Monitoring repeat evaluations of serum human chorionic gonadotropin (hCG) levels is a clinical strategy used to gain important information about diagnosis and prognosis of early pregnancies. The trend of hCG levels is particularly valuable in the first several weeks of gestation, when ultrasonography is nondiagnostic. With levels less than 1,500 mIU/mL, the positive predictive value of ultrasonography for detection of intrauterine pregnancies is poor.1 At levels less than this discriminatory zone, monitoring serially measured serum levels becomes the essential diagnostic tool.
When evaluating the patient who presents with increasing levels of hCG, clinical decisions are facilitated by the well-defined curve of expected hCG rise. This curve, first described by Kadar et al,2 recently has been redefined.3 A 53% increase in hCG concentration in 2 days, rather than a 66% increase, is now considered the lower limit of normal and defines a viable intrauterine pregnancy (K.T. Barnhart et al, in press). A slower rate of increase from this expectation suggests a nonviable pregnancy and prompts intervention to distinguish an ectopic pregnancy from miscarriage.
The evaluation of the patient who presents with decreasing levels of hCG, without medical or surgical intervention, poses more of a clinical conundrum because the expected rate of decline in hCG has never been determined. It is clear that decreasing levels of hCG in the early gestation are indicative of a nonviable pregnancy, either a failed intrauterine pregnancy or an ectopic pregnancy. The management of such cases consists of either observation or intervention. With expectant observation, a spontaneous fall to zero obviates the need for intervention, whereas a “plateau” in the serial hCG levels suggests the need to remove trophoblastic tissue and to evaluate the patient for a possible ectopic pregnancy. Such decisions can be difficult and vary among clinicians likely because guidelines have not been established. Most clinicians feel that they would recognize a plateau or “abnormal” decline when confronted with one but, truly, the expected or “normal” rate of hCG decline is not known (data based on a MEDLINE search in English for the dates January 1962 to January 2004 using the key works “spontaneous abortion,” “miscarriage,” and “human chorionic gonadotropin”).
Studies evaluating patients with ectopic pregnancies who are treated with methotrexate have defined a 15% drop in 7 days as an adequate indication of treatment efficacy.4 Clinically, this rate is considered enough of a decline in hCG to indicate response to treatment and to justify deferring further treatment. Whether this 15% rule is appropriate to apply to the surveillance of untreated suspected spontaneous abortions is not known.
By investigating the hCG patterns in patients known to have spontaneous abortions, we propose to estimate a standard curve of expected hCG decline for gestations later confirmed to have a completed spontaneous abortion. This information will allow clinicians to better predict which failing pregnancies can be safely managed expectantly and thereby avoid unnecessary surgical procedures. It also may facilitate the development of guidelines for the management of early pregnancies with decreasing levels of hCG.
MATERIALS AND METHODS
The Hospital of the University of Pennsylvania has a computerized database dating back to 1990 of all pregnant women who presented to the emergency room in the first trimester with symptoms of pelvic pain and/or vaginal bleeding. Using this database, we retrospectively evaluated a cohort of women who presented between January 1, 1990, and July 31, 1999. If a woman had 2 (or more) pregnancies recorded in the database during this time period, 1 was chosen at random for inclusion in the analysis and the other(s) excluded. This study was approved by the Institutional Review Board of the University of Pennsylvania (Risk Factors of Predictors of Ectopic Pregnancy, #103700).
For inclusion, each patient had to have nondiagnostic transvaginal ultrasonography on presentation (no evidence of an intrauterine or extrauterine gestational sac) and follow-up with serial hCG measurements to determine the definitive diagnosis. For the current study, we extracted those women with decreasing levels who were ultimately diagnosed as having a spontaneous abortion, which is defined as a spontaneous decline of hCG levels to less than 5 mIU/mL in the absence of surgical or medical intervention or as the presence of products of conception on dilation and curettage. The data from women in the latter group were truncated at the time of surgery to look only at the portion of the curve that represented a natural fall. Additional criteria included an initial hCG value of less than 10,000 mIU/mL and a serum sample with undetectable hCG concentration within 84 days from presentation. This initial level was chosen because we were interested in studying patients who were candidates to follow serial hCG values as potentially having a “completed abortion.” We included subjects with hCG levels greater than a discriminatory zone to identify an intrauterine pregnancy because some women may present with the clinical situation suggestive of (but not definitive of) the passage of products of conception. Subjects with hCG values measured at duration of more than 84 days represented loss to follow-up and were excluded to prevent distortion of results. For example, most cases that resulted in greater than 84-day follow-up were the result of extended time between the penultimate and final hCG value (confirming that the value was undetectable). In this case, the true date that the level became undetectable could not be ascertained. Inclusion of these data would bias the results toward a longer resolution time.
