Chapter  18:  Management of Preterm Labor: Evidence Report/Technology Assessment Number 18

Databases, Inclusion/Exclusion Criteria, and Search Terms

We conducted a detailed search of the relevant literatures using the following databases: MEDLINE; EMBASE; the Cochrane Collaboration and its related York Database, International Pharmaceutical Abstracts; the Health Economic Evaluations Database; Genderwatch; and the Population Index. We also conducted an extensive search of the gray literature, chiefly government documents, with a focus on tocolytics and biologic markers.

We examined literature meeting the following criteria:

• Study populations: humans; pregnant females.

• Condition: signs and symptoms of preterm labor, not including articles in which all research subjects experienced preterm premature rupture of membranes, medically indicated preterm delivery, or multiple gestation.

• Interventions: biologic markers as noted, tocolytics, antibiotics, and home uterine activity monitoring.

• Study settings: essentially all inpatient and outpatient settings including patients' homes.

• Outcomes: three major categories of outcomes-birth, maternal morbidities, and infant health.

• Time period: for pharmacotherapies and biologic markers, 1966 through 1999, depending on the date of approval; for home uterine activity monitoring, 1980 through 1999.

• Languages and geographic sites: English, French, and German (English only for home uterine activity monitoring), and excluding locations based on language of publication other than these.

• Admissible evidence: efficacy studies in which health care is delivered under ideal settings, identified as randomized controlled trials (RCTs) (double and single blinded); effectiveness studies (non-RCTs, prospective and retrospective cohort studies, and case control studies) in which health care is delivered under ordinary circumstances; only RCTs were used for home uterine activity monitoring; other information included meta-analyses, review articles for reference lists, and cost-effectiveness studies; finally, sample sizes of 40 or more subjects.

The Medical Subject Headings (MeSH) used for the searches were limited by the subject heading "premature labor." With respect to study designs, we included a variety of terms in "epidemiologic study characteristics" (exploded). In addition, we searched for specific therapies, including "biologic markers," "antibiotics," and "tocolytic agents." We searched the intersection of "premature labor" and "diagnosis," and finally, we searched under "costs and cost analysis and cost-benefit analysis."

Data Collection and Abstraction Process

Grading Article Quality and Overall Strength of Evidence

We first graded the quality of individual articles using a specific scale that considered various aspects of the internal and external validity of each study. It encompassed, among other issues, study design, measurement issues, statistical analyses, and the appropriateness of the conclusions being drawn. To assess the internal validity of a study (i.e., the likelihood that the design and conduct of the study minimize systematic error [bias]), we evaluated the following: recruitment strategy; masking of the treating clinician and patient; inclusion of baseline measurements (e.g., contraction frequency, cervical dilation, effacement, and estimated gestational age); and definition of outcome measures, attrition rates, and confounding variables. We evaluated the adequacy of the methods used and whether the investigators assessed the clinical relevancy of the statistical findings. The quality rating instrument assessed whether all three major categories of outcomes were included in a study: delivery, maternal, and infant. With respect to external validity (i.e., whether the findings of the study can be generalized to populations that did not participate in the study), we determined whether the clinical setting was clearly specified and whether conclusions apply to pregnant women in the United States. The quality scores from each abstractor are included in the evidence tables.

We also rated the quality or strength of the collective evidence on each topic according to a categorical, adjudicated rating developed by the Project Director, Scientific Director, and Clinical Methodologist. The quality of the collective evidence takes into account the design quality of the individual studies and the efficacy or effectiveness of reported outcomes. The ratings are as follows:

We used two designations to grade the overall evidence about harms: "high" if the side effects were life-threatening and their frequency was substantially different from that for alternative treatments or no treatment, and "low" if the side effects were short-term and not life-threatening and their frequency was not substantially different from the frequency for an alternative tocolytic treatment or no treatment.

Development of Evidence Tables

We presented information from individual studies into pairs of evidence tables-one for study design and the other for outcomes. The content of the study design tables covers the following variables (when available from the publications): statement of the research objective, definition of preterm labor used by the authors, patient inclusion/exclusion criteria, patient demographic characteristics, description of the experimental intervention, total number of participants, and number of participants in each arm of the study at its conclusion. The outcomes tables generally include data on delivery, and maternal and infant outcomes. For the biologic marker outcome tables, we present delivery outcomes in terms of sensitivity, specificity, positive predictive value, and negative predictive value.

Supplemental Analyses

In addition to the systematic review of the literature, we conducted meta-analyses focusing on treatment effects of first-line tocolytics by class, maintenance tocolytics by class, antibiotics, and home uterine activity monitoring. The efficacy of treatment was measured in relation to the following outcomes: (1) prolongation of pregnancy, (2) estimated gestational age at delivery, and (3) birth weight. The literature used in the meta-analyses was restricted to RCT studies included in the evidence tables and data concerning women with intact membranes.

Quality of the Literature

The quality of this literature is questionable in many respects, thereby diminishing the confidence with which one can draw conclusions. Among the issues of concern are the definition of preterm labor, the size of the trials, confounding of results because of use of cointerventions, and failure to analyze separately women who have conditions associated with medically indicated preterm births. In addition, we note the general absence (except in a few cases) of the use of survival analytic techniques in preference to the use of dichotomous outcomes (e.g., preterm birth or not) or other categorical outcomes that do not adequately reflect how long a pregnancy "survived" without a preterm birth. Survival analysis allows for the stratification of outcomes by such factors as the gestational age at the onset of preterm labor.

Biologic Markers

Overall, fFN and EVUSD present strong evidence of effectiveness as a diagnostic tool for assessing the risk of preterm birth in women with symptoms of preterm labor. Both tests only moderately were successful in predicting which women with a positive test would deliver before term, but they consistently exhibited strong negative predictive values, thereby identifying women at low risk of preterm birth.

Tocolytics

Antibiotics

Results of our review concerning therapy with antibiotics for treating occult in utero infections associated with preterm labor were mixed. Two RCTs found improvements in all three delivery outcomes of interest-prolongation of pregnancy, mean gestational age at birth, and birth at a particular number of weeks- but all other RCT studies showed mixed results or no significant difference from placebo/control group assignment. Results from survival analyses included in three RCTs were also mixed. Meta-analysis showed the following: a marginally significant increase in length of pregnancy of about 6 days; a marginally significant increase of about 0.60 of a week in gestational age resulting from antibiotic treatment; and a small increase in birth weight that was not statistically significant. The array of agents, routes of administration, and durations of therapy preclude us from making a generalization about the optional antibiotic regimen to achieve these benefits.

Home Uterine Activity Monitoring

Of the four RCTs reviewed, none of the three that controlled for nursing support as an element of the monitoring approach found a significant effect from home uterine activity monitoring; meta-analysis confirmed this "no-effect" conclusion in relation to gestational age at birth and birth weight. A fourth RCT reported that home uterine activity monitoring significantly prolonged delivery to 37 weeks, but the approach in this study included nursing support; the separate effect of monitoring was not analyzed.

Overview and Burden of Disease

Although preterm labor is the most common cause of antenatal hospitalization, the number of pregnancies in which preterm labor occurs is documented inadequately. Only two studies that specifically address the natural history of preterm labor were identified. In a study designed to determine how warning signs of preterm labor relate to eventual diagnosis of preterm labor and preterm birth, Copper, Goldenberg, Davis, et al. ⁷ followed 329 indigent, high-risk patients. In this small cohort, 27 percent of the participants met the research criteria for diagnosis of preterm labor: "contractions occurring between 20 and 36 weeks' gestation at a rate of four in 20 minutes or eight in 1 hour with at least one of the following: ruptured membranes, cervical change over time, dilatation > 2.0 cm, or cervical length < 1.0 cm." Twelve percent of the cohort had spontaneous preterm births. ⁷ Also, West, Yawn, Thorp, et al. ⁸ found that the cumulative incidence of physician diagnosis of preterm labor during the late 1980s and early 1990s was 3.6 to 6.4 percent in a population-based sample of live births in Olmstead County, Minnesota.

Such prospective enumeration of populations of pregnant women is required to accurately measure incidence, the rate of new diagnoses of preterm labor in pregnant women, and prevalence, the proportion of pregnant women receiving treatment for preterm labor, at a given time. Without a denominator from an identified population at risk for developing preterm labor, the majority of existing studies provide "numerator" data that poorly capture the impact of preterm labor on the health care system or on the lives of pregnant women. Each year at least 200,000 infants are born before term after spontaneous preterm labor, ^{5 9} but the number of other pregnancies in the United States affected by the occurrence of preterm labor that do not result in preterm birth has not been projected accurately.

