- Email: [email protected]
The Preterm Prediction Study: Toward a multiple-marker test for spontaneous preterm birth
Fetus-Placenta-Newborn
The Preterm Prediction Study: Toward a multiple-marker test for spontaneous preterm birth Robert L. Goldenberg, MD, Jay D. Iams, MD, Brian M. Mercer, MD, Paul J. Meis, MD, Atef Moawad, MD, Anita Das, PhD, Menachem Miodovnik, MD, Peter J. VanDorsten, MD, Steve N. Caritis, MD, Gary Thurnau, MD, and Mitchell P. Dombrowski, MD, for the Maternal-Fetal Medicine Units Network
OBJECTIVE: The Preterm Prediction Study evaluated 28 potential biologic markers for spontaneous preterm birth in asymptomatic women at 23 to 24 weeks gestational age. This analysis compares those markers individually and in combination for an association with spontaneous preterm birth at <32 and <35 weeks gestational age. STUDY DESIGN: With the use of a nested case-control design from an original cohort study of 2929 women, results of tests from 50 women with a spontaneous preterm birth at <32 weeks and 127 women with a spontaneous preterm birth at <35 weeks were compared with results from matched-term control subjects. RESULTS: In the univariate analysis, the most potent markers that are associated with spontaneous preterm birth at <32 weeks by odds ratio were a positive cervical-vaginal fetal fibronectin test (odds ratio, 32.7) and <10th percentile cervical length (odds ratio, 5.8), and in serum, >90th percentiles of α-fetoprotein (odds ratio, 8.3) and alkaline phosphatase (odds ratio, 6.8), and >75th percentile of granulocyte colony-stimulating factor (odds ratio, 5.5). Results for spontaneous preterm birth at <35 weeks were generally similar but not as strong. Univariate and multivariate logistic regression analyses demonstrated little interaction among the tests in their association with spontaneous preterm birth. Combinations of the 5 markers were evaluated for their association with <32 weeks spontaneous preterm birth. Ninety-three percent of case patients had at least 1 positive test result versus 34% of control subjects (odds ratio, 24.0; 95% CI, 6.4-93.4). Among the case patients, 59% had ≥2 positive test results versus 2.4% of control subjects (odds ratio, 56.5; 95% CI, 7.1-451.7). If a cutoff of 3 positive test results was used, 20% of case patients and none of the control subjects had positive test results (P < .002). With the use of only the 3 serum tests (alkaline phosphatase, α-fetoprotein, and granulocyte colony-stimulating factor), any positive test identified 81% of cases versus 22% of control subjects (odds ratio, 14.7; 95% CI, 5.0-42.7). For spontaneous preterm birth at <35 weeks gestation, any 2 positive tests identified 43% of cases and 6% of control subjects (odds ratio, 11.2; 95% CI, 4.8-26.2). CONCLUSION: Overlap among the strongest biologic markers for spontaneous preterm birth is small. This suggests that the use of tests such as maternal serum α-fetoprotein, alkaline phosphatase, and granulocyte colony-stimulating factor as a group or adding their results to fetal fibronectin test and cervical length test results may enhance our ability to predict spontaneous preterm birth and that the development of a multiplemarker test for spontaneous preterm birth is feasible. (Am J Obstet Gynecol 2001;185:643-51.)
Key words: Preterm birth, prediction, markers, multiple-marker test
Preterm birth is the major cause of perinatal morbidity and death in developed countries.1 In recent decades, substantial progress has been made in the reduction of From the Maternal-Fetal Medicine Units Network. Supported by the following grants from the National Institute of Child Health and Human Development (NICHD): HD21410, HC21414, HC21434, HD27860, HD27861, HD27869, HD27883, HD27889, HD27905, HD27915, HD27917, and HD19897. Received for publication September 19, 2000; revised January 23, 2001; accepted April 24, 2001. Reprints not available from the authors. Copyright © 2001 by Mosby, Inc. 0002-9378/2001 $35.00 + 0 6/1/116752 doi:10:1067/mob.2001.116752
deaths that were associated with preterm birth, but little or no progress has been made in the reduction of the preterm birth rate itself.2 The reasons for this failure are not clear, but evidence is accumulating that most strategies that are aimed at the reduction of preterm birth in the general population are not effective.2 Preventive strategies that are targeted at populations with specific risk factors or preterm birth associated with specific causes are therefore a focus of current research. Risk factors for preterm birth can be categorized into several groups that include (1) demographic information, (2) previous pregnancy history, (3) current pregnancy findings, (4) nutritional/body size in origin, (5) 643
644 Goldenberg et al
associated biophysical markers (such as contractions and cervical length), and (6) measurements of analytes that were made in various biologic fluids.3 Among the first 4 groups, multiple pregnancy, previous preterm birth, black race, and second- or third-trimester bleeding have been associated most consistently with preterm birth. Pregnancies with ≥2 fetuses have an extremely high risk of preterm birth; 50% of all twin pregnancies and virtually all of the higher-order multiple pregnancies are born preterm. Among women with singleton pregnancies, a previous preterm birth is the strongest historic predictor, with relative risks reported in the range of 2 to 6.4, 5 Among the biophysical and biologic fluid tests evaluated to date, cervical length as measured by ultrasound and fetal fibronectin levels in cervical or vaginal fluid have been most strongly and consistently associated with subsequent spontaneous preterm birth.4 The Preterm Prediction Study of the National Institute of Child Health and Human Development Maternal-Fetal Medicine Units Network was initiated to evaluate many potential predictors of spontaneous preterm birth to determine their relative strengths and to evaluate whether several of the tests could be used together (as a multiplemarker test) to predict spontaneous preterm birth more effectively. Evaluations of many of the individual tests have been presented previously.4-20 In this analysis, we compare the tests against 1 another and evaluate groups of tests to determine the feasibility of the development of a multiple-marker test for spontaneous preterm birth. We chose to focus this study on data available at 23 to 24 weeks in asymptomatic women who were undergoing routine prenatal care. At 23 to 24 weeks’ gestational age, the fetus with a weight of approximately 600 g is at the margin of viability. The preponderance of neonatal deaths and much of the birth-associated long-term handicap will be found in those infants who are born over the next 4 to 8 weeks. Reducing the number of births between 24 and 32 weeks’ gestational age would have a major impact on adverse pregnancy outcomes in the United States. Methods The relationship between spontaneous preterm birth and many of the individual markers that were evaluated at 23 to 24 weeks in the Preterm Prediction Study have been presented previously.4-20 Because of the large number of women in the original cohort (n = 2929), the large number of potential tests to be studied, and their costs, we originally planned to complete a nested case-control study of the tests after the major outcomes (such as spontaneous preterm birth) were determined. Women who at the time of the enrollment had a known twin pregnancy were not included in this analysis. In addition, multiparous women who had ≥3-cm cervix dilation, nulliparous women who had ≥2-cm cervix dilation, women who
September 2001 Am J Obstet Gynecol
had membranes that bulged from the cervical os, or women who had a known placental previa before enrollment at 23 to 24 weeks were excluded from participation. After exclusions, the study population consisted of 2929 women who were receiving routine prenatal care. They were not selected because they were at high risk for preterm birth. Of this population, 63% of the women were black; 37% of the women were nonblack; 58% of the women were multiparous; and 42% of the women were nulliparous. The mean age of the women in the study population was 23.7 ± 5.5 years. Of the 2929 women who were enrolled, 127 of the women (4.3%) had a spontaneous preterm birth at <35 weeks, and they and their matched control subjects are the women who were considered in this analysis. Extensive data were collected on every woman in the cohort; the data included the demographic information, previous pregnancy outcomes, health history, and current pregnancy data. In addition, because we were interested in studying sequential cervical/vaginal fetal fibronectin testing, fetal fibronectin samples were collected and analyzed prospectively on each woman at 24, 26, 28, and 30 weeks.6, 7 We had available both cervical and vaginal fetal fibronectin tests, and if either 1 or both tests were positive at 24 weeks, we considered that the test for fetal fibronectin was positive. Also, a Gram stain and pH test for bacterial vaginosis and an ultrasound cervical examination were performed for each woman at 24 and 28 weeks.8, 9 Fluids that were available for analysis after being frozen at –70°C included serum, plasma, and urine that had been obtained at 24 and 28 weeks and cervical and vaginal fluid in buffer that had been originally collected for fetal fibronectin testing. At the start of the study, we planned to measure 10 serum analytes that previously had been associated with spontaneous preterm birth. However, as the study progressed, additional analyses were proposed on the various fluids. These were encouraged, as long as there was reasonable biologic plausibility for an association with spontaneous preterm birth, the fluid was appropriate for the test, the resources were available for the analyses, and the volume of fluid that was needed for the test was not overly large and sufficient fluid remained. Although some of the analytes were assayed at both 24 and 28 weeks, as a general rule, we evaluated the 24-week fluid alone. This strategy was chosen because we were especially interested in the association between various markers and very early spontaneous preterm birth. Also, we often did not have the resources to analyze both sets of samples. The samples were always analyzed in a blinded fashion; the individual laboratories were not aware which samples were from the case patients and which samples were from the control subjects. The latter were defined as women who delivered at term (≥37 weeks) and were individually matched to the case patients by race, parity, and center. The data were entered into a relational database at the
Volume 185, Number 3 Am J Obstet Gynecol
