Abstract
Objective:
To estimate probability of clinical or multiple pregnancy during ovulation induction/ovarian stimulation (OI/OS).
Design:
Secondary analysis of two multicenter randomized clinical trials (combined).
Setting:
Multicenter.
Patients:
750 women with polycystic ovary syndrome (PCOS) and 900 women with unexplained infertility (UI).
Interventions:
OI/OS with either timed-intercourse (PCOS) or intra-uterine insemination (IUI).
Main Outcome Measures:
Clinical and multiple pregnancy rates/cycle, cumulative pregnancy rates.
Statistics:
Age, body mass index (BMI), parity, diagnosis, medication, markers of ovarian reserve and ovarian response were considered in multivariable regression models for clinical, multiple, and cumulative pregnancy rates. Receiver operating characteristic (ROC) curves were created for clinical and multiple pregnancy rates.
Results:
Younger patient (p=0.016) and partner age (p=0.004), treatment type (p=0.002), lower BMI (p=0.002) and medication dose (p<0.001) were all associated with clinical pregnancy. Variables associated with multiple pregnancy included the above (except age), in addition to diagnosis (p=0.002), parity (p=0.037), higher AFC (p<0.001), AMH (p=0.001), and ovarian response (p<0.05). Gonadotropin use was associated with multiple pregnancy (p<0.005), with progressively increasing odds ratios (cycles:1–4). ROC-curves indicated the model’s predictive power to be fair for clinical pregnancy [AUC(95%CI): 0.78(0.75–0.81) for cycle 1 and 0.70(0.64–0.75) for cycle 4] and good-to-excellent for multiple pregnancy [AUC(95%CI): 0.78(0.72–0.84) for cycle 1 and 0.86(0.78–0.93) for cycle 4]. Partner age, lower medication dosage, parity, AMH and diagnosis were associated with cumulative pregnancy rates.
Conclusions:
Utilizing the majority of the factors known to predict the outcome of OI/OS cycles, we constructed an easy-to-use formula that may predict individualized chances of clinical and multiple pregnancy for commonly used fertility treatments (https://pregnancyprediction.medicine.yale.edu/CalDirect.html).
Keywords: ovulation induction, ovarian stimulation, probability of clinical pregnancy, probability of multiple pregnancy, individualized prediction models
CAPSULE:
factors known to predict ovulation induction/ovarian stimulation cycle outcomes were used to construct an easy-to-use formula that can estimate individualized chances of clinical and multiple pregnancy for commonly used fertility treatments.
Introduction
Ovulation induction (OI) and controlled ovarian stimulation (OS) with intrauterine insemination (IUI) are often first-line treatment for many couples with infertility. Historically, and depending on the diagnosis, the age of the partners, as well as the cost and insurance coverage status, treatments may start with oral agents [clomiphene citrate (CC) or letrozole (LTZ)] coupled with either timed intercourse (TIC) or IUI and then progress in a stepwise fashion through gonadotropin/IUI to more aggressive and efficient, but expensive treatments, such as In-Vitro fertilization (IVF). For IVF, counseling of infertile couples regarding relevant success and complications is greatly facilitated by the availability of an annually-updated registry produced by the Society of Assisted Reproductive Technology (SART). In contrast, counseling about potential success of OI and/or OS/IUI treatments seems to rely on results from isolated studies from specific patient populations, such as those with unexplained infertility (UI) or polycystic ovary syndrome (PCOS). Quoted pregnancy rates/cycle range from 8.3%−9.6% for CC/IUI, and 14%−17.1% for gonadotropin/IUI (1–3) to cumulative live birth rates as high as 32.2%, 23.3%, and 18.7%, for gonadotropins, CC, and LTZ treatments, respectively (4, 5).
Over the last decade, the advent of precision medicine has rapidly permeated all areas of medicine. It is now an expectation that counseling and treatment planning be individualized and hopefully accurate. In the case of OI and/or OS/IUI, whether with oral agents or gonadotropins, the incorporation of a wide variety of individual data (such as demographics, biomarkers of ovarian reserve, and clinical and treatment characteristics) can be used to develop algorithms to successfully predict the likelihood of success for each patient, thus guiding decisions towards most effective treatments. The eventual clinical goal is to identify the subgroup of patients that will benefit most from specific types of treatment. This spares those who might not benefit from a non specific algorithm, the expense, the time commitment, side effects, and complications. For IVF, SART developed an easy-to -use on-line calculator based on N >500,000 cycles of therapy (available at https://www.sartcorsonline.com/Predictor/Patient)(6). This tool helps patients and physicians alike predict a particular patient’s individual chances of pregnancy (singleton or multiple) after one or two IVF cycles, when transferring one or more embryos, thus facilitating individualized decision making. Yet, despite the fact that clinical studies have identified some individual patient factors and predictors of clinical and multiple pregnancy among OI and/or OS/IUI cycles (7–9), (10) a tool for effective counseling and individualized treatment planning, similar in magnitude to the SART calculator, is not widely available.
The goal of the present study was to help develop an individualized prediction model for the probability of pregnancy (singleton or multiple) after OI and/or OS among women with either PCOS or unexplained infertility using readily available databases. Our aims were to identify patient and cycle-specific characteristics associated with success in the above-mentioned populations, in order to calculate the probability of clinical and/or multiple pregnancy per OI/OS cycle as well as cumulative chances of pregnancy, based on the previously identified success-predictive factors. Our ultimate goal was to build a predictive model and implement it in an easy-to-use calculator that can be applied to patient counseling to guide clinical decision making.