The incidence and burden of preterm birth is comparatively well understood. Conservatively estimated, 11 percent of all live births, or approximately 440,000 births, occur before term each year in the United States. ^{5 9} Preterm birth is responsible for 70 to 85 percent of fetal, neonatal, and infant deaths ¹⁰ and is second only to fetal deformity in causing perinatal morbidity. Up to one-half of the total burden of neurologic disability in children, including cerebral palsy, is attributable to prematurity. ^{1 11} This risk of neonatal mortality and morbidity is related directly to gestational age and weight at birth: the more premature the infant, the greater the risk. Infants born before 32 weeks, 2 percent of births, experience the majority of neonatal deaths and congenital neurologic disorders. ¹² Even late in the preterm period (35 or 36 weeks gestation), when the overall risk of mortality is modest, ¹³ an increased risk of infant mortality, ¹⁴ respiratory distress syndrome, ^{12 13} and impaired intellectual development ¹⁵ can still be discerned.

The economic costs of caring for preterm infants are substantial. Total care costs are, on average, tenfold higher for preterm infants than for normal-weight infants born at term. ¹⁶ Preterm birth accounts for 35 percent of all health care spending on infants and 10 percent of all spending for children. ¹⁶ The ongoing medical, educational, and special needs of children born at low birth weights, the majority after preterm birth, are estimated to exceed $5 billion each year. ¹⁶ These costs reflect only the care of infants born before term. Additional personal and health care costs accrue from the evaluation and care of women who experienced preterm labor.

To narrow the focus to spontaneous preterm labor potentially preceding spontaneous preterm birth, we first consider the major routes to preterm birth. Preterm births can be classified as spontaneous or medically indicated. Spontaneous preterm births are preceded by spontaneous preterm labor or preterm premature rupture of membranes, or both. ^{17 18} Medically indicated preterm births follow a decision to allow or cause delivery for maternal or fetal indications. Preeclampsia and presumed fetal growth restriction cause the majority of medically indicated preterm deliveries. ¹⁹ Spontaneous preterm labor precedes approximately 43 percent of preterm births; preterm premature rupture of membranes, approximately 32 percent; and medical indications for delivery, the remaining 25 percent. ¹⁷

The subtypes of spontaneous preterm birth, preterm labor, and preterm rupture of membranes share many risk factors. ^{20 21} In fact, 15 to 20 percent of women successfully treated for preterm labor subsequently experience preterm premature rupture of membranes. ^{21 22} Although this interrelationship is clear, the management of these conditions is divergent. This report focuses exclusively on the management of preterm labor.

The Epidemiology of Preterm Birth

The epidemiologic literature focuses on preterm birth, rather than preterm labor, as an outcome. Because the majority of women presenting for care with preterm labor signs or symptoms do not have the more serious outcome of giving birth prematurely, experts believe that examining the association of factors with spontaneous preterm birth itself (i.e., a "hard" outcome) reduces misclassification bias and yields more precise estimates of the importance of specific maternal characteristics and exposures. Research on determinants of preterm birth have three dominant foci:

• Description of risk factors for individuals and populations to guide prenatal surveillance and intervention programs.

• Identification of behaviors and conditions that are amenable to change or treatment.

• Description of physiologic processes that precede preterm birth and may elucidate etiologic pathways or serve as biologic markers of impending birth.

The last category includes uterine activity, cervical change, and biologic markers such as fFN and E3, which will be reviewed later.

Risk Factors for Preterm Birth

The first category of risk factors consists predominantly of factors for which the mechanism of risk modification is unknown or difficult to separate definitively from socioeconomic, behavioral, or nutritional factors. Many of these "exposures" are relatively fixed (i.e., not amenable to modification after a woman becomes pregnant). The most robust and consistent factor is uniquely informative: women who have had a prior spontaneous preterm birth have a threefold increase in risk of preterm birth in a subsequent pregnancy compared with women who have had only term deliveries. ²³ This risk increases with the number of previous preterm births, and the outcome of the most recent pregnancy has the greatest influence. ²⁴ Depending on the population, women with a history of preterm birth deliver before term in 15 to 80 percent of subsequent pregnancies. ²³ Presumably this reflects the constancy of unobserved, underlying factors.

In the United States, black race is a risk factor. Adjusting for known confounders, including income and educational status, being black confers a relative risk of 2.0 to 3.5 of having a preterm baby when compared with being white. ^{13 25} This divergence of risk is most pronounced at the lowest gestational ages (i.e., below 27 weeks). ²⁶ The risk of other minorities in the United States is characterized too poorly to generalize.

Additional reported factors associated with increased risk include both very young and advanced maternal age, single marital status, low socioeconomic status, and some of its isolated correlates such as low income, poor access to and quality of care, and limited educational attainment. Poor nutritional status and low pre-pregnancy body weight also are related to increased risk. ^{27 28} These last two examples underscore the overlap between descriptive markers of risk and the possibility that deprivation plays a role in some cases, whereas knowledge and intentional behavior factors, such as nutritional habits, play a role in others.

Problems in the Diagnosis of Preterm Labor

Labor at term occurs when uterine contractions lead to progressive cervical dilation and effacement. By extension, the hallmarks of preterm labor are uterine activity and cervical change; however, uniformly accepted standards for defining preterm labor do not exist. Therefore, diagnosis of preterm labor invokes an essential circularity: this patient has the kind of uterine activity that may lead to preterm birth if not treated. Thus, intervention precludes definitive diagnosis. As a surrogate, operational definitions of preterm labor seek to quantify how many contractions over what period of time are occurring (e.g., more than four contractions in 20 minutes or five to eight contractions in 1 hour) they also often require documentation of change in dilation or effacement of the cervix.

In practice, clinicians weigh many highly individual factors not captured in formal definitions: Are the contractions painful? Does the mother recognize the contractions? Does she report other symptoms of labor such as back pain, pelvic pressure, vaginal bleeding, or vaginal discharge? What is her individual risk profile? Has she had prior preterm births? Are other processes at work that may be provoking uterine activity, such as a urinary tract infection, substance use, or placental compromise, including abruption (premature separation of the placenta)?

Once other explanations are excluded, two factors drive intervention for preterm labor: (1) estimates of the probability of progressive labor and (2) the degree of immaturity of the fetus. Higher probability and younger gestational age may lead to intervention based on less strict criteria than would otherwise be used. For example, allowing a mother at 28 weeks' gestation to continue to contract with the goal of documenting cervical change simply to meet diagnostic criteria is problematic at best. Because (in theory) achieving uterine quiescence is easier earlier in the cascade of labor events, early intervention is believed to be desirable.

Symptoms, including regular contractions before term, are common; hence, their presence is not highly predictive for discerning who will have progressive cervical change or go on to have a preterm delivery. In a study of 763 women who made unscheduled visits for symptoms suggestive of preterm labor, only 18 percent delivered before 37 weeks; moreover, only 3 percent delivered within the next 2 weeks. ⁴⁶ Early detection of symptoms ^{47 48 49 50 51} and actual measurement of uterine contractions with home monitors ^{49 50 51 52} have failed to lower preterm delivery rates.

These findings suggest that women with contractions and symptoms that will lead to preterm birth are indistinguishable from the larger group who has incidental uterine activity, making it difficult to target surveillance and treatment efficiently. As the studies presented in the evidence tables show, the great majority of women diagnosed with preterm labor will not deliver before term.

Physiology of Preterm Labor

The cervical changes caused by labor (either term or preterm) reflect the culmination of a complex set of molecular changes in cervical connective tissue clinically known as "ripening" or "softening." ^{53 54} Although the sequence of events leading to cervical change, especially the triggers, is less clear for preterm labor than for term labor, the basic physiology is considered to be similar. The uterine wall consists primarily of smooth muscle or myometrium. Contractions result from shortening of the muscle fibers provoked by a stimulus, such as oxytocin or prostaglandins. At the cellular level, regulated release of intracellular calcium is required to set in motion the cascade that allows myosin and actin filaments within the muscle cell to generate force by shortening. To convert from the irregular, disorganized contractions common throughout pregnancy to coordinated contractions, the cells in the uterus create new channels called gap junctions to facilitate the flow of calcium and other intercellular signals. This process results in contractions in which the upper segment of the uterus provides downward expulsive force, pressing the presenting part of the infant against the cervix.

In addition to the mechanical force of uterine contractions, structural changes take place in the cervical tissue itself. White blood cells infiltrate the cervix and release various enzymes (proteases, collagenases, and elastases) that weaken the collagen and protein matrix of the cervix. This weakening allows the stretching and distortion of the cervix required for dilation and effacement.

Many types of stimuli may contribute to the initiation and perpetuation of contractions and to recruitment of inflammatory factors required for cervical change, which potentially provides multiple potential points for therapeutic intervention. In the discussion of management, we review how biologic markers can be used to detect important physiologic changes associated with preterm labor, the mechanism of action for each category of pharmacologic agents, potential harms associated with the use of tocolytics, and the rationale for use of home uterine activity monitoring.

Treatment Mechanisms

Clinicians frequently use a combination of treatments in caring for women with preterm labor. As noted earlier, the chief focus of this report centers on tocolytics and antibiotics. We do not review combinations of treatment that might be employed because the number of different combinations of pharmaceutical and nonpharmaceutical interventions is very large. In addition, frequent cointerventions that are not reviewed specifically in this report are bed rest and hydration.

