Pii: s0029-7844(99)00480-9

Progesterone, Inhibin, and hCG Multiple MarkerStrategy to Differentiate Viable From NonviablePregnancies MAUREEN GLENNON PHIPPS, MD, JOSEPH W. HOGAN, ScD,JEFFREY F. PEIPERT, MD, MPH, GERALYN M. LAMBERT-MESSERLIAN, PhD,JACOB A. CANICK, PhD, AND DAVID B. SEIFER, MD Objective: To determine whether a combination of serum
Conclusion: Serum progesterone appeared to be the single
and urine biomarkers drawn from symptomatic pregnant
most specific biomarker for distinguishing viable from non-
women will help early differentiation of viable from nonvi-
viable pregnancies. When a dual-biomarker strategy was
able pregnancies.
applied, combining serum progesterone with hCG, specific-
Methods: We conducted a prospective cohort study of 220
ity improved significantly, which suggests that a multiple
women who presented in the first trimester of pregnancy
biomarker strategy might help distinguish viable from non-
with complaints of pain, cramping, bleeding, or spotting.
viable pregnancies in early gestation. (Obstet Gynecol 2000;
Serum samples for progesterone, inhibin A, and hCG, and
95:227–31.
2000 by The American College of Obstetri-
urine beta-core hCG, were collected at presentation. To
cians and Gynecologists.)
evaluate whether those biomarkers could predict viable and
nonviable outcomes in pregnancy, we used likelihood ratios
to compare operating characteristics of single and multiple
biomarker strategies.
Many biomarkers in serum, including hCG, progester- Results: Of 220 pregnancies studied, 98 were viable and
one, estradiol (E2), alpha-fetoprotein, fetal fibronectin, 122 nonviable. Among single biomarkers, progesterone
and inhibin A, have been studied to determine whether alone appears to have the greatest utility (area under the
they could help diagnose ectopic pregnancies.1–4 Pro- receiver operator characteristic curve ؍ 0.923). Among dual-
gesterone’s utility as a biomarker has been well dem- biomarker strategies, progesterone plus hCG and progester-
onstrated.5–7 Inhibin A was shown to be lower in one plus inhibin A improved specificity but not sensitivity.
ectopic pregnancies compared with intrauterine preg- At 95% sensitivity, the combination of progesterone and
nancies.3,8 Urine ␤-core hCG, the major metabolite of hCG improved specificity from 0.29 to 0.66 (improvement ؍
hCG in maternal urine, has been studied as a potential 0.37 [95% confidence interval 0.23, 0.52]). A triple-biomarker
combination did not show substantial improvement over the
biomarker for determining ectopic pregnancies com- dual-biomarker strategy. Also, combinations that used urine
pared with normal pregnancies.9 Combinations of bi- beta-core hCG did not improve diagnostic accuracy.
omarkers have also been studied to support rapiddiagnosis of ectopic pregnancies.1 The purpose of this study was to determine whether From the University of Michigan Health System, Robert Wood a combination of multiple serum and urine biomarkers Johnson Clinical Scholars Program and Department of Obstetrics and from symptomatic women at first-trimester clinical pre- Gynecology, Ann Arbor, Michigan; the Center for Statistical Sciences, sentation could differentiate a viable from a nonviable Department of Community Health, Brown University, Providence,Rhode Island; Department of Obstetrics and Gynecology and Depart- pregnancy. We conducted a prospective cohort study of ment of Pathology and Laboratory Medicine, Women and Infants women who presented with complaints of pain, cramp- Hospital of Rhode Island, Providence, Rhode Island; and the University ing, bleeding, or spotting. Serum quantitative hCG, of Medicine and Dentistry of New Jersey-Robert Wood Johnson MedicalSchool, Department of Obstetrics and Gynecology, New Brunswick, serum progesterone, serum inhibin A, and urine ␤-core hCG were evaluated independently and in combination The following companies provided assay reagents for this study: to determine their accuracy in predicting viability of a Diagnostic Products Corp., Los Angeles, California and Chiron Diag-nostics Corp., Alameda, California. nonparametric correlation was determined for eachmarker combination.
