using threedimensional ultrasonography. Fertil Steril 1998;70: Nasseri A, Mukherjee T, Grifo JA, Noyes N, Krey L, Copperman AB. Elevated day 3 serum follicle stimulating hormone and/or estradiol may predict fetal aneuploidy. Fertil Steril 1999;71: Pache TD, Wladimiroff JW, de Jong FH, Hop WC, Fauser BC. Growth patterns of nondominant ovarian follicles during the normal menstrual cycle. Fertil Steril 1990;54: Scheffer GJ. Assessment of reproductive aging in normal women [thesis]. University of Utrecht, Utrecht, The Netherlands, Roest J, van Heusden AM, Mous H, Zeilmaker GH, Verhoeff A. The ovarian response as a predictor for successful in vitro fertilization treatment after the age of 40 years. Fertil Steril 1996;66: Fanchin R, De Ziegler D, Olivennes F, Taieb J, Dzik A, Frydman R. Exogenous follicle stimulating hormone ovarian reserve test (EFORT): a simple and reliable screening test for detecting poor responders in in-vitro fertilization. Hum Reprod 1994;9: Loumaye E, Billion JM, Mine JM, Psalti I, Pensis M, Thomas K. Prediction of individual response to controlled ovarian hyperstimulation by means of a clomiphene citrate challenge test. Fertil Steril 1990;53: Padilla SL, Bayati J, Garcia JE. Prognostic value of the early serum estradiol response to leuprolide acetate in in vitro fertilization. Fertil Steril 1990;53: Bancsi et al. Predictors of poor ovarian response in IVF Vol. 77, No. 2, February 2002