Materials and Methods
Design and Study Population
Data collected from 1650 patients participating in two multicenter, randomized controlled trials (PPCOS II and AMIGOS, clinicaltrials.gov: NCT00719186 and NCT1044862, respectively) performed by the Reproductive Medicine Network of the Eunice Kennedy Shriver National Institute of Health and Human Development were used for this secondary analysis. Seven hundred and fifty and 900 participants were from the Pregnancy in Polycystic Ovary Syndrome II (PPCOSII) and the Assessing Multiple Inrauterine Gestations after Ovulation Stimulation (AMIGOS) trial, respectively. Details regarding the trials’ design, interventions and participants’ characteristics have been previously published [AMIGOS: (4, 11), PPCOS II: (5, 12)]. At each participating site, institutional review board was obtained for both trials. All participants signed an informed consent.
Briefly, PPCOS II included women, diagnosed with Polycystic Ovary Syndrome [according to modified Rotterdam criteria (anovulation with either hyperandrogenism or polycystic ovaries), (13)] who were randomized to either CC or LTZ to determine which treatment was more likely to result in a live birth. AMIGOS included couples with unexplained infertility and this trial was originally designed to determine whether treatment with LTZ would result in a lower multiple pregnancy rate than standard OS regimens for unexplained infertility in use at the time (either CC or gonadotropins). In both studies, participating women were ≥ 18 and ≤ 40 years of age and had to have a normal uterine cavity and at least one patent fallopian tube. Women in the AMIGOS trial also had to have evidence of regular ovulation and normal ovarian reserve. Male partners had semen parameters that permitted either IUI or TIC (at least 5 and 14 million motile sperm per milliliter for AMIGOS and PPCOSII, respectively). Other diagnoses, not in conflict with the above inclusion/exclusion criteria, assigned to these patients though previous fertility evaluations and recorded at study enrollment, are included in the current analysis, are referred to as “other diagnoses” and are considered “historic” compared to the primary diagnoses. Couples were monitored for up to four (AMIGOS trial) or five cycles of treatment (PPCOSII trial) and throughout pregnancy to determine outcomes. In summary, 750 women with PCOS were randomized to up to five cycles of OI/timed intercourse (OI/TIC) using either LTZ or CC (PPCOSII) and 900 women with unexplained infertility were randomized to LTZ, CC or gonadotropins for up to four OS/IUI cycles. For the purposes of this manuscript, the terms ovulation induction and ovarian stimulation are utilized as suggested by the American Society for Reproductive Medicine (ASRM) (14). The fomer term refers to the pharmacological treatment of anovulatory women to induce mono-ovulatory response, while the latter is reserved for the pharmacological treatment of ovulatory women with the intent of inducing multi-follicular development.
Outcomes
Outcomes of interest for this study were clinical and multiple pregnancy rates per treatment cycle, as well as cumulative rates of clinical pregnancy. The decision to estimate both individual (per cycle) and cumulative probabilities of success was based on the fact that both outcomes are of interest to the couple and the counseling clinician.
Clinical pregnancy was defined as an intrauterine pregnancy with cardiac motion identified by early first trimester ultrasonography (at approximately 6 weeks of gestation).
Multiple pregnancy was defined as an intrauterine pregnancy with more than one fetal pole with cardiac activity detected via first trimester ultrasonography. To define the number of fetuses in a multiple pregnancy the highest number of identified fetal heartbeats was used.
Cumulative pregnancy included any intrauterine pregnancy with cardiac motion achieved by either the 4th (AMIGOS) or the 5th (PPCOSII) treatment cycle.
Statistical analysis
The Collaborative Center for Statistics in Science at Yale University provided oversight for both RCT trials, was responsible for data entry and management and performed all the analyses. Baseline characteristics were compared between cycles, between clinical pregnancy and between multiple pregnancies. Students t-test, chi-square or Fisher’s exact tests were performed to compare outcome measures to the putative predictors depending on the data type (continuous or categorical) and distribution (normal or not) of a predictor. ANOVA was used as appropriate. A multivariable logistic regression model was created using baseline characteristics considered as putative predictor variables. Variables included: patient’s and partner’s age, body-mass-index (BMI), prior parity, infertility diagnosis (Dx), treatment type (CC, LTZ, gonadotropins), ovarian reserve measures [serum levels of anti-Mullerian hormone (AMH), basal follicle stimulating hormone (FSH) and estradiol (E2), and antral follicle count (AFC)], endometrial thickness, medication dose and maximum daily dose (MaxMed), peak E2 levels (when available), total number of pre-ovulatory follicles (≥14 mm, when available), and total number of treatment cycles. Peak E2 levels and total number of pre-ovulatory follicles (≥14 mm) were not introduced into the multivariable logistic regression analysis, since these two variables were available for AMIGOS patients only. Variables were introduced to a multivariable logistic regression analysis in a stepwise fashion, with a univariate analysis p-value of <0.30 to enter, and were retained in the multivariable model when p-value was <0.35 (the latter p-value was used to be as inclusive as possible of all potential determinants of pregnancy outcomes). Tables are presented with odds ratios (OR) and the corresponding 95% confidence intervals (95% CI) for the predictors in the adjusted logistic regression analysis (Supp. Tables: 1–4). Of note, in treatment cycle # 5, none of the variables had a p-value < 0.05. Therefore, in cycle number 5, these variables were not predictive for chance of having a clinical or a multiple pregnancy.