We used an observation cohort of 238 pregnant women We estimated the area under the receiver operator who presented to the Women and Infants Hospital of characteristic (ROC) curve based on the likelihood Rhode Island urgent care unit between June 1996 and ratios calculated for each screening strategy. To under- March 1997. Women were eligible if they presented stand the differences between strategies, we compared with complaints of bleeding, spotting, pain, or cramp- specificity for given values of sensitivity and compared ing in the first trimester (less than 13 weeks’ gestation).
sensitivity for given values of specificity. For example, We limited our sample to spontaneously conceived because progesterone alone was a very sensitive test, pregnancies. Pregnancy outcomes determined by re- we evaluated various multiple biomarker strategies by view of the medical records were known in all subjects comparing their specificity at 95% sensitivity. That included in analysis. Of 238 women initially enrolled, involved determining the likelihood ratio cutoff values 18 were excluded from analysis because pregnancy that corresponded to 95% sensitivity, estimating speci- outcomes could not be definitively established in 12 and ficity for the cutoff, and estimating the difference in gestational age was unknown in six. Seven women who specificity (and associated 95% confidence interval [CI]) had therapeutic abortions were included in analysis between biomarker combinations. Comparisons of sen- because there was documentation of a viable pregnancy sitivity at given values of specificity were done simi- before termination. The Women and Infants Hospital larly. Confidence intervals for the differences in sensi- Institutional Review Board approved the research in tivity and specificity were calculated using the normal approximation to the sampling distribution of each To determine marker values each eligible subject had estimated difference, making proper adjustment for serum drawn and a urine sample collected at presenta- tion to the urgent care unit. Serum and urine samples An important component of our analysis is the calcula- were placed in aliquots and frozen at Ϫ20C until assays tion of maternal- and gestational-age-adjusted likelihoodratios. Conditional on maternal and gestational age, the were done. Commercially available assays were used to likelihood ratio for a woman with a particular set of analyze samples for quantitative hCG (Immulite; Diag- marker values is the odds that she will have a nonviable nostic Products Corp., Los Angeles, CA), progesterone pregnancy, denoted by LR ϭ Pr(NV͉M, C)/Pr(V͉M, C), (Immulite; Diagnostic Products Corp.), inhibin A (en- where Pr(A͉B) is the conditional probability of A given B, zyme-linked immunosorbant assay by Serotec, Ltd., NV and V, respectively, denote nonviable and viable, M Oxford, United Kingdom) and urine ␤-core hCG represents a set of marker values and C denotes individ- (Titron; Chiron Diagnostics Corp., Alameda, CA). As- ual characteristics that might affect marker values (mater- says were done without knowledge of pregnancy out- nal and gestational age). Likelihood ratios can be com- comes and results did not influence treatment. Besides puted by logistic regression,13 but using correlated pregnancy status and marker data, we recorded several markers can lead to loss of efficiency and problems with demographic and historical variables including mater- collinearity. Instead, we computed likelihood ratios by nal age, gestational age, gravidity, parity, race, insur- modeling marker distribution separately by viability sta- ance status, and reproductive history.
tus. Using Bayes theorem, the likelihood ratio can be For our primary analysis, we compared operating expressed in terms of the (multivariate) marker distribu- characteristics (ie, measures of diagnostic accuracy) of tions, so that LR ϭ [Pr(M͉NV, C) Pr(NV͉C)]/[Pr(M͉V, C) progesterone only (P), progesterone plus inhibin A Pr(V͉C)]. In general, covariates such as maternal and (PϩI), progesterone plus serum hCG (PϩH), and pro- gestational age might affect marker distribution and via- gesterone plus serum hCG plus inhibin A (PϩHϩI).
