For women in early pregnancy with symptoms of abdominal pain, with or without vaginal bleeding, ultrasonography can diagnose intrauterine pregnancy in 70–90% of cases. However, in the 10–30% of cases1,2 in which ultrasonography cannot definitively diagnose an ongoing pregnancy, miscarriage, or ectopic pregnancy, diagnosis is difficult when the presenting human chorionic gonadotropin (hCG) level is below the discriminatory zone (less than 2,000–3,000 milli-international units/mL).3–7 Incorrect diagnosis in these patients has a heavy “cost” of interruption of a desired, viable intrauterine pregnancy or rupture of an ectopic pregnancy.
Although trends in hCG levels are clinically used to aid in the prediction of a viable intrauterine pregnancy in women symptomatic of possible miscarriage or ectopic pregnancy, mathematical and clinical models are not completely accurate.8,9 Our previous research10 demonstrated that the minimal hCG rise for a viable intrauterine pregnancy was 53% at 2 days and that 99% of viable intrauterine pregnancy demonstrated this rise when the hCG was less than 5,000 milli-international units/mL. However, this model was developed in a relatively homogenous racial and ethnic population. Moreover, women with uncertain gestation (based on uncertain last menstrual period) were not included in the model. This factor may have limited the generalizability of the findings.
Accurate diagnosis of women presenting to the emergency department in early pregnancy with pain, with or without bleeding, is critical. If treated as a nonviable pregnancy, interruption of a potentially viable intrauterine pregnancy may result. Failure to diagnose an ectopic pregnancy increases risks of maternal morbidity and mortality if it later ruptures. Relying only on the change in hCG values can lead to misdiagnosis of women with an early symptomatic gestation.11,12
The current research took a diverse cohort of U.S. women and assessed whether variation in hCG measures, used to assess early gestation viability, was associated with differences in patient risk factors and demographics. We evaluated the effect of differences in race, ethnicity, and other clinical variables such as prior miscarriages, bleeding, or pain at presentation on the profiles of serial hCG measurements.
MATERIALS AND METHODS
This retrospective cohort study was approved by the institutional review boards of the University of Pennsylvania, the University of Southern California, and the University of Miami.
The cohort for this study consisted of pregnant women who presented between 2007 and 2011 to the emergency department of these three academic institutions with pain with or without bleeding. Patients were considered to be eligible for the study if the presenting hCG value was 10,000 milli-international units/mL or less, at least two hCG values were available separated by no more than 7 days, and there was a final diagnosis of a viable intrauterine pregnancy defined as one having fetal heart motion noted on ultrasonography at 12 weeks or later. The ultrasound scans of all patients in this cohort were reviewed and interpreted by the radiology physicians, and the ultrasound findings were communicated to the emergency department attending physician before the gynecology consultation. Serum hCG concentrations were quantified using a chemoluminescent technique (AxSYM) and reported in milli-international units/mL using the Third International Reference Preparation (average interassay coefficient of variation at the three institutions 7.8%). Information on last menstrual period, race, ethnicity (self-reported), maternal age, pain, bleeding, and obstetric history was collected at the time of the initial visit. The sample size for this project represented all of the data available that meet these criteria after 4 years of collection.
All analyses were conducted in SAS 9.2. Analyses were conducted on the natural log transformation of the hCG values to reduce the skewness of the distribution. Data presented in the tables and figures have been converted back to normal values (ie, untransformed values) to aid in clarity of presentation.
The curves were modeled using two methods. As in the prior article,10 we used a linear fixed and random-effects model for natural log-transformed hCG to generate the slope per respondent and the upper and lower standardized percentiles for those predicted slopes. This methodology was used to validate our previously reported findings. This method demonstrated reliability of the data from the University of Pennsylvania site, but this model did not validate well with the additional study sites, detailed subsequently.