Finally, to assess the predictive power of the final models, Receiver Operating Characteristic (ROCs) curves were created for both clinical and multiple pregnancy rates and the areas under the curve (AUC) were calculated (an area of 1.0 represents a perfect test; and areas of: 0.9–0.99, 0.8–0.89, 0.7–0.79 represent an excellent, good, and fair test, respectively). SAS 9.4 (Cary, NC) was used for all the analyses. P< 0.05 was considered statistically significant.
Results
The baseline demographic and treatment cycle characteristics of the study population are shown in Table 1, stratified by outcome (clinical pregnancy and/or multiple pregnancy). Table 2 shows the demographic and cycle characteristics, as well as pregnancy outcomes, stratified by cycle number (cycles 1 through 5). Patients with PCOS were younger and had a higher BMI than patients with UI (mean±SD: 28.9 ±4.3 vs. 32.2±4.3 years, p< 0.001; 35.2±9.3 vs. 26.8±6.4 kg/m2, p< 0.001). Unexplained infertility patients reported a higher level of education and household income than PCOS patients, and African-American and Hispanic race was more commonly reported among patients with PCOS (p< 0.001 for all comparisons). The percentage of patients conceiving a clinical pregnancy was comparable between patients with UI and PCOS (28.8% and 26.4%, p: 0.29, for UI vs. PCOS, respectively). However, a significantly higher proportion of multiple pregnancies was noted among women with UI (5.7% vs. 1.34%, p<0.0001, for UI vs. PCOS, respectively), probably due to use of of gonadotropins only in UI patients (AMIGOS). The majority (1349/1650: 81.7%) of the patients were treated with oral agents (41.0% and 40.8% for CC and LTZ, respectively), while the remaining 18.2% received gonadotropins (Table 1). Most pregnancies (79.6%), occurred within the first three cycles of treatments (Table 2).
Table 1.
Baseline and Cycle Characteristics of 1650 Patients Stratified by Pregnancy Outcome*
| Variable | Clinical Pregnancy (N=457) | Without Clinical Pregnancy (N=1193) | P value ^ | Multiple Pregnancy (N=61) | Without Multiple Pregnancy (N=1589) | P value ^ |
|---|---|---|---|---|---|---|
| Participant’s Age (yrs) | 30.24±4.44(457) | 30.85±4.60(1193) | 0.016 | 30.18±4.09(61) | 30.70±4.58(1589) | 0.384 |
| Partner’s Age (yrs) | 32.36±5.46(448) | 33.27±5.83(1160) | 0.004 | 32.69±5.26(59) | 33.03±5.76(1549) | 0.660 |
| BMI (kg/m 2) | 29.58±8.37(457) | 31.08±9.04(1193) | 0.002 | 26.70±8.07(61) | 30.81±8.88(1589) | <0.001 |
| Prior parity | 108/457(23.6) | 231/1193(19.4) | 0.055 | 19/61(31.1) | 320/1589(20.1) | 0.037 |
| Ovarian reserve measures | ||||||
| Baseline AMH (ng/ml) | 5.03±5.10(452) | 5.12±5.84(1170) | 0.750 | 3.58±3.40(60) | 5.15±5.70(1562) | 0.001 |
| Baseline FSH (IU/L) | 6.84±2.40(452) | 6.60±3.09(1171) | 0.105 | 6.72±2.40(60) | 6.67±2.93Q563) | 0.858 |
| Baseline E2 (pg/ml) | 42.11±39.41(452) | 43.61±33.91(1171) | 0.476 | 34.51±17.10(60) | 43.53±36.00(1563) | <0.001 |
| Antral Follicles Count ∞ | 20.57±9.64(344) | 20.95±9.50(888) | 0.526 | 18.75±9.09(53) | 20.94±9.55(1179) | 0.103 |
| Peak E2 (pg/ml) (AMIGOS only) | 728.73±630.19(257) | 753.78±755.57(629) | 0.613 | 954.46±698.71(51) | 733.81±721.00(835) | 0.034 |
| Total number pre-ovulatory follicle (AMIGOS only) | 4.83±2.68 (253) | 4.52±2.59 (627) | 0.117 | 6.12±3.25 (50) | 4.52±2.55 (830) | 0.001 |
| Infertility Diagnosis | 0.283 | <0.002 | ||||
| Unexplained infertility | 259/457(56.7) | 641/1193(53.7) | 51/61(83.6) | 849/1589(53.4) | ||
| Polycystic ovary syndrome (PCOS) | 198/457(43.3) | 552/1193(46.3) | 10/61(16.4) | 740/1589(46.6) | ||
| Treatment | 0.002 | <0.001 | ||||
| Clomiphene | 166/457(36.3) | 510/1193(42.7) | 14/61(23.0) | 662/1589(41.7) | ||
| Letrozole | 184/457(40.3) | 489/1193(41.0) | 13/61(21.3) | 660/1589(41.5) | ||
| Gonadotropin | 107/457(23.4) | 194/1193(16.3) | 34/61(55.7) | 267/1589(16.8) | ||
| Maximum Medication Dose | <0.001 | <0.001 | ||||
| 1 tablet Clomiphene | 38/446 a(8.52) | 69/1157(5.96) | 3/61(4.92) | 104/1542(6.74) | ||
| 2 tablets Clomiphene | 97/446 a(21.75) | 178/1157(15.38) | 10/61(16.39) | 265/1542(17.19) | ||
| 3 tablets Clomiphene | 24/446 a(5.38) | 246/1157(21.26) | 1/61(1.64) | 269/1542(17.44) | ||
| 1 tablet Letrozole | 49/446 a(10.99) | 65/1157(5.62) | 1/61(1.64) | 113/1542(7.33) | ||
| 2 tablets Letrozole | 96/446 a(21.52) | 186/1157(16.08) | 7/61(11.48) | 275/1542(17.83) | ||
| 3 tablets Letrozole | 35/446 a(7.85) | 219/1157(18.93) | 5/61(8.20) | 249/1542(16.15) | ||
| Gonadotropin <= 250 IU | 100/446(22.42) | 166/1157(14.35) | 31/61(50.82) | 235/1542(15.24) | ||
| Gonadotropin > 250 IU | 7/446(1.57) | 28/1157(2.42) | 3/61(4.92) | 32/1542(2.08) | ||
Data are presented as mean ± SD (total number) or number of subjects/total number (percentage);
AFC is referring to only those patients with AFC ≤40;
In 11 clinical pregnancies no dose data was available.