Biologic Markers to Assess Risk in Women with Preterm Labor Symptoms

Over the past decade, two biologic markers-fFN in cervicovaginal secretions and EVUSD of the cervix have been investigated extensively. When used as tests to assess women with symptoms of preterm labor, these biologic markers have been shown to increase the precision by which clinicians can assess the likelihood of preterm birth. Our third marker, E3, has not been evaluated in a study restricted to women in preterm labor.

The value of these biologic markers lies in their ability to add predictive value (either positive or negative) to the clinician's diagnostic abilities. In other words, their purpose is to predict more accurately which patients presenting with signs and symptoms of preterm labor are likely to deliver prematurely. If biologic markers have strong positive predictive value, a positive test would indicate which women are more likely to deliver. The converse is equally valuable: if biologic markers have strong negative predictive value, a negative test predicts which women are likely not to progress to delivery and, thereby, enable women and their clinicians to avoid unnecessary treatment.

In clinical practice, when any of the three biologic markers is used in a group of women with preterm labor symptoms to predict the likelihood of spontaneous preterm birth, fFN, ^{46 51} EVUSD, ^{55 56} and E3 ⁴⁰ demonstrate moderate positive predictive value (20 to 40 percent) and good negative predictive value (85 to 95 percent). In logistic regression models, the presence or absence of these biologic markers explains significantly more of the variability in preterm birth in women with preterm labor symptoms than do traditional clinical assessment tools, such as cervical dilation by digital assessment or uterine contraction frequency.

Mechanisms of Tocolytic Drugs

Tocolytic drugs decrease the uterine muscle's ability to contract; the goal is to diminish the force and frequency of contractions and preferably bring them to a halt. Many agents have been used, including ethanol, magnesium sulfate, calcium channel blockers, oxytocin antagonists, NSAIDs, and beta-mimetic agonists. Tocolytics can be administered in two ways: (1) those used at initial presentation grouped under the rubric "first-line therapy", and (2) those given to maintain uterine quiescence after first-line tocolysis are labeled "maintenance therapy." The same classes of drugs are used in both therapeutic approaches, although parenteral administration predominates in first-line treatment and oral administration is the delivery route of choice for maintenance therapy.

Mechanisms of Antibiotics for Treatment of Preterm Labor

Women who present with preterm labor symptoms often have occult infections of their upper genital tract that can be detected by amniocentesis and subsequent culture of their amniotic fluid. ^{84 85} Women in the group who have occult infections are more likely to experience preterm birth than women without evidence of infection. ⁸⁴ This association of infections of the upper genital tract and preterm birth in women with preterm labor symptoms provides the rationale for using antibiotics in the management of women with preterm labor. Antibiotic administration may cure occult infections, thereby lowering the risk of spontaneous preterm birth. ⁸⁴

Organisms cultured from the amniotic fluid or placental-decidual interface of women who give birth before term include enteric gram negatives, gram-positive cocci, anaerobic cocci, and mycoplasma species-a mix of genital flora. Given the variety of potential pathogenic organisms, clinicians have empirically chosen broad spectrum antibiotics in an attempt to eradicate these organisms from the upper genital tract in women with preterm labor. Antibiotics used include second- and third-generation cephalosporins, beta-lactam antibiotics with beta-lactamase inhibitors, synthetic penicillins, metronidazole, and combinations of agents such as "triple coverage" (beta-lactam antibiotics, aminoglycoside, and clindamycin).

It is important to distinguish between antibiotics intended to treat occult in utero infection and those intended to prevent early-onset group B streptococcal sepsis in newborns. Preterm infants are particularly susceptible to acquiring group B streptococcal infections at birth from their mothers whose vaginas may be colonized with this organism and may be developing sepsis. The goal of early-onset sepsis prevention can be readily accomplished with single-agent beta-lactam antibiotic treatment during labor.

Home Uterine Activity Monitoring

The goal of home uterine activity monitoring is to detect early labor symptoms in women deemed to be at high risk of preterm delivery, with the assumption that early detection provides an opportunity for successful treatment. Thus, home monitoring proponents do not claim that monitoring in and of itself can prevent preterm birth, but rather that the technology helps clinicians target preventive therapies earlier for women with preterm labor symptoms who are at risk of preterm birth and who may not recognize their early symptoms. This report focuses exclusively on the use of this technology in the management of women who have been diagnosed with preterm labor and released home for monitoring, generally after some treatment.

Home uterine activity monitoring uses a tocodynamometer to record uterine contractions in conjunction with daily telephone calls from a health care provider, usually a nurse, to offer patient support and advice. ^{86 87} The biologic hypothesis underlying the use of the device is that women will have an identifiable increase in uterine contractions before preterm delivery (i.e., a "crescendo" effect in contractions) and that many women are unable to recognize this prodromal uterine activity. ^{88 89} The premise on which the usefulness of home uterine activity monitoring rests is that by detecting early contractions, health care providers can treat preterm labor and prevent or decrease the likelihood of preterm birth. ⁸⁷

Patients attach the external tocodynamometer over the uterus twice each day for 30 to 60 minutes while they are recumbent. ⁹⁰ Digitized tracings from the tocodynamometer are transmitted over telephone lines for immediate interpretation by a health care provider. The Food and Drug Administration (FDA) has approved a portable tocodynamometer for use in the homes of women with high-risk pregnancies, whose fetuses are of more than 24 weeks' gestation, and who have a history of preterm birth. ⁹¹ The device has been shown to record uterine contractions with precision in both hospitalized and ambulatory gravidas. ⁹²

Final Key Questions

The first question concerns the effectiveness of using three biologic markers in assessing the risk of preterm delivery: fFN found in cervicovaginal secretions, EVUSD of the cervix, and E3. The second and third questions relate to pharmacotherapies for arresting the condition: tocolytics to stop uterine contractions, and antibiotics for treating possible underlying infections. The tocolytic review separates studies based on whether the aim was treatment during the episode of preterm labor and uterine activity (first-line use) or promotion of continued quiescence of the uterus (maintenance use). By contrast, although various antibiotics are used in treating preterm labor, they are discussed as one group. The fourth question concerns using home uterine activity monitoring for detecting early labor uterine contractions.

Our specific key questions are as follows:

• 1

  What are the appropriate criteria for the diagnosis of preterm labor? Relatedly, how much positive or negative predictive value does the use of biologic markers add to clinical opinion in diagnosing preterm labor?

• 2

  What is the efficacy and/or effectiveness of tocolytics in managing preterm labor?

• 3

  What is the efficacy and/or effectiveness of antibiotics in managing preterm labor?

• 4

  What is the efficacy of home uterine activity monitoring in decreasing adverse maternal and neonatal outcomes in women who have experienced an episode of preterm labor in the current pregnancy?

In the context of managing preterm labor, the efficacy or effectiveness of treatment refers to three major categories of outcomes: delivery, maternal morbidities, and infant health. From a clinical perspective, the outcomes of greatest concern are infant morbidities and mortality associated with preterm birth; however, because these are rare events and difficult to quantify except in large samples, and because prolonging the pregnancy is believed to be correlated with improved infant outcomes, studies also include birth outcomes as intermediate or proxy measures of efficacy.

Our search included both efficacy results from RCTs and effectiveness outcomes from observational research to ensure the greatest breadth of understanding of the research in this area. Efficacy results are obtained from studies where health care is delivered under relatively ideal conditions, such as controlling for confounding interventions. In contrast, effectiveness results are gathered under more normal conditions. The potential harms associated with treatment are presented in relation to tocolytic therapy. Our review of the literature concerning diagnostic testing includes only cohort studies because those were the only ones available. Our review of the use of home uterine activity monitoring includes only English language efficacy studies, because our search was limited based on time and resource considerations.

Causal Pathways

1

2

3

4

At this point, the clinician faces two challenges. The first challenge is to decide how to diagnose the preterm labor itself, which may include assessing the likelihood of preterm birth using biologic markers. The second is to determine whether the preterm labor stems from a treatable cause (e.g., an infection) that might be addressed through managing the condition or whether it cannot be attributed to a likely treatable cause and so should be limited to managing the symptoms (e.g., through the use of tocolytics).

Figure 1

2

3

Figure 2

Figure 3

These causal pathways can become quite complicated as the possible options and outcomes multiply. In our presentation, the treatment pathways are somewhat simplified because they do not reflect the possibility that clinicians might use more than one type of tocolytic when the first agent that is tried fails or the possibility that different practitioners might invoke the pharmacologic agents at different times in the course of an initial or subsequent preterm labor episode. Similarly, the "diagnosis" part of the causal pathway does not reflect the fact that different practitioners use different clinical criteria for diagnosing preterm labor, including different tests or using the tests in a different order if biologic testing is selected.

Figure 4

Inclusion and Exclusion Criteria for the Literature Search

Table 2. Management of preterm labor: inclusion/exclusion criteria (more...)