To accommodate data from all sites, we used a curve registration approach.13 This is a nonlinear fixed and random-effect model that included a random effect on the time axis. This was used to estimate a random shift in the time axis at each site and by each person relative to each other. This model shifted the curve for each respondent by an amount of days from the presentation date to generate the best fitting overall curves. By using this model, we did not need to know the date of the last menstrual period. The model computed a correction factor to adjust the starting point of the projected hCG curve on the Y axis (ie, time) to reflect gestational age rather than stating all curves from the date each patient presented for care. The variable, called “tau” or “normalized days since presentation,” shifted the time of the start of the curve (in days) for each respondent. The variable was composed of a specific fixed effect for their site plus a random effect attributable to each individual. This new day variable “lined up” the respondents based on their values and the shape of their individual curve. Because the rise of hCG is not linear, this minimized error that may be introduced if some women presented for care earlier or later in gestation, when the curve shape may have a different rate of rise.
Tau was defined as tau=days from presentation+a shift for site+a per participant shift (ie, random effect). Using this shift, we tested both a linear model with tau and a quadratic model:
- Linear model: log_hCG=intercept+b1×tau
- Quadratic model: log_hCG=intercept+b1×tau+b2×tau2
Versions of these models were compared to determine whether the time shift could account for the effects of sites and whether a linear model would be adequate over a short range of time from presentation. In addition, covariates including race, the presence of bleeding, and other background variables were explored to determine whether different predictive models would be needed for patients with these covariates.
To determine the best method for combining the data from the three sites, we created a model with a separate parameter for each of the three sites. In a second model, we added a time shift for each site. When we added the time shift, site was no longer a significant confounder. To determine whether a linear model for these data would be adequate over a restricted range instead of a quadratic model, we compared the fit of the linear and quadratic model using likelihood ratio statistics. The quadratic model fitted the data better (see subsequently) and was used to generate the estimates.
To create tables of predicted hCG values, we first reran the models taking out the fixed effects for the individual sites. The rationale for this was that when we apply this modification to the prediction of hCG values for a new site, the effect of that site would be unknown. Doing so slightly increased the variance of the prediction. We then computed the starting value of tau based on the initial hCG value. Higher hCG values shifted a person further in time from the presentation date, thus moving them into an area of the curve that had slower growth. Second, we computed the expected change in hCG values across time periods of 1, 2, 3, 4, and 7 days. Third, we estimated the percentile ranges in these values assuming a normally distributed per-person shift variable. Thus, the increase listed for the first percentile meant than only 1% of the population of patients with viable intrauterine pregnancies would have an increase in hCG this small or smaller.
We identified 1,165 symptomatic women who presented with early pregnancy pain, with or without bleeding, to the emergency department at the three sites: the University of Pennsylvania, the University of Southern California, and the University of Miami. We excluded 225 women with ectopic pregnancy and 561 with miscarriage, and 94 were lost to follow-up. A total of 285 (24.5%) women were diagnosed with viable intrauterine pregnancies and were included in this analysis. Of these women, 104 (36.5%) had two hCG values, 99 (34.7%) had three, 58 (20.4%) had four, and the remaining 24 (8.4%) had more than four serial hCG values.
Table 1 shows patient characteristics at each of the three sites. The population of women included in this study was diverse. Overall, African American women formed 54% of the study cohort and white women 38%; 8% described themselves as “other.” As expected, the three sites differed significantly (analysis of variance or χ2 test) in terms of race and ethnicity. The University of Pennsylvania and the University of Miami had higher proportions of African American women than did the University of Southern California, and the University of Southern California and the University of Miami had significantly higher numbers of Hispanic women than did the University of Pennsylvania. The sites showed differences in initial clinical presentation: pain at presentation (more pain at the University of Southern California and the University of Miami), last menstrual period certainty (higher certainty at the University of Miami and the University of Pennsylvania), lower gestational age noted in patients at the University of Pennsylvania than at the University of Miami and the University of Southern California (9.3 and 12.9 days longer, respectively), and lower mean hCG at presentation at the University of Pennsylvania (446 milli-international units/mL) followed by the University of Miami and the University of Southern California (1,480 and 1,636 milli-international units/mL, respectively). The significantly lower hCG levels at the University of Pennsylvania corresponded to the lower gestational age at presentation of these patients. Maternal age, history of prior miscarriage or ectopic pregnancy, and bleeding at presentation did not vary significantly between sites.
