Difference between mean gestational sac diameter and crown-rump length as a marker of first-trimester pregnancy loss after in vitro fertilization.

A total of 1,970 embryo transfers resulting in a clinical pregnancy occurred during the study period. Of these, 89 were excluded because outcome data were not available, 623 for missing CRL or GSD measurements, and 15 because of a twin gestations. The remaining 1,243 singleton pregnancies were retained for analysis. Demographics of the study population are represented in Table 1 . Participants were primarily white (92.5%) with a mean maternal age of 32.6 ± 4.24 years and an average BMI of 26.9 kg/m 2 . The most common infertility diagnosis was male factor (43.8%), followed by ovulatory dysfunction (27.8%) and tubal factor (18.7%). Thirty-eight percent of the study population had more than one diagnosis. A majority of embryo transfers were performed on day 5 (71.8%) with the use of fresh embryos (66.5%). One-hundred eighty-five pregnancies resulted in a miscarriage before 20 weeks’ gestation, for an overall incidence of 14.9%. Maternal age was higher (34.7 ± 4.45 versus 32.2 ± 4.10 years) and early ultrasound more commonly performed (36.2% versus 17.5%) in pregnancies that ended in loss. Pregnancy complications were rare, with hypertensive disorders (8.5%) and glucose intolerance (5.5%) being the most common. The majority of ultrasounds (85%) were performed from 6 weeks 6 days to 7 weeks 2 days (man 7 weeks 0.5 days, SD 1.61 days). Univariate analyses demonstrated significantly higher rates of first-trimester pregnancy loss among patients with a mGSD-CRL <5mm compared with the referent mGSD-CRL of 5–9.99 mm (43.7% vs. 15.8%, p<.0001), as seen in Figure 1 . In contrast, miscarriage was significantly less common in patients with mGSD-CRL 10–14.99mm (9.9% vs. 15.8%; p<.007) or ≥15mm (7.1% versus 15.8%; p<.018) compared to the referent mGSD-CRL of 5–9.99mm. Multivariate analyses controlling for maternal age, cycle number, day of transfer, performance of an early ultrasound, and presence of subchorionic hemorrhage revealed a fourfold higher odds of first-trimester loss in pregnancies with mGSD-CRL <5 mm (odds ratio [OR] 4.27, 95% confidence interval [CI] 2.57–7.08; Table 2 ) and a decreased odds of loss with mGSD-CRL of 10–14.99 (OR=0.67, 95%CI 0.45–0.99) and ≥15 (OR 0.43, 95% CI 0.21–0.91). There were no significant associations between high or low mGSD-CRL and hypertensive disorders, glucose intolerance, or other pregnancy complications. No correlations were found between mGSD-CRL and infant birth weight (Spearman rho= 0.07) or gestational age at delivery (Spearman rho=0.04). The model favored by BIC for predicting pregnancy loss included CRL, mGSD, and an unspecified interaction effect ( Table 3 ). The area under the receiver operating characteristic curve for this model was 0.996. The second-best model included both CRL and mGSD without an interaction effect, followed by CRL alone and mGSD alone. The model for mGSD-CRL was a less favorable predictor of pregnancy loss ( Table 3 ). Although the interaction effect between mGSD and CRL in the favored model is unknown, it is likely not a simple difference, because the model considering mGSD-CRL was less favored by the BIC by nearly 200 points. Because a ratio is generally more consistent from a mathematical perspective than a simple difference, the mGSD/CRL ratio was examined. The log(mGSD/CRL) ratio was also included, because the distribution of the mGSD/CRL was skewed to the right. The two ratios were the least favorable predictor models.