Table 2. Management of preterm labor: inclusion/exclusion criteria

Category	Criteria
Study population	Humans Pregnant females
Condition	Signs and symptoms of preterm labor Exclude if all research subjects experience preterm premature rupture of membranes, medically indicated preterm birth, or multiple gestation
Interventions	Biologic markers (restricted to E3, EVUSD, and fFN), antibiotics, tocolytics, and HUAM
Study settings	Inpatient and outpatient settings, including patient homes
Outcomes	Preterm birth must be measured
Time period	1966 and later HUAM: 1980 and later Depends on date of approval for pharmaceutical products or biologics
Geographic site of study	Exclude locations based on language of publication
Publication languages	English, French, and German HUAM: English only
Admissible evidence	Study design: RCT-double- and single-blinded; non-RCT-prospective and retrospective cohort studies, case control Study design HUAM: RCT only Other designs: meta-analyses, review articles for reference list searches, cost-effectiveness studies Sample size: >40 subjects

Search Terms

The literature review specialist in cooperation with the study director and the scientific director developed the MeSH search terms used in the analysis. Searches in the National Library of Medicine (NLM) databases (e.g., MEDLINE) were limited by the subject heading: explode premature labor. This phrase is defined in the NLM databases as the onset of labor before term but after the fetus has become viable, usually sometime during estimated gestational ages of 29 through 38 weeks. Both preterm labor and premature delivery default to this subject heading. Preterm premature rupture of membrane, which is excluded from this analysis, is not included under this heading.

To ensure inclusion of relevant study designs, we searched within the MeSH tree under epidemiologic study characteristics, which are defined as types and formulations of studies used in epidemiologic and clinical research. Relevant MeSH key words included under this general topic and searched concurrently through the "exploded" term are clinical trials (Phase I-IV, controlled clinical trials, multicenter studies, and randomized controlled trials), analytic studies, epidemiologic studies, retrospective studies, prospective studies, longitudinal studies, followup studies, feasibility studies, pilot studies, sampling studies, intervention studies, cross-over studies, cohort studies, case control studies, and cross-sectional studies.

In addition, we searched the intersection of the terms premature labor and diagnosis to capture review articles. Bibliographies in review articles were examined as a double check to our electronic literature search to ensure that all relevant articles were included in the study. We searched for economic and cost analyses using the search terms costs and cost analysis and cost-benefit analysis.

We also searched for specific therapies. First, we searched on the relevant diagnostic tests within the term biologic markers, where the scope is defined as measurable and quantifiable biologic parameters that serve as indexes for health- and physiology-related assessments. Specific biologic markers included in the study are salivary estriol (estriol, saliva), endovaginal ultrasound, and fetal fibronectin (fibronectins). Second, we searched exploded antibiotics, which include ampicillin, amoxicillin, metronidazole, erythromycin, ceftizoxime, ampicillin-salbactam. Finally, we exploded tocolytic agents, which include ritodrine, terbutaline, isoxsuprine, magnesium sulfate, nifedipine, nicardipine, indomethacin, fenoterol, ethanol, nylidrin, and salbutamol.

Our primary literature search was conducted through MEDLINE. The MEDLINE database is widely recognized as the premier source for bibliographic coverage of biomedical literature. MEDLINE encompasses information from Index Medicus, Index to Dental Literature, andInternational Nursing, as well as other sources of coverage in the area of communication disorders, population biology, and reproductive biology.

In addition, we searched EMBASE (the electronic version of Excerpta Medica) because it provides a more complete coverage of the literature published in Europe, Latin America, and Asia. Apart from being a database that AHRQ has specified as being of interest, the EMBASE search is important because, historically, drugs have been approved for clinical application earlier in Europe than in the United States. Clinical findings regarding a drug's efficacy may have been published in Australia or Canada before reaching the U.S. market. We used a similar search strategy and terms for EMBASE as were employed for MEDLINE.

Additional databases searched include: (1) the Cochrane Collaboration database to examine relevant meta-analyses and systematic reviews closely related to our topic; (2) the related York database of the National Health Science Center for Reviews and Dissemination; (3)International Pharmaceutical Abstracts, a database devoted exclusively to pharmaceuticals and produced by the American Health-System Pharmacist Association; (4) the Health Economic Evaluations Database for literature concerning cost and economic analyses; (5)Genderwatch, a search engine that covers international journals, magazines, newsletters, regional publications, special reports, and conference proceedings devoted to women's and gender issues; and (6) the Population Index, a reference tool that focuses on the world's population literature. Of all these databases, Genderwatch and the Population Index uncovered no matches with our topics.

Gray Literature Search

In its first task order (Pharmacotherapies of Alcohol Dependence), the RTI-UNC EPC outlined a search strategy for uncovering the so-called "gray literature" (i.e., potentially useful literature and documents that are to be found in outlets other than peer-reviewed journals). For this evidence report on the management of preterm labor, the RTI-UNC EPC has pursued a similar strategy.

Ideally, gray literature searches include reviews of government documents and monographs; industry reports and studies; unpublished studies and works in progress; proceedings and technical bulletins; and FDA Medical Officer's Reviews of original or extended (change of label) New Drug Approval applications, including advisory committee minutes for drugs, and similar reviews of Pre-Market Approval applications for devices. In reality, constraints on time and resources necessitate strategic reductions in the scope of the gray literature search.

We made two important decisions in restricting our gray literature work. Our earliest strategic decision, in consultation with the task order officer, was not to conduct gray literature searches for home uterine activity monitoring devices because of time and resource limitations. We did search for and review Pre-Market Approval documents archived at the FDA Web site (http://www.fda.gov) for home uterine activity monitoring devices, the E3 test, and the fFN test. We identified and reviewed research information offered by the manufacturer of the E3 test SalEst ^® (http://www.biex.com), and fFN test (http://www.adeza.com). Both manufacturers provided information such as clinical monographs; however, with the exception of one brief four-person case study that appeared to have been written expressly for the manufacturer, all research to which they referred is available already as journal articles or abstracts.

Our second significant strategic decision was sharply to restrict our search for gray literature on antibiotics, for several reasons that reflect an initial investment of study resources. First, thousands of antibiotics are marketed in the United States alone. For example, just one of the earliest antibiotics, erythromycin, is marketed for human use by almost 20 different pharmaceutical firms; most of them market more than one version (oral and intravenous and/or intramuscular). Multiplying all possible antibiotics by all firms marketing them would create a search the size of which no EPC could undertake.

Second, and more important, in no case is an antibiotic specifically labeled for use in the treatment of preterm labor. Only two antibiotics are approved currently by the FDA for the treatment of bacterial vaginosis in pregnancy, metronidazole and clindamycin (Anti-Infective Drugs Advisory Committee Meeting, 64th Meeting, Issue: Guidance documents on developing antimicrobial drugs: General considerations and individual indications. Wednesday, July 29, 1998.) The RTI-UNC EPC's experience with the gray literature search for alcohol suggests that pharmaceutical firms do not engage in research into off-label uses unless they are preparing an application for a new or changed labeling approval, and they will not otherwise comment on off-label uses for their drugs. Thus, although antibiotics are prescribed routinely in the management of preterm labor, all such prescriptions are "off label."

Additionally, most antibiotics in use, except for the latest generation of cephalosporins, could not have been investigated in pregnant human subjects because the FDA explicitly prohibited the use of "women of childbearing potential" from clinical trials beginning in 1977, and an informal practice of exclusion had been the rule before then. Only in 1993, did the FDA issue a new guideline emphasizing the need for drug research using women as subjects, and the question of conducting drug research in pregnant women is still exceedingly controversial. Most antibiotics, like most drugs, are labeled Pregnancy Category C (harms to fetal development are shown in animal studies but no adequate, well-controlled human studies exist); an antibiotic example includes clarithromycin, which also must carry a Boxed Warning strongly advising against use during pregnancy. Other common antibiotics, such as the aminoglycosides, including streptomycin, fall under Pregnancy Category D (there is evidence of harm to the fetus from human studies, and patients should be advised of the potential hazards). (See, for example, Public Hearing, Content and Format of Labeling for Human Prescription Drugs, Part 15 - Pregnancy Labeling Categories, Friday, September 12, 1997, Bethesda, MD, which we have reviewed as a part of the gray literature search.)

In other words, no pharmaceutical company could have tested its new antibiotic in a sample of pregnant women, even had it wished to do so. Thus, it is neither reasonable nor efficient to devote scarce resources to such a search. The one exception may be in the case of the latest generation of cephalosporins, which are classed in Pregnancy Category B (no evidence of harm from animal studies, but no adequate, well-controlled human studies have been conducted, and the drug should be used only if clearly indicated). Anecdotally, manufacturers have sometimes expressed off-the-record interest in research into cephalosporin use during pregnancy, but we have found no evidence that any manufacturer has petitioned the FDA recently for a change in labeling for use in pregnant women.