The figure in Appendix 1 (available online at http://links.lww.com/AOG/A836) shows the linear and quadratic models and their 95% confidence intervals (CIs). The scale of the graph was in the “normalized days from presentation.” The best fitting quadratic curve had an intercept of 6.24 (95% CI 6.04–6.45), a slope (b1) of 0.39 (95% CI 0.38–0.41), and a quadratic term (b2) of −0.008 [95% CI −0.008 to −0.007]. The per-person shift (the random effect) had a standard deviation of 3.45 days. The time shift for the University of Miami was 3.02 days (95% CI 2.07–3.97) and for the University of Southern California 3.21 (95% CI 1.74–4.67) days. Correspondingly, the quadratic model also fitted significantly better than the linear model for data from 0 to 7 days (χ2 =155.3, P>.001) and from 0–21 days (χ2 =373.9, P<.001). Thus, we used the quadratic model to generate the estimates.
The serial hCG curves showed significant variation (across their intercept, linear, and quadratic terms simultaneously, assessed by χ2 test) associated with race (African American compared with non–African American, P<.001), prior miscarriage (P=.014), presence of bleeding at presentation (P<.001), presence of pain at presentation (P=.002), and maternal age older than 34 years at presentation (P<.001). The effects for Hispanic ethnicity, smoking, and prior ectopic pregnancy were not significant (P>.05).
Table 2 shows the overall minimum (first percentile) increases in hCG levels over time. The overall minimum (first percentile) rise for a viable intrauterine pregnancy at 1 day and 2 days was faster when initial hCG was low (29% and 64% at hCG=500 milli-international units/mL, respectively) and was slower when the presenting hCG was high (16% and 33% at hCG=5,000 milli-international units/mL, respectively). The expected minimal (first percentile) rise in hCG for a viable intrauterine pregnancy at 2 days was 49% (or faster) when the initial hCG was 1,500 milli-international units/mL or less. When initial values were between 1,500 and 3,000 milli-international units/mL, the minimal rise was 40% and when the initial hCG value was greater than 3,000 milli-international units/mL, the minimal rise for a viable intrauterine pregnancy was as low as 33%.
Figure 1 shows the model for African American and non–African American women. A steeper rise was observed in African American women compared with non–African American women. Figure 2 depicts the hCG curves for the other factors associated with significantly different curves. Note that prior miscarriage, presence of bleeding, and presence of pain at presentation were associated with higher hCG levels over time, whereas maternal age older than 34 years was associated with a slower slope of hCG increase over time. As shown in Figure 2, some of the differences are subtle, consisting of a different initial intercept or differences in the slope resulting in statistical significance but unlikely to reflect clinical significance.
Table 3 shows the minimum (first percentile) percentage increases in hCG levels over time for African American women compared with women of other ethnicities. The difference in relative increase between African American women and non–African American women was more pronounced at low initial hCG values (500 milli-international units/mL) than at high initial hCG values (2,500 milli-international units/mL). However, the 2-day minimum (first percentile) rise in hCG for African American and non–African American women at low initial hCG (500 milli-international units/mL) was 65% and 62%, respectively, suggesting that there was not a large clinical difference.
We computed a linear slope for hCG rise by days from presentation separately for each site using a random intercept and slope model, as was done in our previous research.10 We used only days 0–10 to make the analysis more congruent with the analyses in the prior article. The slope for the University of Pennsylvania was 0.397 (95% CI 0.381–0.413), close to the 0.402 (0.389–0.415) value reported previously.10 The standard deviation of the random effect for slope was 0.074 close to the standard deviation of 0.083 found previously. However, the slopes for the University of Miami (0.282 [0.260–0.304]) and the University of Southern California (0.262 [0.223–0.303]) were only approximately two thirds of the slopes for the University of Pennsylvania, a finding consistent with a model of decreasing slope of hCG increase as the time from last menstrual period (gestational age and level of hCG) increases.