Student t-test was used for testing differences between the two groups for continuous variables; Chi-square or Fisher’s exact test was used for categorical variables.
Table 2.
Baseline Demographic & Cycle Characteristics and Outcomes, Stratified by Treatment Cycle Number *
| Variables | Cycle 1 | Cycle 2 | Cycle 3 | Cycle 4 | Cycle 5 | P value ^ |
|---|---|---|---|---|---|---|
| Number of Subjects | 1600a | 1347 | 1129 | 919 | 380 | |
| Participant’s Age (yrs) | 30.71±4.55(1600) | 30.66±4.55(1347) | 30.67±4.51(1129) | 30.67±4.44(919) | 29.15±4.10(380) | <0.001 |
| Partner’s Age (yrs) | 33.05±5.72Q565) | 33.01±5.67(1317) | 33.07±5.65(1103) | 33.11±5.61(899) | 31.92±5.42(370) | 0.016 |
| BMI (kg/m 2) | 30.58±8.85Q600) | 30.86±8.94(1347) | 31.17±9.10(1129) | 31.28±9.12(919) | 35.81±9.15(380) | <0.001 |
| Prior parity | 333/1600(20.81) | 269/1347(19.97) | 217/1129(19.22) | 178/919(19.37) | 75/380(19.74) | 0.857 |
| Ovarian reserve measures | ||||||
| Baseline AMH (ng/ml) | 5.05±5.65(1580) | 5.25±5.87(1331) | 5.34±5.82(1114) | 5.39±5.92(907) | 8.13±7.32(379) | <0.001 |
| Baseline FSH (IU/L) | 6.68±2.86Q581) | 6.61±2.82(1331) | 6.55±2.85(1114) | 6.54±3.01(907) | 6.21±3.54(379) | 0.001 |
| Baseline E2 | 42.53±34.09(1581) | 42.75±31.44(1331) | 43.18±31.54(1113) | 43.75±32.45(914) | 52.94±38.90(431) | <0.001 |
| Antral Follicle Counts © | 20.84±9.49(1194) | 21.13±9.55(993) | 21.38±9.51(828) | 21.49±9.38(663) | 25.08±9.76(192) | <0.001 |
| Peak E2 (pg/ml) (AMIGOS only) | 531.5±531.3 (886) | 545.18±528.81(709) | 503.08±442.47(579) | 553.96±647.72(460) | (−)∞ | 0.762 |
| Total number preovulatory follicles (AMIGOS only) | 4.47±3.71 (886) | 4.69±3.62 (709) | 4.57±3.83 (579) | 4.80±3.61 (460) | (−)∞ | 0.177 |
| Infertility Diagnosis | <0.001 | |||||
| Unexplained infertility | 886/1600(55.4) | 709/1347(52.6) | 579/1129(51.3) | 460/919(50.0) | (−)∞ | |
| Polycystic ovary syndrome (PCOS) | 714/1600(44.6) | 638/1347(47.4) | 550/1129(48.7) | 459/919(50.0) | 380/380(100.0) | |
| Treatment Type | <0.001 | |||||
| Clomiphene | 652/1600(40.75) | 560/1347(41.57) | 482/1129(42.69) | 407/919(44.29) | 204/380(53.68) | |
| Letrozole | 651/1600(40.69) | 563/1347(41.80) | 479/1129(42.43) | 393/919(42.76) | 176/380(46.32) | |
| Gonadotropin | 297/1600(18.56) | 224/1347(16.63) | 168/1129(14.88) | 119/919(12.95) | (−)∞ | |
| Maximum Medication Dose | <0.001 | |||||
| 1 tablet Clomiphene | 107/1599(6.69) | 73/1347(5.42) | 49/1129(4.34) | 41/919(4.46) | 33/380(8.68) | |
| 2 tablets Clomiphene | 275/1599(17.20) | 217/1347(16.11) | 170/1129(15.06) | 128/919(13.93) | 36/380(9.47) | |
| 3 tablets Clomiphene | 270/1599(16.89) | 270/1347(20.04) | 263/1129(23.29) | 238/919(25.90) | 135/380(35.53) | |
| 1 tablet Letrozole | 114/1599(7.13) | 73/1347(5.42) | 53/1129(4.69) | 35/919(3.81) | 28/380(7.37) | |
| 2 tablets Letrozole | 282/1599(17.64) | 236/1347(17.52) | 181/1129(16.03) | 141/919(15.34) | 48/380(12.63) | |
| 3 tablets Letrozole | 254/1599(15.88) | 254/1347(18.86) | 245/1129(21.70) | 217/919(23.61) | 100/380(26.32) | |
| Gonadotropin <= 250 IU | 262/1599(16.39) | 190/1347(14.11) | 137/1129(12.13) | 93/919(10.12) | (−)∞ | |
| Gonadotropin > 250 IU | 35/1599(2.19) | 34/1347(2.52) | 31/1129(2.75) | 26/919(2.83) | (−)∞ | |
| Outcome | ||||||
| Live Birth (%) | 133/1600(8.31) | 94/1347(6.98) | 79/1129(7.00) | 58/919(6.31) | 25/380(6.58) | 0.354 |
| Clinical Pregnancy (%) | 150/1600(9.38) | 110/1347(8.17) | 95/1129(8.41) | 63/919(6.86) | 28/380(7.37) | 0.246 |