Additional strategic decisions concerned tocolytics. The gray literature search on tocolytics appears to reveal that discrepancies arise not from differences between published and unpublished literature but from differing emphases on and interpretations of outcomes and, perhaps, differing ideological approaches to the treatment of women in preterm labor.

This task order includes an evaluation of the tocolytic drug atosiban, an oxytocin-receptor antagonist, approval for which was denied by the FDA's Advisory Committee for Reproductive Health Drugs on April 20, 1998. Despite the failure of the R.W. Johnson Pharmaceutical Research Institute (Johnson & Johnson) to gain approval for atosiban, we have nonetheless located an unpublished manuscript describing the multicenter RCT of atosiban (forthcoming in the American Journal of Obstetrics and Gynecology). We have also obtained and reviewed the minutes of the 1998 Reproductive Health Drugs Advisory Committee meeting. From these minutes, it seems clear that the committee's failure to approve atosiban resulted not from its doubts about the drug's efficacy to halt labor as the endpoint, but rather from its concern as to whether the endpoint of interest ought to be improved perinatal outcome. The Advisory Committee was not satisfied that the manufacturer had collected data addressing such concerns, but it was not in agreement about the kind of perinatal outcomes trial it wanted to see. The manufacturer, Johnson & Johnson, maintains that it proceeded correctly in the initial application. Thus, for the present in the United States, there will be no further efforts to seek approval for atosiban, although it is expected to be approved very soon in 15 European Nations, and the scientific basis for its approval will be published shortly on the European Agency for the Evaluation of Medicinal Products Web site (personal communication, between Sue Tolleson-Rinehart and Johnson & Johnson official, September 23, 1999).

Ritodrine and terbutaline are both beta-mimetic agents. Terbutaline is approved only for treatment of asthma, but it has become the most frequently used tocolytic in the United States. Ritodrine, under the brand name Yutopar ^® , was the only drug to receive FDA approval as a tocolytic. The drug, approved in 1980, was available in both oral and intravenous forms until a 1992 meeting of the Fertility and Maternal Drug Advisory Committee, resulting in an FDA request for more testing of high doses of oral ritodrine for tocolytic maintenance. In response, the manufacturer removed oral ritodrine from the market. Controversy still revolves around the clinical information presented to the Drug Advisory Committee at that time from the Canadian Ritodrine Trial; the results of the trial and criticisms of it are in the published literature. ⁹⁶

Ritodrine in intravenous form remained available as a tocolytic treatment until its two manufacturers, Abbott and AstraZeneca, stopped manufacture. Abbott discontinued the intravenous form before December 1997; AstraZeneca stopped manufacturing it on September 1, 1998, but the last expiration date of previously shipped product was August 1999, meaning that ritodrine could have remained in use until that time. Terbutaline has been employed in an off-label use for tocolysis for the same period and in fact, has surpassed ritodrine in use. In a Fertility and Maternal Drug Advisory Committee meeting in 1993, committee members recommended that the FDA solicit a specific application for change of label to include tocolysis for intravenous terbutaline; also discussed was the possibility of a literature-only application because the literature on terbutaline as a tocolytic was already voluminous.

At roughly the same time, the use of terbutaline in a subcutaneous infusion pump delivery system was increasing. Its advocates (including pregnant women) have generally argued that the terbutaline infusion pump effectively maintains tocolysis at much lower doses than is possible with other delivery systems and that it permits pregnant women to continue tocolysis maintenance at home, rather than in the hospital.

In 1996, the National Women's Health Network presented a Citizen's Petition to the FDA, requesting prohibitory action on the terbutaline subcutaneous infusion pump allegedly because of two maternal deaths that could be attributed to terbutaline use (Docket No. 96P-0258/CP1). On November 13, 1997, Associate Commissioner for Health Affairs Stuart Nightingale issued a "Dear Colleague" letter that said, in part:

Based on information available to the Agency, as well as a review of the published literature, it is clear that the demonstrated value of tocolytics in general is limited to an initial, brief period of treatment, probably no more than 48-72 hours. No benefit from prolonged treatment has been documented. In addition, the safety of long-term subcutaneous administration of terbutaline sulfate, especially on an outpatient basis, has not been adequately addressed.

Published reports on the safety of this use are seriously hampered by methodologic inadequacies. It appears that women receiving continuous subcutaneous terbutaline sulfate infusions experience side-effects and complications similar to those experienced by women receiving terbutaline and other beta-sympathomimetics intravenously. Complications include chest pain, tachycardia, dyspnea, and pulmonary edema. At least one maternal death occurred during outpatient use of a continuous infusion of terbutaline sulfate by subcutaneous pump. The impact of long-term use on maternal glucose metabolism and the risks of prolonged exposure of the fetus are largely unknown.

The "Dear Colleague" letter provoked considerable response from clinicians, researchers, and patient advocates; it apparently induced Matria, Inc., the marketer of the terbutaline pump system, to bring suit against the National Women's Health Network. Advocacy also included a petition from obstetricians to the FDA (Docket No. 98P-0218/CP) to approve subcutaneous terbutaline as a tocolytic treatment. The RTI-UNC EPC has no evidence to date that the FDA has responded to the substance of the petition.

This review for tocolytics suggests that its gray literature actually refers to or summarizes the published literature for policy purposes, rather than presenting unpublished literature. The remaining tocolytic in this evidence report, magnesium sulfate, is not subject to FDA approval as a drug.

As we discussed at the beginning of this section on gray literature, we also examined government documents and manufacturers' information associated with approval and use of assays for E3 and fFN. The first fFN assay was approved on September 21, 1995; a subsequent "Rapid System" by the same manufacturer was approved on August 14, 1998. An E3 assay was approved on April 29, 1998. The available information suggests that most (if not all) of the data relied on by the manufacturers now can be found in published articles and abstracts. Similarly, Pre-Market Approval information for ultrasound devices, including a portable device designed by General Electric expressly for use in obstetrics and gynecology, is straightforward and presents no unusual data. It may be worth noting, however, that the FDA suggests in its guidelines for labeling home uterine activity monitoring devices that they include the following precaution: "No widely accepted controlled studies have been conducted that show that this device is effective at the early detection of preterm labor other than in patients with a previous preterm delivery."

Finally, our efforts to identify unpublished work or other work in progress produced only the atosiban manuscript cited earlier in this chapter.

Basic Approach to Article Selection

Once articles were identified through the electronic database search or review article bibliographies, we examined abstracts of the articles to determine whether studies did, in fact, meet our criteria. Two reviewers (one clinician and one methodologist) reviewed abstracts independently for inclusion or exclusion. If both reviewers agreed, an article was retained (or excluded). If the two reviewers disagreed (one indicating eligible and the other judging ineligible), the scientific director made the final decision.

To ensure that all appropriate articles were included and no significant ones were missed, the scientific director rereviewed a 20 percent sample of abstracts determined to be eligible for inclusion as well as all those determined to be ineligible. Generally speaking, for all abstracts, we erred on the side of inclusion rather than exclusion. When abstracts were not available from the electronic search or when there was a question of relevance, the project director or scientific director subjected the full articles to additional scrutiny. In the end, we excluded a study once the full article was examined if, upon this closer scrutiny, we found that it did not conform to the inclusion/exclusion criteria.

Figure 5. Article eligibility screen

Screen (If any response is "No," then stop. If any response is "Can't Tell," bring it to the attention of the research coordinator.)

Preterm birth an outcome?	Yes	No	Can't Tell
40 or more subjects at completion?	Yes	No	Can't Tell
Intervention used as part of treating subjects with signs/symptoms of preterm labor (not a preventive intervention used among asymptomatic patients)?	Yes	No	Can't Tell

Productivity and Results of the Literature Search and Retrieval Process

Table 3. Search results by topic

Table 3. Search results by topic

Topic	Retrieved	Retained
Biologic markers	106	17
Tocolytics Benefits Harms	256 290	65 85
Antibiotics	107	15
Home uterine activity monitoring	37	4
Costs and cost analyses	1	0
Totala	797	95

a

Totals reflect a double count of some articles retrieved (retained) for both the tocolytics benefits and harms reviews.