We conducted a retrospective cohort study of in vitro fertilization (IVF) cycles performed at the University of Iowa Hospitals and Clinics (UIHC) from January 2005 through December 2014. We included all IVF cycles with a fresh or cryopreserved autologous embryo transfer resulting in a singleton gestation with transvaginal ultrasound measurements of mGSD and CRL taken at UIHC at a gestational age of 6 weeks 3 days and 8 weeks (standard IVF protocol). The first ultrasound within the specified gestational age range was included in the analysis, regardless of pregnancy viability. Cycles were excluded if they did not meet the above criteria, if a multiple gestation pregnancy resulted, or if pregnancy outcome data were not available. The Institutional Review Board at the University of Iowa approved this study. The primary exposure was mGSD-CRL (categorized as <5mm, 5–9.99mm, 10–14.99mm and ≥15.00mm for the purposes of analysis). Primary outcomes were first trimester pregnancy loss (based on previously established criteria ( 3 )) and gestational age at delivery. Secondary outcomes were infant birth weight and pregnancy complications, including hypertensive disorders, glucose intolerance/diabetes, preterm labor <37 weeks gestational age, premature rupture of membranes, and other pregnancy complications. Cycle data were abstracted from the UIHC IVF database, electronic medical records, and the UIHC’s ultrasound imaging and storage software (R4 V4.20 [Hyland Software]). All ultrasounds were performed at the University of Iowa by a limited number of experienced sonographers. Data were analyzed with the use of SPSS v22 (IBM Corp.). Generalized estimating equations were used to control for patients with multiple cycles within the data set. A priori power analysis indicated a sample size of 1,217 (100 participants with a mGSD-CRL <5mm and 1117 with a mGSD-CRL ≥5mm) would be sufficient to detect a 10% absolute increase in first trimester pregnancy loss among participants with a mGSD-CRL <5 compared to those with a mGSD-CRL ≥5mm, assuming a 15% baseline prevalence of first trimester loss, at an alpha level of 5% and a power of 80%. Power would be greater for pregnancy complications with baseline prevalence lower than 15%. Assuming an 11.4 % prevalence of preterm birth ( 13 ), 9% prevalence of hypertension ( 14 ), 5% prevalence of gestational diabetes ( 13 ), and 3% prevalence of premature rupture of membranes ( 15 ), the calculated sample size was determined to provide sufficient power to assess a 10% absolute difference in these secondary outcomes among participants with a mGSD-CRL <5 compared with ≥5mm, a degree we felt would be clinically significant. A sample of 1,217 was determined sufficient to detect a 10% correlation between mGSD-CRL and birthweight or gestational age at delivery, because a power analysis for Pearson’s correlation with the use of an alpha of 0.05 and a power of 0.80 for a two-tailed test indicated a sample size of 782. Because Spearman rank correlation is computationally identical to Pearson product-movement coefficient, this would give sufficient power for either test. The range of mGSD-CRL in the study population followed a normal distribution (mean ± SD 10.05 ± 3.89mm). When examined categorically, the most common mGSD-CRL was 5–9.99mm (41.8%), and this was used as the referent population for multivariate analyses. Individual regression models were run for demographic and clinical values. Covariates included maternal age, parity, race (white, non-white, unknown), body mass index (BMI) reported at cycle start, infertility diagnosis (ovulatory dysfunction, male factor, diminished ovarian reserve, endometriosis, tubal factor, unexplained, combined, other/unknown), history of recurrent pregnancy loss, presence of concurrent subchorionic hemorrhage, use of preimplantation genetic testing, use of intracytoplasmic sperm injection, cycle number, fresh versus frozen transfer, and day of transfer (day 3 vs. day 5). The presence of an early ultrasound before 6 weeks 3 days was also accounted for because patients with abnormal rises in hCG or early bleeding are often referred for an early ultrasound and these factors may independently predict abnormal pregnancies. Covariates with significant associations (p<.05) were included in the final combined regression model; these included maternal age, cycle number, day of transfer, performance of an early ultrasound, and presence of subchorionic hemorrhage. The relationship between mGSD-CRL and birth weight, gestational age at delivery, and gestational age at loss was assessed with the use of Spearman rank order correlation, because preliminary analyses indicated violation of the parametric assumption of normality for these outcome variables. To determine if mGSD-CRL is a better predictor than either mGSD or CRL alone, the generalized linear mixed modeling framework was used with a logit link function. Seven main fixed effect predictor sets- 1) CRL alone; 2) mGSD alone; 3) mGSD-CRL difference; 4) mGSD/CRL ratio; 5) log(mGSD/CRL) ratio; 6) CRL, mGSD; and 7) CRL, mGSD- with an interaction effect were compared using the Bayesian information criterion (BIC) in order to determine the most favorable predictor set ( Table 3 ). Log(mGSD/CRL) was included because the distribution of the mGSD/CRL ratio was skewed to the right. The model does not specify the interaction. Each predictor set included a unique participant identifier as a random effect to control for patients with multiple cycles within the data set.