We have reassessed the minimal rise of hCG in a viable symptomatic intrauterine pregnancy in a diverse group of women. The expected minimal rise for a viable intrauterine pregnancy was faster when the initial hCG levels were low, suggesting that one predicted slope cannot be used for all women at all times. To limit interruption of a potential desired intrauterine pregnancy, a more conservative “cutoff” (slower rise) is needed when hCG values are high.
A new, statistical methodology was used to address the variation in modeling hCG curves when women are unsure of their last menstrual period (uncertain last menstrual period occurred 10–37% across the three sites). This methodology innovatively shifts the hCG curve for each woman in time based on the shape of the curve and the initial hCG value. By doing so, we obtained a unique predicted hCG rise for each woman based on a curve normalized to gestational age.
This analysis addressed issues from our previous research10 that estimation of the normal rise in hCG from initial clinical presentation may have included variation and uncertainty because the hCG curve may have a different slope at different initial values of hCG (or at different gestational ages). We have adjusted the modeled hCG curves to account for differences in hCG values by normalizing the “time of presentation” such that specific patient data contributed to the appropriate portion of the hCG curve. Women who presented early contributed to the modeling of the early portion of the curve, and women who presented late contributed to modeling of the later portion of the curve. Thus, the data presented can be applied to all women who present for care, including those unsure of their last menstrual period.
It has been previously suggested10 that a 53% rise after 2 days will identify 99% of pregnancies that are viable when the initial hCG is less than 5,000 milli-international units/mL. The current data suggest that this is likely true for women with an initial hCG value below 1,500 milli-international units/mL. A less stringent value, however, is needed for women with higher initial hCG values. The minimal rise of hCG in 2 days for women with an hCG value at 3,500 milli-international units/mL was predicted to be as low as 38%, and when the value was 5,000 milli-international units/mL, the rise was predicted to be as low as 33%. Based on these data, it is suggested that a minimal (first percentile) rise of 49% (or faster) is expected for initial hCG values below 1,500 milli-international units/mL, a rise of 40% for initial hCG values 1,500–3,000 milli-international units/mL, and a rise of 33% for initial hCG values above 3,000 milli-international units/mL. Note, these values are rounded from the raw data to establish whole numbers that can be used as clinical rules of thumb.
Although the diagnosis of the location and viability of a gestation is often made when hCG levels are high, it has been suggested12,14 that more time should be taken to confirm viability and location before intervention to prevent the interruption of a desired, viable intrauterine pregnancy. The concept that there is a value above which a nonviable gestation (miscarriage or ectopic pregnancy) can be diagnosed if an intrauterine gestation sac is not visualized with transvaginal ultrasonography (often called a discriminatory zone) has been challenged. If a discriminatory zone is to be used, the hCG value should be conservative (high) with a suggested level of 3,000 milli-international units/mL or greater.14 The current data are consistent with our previous studies (that included suspected ectopic pregnancies), which identified viable intrauterine pregnancy with curves less than 53% and as low as 35%.11,12 Importantly, the application of this slower threshold for minimal rise may assist in preventing the inadvertent interruption of viable intrauterine pregnancies.
Repeating our analysis on a new sample of patients at our clinical site demonstrates the reliability of our previous findings, and inclusion of other sites enhanced generalizability to help us develop the minimal rise curve for viable intrauterine pregnancy that can be applied to racially and ethnically diverse sites in the United States. In doing so, we have established that a hCG rise is not the same for all races in women with symptomatic early pregnancy. Overall, African American women had a faster rise in hCG compared with non–African American women. This difference was most notable at low initial hCG values. Previous studies have documented racial and ethnic differences in hCG levels during the first and second trimesters.15 These differences may reflect differential trophoblast development and differentiation and may account for known disparities in intrauterine growth and perinatal morbidity.16 This previous study showed that slower early first-trimester hCG rise was associated with neonates who were of low birth weight or small for gestational age.