bility status. We assume that maternal and gestational age Our goals were to quantify information gained using do not affect belief about viability status before collecting multiple biomarker strategies compared with proges- marker data, the likelihood ratio is proportional to Pr␪1 terone alone and to determine the nature of differences (M͉NV, C)/Pr␪ (M͉V, C). The parameters ␪ between single- and multiple-biomarker strategies. Fol- phasize that marker values follow separate models by lowing Haddow et al10, we used likelihood ratios to quantify the maternal- and gestational-age-adjusted Individual likelihood ratios are calculated by estimat- odds of nonviable pregnancies for a given combination ing ␪1 and ␪0, then evaluating Pr␪ (M͉NV, C)/Pr␪ of marker values. We used multivariate normal regres- (M͉V,C) using individual marker values, maternal age, sion models11 to adjust for variability in biomarkers due and gestational age. Models for the numerator (NV) and to differences in maternal and gestational age and to denominator (V) of the likelihood ratios were fit under account for correlations between biomarkers. Spearman the assumptions that after a suitable transformation, 228 Phipps et al
Table 1. Demographic and Clinical Characteristics
polynomials were used to avoid overfitting of the marker distributions to covariates. All analyses were done using SAS Version 6.12 (SAS Institute, Cary, NC);multivariate models were fit using SAS Proc Mixed.
Pregnancy outcomes included 98 viable intrauterine pregnancies, 85 first-trimester spontaneous abortions, and 37 ectopic pregnancies. Viable intrauterine preg- nancies were defined as viable pregnancies (45%); spon- taneous abortions and ectopic pregnancies were de- fined as nonviable pregnancies (55%). The mean age of subjects with viable pregnancies was lower than that of subjects with nonviable pregnancies (24.1 versus 27.9 years, P Ͻ .01). Other demographic information and clinical characteristics are given in Table 1. There were no significant differences in race, gravidity, or parity.
The overall distribution for each biomarker, given outcomes of viable or nonviable pregnancy, was di- Table entries are total n (%) except for age, which is mean Ϯ vided into percentiles (Table 2). For progesterone, theviable pregnancy value for the 25th percentile was wellabove the nonviable pregnancy value at the 75th per- each set of marker values follows a multivariate normal centile. There was more overlap between other biomar- distribution and that the mean marker value varies kers. The greatest disparity between nonviable and linearly with maternal age, possibly up to a cubic viable pregnancies was in the distribution of progester- function of gestational age. We used natural log trans- one values. As a single marker, progesterone had the formations for progesterone and inhibin A, and 1/6 greatest area under the ROC curve (0.923, compared power transformation for hCG. The estimated parame- with 0.795 for inhibin A, 0.646 for urine ␤-core hCG, and ters ␪ˆ1 and ␪ˆ0 contain regression parameters, marker standard deviations, and marker correlations for the In formulating multiple-marker strategies, we chose fitted multivariate models. Each woman’s likelihood serum hCG rather than urine ␤-core hCG because the two correlated so strongly (Spearman rank correlation (M ͉iV, Ci), where i indexes subject and f is the multivar- 0.90) and because serum hCG is the biomarker used iate normal density function, with dimension equal to most commonly by practitioners for evaluating preg- the number of markers. Under our assumptions, LRi is nancy viability. For other biomarker combinations, the proportional to (ie, has the same rank-ordering across pairwise rank correlation ranged from 0.32 for proges- individuals as) the odds of having a nonviable preg- terone and urine ␤-core hCG to 0.73 for serum hCG and nancy for a given set of marker values, adjusted for maternal and gestational age. The implication is that it Multiple-marker strategies resulted in improvement can be used for nonparametric calculations of sensitiv- in the area under the ROC curve. After incorporating maternal and gestational age–adjusted likelihood ratios, Fitted models used for likelihood ratio calculations the area under the ROC curve for P was .91. When were checked using residual plots, and orthogonal serum hCG was added it increased to .95 (PϩH), withinhibin A to .94 (PϩI), and the triple marker combina-tion had an area of .95 (PϩIϩH).