Data regarding the date and value of all hCG determinations were exported for analysis. We evaluated the curve created by the serial hCG values. All analyses were performed on the natural log transformation for hCG values. To better approximate normality, this transformation was necessary to alleviate the skewness of the distribution of hCG and to reduce the influence of large values. The starting point of the prediction curve was designated as the hCG at presentation, with serial levels being evaluated until the resolution of pregnancy.
To characterize the shape of the curves, both semiparametric (method 1) and parametric (method 2) random effects models were explored. Recently developed semiparametric statistical techniques, splines5–7 used in method 1, enable us to consider the appropriateness of a linear relationship between the rate of change in log hCG and number of days since presentation or whether the function through time appears more complicated (eg, falling off or declining). Additionally in method 2, various parametric models were considered: linear, quadratic, cuboidal, and change-point models for this association. Whereas a quadratic decrease has a smooth decline with a faster initial rate, the change-point model assumes 2 straight lines that join at the change-point. A likelihood ratio test8 is used to detect the presence of a change-point.
To determine the influence of the starting value, we sorted the data into 3 groups based upon initial hCG value to see whether the model would be the same for different groups. The 3 groups were hCG levels of 500 mIU/mL or less at presentation, hCG levels between 500 and 2,000 mIU/mL at presentation, and hCG levels more than 2,000 mIU/mL at presentation. From the semiparametric evaluation, method 1, it appeared that initially there was a steep decline that tapered off. For lower initial values (< 500) the slope was less steep, which indicated the need to incorporate the initial level.
Both methods used random effects to account for the repeated hCG measurements contributed by each subject and are quite flexible even when the number and timing of observations are unbalanced.9 Both estimate a population average curve by aggregating the curves estimated for each subject individually. The polynomial spline model is a modern statistical approach to fit smooth curves without being restrictive on the shape of the curves.10 Curves generated using both methods were overlaid, and the graphs of the estimated population profile along with 95% confidence intervals for the average profile were compared. An optimal parametric model, method 2, was selected based upon substantial overlap in confidence intervals between the 2 curves (figure not shown).
Population average values for slope, standard errors, and upper and lower confidence bounds on the rate of decline were estimated from the log-linear model, which assumed that log (hCG) follows a normal distribution. This predictive model was used to calculate expected values at important clinical time points after presentation. We present slopes for the decline in log hCG as well as the percentage decline for a 2- to 28-day range. In addition to the average/median value for decline, we present 90% and 95% confidence estimates (percentiles) on the rate of decline.
Of the 1,543 patients for whom a definitive diagnosis could not be made at presentation, 838 were ultimately diagnosed with miscarriage. There were 719 patients who had decreased hCG levels from the time of presentation. A total of 9 patients did not meet the enrollment criteria and were excluded (hCG > 10,000 mIU/mL or duration of follow-up > 84 days), yielding a final sample of 710 patients who contributed 2,150 hCG measurements. Descriptive statistics regarding the number of visits per patient, days until diagnosis, and starting hCG values are presented in Table 1. A graphic representation of the data is shown in Fig. 1.
The optimal model was the parametric random effects model of log (hCG) with fixed quadratic effect and random linear effect for number of days since presentation with an interaction between the rate of decline (linear and quadratic slope parameters) and initial value of log (hCG). The average profile for log (hCG) at time since presentation, t, is described by the following equation:
The interaction between initial hCG and the slope (quadratic and linear) fixed effects is necessary to describe that faster rates of fall were associated with higher starting concentrations.