| Multiple Pregnancy (%) | 19/1600(1.19) | 20/1347(1.48) | 13/1129(1.15) | 8/919(0.87) | 1/380(0.26) | 0.324 |
Data are presented as mean ± SD (total number) or number of subjects/total number (percentage);
In 50 subjects no cycle data was available.
One-way ANOVA was used for testing differences between groups for continuous variables; Chi-square or Fisher’s exact test was used for categorical variables.
Only PPCOS-II women underwent a 5 cycle, AMIGOS patients were randomized up to a maximum of four cycles.
AFC is referring to only those patients with AFC ≤40
Clinical Pregnancy
Four hundred and fifty seven (27.7%) patients conceived a clinical pregnancy with 40.3%, 36.3%, and 23.4% of the conceptions resulting from LTZ, CC, or gonadotropins, respectively. As expected, patients who conceived were younger (p=0.016), with lower BMI (p=0.002), and more likely to have a younger partner (p=0.004). Mean serum AMH and antral follicle counts (≤40) were comparable between patients who conceived and those who did not (p=0.750 and p=0.526, respectively, Table 1). Among patients with extremely high AFC (>40), those conceiving had lower mean values compared to those not achieving pregnancy (p=0.026), (data not shown in Table 1). Overall, diagnosis was not related to the chance of pregnancy (except in the first cycle, where patients with PCOS had a lower chance to achieve a clinical pregnancy than UI patients, Supp. Table 1), but the type of treatment and medication doses were. At least one-half of the patients undergoing a 4th or a 5th cycle [49.5% (455/919) and 61.8% (235/380)], respectively, Table 2] were at maximum dose for either CC or LTZ (150mg or 7.5mg, respectively). However, most patients who conceived on either CC (135/159: 84.9%) or LTZ (145/180: 80.6%) did so with either one or two tablets/day (50 and 100 mg or 2.5 and 5 mg, for CC and LTZ respectively). A significantly lower proportion of women conceived on three tablets of either medication (Table 1). This finding was most prominent in the 2nd cycle where the odds of achieving a pregnancy were significantly reduced on either 3 tablets of CC or LTZ compared to one [OR (95%CI): 0.040 (0.011, 0.151), p<0.001; and 0.028 (0.006, 0.134), p<0.001 for CC and LTZ, respectively] and evident in the 3rd cycle for 3 tablets of CC compared to one [OR (95%CI): 0.2 (0.051, 0.787), p: 0.021; (Supp. Tables 2 & 3)]. The adjusted odds ratios (95% CI) for all predictors of clinical pregnancy were calculated per cycle and the results are summarized in supplemental tables 1–4. Of note, in treatment cycle # 5, none of the predictor variables had a p-value < 0.05 (data not shown). Predictor variables were associated with clinical pregnancy, but were inconsistent across all cycles. For example, in treatment cycle 1, women with a diagnosis of PCOS has significantly lower odds for a clinical pregnancy [OR: 0.43 (0.29, 0.62), p<0.001] but not in cycles 2 through 4 (Supp. Tables: 1–4).
Figure 1 depicts the ROC curves and the AUCs of the final model for each cycle for the outcome of clinical pregnancy. In the case of clinical pregnancy, the ROC curves and the AUC demonstrate that the model is fair for predicting clinical pregnancy rates for cycles 1 though 4.
Figure 1:
depicts the ROC curves and the AUC of the final model for each cycle in clinical pregnancy. Top panel shows cycles 1–3 and the bottom panel cycles 4–5.