Table 2. Biomarker Distribution Percentile
Area under the ROC curve is a global measure for characterizing utility, and interpreting clinical implica- tions of differences in area under the ROC curve can be difficult. In more detail, we compared sensitivities at fixed specificity, and specificity at fixed sensitivities.
Table 3 compares specificity for fixed sensitivities. At 95% sensitivity, the specificity for P was 29%. At the same sensitivity, using the dual biomarker strategy of Phipps et al
Table 3. Estimated Specificity at Given Sensitivity for
Table 5. Estimated Sensitivity at Given Specificity for
P ϭ progesterone alone; PϩI ϭ progesterone plus inhibin A; PϩH ϭ progesterone plus serum hCG; PϩIϩH ϭ progesterone plus inhibin Aplus serum hCG.
nonviable pregnancy, or at least 25 ng/mL, suggestinga viable pregnancy.4,5 However, in our study, 41% of PϩI, the specificity increased to 57% and the specificity subjects had progesterone levels in the range of 10 –25 for PϩH was 66%. Based on 95% confidence intervals for difference in specificity, those combinations repre- A highly sensitive and specific test to determine sent statistically significant gains in specificity over P.
viability would be useful in a clinical setting when Specificity for the combination PϩH was 9% higher women present with acute symptoms and a decision than PϩI (95% CI Ϫ0.02, 0.20). Table 4 shows compar- regarding treatment must be made quickly. A highly isons between specificity for P and combinations with sensitive biomarker more accurately identifies viable the other biomarkers when sensitivity is 85% and 95%, pregnancies (true positives) and a highly specific bi- which indicated that PϩH is the superior choice. At omarker more accurately identifies nonviable pregnan- 95% sensitivity, PϩIϩH did not show an appreciable cies (true negatives). Treatment of women who present gain in specificity over PϩH (0.69 compared with 0.66).
with cramping and spotting in the first trimester of At 95% specificity, P had sensitivity of 83%. Table 5 pregnancy would be better guided by a sensitive and shows minimal gains in sensitivity among the various specific test that would reliably categorize prognoses biomarker strategies at fixed specificity values between for pregnancies. Thus far, there is no single test that accurately predicts pregnancy outcome in urgent situ-ations.
Of the single biomarkers, progesterone has the great- est utility as measured by area under the ROC curve.
We evaluated diagnostic usefulness of biomarkers for Although dual-biomarker strategies improved the area distinguishing viable from nonviable early pregnancies under the ROC curve, we found that applying the in symptomatic women who presented for urgent care.
multiple-biomarker strategies improved specificity but It has been well established that measuring serial serum not sensitivity. For improving specificity, the dual bi- quantitative hCG is helpful in treating symptomatic omarker combination progesterone plus serum hCG women in early gestation.14 In clinical practice, the time (PϩH) is better than progesterone plus inhibin A (PϩI) delay necessary for distinguishing a viable from a and as good as the triple biomarker combination pro- nonviable pregnancy is often distressing to women and gesterone plus inhibin A plus serum hCG (PϩIϩH).
practitioners when women present with symptoms in The addition of the urine ␤-core hCG was not helpful in an emergency setting. Adding serum progesterone to the diagnostic tests can be helpful in clinical manage- Our study had several limitations. It was conducted ment if the level is under 10 ng/mL, suggesting a from a convenience sample of symptomatic women inearly gestation who conceived spontaneously. Thus, theresults cannot be generalized to asymptomatic pregnant Table 4. Difference in Specificity
women or women who pursued ovulation inductionand assisted reproductive technologies to achieve preg- nancy. Our computation of likelihood ratios relied on a (carefully constructed) parametric model to adjust for differences in maternal and gestational ages, both of which have underlying associations with various marker values. The assumptions are needed because of P ϭ progesterone alone; PϩI ϭ progesterone plus inhibin A; PϩH ϭ limited sample size. Having data from large, popula- tion-based samples would reduce the need to rely on 95% confidence interval widths adjusted for the six multiple com- parisons using the Bonferroni method.