Important clinical differences among the curves may be better demonstrated as actual hCG values and percent reductions. Table 2 depicts the mean and the 90th and the 95th percentile of hCG values during the first 7 days after presentation. Table 3 depicts the fall in hCG values during the 4 weeks after presentation. The mean number of days until the hCG level was negligible ranged from 12 to 16 days. Tables 4 and 5 convert the hCG values into percent decline and clearly show that the rate of fall varies with the initial hCG level. For example, patients with presenting levels of 2,000 mIU/mL experienced a mean reduction of 74% in 2 days, whereas those with initial levels of 500 had a 71% mean decline in 2 days. The slowest rate of decline was observed in the group of patients who presented with hCG levels of less than 250 mIU/mL, in which the mean reduction was 70% in 2 days and the 95th percentile experienced a 21% decline in 2 days. By 7 days after initial presentation, the 95th percentile reduction in hCG ranged from 60% for patients with the lowest starting hCG concentrations of 250 to 84% for patients with the highest starting levels of 5,000. This pattern was consistent for the 95th percentile in each hCG category and for each subsequent day of observation.
The normal or expected rate of hCG decline for spontaneous abortions has not previously been reported (data from a MEDLINE search in English for the dates January 1962 to January 2004 using the key works “spontaneous abortion,” “miscarriage,” and “human chorionic gonadotropin”). The lack of this important information has created difficulties for the clinician in determining what should be considered an abnormal decline in hCG levels and at what point intervention should be considered to differentiate a spontaneous abortion from an ectopic pregnancy. The proper selection of cases for expectant management is essential to avoid consequences of the undiagnosed ectopic pregnancy. Similar to the well-known curve of normal hCG rise, we demonstrate that a log linear pattern best describes the normal decrease in hCG. However, unlike the increase in hCG, these curves of decline are dependent on the initial hCG level.
Several authors have investigated the rates of disappearance of hCG from maternal blood, with conflicting results. A biexponential pattern of a rapid elimination followed by a gradual decrease in serum levels was found after term delivery.11 In patients undergoing elective first-trimester terminations, Marrs et al12 observed that the mean clearance time was 37.5 ± 5.1 days, but the trend of decline was not described. Others, however, found that rates varied depending upon precurettage hCG levels, with longer median disappearance times associated with higher initial levels.13,14 It is possible that there is a biexponential decline of hCG value that start at levels greater than 5,000 mIU/mL, but that was not the focus of our investigation.
It has also been demonstrated previously that there is a relatively rapid decline in hCG values after removal of trophoblastic tissue in the first trimester. Steier et al13 compared clearance curves between spontaneous abortions, induced abortions, and ectopic pregnancies. They found that, after surgical removal, ectopic pregnancies were characterized by the shortest elimination time, followed by spontaneous abortions, whereas induced abortions were associated with the longest, with hCG detectable up to 60 days after termination. We also have recently demonstrated that there is a rapid decline of serum hCG after medical treatment of spontaneous abortion.15 Data presented in this article focus the spontaneous resolution of hCG in women at risk for an ectopic pregnancy. It is anticipated that there may be a slower rate of decline in hCG levels in this group of patients because some trophoblasts are still in situ without surgical or medical intervention. In other words, although there is elimination of hCG from the serum, there may be some production of hCG from the residual trophoblast.
One of the objectives of this work was to provide evidence of when the decline in hCG levels is “too slow.” Our data revealed that the rate of clearance is dependent on the initial concentration of hCG. Higher starting concentrations were associated with a more rapid decline. We report that the slowest rates of decline for each hCG group (as represented by the 95th percentile) ranged from 21% to 35% reduction in 2 days and 60% to 84% in 7 days. This suggests that the “15% drop in 7 days” rule that has been adopted from the studies of treated ectopic pregnancies is too simplistic and too conservative to apply to monitoring of presumed spontaneous abortions. For those patients who are being managed expectantly with the hope that their pregnancy will resolve spontaneously, a 15% decline in 7 days is slower than the expected rate and should prompt intervention to eliminate the possibility of an ectopic pregnancy.
The data presented in the tables can be used as a rule of thumb to compare the drop in hCG over the course of time for a current gestation compared with what is expected in those that have undergone complete spontaneous resolution. We present both the 90th and 95th percentile of the fall in hCG levels that characterized a complete spontaneous abortion as the clinical situation may dictate an aggressive or conservative approach to establish the final diagnosis of miscarriage or ectopic pregnancy.