Multiple Pregnancy
Sixty-one pregnancies (13.4%) involved multiples. Over half of those (55.7%) were conceived in gonadotropin stimulated cycles, while another 23.0% and 21.3% resulted from either CC or LTZ treatments, respectively. The majority of patients conceiving multiples on gonadotropins did so at a max daily dose ≤ 250 IU and not > 250 IU (50.8% vs. 4.9%). Variables including prior parity (p=0.037) and infertility diagnosis (p=0.002), as well as the patient’s BMI (p<0.001), baseline AMH (p=0.001), antral follicle count (p<0.001, data not shown in Table 1), type of treatment (p<0.001), maximum daily medication dose (p<0.001), peak E2 levels (p=0.034), and total number of pre-ovulatory follicles (p=0.001 were all associated with whether the participant conceived a multiple pregnancy (Table 1).
Few of the predictor variables were significantly associated with the outcome of multiple pregnancy (i.e.: age, prior parity etc.) for one or more cycles, however gonadotropin use had the strongest association and this persisted through cycles 1–4. Overall, the odds of conceiving multiples were higher with gonadotropins than with CC. The ORs increased progressively from the 1st to the 4th cycle [OR (95%CI): 4.6 (1.2, 17.2), p<0.02; 6.2 (1.2, 21.0), p=0.003; 6.3 (1.7, 24.2), p=0.007; 8.5 (1.4, 50.5), p=0.019; for cycles 1 through 4, respectively).
Supplemental Figure 1 depicts the ROC curves and the AUCs of the final model for each cycle for the outcome of multiple pregnancy. In the case of multiple pregnancy, ROC curves and the AUCs suggested the model’s fair to good association for cycles 1 and 2, and good to excellent for cycles 3 through 4 with the model’s ability increasing with a higher number of cycles [AUC (95%CI): 0.78 (0.72–0.84) for cycle 1 and 0.86 (0.78–0.93) for cycle 4].
Cumulative Pregnancy Rate
Most of the variables associated with clinical and multiple pregnancy rates per cycle were also associated with the cumulative pregnancy rates across the trials. Partner age (p<0.001), prior parity (p=0.003), AMH level (p=0.037), lower medication dosage and a diagnosis of UI as opposed to PCOS (p=0.001) were all associated with a higher cumulative pregnancy rate (Supp. Table 5).
Finally, utilizing the putative predictors and the models described earlier, we asssociated the probability of having a clinical or a multiple pregnancy from the variables we assessed, and constructed an easy-to-use calculator.
The calculator is available on the following website: https://pregnancyprediction.medicine.yale.edu/CalDirect.html, (Fig. 2). For example in the case of a 26 year-old female parous woman with the diagnosis of PCOS, whose BMI is 36.5 kg/m2, an AMH of 5.0 ng/mL, AFC: 29, day-3 FSH: 5.5, baseline E2: 62 pg/mL and endometrial thickness of 5 mm, with a 26 year-old partner who is undergoing her 1st cycle of LTZ at a daily dose of 2.5 mg/d, the probability of a clinical pregnancy for her is estimated to be 21.4%, and the risk of multiple pregnancy is 2.0% for her.
Figure 2:
snapshot of pregnancy calculator for pregnancy outcomes. All variables are based on baseline or screening values from the two studies.
Discussion
Using clinically available, prospectively derived, patient-specific data, we demonstrate that the probability of a clinical pregnancy and the risk of a multiple pregnancy can be estimated for patients undergoing treatment with LTZ, CC or gonadotropins. Partner’s age, patient’s age and BMI, type of treatment, and maximum medication dose were all predictive of a clinical pregnancy; while BMI, AMH, AFC, parity, diagnosis, treatment type, maximum medication dose, peak E2 and the total number of pre-ovulatory follicles were all predictive of multiple pregnancy.
For multiple pregnancy, the most consistent predictor among all considered factors was gonadotropin use (AMIGOS only). By combining data from two, multi-center RMN RCTs, we created an easy-to-use calculator that might predict with reasonable accuracy, among patients with either UI or PCOS, the chances of a clinical or a multiple pregnancy taking into consideration patient specific characteristics. This hypothesis needs to be tested further in additional diagnoses.
With the advent of precision medicine, the advances in bioinformatics, artificial intelligence, and the availabitility of large biomedical data, the concept of delivering more individualized treatments by integrating the characteristics (demographic, diagnostic, genetic, psychosocial, lifestyle, treatment, etc.) that distinguish one patient from another, has gained ground in reproductive medicine. Recognizing the need for the availability of such tools in the field of reproduction, the Society for Assisted Reproductive Technology created an easy-to-access, online calculator that can individualize predictions for pregnancy (singleton and/or multiple) and live birth after IVF treatments. Similarly, recent efforts are focusing on integrating emerging ART, time-lapse imaging, and -omic technologies (genomics, transcriptomics, proteomics and metabolomics), to improve outcomes (15). Despite recoginizing the need for the availability of similar predictive models in OI/OS and/or IUI treatments, no such tools are widely available in reproductive medicine. Patient counseling and treatment planning is based on the results of previously published studies (2, 4, 7–9) but efforts to integrate this information and to provide predictive models with individual patient data have either been isolated, or are rudimentary (9). The current project is an attempt to bridge this gap by developing an easy to use, individualized prediction model to estimate the probability of pregnancy (singleton or multiple) after OI/OS and/or IUI for women with either polycystic ovary syndrome or unexplained infertility. Utilizing a well-characterized patient sample drawn from multiple sites throughout the US, we suggest it is possible to create a tool that can aid in the counseling and clinical decision making of patients with these types of infertility who are weighing treatment options. We believe this is an improvement over current practice, which seems to rely on previously published data, most of which is single-site, or internal, with clinic specific metrics that can suffer from smaller numbers and time related variations in outcomes. More precise estimation of the odds of pregnancy based on pooling data can be used to identify the subpopulations likely to benefit most from a given proposed treatment or who likely will fare better by moving earlier to IVF, and sparing expenss, side effects and psychological stress.