George Washington University Biostatistics Center and analyzed only after all measurements of a specific analyte were complete. Each woman provided informed consent, and each center obtained approval for the study by its Institutional Review Board. In the original study, the primary outcome was a spontaneous preterm birth before the 35th week (defined as after the onset of spontaneous labor or rupture of membranes). We hypothesized that approximately 3.5% of our population of 3000 women would have this outcome (n = 105). In a case-control study in which 10% of the population is positive for a given screening test, 100 case patients and 100 control subjects would give 62% power to detect an association with spontaneous preterm birth characterized by an odds ratio of ≥2.5 and 80% power for an odds ratio of 3.0. Because we were also interested in predicting earlier spontaneous preterm birth, we chose to evaluate the births that occurred at <32 weeks in a similar casecontrol fashion. If 10% of the population is positive for a specific risk factor, with 45 case patients and control subjects, we would have a 59% power to detect an odds ratio of 3.5 and 69% power for an odds ratio of 4.0. Various factors or positive tests are defined as a previous spontaneous preterm birth that occurred 20 through 36 weeks after the onset of spontaneous labor or rupture of membranes, regardless of the method of delivery. Body mass index was defined as weight in kilograms divided by height in meters squared. Vaginal bleeding was defined as any noted by the woman after her last menstrual period. The presence of contractions was defined as any contractions appreciated by the patient in the 2 weeks before her study visit. Many of the methods for the various tests and the 90th percentile cutoffs have been published previously and are referenced.4-20 The methods for those not published can be obtained from the authors. For consistency, for those tests that did not have a previously defined percentile cutoff, we used values greater than the 90th percentile of the control subjects to define a positive test. For fetal fibronectin, we used a cutoff of ≥50 ng/mL. Also, because such a high proportion of this population had a vaginal Gram stain score of ≥7 and a pH of ≥4.5, to obtain a smaller percent of positives, we present data using a Gram stain score of ≥9 and a pH of ≥5. As described in the “Results” section of this article, for each test we also evaluated >75th and >95th percentile of the control subjects as a predictor of spontaneous preterm birth. Because this was a case-control and not a cohort study, creating a standard receiver-operator curve for each analyte was deemed inappropriate, as was presenting the positive and negative predictive values. Results Tables I and II show each of the factors included in the case-control analysis, the source of data or fluid from which the measurement was made, the cutoff value for a
Goldenberg et al 645
positive test, the percent of test results that were positive in the case patients and in the control subjects, and the crude odds ratio for spontaneous preterm birth at <32 weeks (Table I) and at <35 weeks (Table II). Significance of the association is also noted. As can be seen for both spontaneous preterm birth at <32 and <35 weeks, a positive fetal fibronectin test result had the greatest odds ratio for its association with spontaneous preterm birth, but many other factors were also significantly associated with spontaneous preterm birth in 1 or both gestational age categories. For both <32- and <35-week spontaneous preterm birth, these positive test results included alkaline phosphatase level, α-fetoprotein level, vaginal pH, previous spontaneous preterm birth, and short cervical length. For spontaneous preterm birth at <32 weeks alone, interstitial cell adhesion molecule, defensins, and cervical IL-6 were significantly associated; for spontaneous preterm birth at <35 weeks alone, cervical lactoferrin and chlamydia, bleeding, and low body mass index were significantly associated. In addition to evaluating the 90th percentile as a cutoff, we also examined the effect on spontaneous preterm birth of the use of the 75th and 95th percentile as cutoffs for a positive test. The only factor that was previously found to have an insignificant relationship that consistently emerged as significant in this analysis was serum granulocyte colony-stimulating factor at the 75th percentile for both <35- and <32-week spontaneous preterm birth (odds ratio, 2.8 for <35 weeks and 5.5 for <32 weeks). For this reason, in subsequent analyses, for granulocyte colony-stimulating factor, we used the 75th percentile as the cutoff for a positive test result. To identify factors to consider for use in a multiplemarker test, we performed regression analyses for <32- or <35-week spontaneous preterm birth using all factors that were significantly associated with spontaneous preterm birth in univariate analyses (Table III). For spontaneous preterm birth at <32 weeks, fetal fibronectin, granulocyte colony-stimulating factor, defensins, and α-fetoprotein remained significantly associated. For <35 weeks, fetal fibronectin, short cervix, α-fetoprotein, alkaline phosphatase, granulocyte colony-stimulating factor, vaginal bleeding, and previous spontaneous preterm birth remained significantly associated. We were interested in the stability of the tests that were performed at both 24 and 28 weeks. Table IV shows the correlation coefficients and probability values between the 2 measurements, when data at 28 weeks were available. For specific tests, values obtained at both 24 and 28 weeks always correlated significantly with 1 another. In addition, the correlation coefficients were generally high (>0.4), which confirmed that values that are high at 24 weeks tended to remain high at 28 weeks and indicated that the results were generally consistent over time. Interestingly, fetal fibronectin results had 1 of the lowest correlations between 24 and 28 weeks.
646 Goldenberg et al
September 2001 Am J Obstet Gynecol
Table I. Risk factors and tests for spontaneous preterm birth at <32 weeks obtained or available at 24 weeks’ gestational age in case patients and control subjects Factor
Corticotropin-releasing hormone α-fetoprotein Alkaline phosphatase Beta2-macroglobulin Ferritin Interstitial cell adhesion molecule-1 Interleukin-6 C-reactive protein Cortisol Lactoferrin Defensins Relaxin Interleukin-10 Granulocyte colony-stimulating factor Activan Interleukin-6 Lactoferrin Defensins Sialidase Short cervix Fetal fibronectin Gram stain score pH Chlamydia Previous spontaneous preterm birth Contractions Bleeding Body mass index *P
Source of data or fluid
Test cutoff
Case Patients (% positive)
Control subjects (% positive)
Odds ratio (case patients vs control subjects)
Serum Serum Serum Serum Serum Serum Serum Serum Serum Serum Serum Serum Serum Serum Serum Cervix Cervix Cervix Cervix Ultrasound Cervix/vagina Vagina Vagina Vagina History History History Measured
90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile Positive 90th %ile 90th %ile 90th %ile 90th %ile ≤25 mm ≥50 ng/mL ≥9 ≥5.0 Positive Positive Positive Positive <19.8
10.6 36.1 22.9 6.3 14.6 23.4 10.4 10.4 10.4 11.6 20.9 10.9 6.4 8.5 8.5 20.4 4.3 10.5 8.2 44.9 40.0 30.0 42.9 15.2 42.0 24.0 36.0 29.2
4.3 6.4 4.2 2.1 2.1 6.4 8.3 6.3 2.1 9.3 4.7 10.9 12.8 10.6 10.6 6.1 0.0 18.4 6.1 12.2 2.0 24.0 18.4 6.5 14.0 32.0 24.0 14.6
2.7 8.3* 6.8* 3.1 8.0 4.5* 1.3 1.7 5.5 1.3 5.4* 1.0 0.5 0.8 0.8 3.9* ∞ 0.5 1.4 5.8* 32.7* 1.4 3.3* 2.6 4.5* 0.7 1.8 2.4
< .05.
Table V shows the correlation coefficients among the tests that were performed at 24 weeks now being considered as part of the multiple-markers panel. The low correlations among the tests indicate that overlap between many of the tests in their association with subsequent spontaneous preterm birth is small. Among the tests considered here, none had a correlation coefficient >0.23, despite the fact that a number of associations were statistically significant. To look for common patterns of test positivity, we next created a matrix that included the distribution of positive tests for each of the case patients and for each of the control subjects (data not shown). For spontaneous preterm birth at <35 weeks, of the 107 cases with all tests available for analysis, the most common pattern was no positive test result, which was found in 13 of 107 women, followed by 10 of 107 women who had only granulocyte colonystimulating factor positivity, 8 of 107 women who had only fetal fibronectin positivity, 7 of 107 women who had only a short cervix, and 6 of 107 women who had only αfetoprotein positivity. Fifty-nine case patients had various combinations of more than 1 positive test, none of which were present in more than 4 women. For the 107 matched control subjects, the most common pattern, by far, was all test results negative in 52 of 107 women, fol-
lowed by 15 of 107 women with only bleeding. Similar scatter in test patterns was seen among the case patients and control subjects who were delivered before 32 weeks. Because any single combination of positive tests was not common enough to be of value, we next evaluated whether counting the number of positive tests would be useful. Table VI shows how individual and then selected combinations of tests relate to the presence of subsequent spontaneous preterm birth at <32 weeks. Specifically, the sensitivity and specificity for spontaneous preterm birth at <32 weeks that were associated with selected patterns of positive tests and the odds ratios are shown. For spontaneous preterm birth at <32 weeks, if a combination of 5 tests (fetal fibronectin, cervical length, defensins, α-fetoprotein, and granulocyte colony-stimulating factor) is used and if a positive result for any test makes the panel positive, 92.7% of those women who had a spontaneous preterm birth had a positive test with an odds ratio of 24.4. However, the specificity was only 65.9%. If 2 of the 5 tests were required to be positive for a “positive” test, 58.5% of those women who delivered preterm had a positive test with an odds ratio of 56.5 and a specificity of 97.6%. Using 3 positive tests to define a “positive test” reduced the sensitivity to 19.5%, but achieved 100% specificity. Especially interesting is that when the 3
Goldenberg et al 647
Volume 185, Number 3 Am J Obstet Gynecol
Table II. Risk factors and tests for spontaneous preterm birth at <35 weeks obtained or available at 24 weeks’ gestational age in case patients and control subjects Factor
Corticotropin-releasing hormone α-fetoprotein Alkaline phosphatase Beta2-macroglobulin Ferritin Interstitial cell adhesion molecule-1 Interleukin-6 C-reactive protein Cortisol Lactoferrin Defensins Relaxin Interleukin-10 Granulocyte colony-stimulating factor Activan Interleukin-6 Lactoferrin Defensins Sialidase Short cervix Fetal fibronectin Gram stain score pH Chlamydia Previous spontaneous preterm birth Contractions Bleeding Body mass index *P
Fluid
Test cutoff
Positive case patients (%)
Positive control subjects (%)
Serum Serum Serum Serum Serum Serum Serum Serum Serum Serum Serum Serum Serum Serum Serum Cervix Cervix Cervix Cervix Ultrasound Cervix/vagina Vagina Vagina Vagina History History History Measured
90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile Positive 90th %ile 90th %ile 90th %ile 90th %ile ≤25 mm ≥50 ng/mL ≥9 >5.0 Positive Positive Positive Positive <19.8
11.7 35.3 14.9 5.1 9.9 16.4 10.0 8.3 12.6 8.8 18.4 13.6 4.3 7.7 14.2 20.0 5.0 16.7 9.6 36.8 22.8 22.8 38.1 12.8 43.3 31.5 35.4 30.9
8.3 13.5 3.3 6.8 7.4 9.1 9.2 6.7 7.6 11.4 10.5 8.2 10.3 7.7 11.7 9.6 0.0 10.0 9.6 9.6 3.2 15.0 21.4 5.1 15.0 31.5 20.5 17.9
Odds ratio (case patients vs control subjects) 1.5 3.5* 5.1* 0.7 1.4 2.0 1.1 1.3 1.8 0.8 1.9 1.8 0.4 1.0 1.3 2.4 ∞ 1.8 1.0 5.5* 9.1* 1.7 2.3* 2.7* 4.3* 1.0 2.1* 2.1*
< .05.