Table 4. Biologic markers search strategy

Table 4. Biologic markers search strategy

Biologic Markers	Results
MEDLINE (Index Medicus)
1. Explode labor, premature, and explode epidemiologic studies, and limit to humans	1,039
2. Explode biologic markers; or estriol; or fibronectins; or saliva; or ultrasonography, or endovaginal ultrasound	165,902
3. 1 and 2	35
EMBASE (Excerpta Medica)
1. Premature labor; or premature delivery	3,764
2. Biologic with marker	12,621
3. 1 and 2	99
International Pharmaceutical Abstracts
1. Premature labor; or premature delivery	162
2. Biologic markers; or estriol; or fibronectin	85
3. 1 and 2	3
Total unduplicated count from bibliographic databases	99
1. Articles found from bibliographies of review articles	7
Total possible articles	106
Reductions Based On Abstract and Article Reviews
1. Opinion or editorial	5
2. Study with <40 subjects	6
3. Review article	40
4. Wrong language	2
5. Not preterm labor	18
6. Wrong biologic marker	16
7. Twin study	3
8. Inclusion status not yet obtained	0
Total articles excluded	90
Total articles reviewed	16

Table 5. Tocolytic search strategy

Table 5. Tocolytic search strategy

Tocolytics	Results
MEDLINE (Index Medicus)
1. Explode labor, premature, and explode epidemiologic studies, and limit to humans	1,039
2. Explode tocolytic agents
3. 1 and 2	145
4. Explode labor, premature and, explode nitroglycerin	11
EMBASE (Excerpta Medica)
1. Premature labor; or premature delivery	3,764
2. Tocolytic	438
3. 1 and 2	245
4. Nitroglycerin	2
International Pharmaceutical Abstracts
1. Premature labor; or premature delivery	162
2. Tocolytic	138
3. 1 and 2	32
Total unduplicated count from bibliographic databases	251
1. Articles found from bibliographies of review articles	5
Total possible articles	256
Reductions Based On Abstract and Article Reviews
1. Opinion or editorial	7
2. Study with <40 subjects	45
3. Review articles	39
4. Wrong language	6
5. Not preterm labor	71
6. Not human subjects	1
7. Twins	5
8. Wrong drug	5
9. Unavailable a	12
Total articles excluded	191
Total articles reviewed	65

a

Unavailable: 1 German, 2 French, and 9 English articles were not received in sufficient time for review.

Table 6. Harms related to tocolytic use search strategy (more...)

Table 6. Harms related to tocolytic use search strategy

Harms Related to Tocolytic Use	Results
MEDLINE (Index Medicus)
1. Explode labor, premature	7,001
2. Limit 1 to human and English language	4,299
3. Explode tocolytic agents	43,746
4. Atosiban	24
5. Explode nitroglycerin; or explode nitrous oxide	16,112
6. 3 or 4 or 5	59,383
7. 2 and 6	880
8. Explode morbidity; or explode mortality; or explode heart failure, congestive	217,717
9. Explode arrhythmia; or explode liver diseases; or explode anxiety	310,250
10. Explode glucose intolerance; or explode pregnancy complications; or explode cerebral hemorrhage	196,944
11. Explode enterocolitis	4,553
12. Explode heart hypertrophy; or explode bradycardia; or explode cerebral palsy	33,993
13. Adverse effects	21,287
14. 8; or 9; or 10; or 11; or 12; or 13	719,408
15. 7 and 14	880
16. Limit 15 to clinical trial	149
17. Limit 15 to randomized controlled trial	90
18. Limit 15 to review; or review literature; or review of reported cases; or review, academic; or review, multicase; or review, tutorial	87
19. Limit 15 to meta-analysis	1
20. 16; or 17; or 18; or 19	232
21. 15 and 20	200
22. Case studies	2,799
23. Explode cohort studies	335,464
24. 22 or 23	338,143
25. 15 and 24	88
26. 21 or 25	270
EMBASE (Exerpta Medica)
1. Premature labor; or premature delivery	3,764
2. Tocolytic	438
3. Complications; or adverse effects; or harms; or mortality	182,906
4. 1 and 2 and 3	5
Total unduplicated count from bibliographic databases	290
Total articles excluded	205
Total articles abstracted	85

Table 7. Antibiotic search strategy

Table 7. Antibiotic search strategy

Antibiotics	Results
MEDLINE (Index Medicus)
1. Explode labor, premature, and explode epidemiologic study characteristics, and limit to humans	1,039
2. Explode antibiotics	306,659
3. 1 and 2	20
EMBASE (Excerpta Medica)
1. Premature labor; or premature delivery	3,764
2. Antibiotics	79,377
3. 1 and 2	105
International Pharmaceutical Abstracts
1. Premature labor; or premature delivery	162
2. Antibiotics	10,411
3. 1 and 2	3
Total unduplicated count from bibliographic databases	105
1. Articles found from bibliographies of review articles	2
Total possible articles	107
Reductions Based On Abstract and Article Reviews
1. Opinion or editorial	4
2. Study with <40 subjects	6
3. Review articles	28
4. Wrong language	3
5. Not preterm labor	33
6. Unavailable a	2
7. Antibiotics not primary focus	16
Total articles excluded	92
Total articles reviewed	15

a

Unavailable: 2 German articles were not received in sufficient time for review.

Table 8. Home uterine activity monitoring search strategy (more...)

Table 8. Home uterine activity monitoring search strategy

Home Uterine Activity Monitoring	Results
MEDLINE (Index Medicus)
1. Explode labor, premature; or explode uterine contractions; or explode pregnancy	399,320
2. Explode uterine monitoring; or explode monitoring, physiologic	25,688
3. Explode home care services	17,002
4. Fetal monitoring; or tocodynamometer	4,728
5. 2 or 4	30,315
6. 1 and 3 and 5	54
7. Home uterine activity monitoring	51
8. 6 or 7	84
9. Limit 8 to human, and English language, and year = 1980 to 1999	75
10. Limit 9 to RCTs	30
Total unduplicated count from bibliographic databases	30
1. Articles found through TEAG and scientific director recommendation	7
Total possible articles	37
Reductions Based on Abstract and Article Reviews
1. Not an RCT	2
2. Review article	6
3. Wrong study population	15
4. Cost analysis	1
5. Meta-analysis	1
6. Letter	4
7. Not home uterine activity monitoring	1
8. Twin study	1
9. Study <40 subjects	1
10. Not preterm labor	1
Total articles excluded	33
Total articles abstracted	4

Table 9. Costs and cost analyses search strategy

Table 9. Costs and cost analyses search strategy

Costs and Cost Analyses (1966-present)	Results
MEDLINE (Index Medicus)
1. Explode costs and cost analysis	55,836
2. Explode labor, premature, and explode epidemiologic studies, and limit to humans	1,039
3. Explode antibiotics	306,659
4. 1 and 2 and 3	0
5. Explode tocolytic agents	42,787
6. 1 and 2 and 4	0
7. Explode biologic markers; or estriol; or fibronectins; or saliva	31,610
8. 1 and 2 and 6	1

Data Extraction Form

The project director and scientific director worked with core project staff to develop data extraction forms to use for entry of relevant information from the eligible publications. The data extraction form was developed in an iterative fashion with extensive communication between the methodologists and researchers. We began by outlining common variables across types of studies, including study design features and outcomes. We then created more specialized forms to serve as the most efficient and effective vehicles to extract the necessary information from each of the five bodies of literature included in this review: (1) biologic markers, (2) tocolytic benefits, (3) tocolytic harms, (4) antibiotics, and (5) home uterine activity monitoring.

Included across all studies were the following:

• Description of study designs.

• Description of patient population, including maternal age and race and clinical. inclusion/exclusion criteria such as condition of membranes (ruptured or intact), estimated gestational age, and maternal and fetal indications for birth.

• Definition of preterm labor.

• Description of the test, treatment, or intervention.

• Description of adjunct therapies.

• Description of outcomes measured such as prolongation of pregnancy, gestational age at delivery, maternal morbidities, and infant birth weight.

• Description of secondary analyses performed.

These variables were incorporated into the data extraction forms whose basic structure taken from forms developed for the evidence report on pharmacotherapies for alcohol dependence. The draft forms underwent extensive reviews, including pretesting on several randomly selected articles. After the pretest, we revised the forms to increase their utility and efficiency. The final forms can be found in Appendix B.

The tocolytic and antibiotic data extraction forms were quite similar in design. The tocolytic form had one additional question to capture the definition of tocolysis used in the analysis; therefore, only the tocolytic form is included in Appendix B. We made minor modifications to the form design to accommodate the home uterine activity monitoring literature and made more extensive modifications for the biologic marker form. Rather than mostly capturing the efficacy of treatment based on outcomes of RCTs, the biologic marker literature focuses on the effectiveness of the intervention with outcomes presented generally in terms of predictive values (positive and negative) and sensitivity and specificity. Both the home uterine activity monitoring and biologic marker extraction forms also are found in Appendix B.

Abstractors and Training

The RTI-UNC EPC used abstractors with two types of backgrounds for the data extraction process: individuals with content or clinical expertise and those with strong methodologic skills. Several obstetricians, one pediatrician with training in women's health, and a nurse midwife served as clinican abstractors; all had prior research experience. The methods abstractors were doctoral students in the UNC School of Public Health or the UNC Department of Economics who have extensive training in quantitative methods. In addition, reviewers included a health services researcher with expertise in quantitative methods (the project director) and an obstetrician with expertise in treating women with preterm labor and in conducting clinical research in the topic area (the scientific director).