Our finding that baseline ovarian reserve did not differ between patients achieving pregnancy and those who did not, is likely explained by the inclusion of a large number of PCOS patients with expected high AMH values, which potentially masked an existing difference in prognosis (16). Supporting this, a significantly higher percentage of patients with very high AFCs (>40) was found among patients not achieving a clinical pregnancy. Furthermore, patients with diminished ovarian reserve diagnosis were not included in the study. On the other hand, others have shown that AMH alone is not predictive of reproductive success (defined as the cumulative probability of conception) in a population based sample of regularly cycling women followed until conception (17).
Strengths of this study include the combination of data sets, their multi-center designs which recruited patients from nine different sites with a broad geographic and socioeconomonic distribution across the USA. Our sample contained sufficient racial and ethnic diversity to allow for extrapolation to most of the subpopulations of the United States. A large number of variables identified allowed for further input into the model, to improve prediction. The prediction model and the calculation formula take into consideration most factors known to predict cycle specific outcomes. We opted to estimate individual (per cycle), as well as, cumulative probabilities of success, since both outcomes are of interest to the couple and the counseling clinician.
Random, population-based sampling was not performed however. Patients for both studies were selected based on specific criteria recruited by convenience sampling. Limitations of note include the following: the study population by design was limited to PCOS or unexplained infertility in a multi-site study. The logistic regression model used took into consideration the diagnosis but combined both populations. Other diagnoses, recorded at study enrollment and not in conflict with the two RTCs’ inclusion/exclusion criteria, were considered “historic” compared to the primary diagnoses (UI or PCOS), since there were assigned to the participating patients though previous fertility evaluations. Future validation of the model in subpopulations with other diagnoses will be required. In the meantime, the calculator should only be used in patients with the diagnosis of UI or PCOS, as those were defined in the two RCTs. The present version of the calculator is not built to predict the outcomes of a 5th cycle, although when data are available in the future, an extension is possible. Outcomes observed here were based on strict study protocols and specific center- and study-related variables. Patients that participated in either clinical trial might have otherwise differed from those that do not, thus limiting the generalizability of the findings. Identification and inclusion of genomic data that might further characterize patient’s response to treatment might prove to be useful and should be considered in future predictive models.
The ability to weigh the estimated chance of clinical pregnancy against the risk of a multiple pregnancy, for each individual cycle in real-time, might help guide decisions related to aborting a particular cycle, if risks seem to outweigh the benefits. Predictions can be made prior to the initiation of medication and used for counseling. This may inform medication choice and aid in estimating the cumulative probability of a clinical pregnancy, and facilitate decisions regarding when IVF is likely a better option.
Prior to initiation of the cycle, the predictive model might be used to counsel the patient regarding the risk of a multiple pregnancy, and this risk can be balanced against the overall probability of achieving clinical pregnancy in a shared decision making paradigm.
Conclusion
In summary, we observed that partner’s age, patient’s age and BMI, their type of treatment and the maximum medication dose were all predictor’s of a clinical pregnancy. BMI, AMH, AFC, parity, diagnosis, type of treatment, maximum medication dose, peak E2 and total number of pre-ovulatory follicles were all predictors of having a multiple pregnancy. For multiple pregnancies the most consistent predictor was gonadotropin use for OS. Our model combines the above the risks of having a multiple pregnancy with common fertility treatments among patients with either PCOS or UI. Hopefully this tool, when further modified to include other diagnoses, can assist in counseling and provide more precise therapeutic strategies “tailored” to each patient’s individual needs.
Supplementary Material
Supplemental Table 5: summarizes the adjusted odds ratios (OR) with 95% confidence intervals (95%CI) for the various predictive variables for cumulative clinical pregnancy and cumulative multiple pregnancy.
Supplemental Tables 1-4: summarize the adjusted odds ratios (OR) with 95% confidence intervals (95%CI) for the various predictive variables, stratified by cycle number
Supplemental Figure 1: depicts the ROC curves and the AUC of the final model for each cycle in multiple pregnancy. Top panel shows cycles: 1–3 and the bottom panel cycles: 4–5.
ACKNOWLEDGEMENTS
The authors would like to acknowledge the following individuals for their help in the development of the calculator: Jiuzhou Wang (PHDstudent, Division of Biostatistics, School of Public Health, University of Minnesota, USA), Yajie Duan (PHD student, Department of Statistics, Rutgers, the State University of New Jersey, Piscataway, USA), Zhe Cai (MPS student in Analytics, College of Professional Studies, Northeastern University, Boston, MA, USA).
Supported by grants from the National Institutes of Health, the Eunice Kennedy Shriver National Institute of Child Health and Human Development [R25HD075737 (to N.S.); U10 HD39005 (to M.P.D.); U10 HD38992 (to R.S.L.); U10HD055925 (to H.Z.); U10 HD077680 (to K.R.H)].