Table III. The adjusted odds ratio for a spontaneous preterm birth at <32 or <35 weeks that was associated with a specific risk factor, based on regression analysis Preterm birth Factor α-fetoprotein (>90th %ile) Alkaline phosphatase (>90th %ile) Granulocyte colony-stimulating factor (>75th %ile) Defensins (>90th %ile) Fetal fibronectin (≥50 ng/mL) Cervical length (≤25 mm) History of previous spontaneous preterm birth Vaginal bleeding
<32 wks Odds ratio (%) 33.7 — 12.7 13.1 12.2 3.7 2.1 —
serum tests were considered as a group, with a requirement for a “positive test” being any positive serum test, a “positive test” was found in 80.5% of those women who delivered preterm, with an odds ratio of 14.7 and a specificity of 78.1%. Overall, various combinations of tests at 24 weeks do not appear to be as strongly predictive of <35-week spontaneous preterm birth as of <32-week spontaneous preterm birth (Table VII). However, if the 3 serum tests (α-fetoprotein, granulocyte colony-stimulating factor, and alkaline phosphatase) were used, 1 positive test
95% CI 5.0-229.7 — 2.8-56.5 1.6-104.5 1.1-135.2 0.7-19.4 0.5-9.4 —
<35 wks Odds ratio (%) 3.9 4.0 3.1 — 6.6 3.9 4.0 2.2
95% CI 1.7-8.7 1.1-14.9 1.4-6.9 — 1.7-25.5 1.7-9.2 1.9-8.4 1.1-4.4
among them had a sensitivity of 60.2%, a specificity of 72.6%, and an odds ratio of 4.0, although 2 positive tests will have a sensitivity of 18.6%, a specificity of 97.4%, and an odds ratio of 8.4. Adding previous spontaneous preterm birth to the panel (data not shown) increased the sensitivity to 78.8% and the odds ratio to 6.0 but reduced the specificity to 62%. Comment To date, the potential benefits in improving pregnancy outcome by predicting spontaneous preterm birth have
648 Goldenberg et al
September 2001 Am J Obstet Gynecol
Table IV. Correlation between tests performed at 24 and 28 weeks
Test Corticotropin-releasing hormone α-fetoprotein Alkaline phosphatase Beta2-macroglobulin Ferritin Interstitial cell adhesion molecule-1 Interleukin-6 C-reactive protein Cortisol Lactoferrin Defensins Relaxin Interleukin-10 Granulocyte colony-stimulating factor Cervical length Cervical/vaginal fetal fibronectin Gram stain score pH
Results that were available at both 24 and 28 wks (n)
Correlation coefficient
P value
154 156 156 146 156 132 156 156 147 148 150 134 144 144 172 164 174 175
0.28 0.69 0.74 0.46 0.63 0.81 0.79 0.58 0.53 0.48 0.51 0.84 0.37 0.33 0.70 0.18 0.63 0.53
.0005 .0001 .0001 .0001 .0001 .0001 .0001 .0001 .0001 .0001 .0001 .0001 .0001 .0001 .0001 .020 .0001 .0001
Table V. The correlations among the strongest predictors of spontaneous preterm birth Alkaline phosphatase Serum alkaline phosphatase Serum α-fetoprotein Serum granulocyte colonystimulating factor Serum defensin Fetal fibronectin Cervical length *P
— 0.10 0.14* 0.09 0.10 –0.01
α-fetoprotein
Granulocyte colonystimulating factor
Defensins
Fetal fibronectin
Cervical length
0.10 — –0.03
0.14* –0.03 —
0.09 0.05 0.10
0.10 –0.04 0.15*
–0.01 –0.03 –0.09
0.05 –0.04 –0.03
0.10 0.15* –0.09
— 0.05 0.17*
0.05 — –0.23*
0.17* –0.23* —
< .05.
not been realized, principally because we have few, if any, effective interventions that can be used once high-risk status has been identified.2 As an example, it is now reasonably clear that the use of the demographic screening tests that focus on age, race, workload, and previous pregnancy outcomes, when used to identify women for various additional testing or interventions such as more intensive observation, home uterine activity monitoring, bed rest, or tocolytic use, has not resulted in a reduction in preterm births.2 One explanation commonly given for this failure, other than lack of effectiveness, is that the risk factors were too nonspecific and, in any case, not strong enough to identify a population that might benefit from the interventions. Based on several promising studies, a number of authors have proposed targeting specific interventions (ie, antibiotics) at women who are high risk (ie, who have had a previous spontaneous preterm birth) and who have another and perhaps treatable risk factor (ie, bacterial vaginosis).21, 22 In fact, this strategy is 1 of the few in which randomized trial data support the use of a particular intervention, although, clearly, the results have not been consistent across all studies.23
We believe there are at least 2 reasons that (even with a history of failure in the use of predictors of spontaneous preterm birth to reduce prematurity) research in this area is important.3 First, the identification of specific markers as predictors may provide important clues to pathways that lead to spontaneous preterm birth and that may, in themselves, suggest interventions. Second, the knowledge of which markers are associated with spontaneous preterm birth will encourage the study of populations of women at especially high risk of spontaneous preterm birth in intervention trials, which will allow the trials to be performed more efficiently and with less risk of harm to women who are destined to deliver at term. We do not yet advocate a third reason commonly suggested for the identification of women at high risk of spontaneous preterm birth: the belief that the identification of risk without having an effective intervention will somehow improve outcome. To date, the screening of populations with any test and the labeling of some women as high risk for spontaneous preterm birth have only led to the increased use of various interventions and to increased costs, without any improvement in outcome.
Goldenberg et al 649
Volume 185, Number 3 Am J Obstet Gynecol
Table VI. Test values for the prediction of spontaneous preterm birth at <32 weeks, comparing single versus multiple markers Test Fetal fibronectin (≥50 ng/mL) Cervix (≤25 mm) α-fetoprotein(>90th %ile) Granulocyte colony-stimulating factor (>75th %ile) Defensins (> 90th %ile ) 1/5 Tests positive 2/5 Tests positive 3/5 Tests positive 1/3 (serum) Tests positive 2/3 (serum) Tests positive *Odds
Sensitivity
Specificity
Odds ratio (%)
95% CI
40.0 44.9 36.2 48.9
98.0 87.8 93.6 85.1
32.7 5.8 8.3 5.5
4.2-256.1 2.1-16.2 2.2-30.9 2.0-14.7
20.9 92.7 58.5 19.5 80.5 26.8
95.4 65.9 97.6 100 78.1 100
5.4 24.4 56.5 ∞* 14.7 ∞*
1.1-26.8 6.4-93.4 7.1-451.7 — 5.0-42.7 —
ratio cannot be calculated with specificity of 100%.
Table VII. Test values for the prediction of spontaneous preterm birth at <35 weeks, comparing single versus multiple markers Test Fetal fibronectin (≥50 ng/mL) Cervix (≤25 mm) α-fetoprotein (>90th %ile) Alkaline phosphatase (>90th %ile) Granulocyte colony-stimulating factor (>75th %ile) 1/5 Tests positive 2/5 Tests positive 3/5 Tests positive 1/3 (serum) Tests positive 2/3 (serum) Tests positive *Odds
Sensitivity
Specificity
Odds ratio (%)
95% CI
22.8 36.8 35.3 14.9 32.5
96.9 90.4 86.6 96.7 85.5
9.1 5.5 3.5 5.1 2.8
3.1-26.8 2.7-11.0 1.8-6.7 1.7-15.6 1.5-5.4
78.8 42.5 15.0 60.2 18.8
63.7 93.8 100 72.6 97.4
6.5 11.2 ∞* 4.0 5.7
3.6-11.8 4.8-26.2 — 2.3-7.0 2.4-29.0
ratio cannot be calculated with specificity of 100%.