History of miscarriage, bleeding, pain, and maternal age at presentation statistically affected the rate of hCG rise, but differences were not clinically significant. For clinical diagnosis of symptomatic early pregnancy, the hCG differences based on race, history, or symptoms are small and no change in the minimal threshold for potential viability is recommended.
1. Shalev E, Yarom I, Bustan M, Weiner E, Ben-Shlomo I. Transvaginal sonography as the ultimate diagnostic tool for the management of ectopic pregnancy: experience with 840 cases. Fertil Steril 1998;69:62–5.
2. Kadar N, Caldwell BV, Romero R. A method of screening for ectopic pregnancy and its indications. Obstet Gynecol 1981;58:162–6.
3. Kirk E, Bourne T. Predicting outcomes in pregnancies of unknown location. Womens Health (Lond England) 2008:4:491–9.
4. Barnhart KT, van Mello N, Bourne T, Kirk E, Van Calster B, Bottomley C, et al. Pregnancy of unknown location: a consensus statement of nomenclature, definitions and outcome. Fertil Steril 2011;95:857–66.
5. Barnhart KT, Casanova B, Sammel MD, Timbers K, Chung K, Kulp JL. Prediction of location of symptomatic early gestation based solely on clinical presentation. Obstet Gynecol 2008;112:1319–26.
6. Casanova BC, Sammel MD, Chittams J, Timbers K, Kulp JL, Barnhart KT. Prediction of outcome in women with symptomatic first-trimester pregnancy: focus on intrauterine rather than ectopic gestation. J Womens Health (Larchmt) 2009;18:195–200.
7. Condous G, Van Calster B, Kirk E, Haider Z, Timmerman D, Van Huffel S, et al. Clinical information does not improve the performance of mathematical models in predicting the outcome of pregnancies of unknown location. Fertil Steril 2007;88:572–80.
8. Barnhart KT. Clinical practice. Ectopic pregnancy. N Engl J Med 2009;361:379–87.
9. Barnhart KT, Katz I, Hummel A, Gracia CR. Presumed diagnosis of ectopic pregnancy. Obstet Gynecol 2002;100:505–10.
10. Barnhart KT, Sammel MD, Rinaudo PF, Zhou L, Hummel A, Guo W. Symptomatic patients with an early intrauterine pregnancy: HCG curves redefined. Obstet Gynecol 2004;104:50–5.
11. Seeber BE, Sammel MD, Guo W, Zhou L, Hummel A, Barnhart KT. Application of redefined human chorionic gonadotropin curves for the diagnosis of women at risk for ectopic pregnancy. Fertil Steril 2006;86:454–9.
12. Morse CB, Sammel MD, Shaunik A, Allen-Taylor L, Oberfoell NL, Takacs P, et al. Performance of human chorionic gonadotropin curves in women at risk for ectopic pregnancy: exceptions to the rules. Fertil Steril 2012;97:101–6.e2.
13. Ramsay JO, Silverman BW. Functional Data Analysis (Springer Series in Statistics). New York (NY): Springer; 2005.
14. Doubilet PM, Benson CB, Bourne T, Blaivas M; Society of Radiologists in Ultrasound Multispecialty Panel on Early First Trimester Diagnosis of Miscarriage and Exclusion of a Viable Intrauterine Pregnancy, Barnhart KT, et al. Diagnostic criteria for nonviable pregnancy early in the first trimester. N Engl J Med 2013;369:1443–51.
15. Spencer K, Ong CY, Liao AW, Nicolaides KH. The influence of ethnic origin on first trimester biochemical markers of chromosomal abnormalities. Prenat Diagn 2000;20:491–4.
16. Morse CB, Barnhart KT, Senapati S, Sammel MD, Prochaska EC, Dokras A, et al. Association of the very early rise of human chorionic gonadotropin with adverse outcomes in singleton pregnancies following in vitro fertilization. Fertil Steril 2016;105:1208–14.
Supplemental Digital Content
© 2016 by The American College of Obstetricians and Gynecologists. Published by Wolters Kluwer Health, Inc. All rights reserved.