We recognize that the curve described by these data may also include the spontaneous resolution of women with an ectopic pregnancy. The number of women with a spontaneous resolution of an ectopic pregnancy in this data set is not possible to estimate. Currently, algorithms designed to aid in the evaluation of women at risk for ectopic pregnancy do not advocate intervention in a woman with an “adequate” decrease in hCG level. The likely inclusion of some women with a spontaneous resolution of an ectopic pregnancy in these data likely makes the “cut off” used more conservative, allowing for a slower rate of decline.
Importantly, it was not anticipated that the rate of decline would distinguish all women with an ectopic pregnancy from those with a miscarriage. As many as one third of ectopic pregnancies are diagnosed with an initial decline in hCG levels,16 and as many as 30% of women with an abnormal decrease in hCG are confirmed to have a chorionic villi detected with dilation and curettage.17 Finally, a decrease in hCG levels within the parameters described in this article does not rule out the possibility of a rupture of a resolving ectopic pregnancy.
1. Barnhart KT, Simhan H, Kamelle SA. Diagnostic accuracy of ultrasound above and below the beta-hCG discriminatory zone. Obstet Gynecol 1999;94:583–7.
2. Kadar N, Caldwell BV, Romero R. A method of screening for ectopic pregnancy and its indications. Obstet Gynecol 1981;58:162–6.
3. Barnhart KT, Sammel MD, Rinaudo PF, Zhou L, Hummel AC, Guo W. Symptomatic patients with an early viable intrauterine pregnancy: HCG curves redefined. Obstet Gynecol 2004;104:50–5.
4. Stovall TG, Ling FW, Carson SA, Buster JE. Nonsurgical diagnosis and treatment of tubal pregnancy. Fertil Steril 1990;54:537–8.
5. Shepherd RW, Patton PE, Novy MJ, Burry KA. Serial beta-hCG measurements in the early detection of ectopic pregnancy. Obstet Gynecol 1990;75:417–20.
6. Lenton EA, Neal L, Sulaiman R. Plasma concentrations of human chorionic gonadotropin from the time of implantation until the second week of pregnancy. Fertil Steril 1982;37:773–8.
7. Brumback BA, Rice JA. Smoothing spline models for the analysis of nested and crossed samples of curves. J Am Stat Assoc 1998;93:961–76.
8. Kim H, Siegmund D. The likelihood ratio test for a change-point in simple linear regression. Biometrika 1989;76:409–23.
9. Wang Y. Smoothing spline models with correlated random errors. J Am Stat Assoc 1998;93:341–8.
10. Zhang D, Lin X, Raz J, Sowers M. Semiparametric Stochastic Mixed Models for Longitudinal Data. J Am Stat Assoc 1998;93:710–9.
11. Midgley AR, Jaffe RB. Regulation of human gonadotropins. II. Disappearance of human chorionic gonadotropin following delivery. J Clin Endocrinol Metab 1968;28:1712–8.
12. Marrs RP, Kletzky OA, Howard WF, Mishell DR. Disappearance of human chorionic gonadotropin and resumption of ovulation following abortion. Am J Obstet Gynecol 1979;135:731–6.
13. Steier JA, Bergsjo P, Myking OL. Human chorionic gonadotropin in maternal plasma after induced abortion, spontaneous abortion, and removed ectopic pregnancy. Obstet Gynecol 1984;64:391–4.
14. Aral K, Gurkan Zorlu C, Gokmen O. Plasma human chorionic gonadotropin levels after induced abortion. Adv Contracept 1996;12:11–4.
15. Barnhart KT, Bader T, Huang X, Frederick MM, Timbers KA, Zhang JJ. Hormone pattern after misoprostol administration for a non-viable first trimester gestation. Fertil Steril 2004;81:1099–105.
16. Barnhart K, Mennuti MT, Benjamin I, Jacobson S, Goodman D, Coutifaris C. The prompt diagnosis of ectopic pregnancy in an emergency department setting. Obstet Gynecol 1994;84:1010–5.
© 2004 by The American College of Obstetricians and Gynecologists. Published by Wolters Kluwer Health, Inc. All rights reserved.
17. Barnhart KT, Katz I, Hummel A, Gracia CR. Presumed diagnosis of ectopic pregnancy, Obstet Gynecol 2002;100:505–10.