In contrast to previous studies showing pregnancy loss rates as high as 94% in spontaneously conceived pregnancies with mGSD-CRL <5 mm on first-trimester ultrasound ( 6 ), we found a much lower rate of pregnancy loss (43.7%) with mGSD-CRL 50% of IVF gestations with mGSD-CRL <5 mm go on to form viable pregnancies. The most important role of a marker such as mGSD-CRL may therefore be to identify patients at increased risk for first-trimester pregnancy loss. This is similar to the previously reported threefold increase in miscarriage noted with an enlarged yolk sac (≥5mm) ( 16 ). Although neither marker can diagnose a failed pregnancy, both warrant closer follow-up when noted. The reason for the discrepancy between our findings and those of Bromley et al. remains unclear but may be related to the fact that our study used slightly different gestational ages (6 weeks 3 days to 8 weeks vs. 5 weeks 5 days to 9 weeks) and included a larger number of women with mGSD-CRL <5 (103 vs. 16). Given that miscarriage rates are generally equivalent in unassisted and IVF pregnancies ( 17 , 18 ), it seems unlikely that assisted reproductive technology provides any inherent protective effect against pregnancy loss. In fact, some authors have suggested that the altered hormonal milieu associated with ovarian stimulation may adversely affect endometrial receptivity and IVF outcomes ( 19 – 23 ) and increase the risk of ectopic pregnancy ( 24 , 25 ). It is more probable that a combination of improved ultrasound technology, adjustment for the confounding effects of maternal age, and the more accurate dating of IVF pregnancies contributed to the lower rates of pregnancy loss seen in our population. Most of the existing studies on mGSD-CRL and miscarriage examined the utility of a specific cutoff value in predicting pregnancy loss. One study found that a mGSD-CRL of ≤10.1 mm could predict spontaneous abortion with a sensitivity of 56.4% and specificity of 82.1% ( 5 ). Another concluded that there was a 95% probability of a viable pregnancy with the use of a cutoff of >10 mm ( 7 ). Our results compare favorably with these data, because we found a combined miscarriage rate with mGSD-CRL <10mm to be 59.5% (43.7% with mGSD-CRL <5mm and 15.8% with mGSD-CRL 5–9.99mm). A miscarriage rate of 7.1% was noted in the mGSD-CRL ≥15mm group. Our study was unique in that it looked at several 5mm increments of mGSD-CRL and found a significant linear decrease in rates of pregnancy loss with each incremental increase in mGSD-CRL. To our knowledge, this finding has not been reported previously in the literature. Importantly, second- and third-trimester pregnancy complications, such as hypertensive disorders, glucose intolerance, preterm labor, and preterm premature rupture of membranes – were not associated with mGSD-CRL. mGSD-CRL was also not predictive of birth weight or gestational age at delivery. Although IVF is associated with a slightly increased risk of poor obstetrical outcomes including preterm delivery, low birth weight, gestational diabetes, and preeclampsia, compared to unassisted conceptions ( 26 , 27 ), these findings can be used to reassure patients with a low mGSD-CRL whose pregnancies progress past 20 weeks of viability that this does not appear to add additional risks. The presence of an early ultrasound prior to 6 weeks 3 days was predictive of early pregnancy loss in our study population. This likely represents confounding by indication, because early signs of miscarriage, such as vaginal bleeding or abnormal rise in serum levels of hCG, would typically prompt an early ultrasound. The decreased pregnancy loss associated with subchorionic hemorrhage appears to be counterintuitive. It is possible that a subchorionic hemorrhage impinging on the gestational sac might distort measurements of the GSD, making the mGSD-CRL appear lower than it actually is. Although day of transfer was associated with increased odds of pregnancy loss in the individual model, the association was not significant in the combined regression model. When multiple models were compared with the use of the BIC, the model for mGSD-CRL was determined to be one of the least appropriate for predicting early pregnancy loss. Models considering mGSD and CRL alone were both better at predicting pregnancy loss than the difference (mGSD-CRL). This is likely due to the fact that mGSD-CRL describes the difference in these variables, but not how the difference was achieved. For example, both increasing CRL and decreasing mGSD would decrease the mGSD-CRL. It seems unlikely that changes in these two variables have equivalent effects on pregnancy loss. In other words, a change in one of these variables might lead to a more significant effect than the other. In addition, these measurements are constantly changing with early fetal growth, and normative values have been established for different gestational ages ( 2 , 28 , 29 ). Therefore, although an association does exist between mGSD, CRL, and pregnancy loss, the true association is complex and involves an interaction other than the difference (or ratio) between the two primary parameters. The simple difference between mGSD and CRL does not appear to offer any significant predictive advantage over using either measurement alone. Because our statistical model did not specify the nature of the interaction, further testing is needed to clarify this relationship. The present study is the first to examine the difference in mGSD-CRL in a large cohort of infertility patients, the first to investigate the relationship between mGSD-CRL and multiple pregnancy outcomes, and the first to compare multiple models using CRL, mGSD, and mGSD-CRL. Because IVF pregnancies are