Footnotes
Clinical Trial Registration Numbers: AMIGOS NCT1044862; PPCOS II NCT00719186
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
References
- 1.Berker B, Kahraman K, Taskin S, Sukur YE, Sonmezer M, Atabekoglu CS. Recombinant FSH versus clomiphene citrate for ovarian stimulation in couples with unexplained infertility and male subfertility undergoing intrauterine insemination: a randomized trial. Arch Gynecol Obstet 2011;284:1561–6. [DOI] [PubMed] [Google Scholar]
- 2.Guzick DS, Sullivan MW, Adamson GD, Cedars MI, Falk RJ, Peterson EP et al. Efficacy of treatment for unexplained infertility. Fertil Steril 1998;70:207–13. [DOI] [PubMed] [Google Scholar]
- 3.Souter I, Dimitriadis Baltagi LM, Meeker JD, Petrozza JC Elevated day 3 follicle-stimulating hormone in younger women: is gonadotropin stimulation/intrauterine insemination a good option? Am J Obstet Gynecol 2014;211:62 e1–8. [DOI] [PubMed] [Google Scholar]
- 4.Diamond MP, Legro RS, Coutifaris C, Alvero R, Robinson RD, Casson P et al. Letrozole, Gonadotropin, or Clomiphene for Unexplained Infertility. N Engl J Med 2015;373:1230–40. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Legro RS, Brzyski RG, Diamond MP, Coutifaris C, Schlaff WD, Casson P et al. Letrozole versus clomiphene for infertility in the polycystic ovary syndrome. N Engl J Med 2014;371:119–29. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Society for Assisted Reproductive Technology SAR. What are my chances with ART? Available at: https://www.sartcorsonline.com/Predictor/Patient. Accessed February 02, 2021 7.
- 7.Hansen KR, He AL, Styer AK, Wild RA, Butts S, Engmann L et al. Predictors of pregnancy and live-birth in couples with unexplained infertility after ovarian stimulation-intrauterine insemination. Fertil Steril 2016;105:1575–83 e2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Huang H, Hansen KR, Factor-Litvak P, Carson SA, Guzick DS, Santoro N et al. Predictors of pregnancy and live birth after insemination in couples with unexplained or male-factor infertility. Fertil Steril 2012;97:959–67. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Goldman RH, Batsis M, Petrozza JC, Souter I. Patient-specific predictions of outcome after gonadotropin ovulation induction/intrauterine insemination. Fertil Steril 2014;101:1649–55 e1–2. [DOI] [PubMed] [Google Scholar]
- 10.Calculate the probability of a spontaneous ongoing pregnancy within 1 year (leading to live birth). Available at: https://www.freya.nl/probability.php. Accessed March 08, 2021. In.
- 11.Diamond MP, Mitwally M, Casper R, Ager J, Legro RS, Brzyski R et al. Estimating rates of multiple gestation pregnancies: sample size calculation from the assessment of multiple intrauterine gestations from ovarian stimulation (AMIGOS) trial. Contemp Clin Trials 2011;32:902–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Legro RS, Kunselman AR, Brzyski RG, Casson PR, Diamond MP, Schlaff WD et al. The Pregnancy in Polycystic Ovary Syndrome II (PPCOS II) trial: rationale and design of a double-blind randomized trial of clomiphene citrate and letrozole for the treatment of infertility in women with polycystic ovary syndrome. Contemp Clin Trials 2012;33:470–81. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Rotterdam EA-SPCWG. Revised 2003 consensus on diagnostic criteria and long-term health risks related to polycystic ovary syndrome. Fertil Steril 2004;81:19–25. [DOI] [PubMed] [Google Scholar]
- 14.Practice Committee of the American Society for Reproductive Medicine. Practice Committee of the American Society for Reproductive M. Evidence-based treatments for couples with unexplained infertility: a guideline. Fertil Steril 2020;113:305–22. [DOI] [PubMed] [Google Scholar]
- 15.Simon C, Sakkas D, Gardner D, Critchley H. Precision medicine in human reproduction. Special Issue. Biol Reprod 2019;101:1075. [DOI] [PubMed] [Google Scholar]
- 16.Mumford SL, Legro RS, Diamond MP, Coutifaris C, Steiner AZ, Schlaff WD et al. Baseline AMH Level Associated With Ovulation Following Ovulation Induction in Women With Polycystic Ovary Syndrome. J Clin Endocrinol Metab 2016;101:3288–96. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Steiner AZ, Pritchard D, Stanczyk FZ, Kesner JS, Meadows JW, Herring AH et al. Association Between Biomarkers of Ovarian Reserve and Infertility Among Older Women of Reproductive Age. JAMA 2017;318:1367–76. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supplemental Table 5: summarizes the adjusted odds ratios (OR) with 95% confidence intervals (95%CI) for the various predictive variables for cumulative clinical pregnancy and cumulative multiple pregnancy.
Supplemental Tables 1-4: summarize the adjusted odds ratios (OR) with 95% confidence intervals (95%CI) for the various predictive variables, stratified by cycle number
Supplemental Figure 1: depicts the ROC curves and the AUC of the final model for each cycle in multiple pregnancy. Top panel shows cycles: 1–3 and the bottom panel cycles: 4–5.