We believe that 1 finding from this study is especially important: that is, the apparently small overlap between many of the tests in their association with spontaneous preterm birth. If all the tests tended to be positive in the same group of women who delivered preterm, this would suggest a single predominant pathway that leads to spontaneous preterm birth. It would also suggest that a multiple-marker test might be less valuable in the prediction of spontaneous preterm birth. Our finding of only a small degree of overlap among the markers in women who deliver preterm suggests that there are several pathways that lead to spontaneous preterm birth4 and also suggests that the use of several biologic markers together might be useful in the creation of a multiple-marker test with high sensitivity and a high odds ratio. In fact, we and other groups have previously contemplated the benefits of using more than 1 biologic marker.4 For example, McLean et al24 suggested the use of corticotropin-releasing hormone and α-fetoprotein together, although other researchers have suggested the use of fetal fibronectin and cervical length or cervical interleukin-6 and fetal fibronectin.18 Many groups have studied individual serum/plasma or cervical/vaginal tests for the prediction of preterm birth,
often with discrepant results. Although there are many potential reasons for differing results (including the populations studied), whether the preterm births were predominantly early or late, the test characteristics, the differences in the gestational age at testing, and the timing of the test relative to the timing of the preterm delivery are important factors often not considered. For example, we have previously shown that plasma alkaline phosphatase is a strong predictor of spontaneous preterm birth at 24 weeks, but not at 17 weeks.25 Serum MMP-9 was a strong predictor of spontaneous preterm birth immediately before delivery, but not earlier.26 In the creation of a multiple-marker test, issues such as patient convenience, ease of test performance, and cost are all of importance, once acceptable levels of predictability are achieved. For this reason, in the Preterm Prediction Study, we chose not to evaluate fluids that were obtained by amniocentesis as tests for spontaneous preterm birth, although there is evidence that the presence of amniotic fluid bacterial products or elevated levels of various cytokines may be of value in the prediction of spontaneous preterm birth. On the other hand, it appears that a vaginal/cervical examination to obtain ultrasound data or to collect a fluid sample is acceptable to
650 Goldenberg et al
most women, even if an examination is not routinely scheduled as part of prenatal care at that time. Similarly, a blood test appears acceptable to most women. If the test is to be of widespread value, the handling of the samples must be relatively convenient. The more special handling that a specimen may require (such as immediate centrifuging and fast freezing for corticotropin-releasing hormone24), the less widely applicable the test is likely to be. For these reasons, the findings that are related to the 3 serum tests are particularly intriguing. If replicable, a single blood drawing at about 24 weeks will identify 78% of the women who will deliver in the next 8 weeks. Because there is already extensive experience with a multiplemarker serum test for the prediction of fetal trisomies and because the individual components are not inordinately costly, some combination of these tests might be deemed appropriate for a multiple-marker test. Alternatively, the addition of 1 or more of these serum tests to the more widely studied fetal fibronectin or cervical length tests might be more appropriate. Further research that is necessary to arrive at a usable multiple-marker test would first include the replication of these findings in a large prospective study in which positive and negative predictive values could be calculated directly. Next, women who test positive should be enrolled in clinical trials that test an intervention with biologic plausibility to reduce the prematurity. Only when these criteria are met should any “multiple-marker test” for spontaneous preterm birth become part of routine prenatal care. The authors acknowledge the participation of the following institutions and others who participated in the study: The University of Alabama at Birmingham (Rachel L. Copper, MSN, CRNP; John C. Hauth, MD; Allison Northen, RN); Wake Forest University (Allison Frye, RN; Eberhard Mueller-Heubach, MD; Melissa Swain, RN); University of Chicago (Phyllis Jones, MPH, RN; Marshall Lindheimer, MD); University of Cincinnati (Nancy Elder, MSN, RN; Tariq A. Siddiqi, MD); George Washington University, The Biostatistics Center (Molly Fischer, MPH, CRNP; Lucy Leuchtenburg; Elizabeth Thom, PhD); Magee Women’s Hospital (Steve N. Caritis, MD; Margaret Cotroneo, RN; James H. Harger, MD; James M. Roberts, MD; Cynthia Stallings, MSN); National Institute of Child Health and Human Development (Charlotte Catz, MD; Mark Klebanoff, MD; Donald McNellis, MD; Sumner J. Yaffe, MD); Ohio State University (Francee Johnson, RN, BSN; Mark B. Landon, MD); University of Oklahoma (J. Christopher Carey, MD; Arlene Meier, RN); Medical University of South Carolina (Beth A. Collins, PhD, RNC; Faye LeBoeuf, MSN, CNM; Kathryn Menard, MD; Roger B. Newman, MD); University of Tennessee (Joyce Fricke, RNC; Risa Ramsey, BSN, RN; Baha Sibai, MD); Wayne
September 2001 Am J Obstet Gynecol
State University (Sydney F. Bottoms, MD [deceased]; Gwendolyn S. Norman, MPH, RN).
REFERENCES
1. McCormick MC. The contribution of low birth weight to infant mortality and childhood morbidity. N Engl J Med 1985;312:8290. 2. Goldenberg RL, Rouse DJ. The prevention of premature birth. N Engl J Med 1998;339:313-20. 3. Goldenberg RL, Andrews WW, Hauth JC. Markers of preterm birth. Prenat Neonat Med 1998;3:43-6. 4. Goldenberg RL, Iams JD, Mercer BM, Meis PJ, Moawad AH, Copper RL, et al. The preterm prediction study: the value of new vs standard risk factors in predicting early and all spontaneous preterm birth. Am J Public Health 1998;88:233-8. 5. Mercer BM, Goldenberg RL, Moawad AH, Meis PJ, Iams JD, Das A, et al. The preterm prediction study: effect of gestational age and cause of preterm birth on subsequent obstetric outcome. Am J Obstet Gynecol 1999;181:1216-21. 6. Goldenberg RL, Thom E, Moawad AH, Johnson F, Roberts J, Caritis SN. The preterm prediction study: fetal fibronectin, bacterial vaginosis and peripartum infection. Obstet Gynecol 1996;87:656-60. 7. Goldenberg RL, Mercer BM, Meis PJ, Copper RL, Das A, McNellis D. The preterm prediction study: fetal fibronectin testing and spontaneous preterm birth. Obstet Gynecol 1996;87:643-8. 8. Meis PJ, Goldenberg RL, Mercer B, Moawad A, Das A, McNellis D, et al. The preterm prediction study: significance of vaginal infections. Am J Obstet Gynecol 1995;173:1231-5. 9. Iams JD, Goldenberg RL, Meis PJ, Mercer BM, Moawad A, Das A, et al. The length of the cervix and the risk of spontaneous premature delivery. N Engl J Med 1996;334:567-72. 10. Andrews WW, Tsao J, Goldenberg RL, Hauth JC, Mercer B, Iams J, et al. The preterm prediction study: mid-trimester cervical sialidase does not predict subsequent spontaneous preterm birth. Am J Obstet Gynecol 1999;180:1151-4. 11. Goldenberg RL, Andrews WW, Guerrant RL, Newman M, Mercer B, Iams JD, et al. The preterm prediction study: lactoferrin, other markers of vaginal infection, and preterm birth. Am J Obstet Gynecol 2000;182:631-5. 12. Goldenberg RL, Andrews WW, Mercer BM, Moawad A, Meis PJ, Iams JD, et al. Granulocyte colony stimulating factor and spontaneous preterm birth. Am J Obstet Gynecol 2000;182:625-30. 13. Goldenberg RL, Iams JD, Das A, Mercer BM, Meis PJ, Moawad AH, et al. The preterm prediction study: sequential cervical length and fetal fibronectin testing for the prediction of spontaneous preterm birth. Am J Obstet Gynecol 2000;182:636-43. 14. Iams JD, Goldenberg RL, Mercer BM, Moawad AH, Meis PJ, Das A, et al. The preterm prediction study: recurrence risk of spontaneous preterm birth. Am J Obstet Gynecol 1998;178:1035-40. 15. Andrews WW, Goldenberg RL, Mercer B, Iams JD, Meis PJ, Moawad AH, et al. The preterm prediction study: association of mid-trimester genital chlamydia infection and subsequent spontaneous preterm birth. Am J Obstet Gynecol 2000;183:662-8. 16. Mercer BM, Goldenberg RL, Das AMS, Moawad AH, Iams JD, Meis PJ. The preterm prediction study: a clinical risk assessment system. Am J Obstet Gynecol 1996;174:1885-95. 17. Goepfert AR, Goldenberg RL. The preterm prediction study: quantitative fetal fibronectin values and the prediction of spontaneous preterm birth. Am J Obstet Gynecol 2000; 183:1480-3. 18. Goepfert AR, Goldenberg RL, Andrews WW, Hauth JC, Mercer BM, Iams JD, et al. The preterm prediction study: association between cervical interleukin-6, fetal fibronectin and spontaneous preterm birth. Am J Obstet Gynecol 2001; 184:483-8. 19. Copper RL, Goldenberg RL, Das A, Elder N, Swain M, Norman G, et al. The preterm prediction study: Maternal stress is associated with spontaneous preterm birth at less than thirty-five weeks’ gestation. Am J Obstet Gynecol 1996;175:1286-92.
Volume 185, Number 3 Am J Obstet Gynecol
20. Goldenberg RL, Iams JD, Miodovnik M, Van Dorsten JP, Thurnau G, Bottoms S, et al. The preterm prediction study: risk factors in twin gestations. Am J Obstet Gynecol 1996;175:1047-53. 21. Hauth JC, Goldenberg RL, Andrews WW, DuBard MB, Copper RL. Reduced incidence of preterm delivery with metronidazole and erythromycin in women with bacterial vaginosis. N Engl J Med 1995;333:1732-6. 22. Morales WJ, Schorr S, Albritton J. Effect of metronidazole in patients with preterm birth in preceding pregnancy and bacterial vaginosis: a placebo-controlled, double-blind study. Am J Obstet Gynecol 1994;171:345-9. 23. Carey JC, Klebanoff MA, Hauth JC, Hillier SL, Thom EA, Ernest JM, et al. Metronidazole to prevent preterm delivery in pregnant
Goldenberg et al 651
women with asymptomatic bacterial vaginosis. N Engl J Med 2000;342:534-40. 24. McLean M, Bisits A, Davies J, Walters W, Hackshaw A, De Voss K, et al. Predicting risk of preterm delivery by second-trimester measurement of maternal plasma corticotropin-releasing hormone and alpha-fetoprotein concentrations. Am J Obstet Gynecol 1999;181:207-15. 25. Goldenberg RL, Tamura T, DuBard M, Johnston KE, Copper RL, Neggers Y. Plasma alkaline phosphatase and pregnancy outcome. J Matern Fetal Med 1997;6:140-5. 26. Tu FF, Goldenberg RL. Plasma matrix metalloproteinase-9 (MMP-9) levels as predictors of spontaneous preterm birth. Obstet Gynecol 1998;92:446-9.