To collect high-quality data, we put necessary instructions directly on the data extraction forms. Based on prior work of the RTI-UNC EPC, we found that any separately distributed written instructions were less reliable sources of help for abstractors than the extraction form itself. In addition, all abstractors attended two formal training sessions. At the first session, we explained the process and goals of the abstraction. Following the training, the abstractors were sent home with an article to review. We then reconvened the group and, through a review of the test article, ensured that the reviewers understood what was expected from their work. For example, we instructed abstractors that when inconsistencies arose between results stated in the text of an article and those presented in tabular form, they were to take data from the text. They also were told to extract precisely what was contained in the article and to reserve opinions concerning the contents to the comments section at the end of the form.

At the completion of abstractor training, the data abstraction process began. The project director, scientific director, and UNC research coordinator monitored progress. Problems or questions that abstractors encountered were routed through the research coordinator to the appropriate senior staff member.

Tocolytic Harms Abstraction

Analysis Methodology

Harms results are presented through graphic representation rather than as evidence table entries. This methodology provides a means of comparing the number and percentage of participants experiencing adverse reactions to treatment or placebo. For clarity in the analysis, we grouped the harms into two categories: maternal harms and fetal/newborn harms. We aggregated individual descriptors into clinically sensible categories within each of these broad categories. The aggregation process may lead to some "double counting" of harms. For example, if an article described 15 percent of the subjects as having chest pain and 10 percent as having arrhythmia, then 25 percent of the patients would be assessed as having serious cardiovascular harms. From the studies, one cannot determine whether the patients with the various types of harms are the same or different patients. We did not allow greater than 100 percent of subjects to have harms within a category. Double counting of harms would result when one patient experienced both chest pain and arrhythmia; therefore, the graphs comparing studies should be viewed as somewhat more valuable for demonstrating the relative rates of harms among the therapeutic categories, rather than the absolute rates. (See results in Chapter 3.)

We performed a sensitivity analysis to examine whether such possible double counting would affect the patterns of harms demonstrated in the graphs. A low estimate of harm rates could be attained by selecting the side effect with the highest percentage within each category rather than by adding the harms within categories. In this low-estimate sensitivity analysis, if 25percent of patients had chest pain and 10 percent arrhythmia, then serious maternal cardiac harms would be reported as 25 percent. When we performed this analysis, we found the same overall pattern of harms.

The analysis of fetal/newborn harms and the evaluation of efficacy in other sections of this report overlap to some extent. Rates of fetal demise, sepsis, or intraventricular hemorrhage (IVH) could be considered as either a therapeutic effect or a harm associated with use of the pharmacologic treatment. This information, then, may also be found in the evidence tables.

Table 10. Categories of maternal and fetal/newborn (more...)

Table 10. Categories of maternal and fetal/newborn harms related to tocolytictreatment

Maternal harms

Serious cardiovascular harms: Respiratory arrest Adult respiratory distress syndrome Ischemic changes on ECG Arrhythmia, including SVT or symptomatic PVCs Chest pain Heart failure or pulmonary edema Abnormal ECG, including nonspecific ST-T wave changes Minor cardiovascular harms Hypertension: Systemic BP >140 Diastolic BP <60 Transient hypertension Palpitations Tachycardia Gastrointestinal harms: Constipation Diarrhea Epigastric pain Nausea Vomiting Dysgeusia Rectal intolerance Heartburn Soft stool Ileus Metabolic harms: Hypokalemia Hypocalcemia Hypermagnesemia Hyperglycemia Abnormal glucose tolerance test Psychologic harms: Itching Restlessness Agitation Anxiety Other emotional reactions Shakiness Tremor Insomnia

Fetal/newborn harms

Fetal tachycardia Increased fetal heart rate above baseline Fetal tachycardia >120 Tachycardia Ductal constriction/tricuspid regurgitation Ductal constriction Tricuspid regurgitation IVH IVH IVH i-ii IVH iii-iv Necrotizing enterocolitis Sepsis Amnionitis Sepsis Infection Meningitis Probable sepsis Pneumonia

Notes: ECG = electrocardiogram; SVT = suproventricular tachycardia; PVC = premature ventricular contraction; BP = blood pressure; IVH = intraventricular hemorrhage.

The aggregation of individual harm variables into categories is presented in Table 10. This categorization, of course, is somewhat arbitrary, but it does reflect a significant advance over the level of categorization in the source articles, which often provided no categorization at all. The placebo category refers to placebo pharmacologic therapy within the context of an RCT. Almost all placebo groups also received standard supportive care such as hospitalization, bed rest, and intravenous hydration.

Quality of Individual Articles

The RTI-UNC EPC approach to assessing the quality of the individual articles considers the various concerns important to the scientific method of conducting research, including the internal and external validity of each study. It encompasses, among other factors, study design, measurement issues, statistical analyses, and the appropriateness of the conclusions being drawn.

Figure 6. Quality of articles form: Categories and points for grading literature Preterm labor: Antibiotics and tocolytics literature

Category 1: Problem or question studied
Category 2: Sampling
Category 3: Measurement
Category 4: Internal validity
Category 5: External validity
Category 6: Construct validity
Category 7: Statistical conclusions
Category 8: Justification for conclusions
		No	Yes
1.	Is the defnition used for the diagnosis of preterm labor clearly stated?	0	2
		No	Yes
2.	Is the recruitment strategy clearly described?	0	1
		No	Yes
3.	Is there random allocation of treatment?	0	2
		No	Yes
4.	Are baseline levels of cervical dilation and effacement assessed?	0	1
		No	Yes
5.	Are patients with ruptured membranes excluded or assessed in a stratified analysis?	0	1
		No	Yes
6.	Are medically indicated or iatrogenic pregnancies excluded or assessed in a stratified analysis (e.g., preeclampsia, severe fetal defects, placenta previa)?	0	1
		Unknown or >20%	<20%
7.	Is the dropout rate of patients invited into the study less than 20%?	0	1
8.	What outcomes are measured?	No	Yes
	1 Proportion of patients with preterm delivery at a defined period (e.g., 32, 34, or 37 weeks).	0	1
	2 Maternal outcomes (e.g., length of hospital stay, chorioamnionitis, endometritis).	0	1
	3 Neonatal outcomes (e.g., APGAR, need for intensive care unit, length of stay, respiratory distress syndrome, birth weight, intraventricular hemorrhage).	0	1
		No	Yes
9.	Are the outcome measurements clinically relevant?	0	1	0.5
		No	Yes	NA
10.	Are potential confounding variables taken into account in the outcome analysis (e.g., age, race, type of preterm labor, parity)?	0	1
		No	Yes	NA
11.	Are cointerventions (e.g., bed rest, hydration, or use of antibiotics when tocolytics are the study question, or vice versa) measured or reported on?	0	1
		No	Yes	NA
12.	Are the treating clinicians masked to treatment administered?	0	1
		No	Yes	NA
13.	Are the patients masked to treatment received?	0	1
		No	Yes
14.	Do the study conclusions apply to pregnant women in the United States or to important subgroups of the U.S. population?	0	1
		No	Yes
15.	Is the clinical setting specified clearly?	0	1
		No	Yes
16.	Are the criteria used to assign gestational age clearly stated?	0	1
		No	Yes	NA
17.	If the study is negative, is a power analysis performed?	0	1
		No	Yes	Some
18.	Are statistically significant (or insignificant) findings clinically significant?	0	1	0.5
		No	Yes	Some
19.	Are statistical tests used appropriate to the data?	0	1	0.5
		No	Yes
20.	Are the conclusions warranted from the data and/or design of the study?	0	2

Figure 7. Quality of articles form: Categories and points for grading biologic marker literature

Category 1: Problem or question studied
Category 2: Sampling
Category 3: Measurement
Category 4: Internal Validity
Category 5: External validity
Category 6: Construct validity
Category 7: Statistical conclusions
Category 8: Justification for conclusions
		No	Yes
1.	Is the description of the test procedure provided?	0	1
		No	Yes
2.	Are the relevant outcomes clearly defined?	0	1
		No	Yes
3.	Is a reference standard for the test clearly referenced?	0	1
		No	Yes
4.	Is the recruitment strategy clearly described?	0	1
5.	Are inclusion/exclusion criteria clearly stated?	0	1
6.	Are baseline levels of cervical dilation and/or effacement assessed?	0	1
7.	Are patients with ruptured membranes excluded or assessed in a stratified analysis?	0	1
8.	Are medically indicated or iatrogenic deliveries excluded or assessed in a stratified analysis (e.g., preeclampsia, severe fetal defects, placenta previa)?	0	1
		Unknown or >20%	<20%
9.	Is the dropout rate of patients invited into the study less than 20%?	0	1
10.	What outcomes are measured?	No	Yes
	1 Proportion of patients with preterm delivery at a defined period (e.g., 32, 34, or 37 weeks).	0	1
	2 Maternal outcomes (e.g., length of hospital stay, chorioamnionitis, endometritis).	0	1
	3 Neonatal outcomes (e.g., APGAR, need for intensive care unit, length of stay, respiratory distress syndrome, birth weight, intraventricular hemorrhage).	0	1
		No	Yes	Some
11.	Are the outcome measurements clinically relevant?	0	1	0.5
		No	Yes	NA
12.	Are potential confounding variables taken into account in the outcome analysis (e.g., age, race, type of preterm labor, parity)?	0	1
		No	Yes	NA
13.	Are cointerventions (e.g., bed rest, hydration, tocolytic use) measured or reported on?	0	1
		No	Yes	NA
14.	Are the treating clinicians masked to test result?	0	1
		No	Yes	NA
15.	Are the patients masked to test result?	0	1
		No	Yes
16.	Is the clinical setting specified clearly?	0	1
		No	Yes
17.	Are the criteria used to assign gestational age clearly stated?	0	1
		No	Yes	NA
18.	Is a power analysis performed?	0	1
		No	Yes	Some
19.	Are statistically significant (or insignificant) findings clinically significant?	0	1	0.5
		No	Yes	Some
20.	Are statistical tests used appropriate to the data?	0	1	0.5
		No	Yes
21.	Do the authors provide either sensitivity/specificity, likelihood ratios or odds ratios?	0	1
		No	Yes
22.	Are the conclusions warranted from the data and/or design of the study?	0	2