routinely followed with early sonograms and outcome data are tracked for all patients, we had the advantage of a large sample size with known dates of conception, allowing us to control for a number of potential confounding variables. Despite this, our study has several limitations. Although ultrasounds were completed at a single institution by a limited number of experienced sonographers, differences in measurement techniques between sonographers may still exist. In addition, advances in ultrasound technology over the study period likely allowed for more accurate measurements of both mGSD and CRL. The impact of several lab changes and procedures remains unknown. Specifically, vitrification was implemented in September 2012, with slow freezing used before that. Assisted hatching is routinely performed on embryos cryopreserved at the blastocyst stage. The impact of this on the mGSD-CRL and subsequent outcomes is not known. Also, our conclusions cannot necessarily be extrapolated to patients who conceived without IVF. Early ultrasounds are not routinely performed for non-IVF patients at our institution, so there is no way to retrospectively examine these measurements in this population. Finally, although pregnancy outcome can be objectively tracked, pregnancy complications were based on patient report, which may be subject to recall bias. Despite these limitations, we were able to demonstrate that a complex relationship exists between mGSD and CRL and that miscarriage rates with mGSD-CRL <5mm in an IVF population are significantly lower than previously reported in unassisted pregnancies. Future studies should prospectively evaluate this relationship in assisted and unassisted pregnancies at multiple institutions with the goal of developing a statistical model to predict the risk of miscarriage based on routine early ultrasound measurements. Serial ultrasounds may also be helpful to determine the optimal timing for performing these measurements.

An inverse relationship exists between mGSD-CRL and miscarriage in the first trimester. Although our results are consistent with previous studies in the general population showing significantly higher early pregnancy loss rates with mGSD-CRL <5mm, the overall risk was much lower than previously reported in our population of IVF patients. In addition, there seemed to be a protective effect with increasing mGSD-CRL differences. A small mGSD-CRL was not associated with an increased risk of complications in those pregnancies that continued beyond 20 weeks. Although the mGSD-CRL difference is important, the relationship between these variables is complex and the individual contribution of mGSD and CRL to the difference appears to be more important in predicting miscarriage than the difference itself.

First trimester ultrasounography is routinely performed after infertility treatment to confirm pregnancy location and assess viability. Mean gestational sac diameter (mGSD; defined as average of measurements taken in three dimensions) ( 1 ) and crown-rump length (CRL; defined as the average of three measurements of the longest fetal length) ( 2 ) are two parameters routinely measured during early sonograms that are part of well-established criteria for diagnosing early pregnancy loss. Specifically, mGSD ≥25mm without an embryo, CRL ≥7mm without a fetal heartbeat ≥2 weeks after identifying a gestational sac or ≥11 days after identifying a gestational sac and yolk sac are all considered to be diagnostic of a nonviable gestation ( 3 ). These values were confirmed in a prospective multicenter observational study ( 4 ). Although these criteria are useful to diagnose a nonviable gestation, clinicians often seek to predict pregnancy loss by identifying at-risk patients in advance of the pregnancy failure. Several studies explored associations between mGSD, CRL, and future pregnancy outcome ( 1 – 3 , 5 – 12 ). Small CRL (>2 SD below the mean) has been associated with small for gestational age infants ( 8 ) and mGSD <50 th percentile has been associated with chromosomal abnormalities such as triploidy and trisomy 16 ( 12 ). Both small CRL and small mGSD have also been independently associated with first trimester pregnancy loss ( 3 ). A few authors examined the difference between mGSD and CRL (mGSD-CRL) as another potential predictor of pregnancy loss after unassisted conceptions, though studies are older and limited by small numbers of participants, variable dating and the resultant inability to account for confounding factors, along with outdated ultrasound technology ( 5 – 7 ). For example, miscarriage rates as high as 94% were reported with a mGSD-CRL <5mm, but conclusions were based on data from only 16 pregnancies ( 6 ). To date, no large-scale studies have examined the predictive value of mGSD-CRL for adverse pregnancy outcomes in an infertile population with a known date of conception with the use of modern ultrasound technology. The primary objective of the present study was to determine if an association exists between mGSD-CRL and first-trimester pregnancy loss. Secondary objectives were to examine the association between mGSD-CRL and pregnancy complications and to compare the predictive value of mGSD-CRL with that of either parameter alone. We hypothesized that a mGSD-CRL <5mm would be associated with an increased risk of pregnancy loss as compared to the general population but that the magnitude of risk may be substantially lower than previously reported. We further hypothesized that mGSD-CRL would not be associated with adverse pregnancy outcomes beyond the first trimester and anticipated that the mGSD-CRL difference would be a more effective predictor of pregnancy loss than either mGSD or CRL alone.