The Preterm Prediction Study: Toward a multiple-marker test for spontaneous preterm birth Robert L. Goldenberg, MD, Jay D. Iams, MD, Brian M. Mercer, MD, Paul J. Meis, MD, Atef Moawad, MD, Anita Das, PhD, Menachem Miodovnik, MD, Peter J. VanDorsten, MD, Steve N. Caritis, MD, Gary Thurnau, MD, and Mitchell P. Dombrowski, MD, for the Maternal-Fetal Medicine Units Network
OBJECTIVE: The Preterm Prediction Study evaluated 28 potential biologic markers for spontaneous preterm birth in asymptomatic women at 23 to 24 weeks gestational age. This analysis compares those markers individually and in combination for an association with spontaneous preterm birth at <32 and <35 weeks gestational age. STUDY DESIGN: With the use of a nested case-control design from an original cohort study of 2929 women, results of tests from 50 women with a spontaneous preterm birth at <32 weeks and 127 women with a spontaneous preterm birth at <35 weeks were compared with results from matched-term control subjects. RESULTS: In the univariate analysis, the most potent markers that are associated with spontaneous preterm birth at <32 weeks by odds ratio were a positive cervical-vaginal fetal fibronectin test (odds ratio, 32.7) and <10th percentile cervical length (odds ratio, 5.8), and in serum, >90th percentiles of α-fetoprotein (odds ratio, 8.3) and alkaline phosphatase (odds ratio, 6.8), and >75th percentile of granulocyte colony-stimulating factor (odds ratio, 5.5). Results for spontaneous preterm birth at <35 weeks were generally similar but not as strong. Univariate and multivariate logistic regression analyses demonstrated little interaction among the tests in their association with spontaneous preterm birth. Combinations of the 5 markers were evaluated for their association with <32 weeks spontaneous preterm birth. Ninety-three percent of case patients had at least 1 positive test result versus 34% of control subjects (odds ratio, 24.0; 95% CI, 6.4-93.4). Among the case patients, 59% had ≥2 positive test results versus 2.4% of control subjects (odds ratio, 56.5; 95% CI, 7.1-451.7). If a cutoff of 3 positive test results was used, 20% of case patients and none of the control subjects had positive test results (P < .002). With the use of only the 3 serum tests (alkaline phosphatase, α-fetoprotein, and granulocyte colony-stimulating factor), any positive test identified 81% of cases versus 22% of control subjects (odds ratio, 14.7; 95% CI, 5.0-42.7). For spontaneous preterm birth at <35 weeks gestation, any 2 positive tests identified 43% of cases and 6% of control subjects (odds ratio, 11.2; 95% CI, 4.8-26.2). CONCLUSION: Overlap among the strongest biologic markers for spontaneous preterm birth is small. This suggests that the use of tests such as maternal serum α-fetoprotein, alkaline phosphatase, and granulocyte colony-stimulating factor as a group or adding their results to fetal fibronectin test and cervical length test results may enhance our ability to predict spontaneous preterm birth and that the development of a multiplemarker test for spontaneous preterm birth is feasible. (Am J Obstet Gynecol 2001;185:643-51.)
Key words: Preterm birth, prediction, markers, multiple-marker test
Preterm birth is the major cause of perinatal morbidity and death in developed countries.1 In recent decades, substantial progress has been made in the reduction of From the Maternal-Fetal Medicine Units Network. Supported by the following grants from the National Institute of Child Health and Human Development (NICHD): HD21410, HC21414, HC21434, HD27860, HD27861, HD27869, HD27883, HD27889, HD27905, HD27915, HD27917, and HD19897. Received for publication September 19, 2000; revised January 23, 2001; accepted April 24, 2001. Reprints not available from the authors. Copyright © 2001 by Mosby, Inc. 0002-9378/2001 $35.00 + 0 6/1/116752 doi:10:1067/mob.2001.116752
deaths that were associated with preterm birth, but little or no progress has been made in the reduction of the preterm birth rate itself.2 The reasons for this failure are not clear, but evidence is accumulating that most strategies that are aimed at the reduction of preterm birth in the general population are not effective.2 Preventive strategies that are targeted at populations with specific risk factors or preterm birth associated with specific causes are therefore a focus of current research. Risk factors for preterm birth can be categorized into several groups that include (1) demographic information, (2) previous pregnancy history, (3) current pregnancy findings, (4) nutritional/body size in origin, (5) 643
644 Goldenberg et al
associated biophysical markers (such as contractions and cervical length), and (6) measurements of analytes that were made in various biologic fluids.3 Among the first 4 groups, multiple pregnancy, previous preterm birth, black race, and second- or third-trimester bleeding have been associated most consistently with preterm birth. Pregnancies with ≥2 fetuses have an extremely high risk of preterm birth; 50% of all twin pregnancies and virtually all of the higher-order multiple pregnancies are born preterm. Among women with singleton pregnancies, a previous preterm birth is the strongest historic predictor, with relative risks reported in the range of 2 to 6.4, 5 Among the biophysical and biologic fluid tests evaluated to date, cervical length as measured by ultrasound and fetal fibronectin levels in cervical or vaginal fluid have been most strongly and consistently associated with subsequent spontaneous preterm birth.4 The Preterm Prediction Study of the National Institute of Child Health and Human Development Maternal-Fetal Medicine Units Network was initiated to evaluate many potential predictors of spontaneous preterm birth to determine their relative strengths and to evaluate whether several of the tests could be used together (as a multiplemarker test) to predict spontaneous preterm birth more effectively. Evaluations of many of the individual tests have been presented previously.4-20 In this analysis, we compare the tests against 1 another and evaluate groups of tests to determine the feasibility of the development of a multiple-marker test for spontaneous preterm birth. We chose to focus this study on data available at 23 to 24 weeks in asymptomatic women who were undergoing routine prenatal care. At 23 to 24 weeks’ gestational age, the fetus with a weight of approximately 600 g is at the margin of viability. The preponderance of neonatal deaths and much of the birth-associated long-term handicap will be found in those infants who are born over the next 4 to 8 weeks. Reducing the number of births between 24 and 32 weeks’ gestational age would have a major impact on adverse pregnancy outcomes in the United States. Methods The relationship between spontaneous preterm birth and many of the individual markers that were evaluated at 23 to 24 weeks in the Preterm Prediction Study have been presented previously.4-20 Because of the large number of women in the original cohort (n = 2929), the large number of potential tests to be studied, and their costs, we originally planned to complete a nested case-control study of the tests after the major outcomes (such as spontaneous preterm birth) were determined. Women who at the time of the enrollment had a known twin pregnancy were not included in this analysis. In addition, multiparous women who had ≥3-cm cervix dilation, nulliparous women who had ≥2-cm cervix dilation, women who
September 2001 Am J Obstet Gynecol
had membranes that bulged from the cervical os, or women who had a known placental previa before enrollment at 23 to 24 weeks were excluded from participation. After exclusions, the study population consisted of 2929 women who were receiving routine prenatal care. They were not selected because they were at high risk for preterm birth. Of this population, 63% of the women were black; 37% of the women were nonblack; 58% of the women were multiparous; and 42% of the women were nulliparous. The mean age of the women in the study population was 23.7 ± 5.5 years. Of the 2929 women who were enrolled, 127 of the women (4.3%) had a spontaneous preterm birth at <35 weeks, and they and their matched control subjects are the women who were considered in this analysis. Extensive data were collected on every woman in the cohort; the data included the demographic information, previous pregnancy outcomes, health history, and current pregnancy data. In addition, because we were interested in studying sequential cervical/vaginal fetal fibronectin testing, fetal fibronectin samples were collected and analyzed prospectively on each woman at 24, 26, 28, and 30 weeks.6, 7 We had available both cervical and vaginal fetal fibronectin tests, and if either 1 or both tests were positive at 24 weeks, we considered that the test for fetal fibronectin was positive. Also, a Gram stain and pH test for bacterial vaginosis and an ultrasound cervical examination were performed for each woman at 24 and 28 weeks.8, 9 Fluids that were available for analysis after being frozen at –70°C included serum, plasma, and urine that had been obtained at 24 and 28 weeks and cervical and vaginal fluid in buffer that had been originally collected for fetal fibronectin testing. At the start of the study, we planned to measure 10 serum analytes that previously had been associated with spontaneous preterm birth. However, as the study progressed, additional analyses were proposed on the various fluids. These were encouraged, as long as there was reasonable biologic plausibility for an association with spontaneous preterm birth, the fluid was appropriate for the test, the resources were available for the analyses, and the volume of fluid that was needed for the test was not overly large and sufficient fluid remained. Although some of the analytes were assayed at both 24 and 28 weeks, as a general rule, we evaluated the 24-week fluid alone. This strategy was chosen because we were especially interested in the association between various markers and very early spontaneous preterm birth. Also, we often did not have the resources to analyze both sets of samples. The samples were always analyzed in a blinded fashion; the individual laboratories were not aware which samples were from the case patients and which samples were from the control subjects. The latter were defined as women who delivered at term (≥37 weeks) and were individually matched to the case patients by race, parity, and center. The data were entered into a relational database at the
Volume 185, Number 3 Am J Obstet Gynecol