230 Phipps et al
Our study suggests the need for formalizing a mul- 8. D’Antona D, Mamers PM, Lowe PJM, Balazs N, Groome NP, tiple-biomarker strategy for distinguishing viable from Wallace EM. Evaluation of serum inhibin-A as a surveillancebiomarker after conservative management of tubal pregnancy.
nonviable pregnancies. Clinically, that approach might be helpful in prognoses for pregnancies or identifying 9. Cole LA, Kardana A, Seifer DB, Bohler HCL. Urine hCG ␤-subunit women with ectopic pregnancy. A study validating our core fragment, a sensitive test for ectopic pregnancy. J Clin multiple-biomarker screening strategy using a large population database might aid in predicting pregnancy 10. Haddow JE, Palomaki GE, Knight GJ, Williams J, Miller WA, Johnson A. Screening of maternal serum for fetal Down’s syn- drome in the first trimester. N Engl J Med 1998;338:955– 61.
11. Crowder MJ, Hand DJ. Analysis of repeated measures. 1st ed.
12. Agresti A. Categorical data analysis. New York: Wiley & Sons, 1. Grosskinsky CM, Hage ML, Tyrey L, Christakos AC, Hughes CL.
hCG, progesterone, alpha-fetoprotein, and estradiol in the identi- 13. Irwig L. Modeling result-specific likelihood ratios. J Clin Epide- fication of ectopic pregnancy. Obstet Gynecol 1993;81:705–9.
2. Ness RB, McLaughlin MT, Heine RP, Bass DC, Mortimer L. Fetal 14. Speroff L, Glass RH, Kase NG. Clinical gynecologic endocrinology fibronectin as a biomarker to discriminate between ectopic and and infertility. 5th ed. Baltimore: Williams & Wilkins, 1994.
intrauterine pregnancies. Am J Obstet Gynecol 1998;179:697–702.
3. Seifer DB, Lambert-Messerlian GM, Canick JA, Frishman GN, Schneyer AL. Serum inhibin levels are lower in ectopic than intrauterine spontaneously conceived pregnancies. Fertil Steril University of Michigan Health System 4. Stovall TG, Ling FW, Andersen RN, Buster JE. Improved sensitiv- Robert Wood Johnson Clinical Scholars Program ity and specificity of a single measurement of serum progesteroneover serial quantitative beta-human chorionic gonadotropin in screening for ectopic pregnancy. Hum Reprod 1992;7:723–5.
5. Cowan BD, Vandermolen DT, Long CA, Whitworth NS. Receiver- operator characteristic, efficiency analysis, and predictive value of serum progesterone concentration as a test for abnormal gesta-tions. Am J Obstet Gynecol 1992;166:1729 –37.
6. McCord ML, Muram D, Buster JE, Arheart KL, Stovall TG, Carson SA. Single serum progesterone as a screen for ectopic pregnancy: Exchanging specificity and sensitivity to obtain optimal test per- Received in revised form June 18, 1999. formance. Fertil Steril 1996;66:513– 6.
7. Dart R, Dart L, Segal M, Page C, Brancato J. The ability of a single serum progesterone to identify abnormal pregnancies in patientswith beta-human chorionic gonadotropin values less than 1,000 Copyright 2000 by The American College of Obstetricians and mIU/mL. Acad Emerg Med 1998;5:304 –9.
Gynecologists. Published by Elsevier Science Inc.
Phipps et al

Source: http://health.bsd.uchicago.edu/thisted/epor/Papers/Hogan-Progesterone.pdf