George Washington University Biostatistics Center and analyzed only after all measurements of a specific analyte were complete. Each woman provided informed consent, and each center obtained approval for the study by its Institutional Review Board. In the original study, the primary outcome was a spontaneous preterm birth before the 35th week (defined as after the onset of spontaneous labor or rupture of membranes). We hypothesized that approximately 3.5% of our population of 3000 women would have this outcome (n = 105). In a case-control study in which 10% of the population is positive for a given screening test, 100 case patients and 100 control subjects would give 62% power to detect an association with spontaneous preterm birth characterized by an odds ratio of ≥2.5 and 80% power for an odds ratio of 3.0. Because we were also interested in predicting earlier spontaneous preterm birth, we chose to evaluate the births that occurred at <32 weeks in a similar casecontrol fashion. If 10% of the population is positive for a specific risk factor, with 45 case patients and control subjects, we would have a 59% power to detect an odds ratio of 3.5 and 69% power for an odds ratio of 4.0. Various factors or positive tests are defined as a previous spontaneous preterm birth that occurred 20 through 36 weeks after the onset of spontaneous labor or rupture of membranes, regardless of the method of delivery. Body mass index was defined as weight in kilograms divided by height in meters squared. Vaginal bleeding was defined as any noted by the woman after her last menstrual period. The presence of contractions was defined as any contractions appreciated by the patient in the 2 weeks before her study visit. Many of the methods for the various tests and the 90th percentile cutoffs have been published previously and are referenced.4-20 The methods for those not published can be obtained from the authors. For consistency, for those tests that did not have a previously defined percentile cutoff, we used values greater than the 90th percentile of the control subjects to define a positive test. For fetal fibronectin, we used a cutoff of ≥50 ng/mL. Also, because such a high proportion of this population had a vaginal Gram stain score of ≥7 and a pH of ≥4.5, to obtain a smaller percent of positives, we present data using a Gram stain score of ≥9 and a pH of ≥5. As described in the “Results” section of this article, for each test we also evaluated >75th and >95th percentile of the control subjects as a predictor of spontaneous preterm birth. Because this was a case-control and not a cohort study, creating a standard receiver-operator curve for each analyte was deemed inappropriate, as was presenting the positive and negative predictive values. Results Tables I and II show each of the factors included in the case-control analysis, the source of data or fluid from which the measurement was made, the cutoff value for a
Goldenberg et al 645
positive test, the percent of test results that were positive in the case patients and in the control subjects, and the crude odds ratio for spontaneous preterm birth at <32 weeks (Table I) and at <35 weeks (Table II). Significance of the association is also noted. As can be seen for both spontaneous preterm birth at <32 and <35 weeks, a positive fetal fibronectin test result had the greatest odds ratio for its association with spontaneous preterm birth, but many other factors were also significantly associated with spontaneous preterm birth in 1 or both gestational age categories. For both <32- and <35-week spontaneous preterm birth, these positive test results included alkaline phosphatase level, α-fetoprotein level, vaginal pH, previous spontaneous preterm birth, and short cervical length. For spontaneous preterm birth at <32 weeks alone, interstitial cell adhesion molecule, defensins, and cervical IL-6 were significantly associated; for spontaneous preterm birth at <35 weeks alone, cervical lactoferrin and chlamydia, bleeding, and low body mass index were significantly associated. In addition to evaluating the 90th percentile as a cutoff, we also examined the effect on spontaneous preterm birth of the use of the 75th and 95th percentile as cutoffs for a positive test. The only factor that was previously found to have an insignificant relationship that consistently emerged as significant in this analysis was serum granulocyte colony-stimulating factor at the 75th percentile for both <35- and <32-week spontaneous preterm birth (odds ratio, 2.8 for <35 weeks and 5.5 for <32 weeks). For this reason, in subsequent analyses, for granulocyte colony-stimulating factor, we used the 75th percentile as the cutoff for a positive test result. To identify factors to consider for use in a multiplemarker test, we performed regression analyses for <32- or <35-week spontaneous preterm birth using all factors that were significantly associated with spontaneous preterm birth in univariate analyses (Table III). For spontaneous preterm birth at <32 weeks, fetal fibronectin, granulocyte colony-stimulating factor, defensins, and α-fetoprotein remained significantly associated. For <35 weeks, fetal fibronectin, short cervix, α-fetoprotein, alkaline phosphatase, granulocyte colony-stimulating factor, vaginal bleeding, and previous spontaneous preterm birth remained significantly associated. We were interested in the stability of the tests that were performed at both 24 and 28 weeks. Table IV shows the correlation coefficients and probability values between the 2 measurements, when data at 28 weeks were available. For specific tests, values obtained at both 24 and 28 weeks always correlated significantly with 1 another. In addition, the correlation coefficients were generally high (>0.4), which confirmed that values that are high at 24 weeks tended to remain high at 28 weeks and indicated that the results were generally consistent over time. Interestingly, fetal fibronectin results had 1 of the lowest correlations between 24 and 28 weeks.
646 Goldenberg et al
September 2001 Am J Obstet Gynecol
Table I. Risk factors and tests for spontaneous preterm birth at <32 weeks obtained or available at 24 weeks’ gestational age in case patients and control subjects Factor
Corticotropin-releasing hormone α-fetoprotein Alkaline phosphatase Beta2-macroglobulin Ferritin Interstitial cell adhesion molecule-1 Interleukin-6 C-reactive protein Cortisol Lactoferrin Defensins Relaxin Interleukin-10 Granulocyte colony-stimulating factor Activan Interleukin-6 Lactoferrin Defensins Sialidase Short cervix Fetal fibronectin Gram stain score pH Chlamydia Previous spontaneous preterm birth Contractions Bleeding Body mass index *P
Source of data or fluid
Test cutoff
Case Patients (% positive)
Control subjects (% positive)
Odds ratio (case patients vs control subjects)
Serum Serum Serum Serum Serum Serum Serum Serum Serum Serum Serum Serum Serum Serum Serum Cervix Cervix Cervix Cervix Ultrasound Cervix/vagina Vagina Vagina Vagina History History History Measured
90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile Positive 90th %ile 90th %ile 90th %ile 90th %ile ≤25 mm ≥50 ng/mL ≥9 ≥5.0 Positive Positive Positive Positive <19.8
10.6 36.1 22.9 6.3 14.6 23.4 10.4 10.4 10.4 11.6 20.9 10.9 6.4 8.5 8.5 20.4 4.3 10.5 8.2 44.9 40.0 30.0 42.9 15.2 42.0 24.0 36.0 29.2
4.3 6.4 4.2 2.1 2.1 6.4 8.3 6.3 2.1 9.3 4.7 10.9 12.8 10.6 10.6 6.1 0.0 18.4 6.1 12.2 2.0 24.0 18.4 6.5 14.0 32.0 24.0 14.6
2.7 8.3* 6.8* 3.1 8.0 4.5* 1.3 1.7 5.5 1.3 5.4* 1.0 0.5 0.8 0.8 3.9* ∞ 0.5 1.4 5.8* 32.7* 1.4 3.3* 2.6 4.5* 0.7 1.8 2.4
< .05.
Table V shows the correlation coefficients among the tests that were performed at 24 weeks now being considered as part of the multiple-markers panel. The low correlations among the tests indicate that overlap between many of the tests in their association with subsequent spontaneous preterm birth is small. Among the tests considered here, none had a correlation coefficient >0.23, despite the fact that a number of associations were statistically significant. To look for common patterns of test positivity, we next created a matrix that included the distribution of positive tests for each of the case patients and for each of the control subjects (data not shown). For spontaneous preterm birth at <35 weeks, of the 107 cases with all tests available for analysis, the most common pattern was no positive test result, which was found in 13 of 107 women, followed by 10 of 107 women who had only granulocyte colonystimulating factor positivity, 8 of 107 women who had only fetal fibronectin positivity, 7 of 107 women who had only a short cervix, and 6 of 107 women who had only αfetoprotein positivity. Fifty-nine case patients had various combinations of more than 1 positive test, none of which were present in more than 4 women. For the 107 matched control subjects, the most common pattern, by far, was all test results negative in 52 of 107 women, fol-
lowed by 15 of 107 women with only bleeding. Similar scatter in test patterns was seen among the case patients and control subjects who were delivered before 32 weeks. Because any single combination of positive tests was not common enough to be of value, we next evaluated whether counting the number of positive tests would be useful. Table VI shows how individual and then selected combinations of tests relate to the presence of subsequent spontaneous preterm birth at <32 weeks. Specifically, the sensitivity and specificity for spontaneous preterm birth at <32 weeks that were associated with selected patterns of positive tests and the odds ratios are shown. For spontaneous preterm birth at <32 weeks, if a combination of 5 tests (fetal fibronectin, cervical length, defensins, α-fetoprotein, and granulocyte colony-stimulating factor) is used and if a positive result for any test makes the panel positive, 92.7% of those women who had a spontaneous preterm birth had a positive test with an odds ratio of 24.4. However, the specificity was only 65.9%. If 2 of the 5 tests were required to be positive for a “positive” test, 58.5% of those women who delivered preterm had a positive test with an odds ratio of 56.5 and a specificity of 97.6%. Using 3 positive tests to define a “positive test” reduced the sensitivity to 19.5%, but achieved 100% specificity. Especially interesting is that when the 3
Goldenberg et al 647
Volume 185, Number 3 Am J Obstet Gynecol
Table II. Risk factors and tests for spontaneous preterm birth at <35 weeks obtained or available at 24 weeks’ gestational age in case patients and control subjects Factor
Corticotropin-releasing hormone α-fetoprotein Alkaline phosphatase Beta2-macroglobulin Ferritin Interstitial cell adhesion molecule-1 Interleukin-6 C-reactive protein Cortisol Lactoferrin Defensins Relaxin Interleukin-10 Granulocyte colony-stimulating factor Activan Interleukin-6 Lactoferrin Defensins Sialidase Short cervix Fetal fibronectin Gram stain score pH Chlamydia Previous spontaneous preterm birth Contractions Bleeding Body mass index *P
Fluid
Test cutoff
Positive case patients (%)
Positive control subjects (%)
Serum Serum Serum Serum Serum Serum Serum Serum Serum Serum Serum Serum Serum Serum Serum Cervix Cervix Cervix Cervix Ultrasound Cervix/vagina Vagina Vagina Vagina History History History Measured
90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile 90th %ile Positive 90th %ile 90th %ile 90th %ile 90th %ile ≤25 mm ≥50 ng/mL ≥9 >5.0 Positive Positive Positive Positive <19.8
11.7 35.3 14.9 5.1 9.9 16.4 10.0 8.3 12.6 8.8 18.4 13.6 4.3 7.7 14.2 20.0 5.0 16.7 9.6 36.8 22.8 22.8 38.1 12.8 43.3 31.5 35.4 30.9
8.3 13.5 3.3 6.8 7.4 9.1 9.2 6.7 7.6 11.4 10.5 8.2 10.3 7.7 11.7 9.6 0.0 10.0 9.6 9.6 3.2 15.0 21.4 5.1 15.0 31.5 20.5 17.9
Odds ratio (case patients vs control subjects) 1.5 3.5* 5.1* 0.7 1.4 2.0 1.1 1.3 1.8 0.8 1.9 1.8 0.4 1.0 1.3 2.4 ∞ 1.8 1.0 5.5* 9.1* 1.7 2.3* 2.7* 4.3* 1.0 2.1* 2.1*
< .05.
Table III. The adjusted odds ratio for a spontaneous preterm birth at <32 or <35 weeks that was associated with a specific risk factor, based on regression analysis Preterm birth Factor α-fetoprotein (>90th %ile) Alkaline phosphatase (>90th %ile) Granulocyte colony-stimulating factor (>75th %ile) Defensins (>90th %ile) Fetal fibronectin (≥50 ng/mL) Cervical length (≤25 mm) History of previous spontaneous preterm birth Vaginal bleeding
<32 wks Odds ratio (%) 33.7 — 12.7 13.1 12.2 3.7 2.1 —
serum tests were considered as a group, with a requirement for a “positive test” being any positive serum test, a “positive test” was found in 80.5% of those women who delivered preterm, with an odds ratio of 14.7 and a specificity of 78.1%. Overall, various combinations of tests at 24 weeks do not appear to be as strongly predictive of <35-week spontaneous preterm birth as of <32-week spontaneous preterm birth (Table VII). However, if the 3 serum tests (α-fetoprotein, granulocyte colony-stimulating factor, and alkaline phosphatase) were used, 1 positive test
95% CI 5.0-229.7 — 2.8-56.5 1.6-104.5 1.1-135.2 0.7-19.4 0.5-9.4 —
<35 wks Odds ratio (%) 3.9 4.0 3.1 — 6.6 3.9 4.0 2.2
95% CI 1.7-8.7 1.1-14.9 1.4-6.9 — 1.7-25.5 1.7-9.2 1.9-8.4 1.1-4.4
among them had a sensitivity of 60.2%, a specificity of 72.6%, and an odds ratio of 4.0, although 2 positive tests will have a sensitivity of 18.6%, a specificity of 97.4%, and an odds ratio of 8.4. Adding previous spontaneous preterm birth to the panel (data not shown) increased the sensitivity to 78.8% and the odds ratio to 6.0 but reduced the specificity to 62%. Comment To date, the potential benefits in improving pregnancy outcome by predicting spontaneous preterm birth have
648 Goldenberg et al
September 2001 Am J Obstet Gynecol
Table IV. Correlation between tests performed at 24 and 28 weeks
Test Corticotropin-releasing hormone α-fetoprotein Alkaline phosphatase Beta2-macroglobulin Ferritin Interstitial cell adhesion molecule-1 Interleukin-6 C-reactive protein Cortisol Lactoferrin Defensins Relaxin Interleukin-10 Granulocyte colony-stimulating factor Cervical length Cervical/vaginal fetal fibronectin Gram stain score pH
Results that were available at both 24 and 28 wks (n)
Correlation coefficient
P value
154 156 156 146 156 132 156 156 147 148 150 134 144 144 172 164 174 175
0.28 0.69 0.74 0.46 0.63 0.81 0.79 0.58 0.53 0.48 0.51 0.84 0.37 0.33 0.70 0.18 0.63 0.53
.0005 .0001 .0001 .0001 .0001 .0001 .0001 .0001 .0001 .0001 .0001 .0001 .0001 .0001 .0001 .020 .0001 .0001
Table V. The correlations among the strongest predictors of spontaneous preterm birth Alkaline phosphatase Serum alkaline phosphatase Serum α-fetoprotein Serum granulocyte colonystimulating factor Serum defensin Fetal fibronectin Cervical length *P
— 0.10 0.14* 0.09 0.10 –0.01
α-fetoprotein
Granulocyte colonystimulating factor
Defensins
Fetal fibronectin
Cervical length
0.10 — –0.03
0.14* –0.03 —
0.09 0.05 0.10
0.10 –0.04 0.15*
–0.01 –0.03 –0.09
0.05 –0.04 –0.03
0.10 0.15* –0.09
— 0.05 0.17*
0.05 — –0.23*
0.17* –0.23* —
< .05.
not been realized, principally because we have few, if any, effective interventions that can be used once high-risk status has been identified.2 As an example, it is now reasonably clear that the use of the demographic screening tests that focus on age, race, workload, and previous pregnancy outcomes, when used to identify women for various additional testing or interventions such as more intensive observation, home uterine activity monitoring, bed rest, or tocolytic use, has not resulted in a reduction in preterm births.2 One explanation commonly given for this failure, other than lack of effectiveness, is that the risk factors were too nonspecific and, in any case, not strong enough to identify a population that might benefit from the interventions. Based on several promising studies, a number of authors have proposed targeting specific interventions (ie, antibiotics) at women who are high risk (ie, who have had a previous spontaneous preterm birth) and who have another and perhaps treatable risk factor (ie, bacterial vaginosis).21, 22 In fact, this strategy is 1 of the few in which randomized trial data support the use of a particular intervention, although, clearly, the results have not been consistent across all studies.23
We believe there are at least 2 reasons that (even with a history of failure in the use of predictors of spontaneous preterm birth to reduce prematurity) research in this area is important.3 First, the identification of specific markers as predictors may provide important clues to pathways that lead to spontaneous preterm birth and that may, in themselves, suggest interventions. Second, the knowledge of which markers are associated with spontaneous preterm birth will encourage the study of populations of women at especially high risk of spontaneous preterm birth in intervention trials, which will allow the trials to be performed more efficiently and with less risk of harm to women who are destined to deliver at term. We do not yet advocate a third reason commonly suggested for the identification of women at high risk of spontaneous preterm birth: the belief that the identification of risk without having an effective intervention will somehow improve outcome. To date, the screening of populations with any test and the labeling of some women as high risk for spontaneous preterm birth have only led to the increased use of various interventions and to increased costs, without any improvement in outcome.
Goldenberg et al 649
Volume 185, Number 3 Am J Obstet Gynecol
Table VI. Test values for the prediction of spontaneous preterm birth at <32 weeks, comparing single versus multiple markers Test Fetal fibronectin (≥50 ng/mL) Cervix (≤25 mm) α-fetoprotein(>90th %ile) Granulocyte colony-stimulating factor (>75th %ile) Defensins (> 90th %ile ) 1/5 Tests positive 2/5 Tests positive 3/5 Tests positive 1/3 (serum) Tests positive 2/3 (serum) Tests positive *Odds
Sensitivity
Specificity
Odds ratio (%)
95% CI
40.0 44.9 36.2 48.9
98.0 87.8 93.6 85.1
32.7 5.8 8.3 5.5
4.2-256.1 2.1-16.2 2.2-30.9 2.0-14.7
20.9 92.7 58.5 19.5 80.5 26.8
95.4 65.9 97.6 100 78.1 100
5.4 24.4 56.5 ∞* 14.7 ∞*
1.1-26.8 6.4-93.4 7.1-451.7 — 5.0-42.7 —
ratio cannot be calculated with specificity of 100%.
Table VII. Test values for the prediction of spontaneous preterm birth at <35 weeks, comparing single versus multiple markers Test Fetal fibronectin (≥50 ng/mL) Cervix (≤25 mm) α-fetoprotein (>90th %ile) Alkaline phosphatase (>90th %ile) Granulocyte colony-stimulating factor (>75th %ile) 1/5 Tests positive 2/5 Tests positive 3/5 Tests positive 1/3 (serum) Tests positive 2/3 (serum) Tests positive *Odds
Sensitivity
Specificity
Odds ratio (%)
95% CI
22.8 36.8 35.3 14.9 32.5
96.9 90.4 86.6 96.7 85.5
9.1 5.5 3.5 5.1 2.8
3.1-26.8 2.7-11.0 1.8-6.7 1.7-15.6 1.5-5.4
78.8 42.5 15.0 60.2 18.8
63.7 93.8 100 72.6 97.4
6.5 11.2 ∞* 4.0 5.7
3.6-11.8 4.8-26.2 — 2.3-7.0 2.4-29.0
ratio cannot be calculated with specificity of 100%.
We believe that 1 finding from this study is especially important: that is, the apparently small overlap between many of the tests in their association with spontaneous preterm birth. If all the tests tended to be positive in the same group of women who delivered preterm, this would suggest a single predominant pathway that leads to spontaneous preterm birth. It would also suggest that a multiple-marker test might be less valuable in the prediction of spontaneous preterm birth. Our finding of only a small degree of overlap among the markers in women who deliver preterm suggests that there are several pathways that lead to spontaneous preterm birth4 and also suggests that the use of several biologic markers together might be useful in the creation of a multiple-marker test with high sensitivity and a high odds ratio. In fact, we and other groups have previously contemplated the benefits of using more than 1 biologic marker.4 For example, McLean et al24 suggested the use of corticotropin-releasing hormone and α-fetoprotein together, although other researchers have suggested the use of fetal fibronectin and cervical length or cervical interleukin-6 and fetal fibronectin.18 Many groups have studied individual serum/plasma or cervical/vaginal tests for the prediction of preterm birth,
often with discrepant results. Although there are many potential reasons for differing results (including the populations studied), whether the preterm births were predominantly early or late, the test characteristics, the differences in the gestational age at testing, and the timing of the test relative to the timing of the preterm delivery are important factors often not considered. For example, we have previously shown that plasma alkaline phosphatase is a strong predictor of spontaneous preterm birth at 24 weeks, but not at 17 weeks.25 Serum MMP-9 was a strong predictor of spontaneous preterm birth immediately before delivery, but not earlier.26 In the creation of a multiple-marker test, issues such as patient convenience, ease of test performance, and cost are all of importance, once acceptable levels of predictability are achieved. For this reason, in the Preterm Prediction Study, we chose not to evaluate fluids that were obtained by amniocentesis as tests for spontaneous preterm birth, although there is evidence that the presence of amniotic fluid bacterial products or elevated levels of various cytokines may be of value in the prediction of spontaneous preterm birth. On the other hand, it appears that a vaginal/cervical examination to obtain ultrasound data or to collect a fluid sample is acceptable to
650 Goldenberg et al
most women, even if an examination is not routinely scheduled as part of prenatal care at that time. Similarly, a blood test appears acceptable to most women. If the test is to be of widespread value, the handling of the samples must be relatively convenient. The more special handling that a specimen may require (such as immediate centrifuging and fast freezing for corticotropin-releasing hormone24), the less widely applicable the test is likely to be. For these reasons, the findings that are related to the 3 serum tests are particularly intriguing. If replicable, a single blood drawing at about 24 weeks will identify 78% of the women who will deliver in the next 8 weeks. Because there is already extensive experience with a multiplemarker serum test for the prediction of fetal trisomies and because the individual components are not inordinately costly, some combination of these tests might be deemed appropriate for a multiple-marker test. Alternatively, the addition of 1 or more of these serum tests to the more widely studied fetal fibronectin or cervical length tests might be more appropriate. Further research that is necessary to arrive at a usable multiple-marker test would first include the replication of these findings in a large prospective study in which positive and negative predictive values could be calculated directly. Next, women who test positive should be enrolled in clinical trials that test an intervention with biologic plausibility to reduce the prematurity. Only when these criteria are met should any “multiple-marker test” for spontaneous preterm birth become part of routine prenatal care. The authors acknowledge the participation of the following institutions and others who participated in the study: The University of Alabama at Birmingham (Rachel L. Copper, MSN, CRNP; John C. Hauth, MD; Allison Northen, RN); Wake Forest University (Allison Frye, RN; Eberhard Mueller-Heubach, MD; Melissa Swain, RN); University of Chicago (Phyllis Jones, MPH, RN; Marshall Lindheimer, MD); University of Cincinnati (Nancy Elder, MSN, RN; Tariq A. Siddiqi, MD); George Washington University, The Biostatistics Center (Molly Fischer, MPH, CRNP; Lucy Leuchtenburg; Elizabeth Thom, PhD); Magee Women’s Hospital (Steve N. Caritis, MD; Margaret Cotroneo, RN; James H. Harger, MD; James M. Roberts, MD; Cynthia Stallings, MSN); National Institute of Child Health and Human Development (Charlotte Catz, MD; Mark Klebanoff, MD; Donald McNellis, MD; Sumner J. Yaffe, MD); Ohio State University (Francee Johnson, RN, BSN; Mark B. Landon, MD); University of Oklahoma (J. Christopher Carey, MD; Arlene Meier, RN); Medical University of South Carolina (Beth A. Collins, PhD, RNC; Faye LeBoeuf, MSN, CNM; Kathryn Menard, MD; Roger B. Newman, MD); University of Tennessee (Joyce Fricke, RNC; Risa Ramsey, BSN, RN; Baha Sibai, MD); Wayne
September 2001 Am J Obstet Gynecol
State University (Sydney F. Bottoms, MD [deceased]; Gwendolyn S. Norman, MPH, RN).
REFERENCES
1. McCormick MC. The contribution of low birth weight to infant mortality and childhood morbidity. N Engl J Med 1985;312:8290. 2. Goldenberg RL, Rouse DJ. The prevention of premature birth. N Engl J Med 1998;339:313-20. 3. Goldenberg RL, Andrews WW, Hauth JC. Markers of preterm birth. Prenat Neonat Med 1998;3:43-6. 4. Goldenberg RL, Iams JD, Mercer BM, Meis PJ, Moawad AH, Copper RL, et al. The preterm prediction study: the value of new vs standard risk factors in predicting early and all spontaneous preterm birth. Am J Public Health 1998;88:233-8. 5. Mercer BM, Goldenberg RL, Moawad AH, Meis PJ, Iams JD, Das A, et al. The preterm prediction study: effect of gestational age and cause of preterm birth on subsequent obstetric outcome. Am J Obstet Gynecol 1999;181:1216-21. 6. Goldenberg RL, Thom E, Moawad AH, Johnson F, Roberts J, Caritis SN. The preterm prediction study: fetal fibronectin, bacterial vaginosis and peripartum infection. Obstet Gynecol 1996;87:656-60. 7. Goldenberg RL, Mercer BM, Meis PJ, Copper RL, Das A, McNellis D. The preterm prediction study: fetal fibronectin testing and spontaneous preterm birth. Obstet Gynecol 1996;87:643-8. 8. Meis PJ, Goldenberg RL, Mercer B, Moawad A, Das A, McNellis D, et al. The preterm prediction study: significance of vaginal infections. Am J Obstet Gynecol 1995;173:1231-5. 9. Iams JD, Goldenberg RL, Meis PJ, Mercer BM, Moawad A, Das A, et al. The length of the cervix and the risk of spontaneous premature delivery. N Engl J Med 1996;334:567-72. 10. Andrews WW, Tsao J, Goldenberg RL, Hauth JC, Mercer B, Iams J, et al. The preterm prediction study: mid-trimester cervical sialidase does not predict subsequent spontaneous preterm birth. Am J Obstet Gynecol 1999;180:1151-4. 11. Goldenberg RL, Andrews WW, Guerrant RL, Newman M, Mercer B, Iams JD, et al. The preterm prediction study: lactoferrin, other markers of vaginal infection, and preterm birth. Am J Obstet Gynecol 2000;182:631-5. 12. Goldenberg RL, Andrews WW, Mercer BM, Moawad A, Meis PJ, Iams JD, et al. Granulocyte colony stimulating factor and spontaneous preterm birth. Am J Obstet Gynecol 2000;182:625-30. 13. Goldenberg RL, Iams JD, Das A, Mercer BM, Meis PJ, Moawad AH, et al. The preterm prediction study: sequential cervical length and fetal fibronectin testing for the prediction of spontaneous preterm birth. Am J Obstet Gynecol 2000;182:636-43. 14. Iams JD, Goldenberg RL, Mercer BM, Moawad AH, Meis PJ, Das A, et al. The preterm prediction study: recurrence risk of spontaneous preterm birth. Am J Obstet Gynecol 1998;178:1035-40. 15. Andrews WW, Goldenberg RL, Mercer B, Iams JD, Meis PJ, Moawad AH, et al. The preterm prediction study: association of mid-trimester genital chlamydia infection and subsequent spontaneous preterm birth. Am J Obstet Gynecol 2000;183:662-8. 16. Mercer BM, Goldenberg RL, Das AMS, Moawad AH, Iams JD, Meis PJ. The preterm prediction study: a clinical risk assessment system. Am J Obstet Gynecol 1996;174:1885-95. 17. Goepfert AR, Goldenberg RL. The preterm prediction study: quantitative fetal fibronectin values and the prediction of spontaneous preterm birth. Am J Obstet Gynecol 2000; 183:1480-3. 18. Goepfert AR, Goldenberg RL, Andrews WW, Hauth JC, Mercer BM, Iams JD, et al. The preterm prediction study: association between cervical interleukin-6, fetal fibronectin and spontaneous preterm birth. Am J Obstet Gynecol 2001; 184:483-8. 19. Copper RL, Goldenberg RL, Das A, Elder N, Swain M, Norman G, et al. The preterm prediction study: Maternal stress is associated with spontaneous preterm birth at less than thirty-five weeks’ gestation. Am J Obstet Gynecol 1996;175:1286-92.
Volume 185, Number 3 Am J Obstet Gynecol
20. Goldenberg RL, Iams JD, Miodovnik M, Van Dorsten JP, Thurnau G, Bottoms S, et al. The preterm prediction study: risk factors in twin gestations. Am J Obstet Gynecol 1996;175:1047-53. 21. Hauth JC, Goldenberg RL, Andrews WW, DuBard MB, Copper RL. Reduced incidence of preterm delivery with metronidazole and erythromycin in women with bacterial vaginosis. N Engl J Med 1995;333:1732-6. 22. Morales WJ, Schorr S, Albritton J. Effect of metronidazole in patients with preterm birth in preceding pregnancy and bacterial vaginosis: a placebo-controlled, double-blind study. Am J Obstet Gynecol 1994;171:345-9. 23. Carey JC, Klebanoff MA, Hauth JC, Hillier SL, Thom EA, Ernest JM, et al. Metronidazole to prevent preterm delivery in pregnant
Goldenberg et al 651
women with asymptomatic bacterial vaginosis. N Engl J Med 2000;342:534-40. 24. McLean M, Bisits A, Davies J, Walters W, Hackshaw A, De Voss K, et al. Predicting risk of preterm delivery by second-trimester measurement of maternal plasma corticotropin-releasing hormone and alpha-fetoprotein concentrations. Am J Obstet Gynecol 1999;181:207-15. 25. Goldenberg RL, Tamura T, DuBard M, Johnston KE, Copper RL, Neggers Y. Plasma alkaline phosphatase and pregnancy outcome. J Matern Fetal Med 1997;6:140-5. 26. Tu FF, Goldenberg RL. Plasma matrix metalloproteinase-9 (MMP-9) levels as predictors of spontaneous preterm birth. Obstet Gynecol 1998;92:446-9.