Figure 8. Quality of articles form: Categories and points for grading literature home uterine activity monitoring

Category 1: Problem or question studied
Category 2: Sampling
Category 3: Measurement
Category 4: Internal validity
Category 5: Construct validity
Category 6: Statistical conclusions
Category 7: Justification for conclusions
		No	Yes
1.	Is the definition used for the diagnosis of preterm labor clearly stated?	0	1
2.	Is the intervention being tested clearly described?	0	1
		No	Yes
3.	Is the recruitment strategy described?	0	1
		No	Yes
4.	Are the inclusion criteria for the study stated?	0	1
		No	Yes
5.	Is the randomization scheme described?	0	1
		No	Yes
6.	Are baseline levels of cervical dilation and effacement presented?	0	1
		Unknown or >20%	<20%
7.	Is the dropout rate of patients invited into the study less than 20%?	0	1
		No	Yes
8.	Are all patients who dropped out of the study accounted for?	0	1
		No	Yes
9.	Are the outcomes clearly described?	0	1
		No	Yes	Some
10.	Are the outcome measurements clinically relevant?	0	1	0.5
		No	Yes	NA
11.	Are potential confounding variables taken into account in the outcome analysis (e.g., age, race, labor, parity)?	0	1
		No	Yes	NA
12.	Are any cointerventions reported on?	0	1
		No	Yes	NA
13.	Are the treating clinicians masked to intervention group?	0	1
		No	Yes	NA
14.	Are the patients masked to treatment received?	0	1
		No	Yes
15.	Are the criteria used to assign gestational age stated?	0	1
		No	Yes	NA
16.	Was a power analysis performed?	0	1
		No	Yes	Some
17.	Are statistical tests used appropriate to the data?	0	1	0.5
		No	Yes	Some
18.	Are statistical tests clearly described?	0	1	0.5
		No	Yes
19.	Are levels of significance and or confidence intervals given?	0	1
		No	Yes
20.	Was an intent-to-treat analysis performed?	0	1
		No	Yes
21.	Are the conclusions warranted from the data?	0	1
22.	Are the conclusions warranted from the study design?	0	1

To assess the internal validity of a study (i.e., the likelihood that the design and conduct of the study minimize systematic error or bias) the following factors were evaluated from each study across all study types:

• Description of the recruitment strategy.

• Masking of the treating clinician and patient.

• Inclusion of baseline measurements including cervical dilation, effacement, and estimated gestational age.

• Definition of outcome measures, attrition rates, and confounding variables.

Statistical considerations included assessing the adequacy of the methods used and whether the authors did a power calculation. The rating scale also addressed the clinical relevancy of the statistical findings.

With regard to diagnostic and outcome measures, the abstractors evaluated whether the standard for a test was clearly referenced. In addition, the quality rating instrument assessed whether studies included the three major categories of outcomes (i.e., those for delivery, mothers, and infants).

External validity (i.e., whether the findings of the study can be generalized to populations that did not participate in the study) also was considered for quality rating. We included in our evaluation whether the clinical setting was specified clearly and whether conclusions apply to pregnant women in the United States.

The quality rating instrument was based on a point scale that was transformed into a percentage score, with the higher percentage score denoting a better study. Each abstractor completed the quality rating for each article reviewed. The individual scores provided in the evidence tables for each article were derived by calculating the following ratio: the points from the responses circled by the abstractor (numerator) divided by the total number of points that could have been awarded (denominator). When the items were determined to not be applicable, "NA," then the total number of points for that item was removed from both the numerator and denominator. Quality rating scores for each article were not averaged across the two abstractors (i.e., clinician and methodologist), so that patterns across studies could be observed.

Grading of the Evidence

According to a report developed by Lohr and Carey,⁹⁷ several systems have been proposed in the past 10 to 15 years for "grading the quality or strength" on a given clinical topic or pathway; however, little or no consensus exists about which specific system is best, and thus little guidance was given about which approach to use for our review.

The RTI-UNC EPC developed a grading scheme that considers the strength and the quality of the evidence based on the work of the RTI-UNC EPC in the Evidence Report on the Pharmacotherapy for Alcohol Dependence⁹⁸ and a report by the U.S. Preventive Services Task Force (1996).⁹⁹ Our determination of the strength of the evidence includes an evaluation of the consistency of study findings over all studies, the quality rating scores for individual pieces of research, the magnitude of important outcomes, and, when available, the results of our meta-analyses. Because we include both randomized and cohort studies in the review, we also record whether the evidence is appropriate for evaluating the efficacy or the effectiveness of the findings. The letter grade given to a body of literature is not necessarily higher because it contains a greater number of RCTs; rather, it rests on the quality of the studies themselves. We did not assign scores for each of the variables considered for the ratings; instead, we based our quality assessment on an adjudicated rating that was initially provided by the project director, scientific director, and clinical methodologist as follows:

Description of the Biologic Marker Literature

Results of the Biologic Marker Literature Review

Figure 9. Use of fFN to adjust clinical estimates of (more...)

   Figure 9. Use of fFN to adjust clinical estimates of risk for preterm birth

Figure 10. Use of EVUSD to adjust clinical estimates (more...)

   Figure 10. Use of EVUSD to adjust clinical estimates of risk for preterm birth

9

10

Aggregate Performance of fFN and EVUSD

Overview

This section presents estimates of sensitivity, specificity, PPV, NPV, and likelihood ratios as a means of clarifying the potential value of these biologic markers in clinical practice. Of the 14 articles evaluating the use of fFN, 12 provided sufficient detail to construct contingency tables describing the relationship between test results and birth before 37 completed weeks' estimated gestational age (see Evidence Table 1). ^{46 101 102 103 104 105 106 107 108 109 113 114} Each study used a cutoff point of greater than or equal to 50 ng/ml for a positive test.

Aggregating across these studies, 1,786 samples from women being evaluated or treated for preterm labor were available for this analysis. In this combined cohort, 488 preterm births occurred, for an overall prevalence of 27 percent. Of the 267 women who delivered before term, 267 had positive test results at their initial evaluation and 221 did not. Of those who delivered at term, 1,111 women had true-negative fFN results and 187 women had false-positive fFN values.

Results

Combined, the sensitivity of fFN (i.e., the ability to identify correctly those who delivered before 37 weeks) was 54.7 percent; specificity (i.e., the ability to identify those who delivered after 37 weeks) was 85.6 percent. The PPV (i.e., the probability that a woman with a positive test result would deliver before term) was 58.8 percent. The NPV (i.e., the probability that a woman with a negative test would deliver at term) was 83.4 percent. The likelihood ratio for a positive fFN is 3.8 (95 percent confidence interval [CI]: 3.3, 4.4) and 0.53 (95 percent CI: 0.48, 0.58) for a negative test result. The clinical application of likelihood ratios is reviewed briefly below.

The majority of these studies enrolled women in a 10-week range of gestational ages, meaning some subjects have longer time at risk for preterm birth than others, typically between 24 and 34 weeks. This factor has the effect of reducing sensitivity of testing at a single point in time for predicting birth before 37 weeks. Presumably, some women with an initial negative fFN test will subsequently develop positive tests. In a sense, these women actually do not reflect false-negative values on their initial test; rather, they reflect the time-dependent nature of using biologic markers and their relationship to outcomes.

Limiting the effect of weeks elapsing between fFN is informative about the utility of fFN as a reliable biologic marker. The five publications that examine the more immediate outcome of delivery within a week are instructive on this point. ^{46 103 104 105 108} The aggregate sensitivity of fFN (from these studies) to identify women who delivered in 7 days after testing was 89.4 percent; the specificity was 83.3 percent; the PPV was 22.9 percent; and the NPV was 99.3 percent. For the outcome of birth within 7 days, the likelihood ratio for a positive fFN is 5.3 (95 percent CI: 4.6, 6.2) and 0.13 (95 percent) CI: 0.06, 0.26) for a negative fFN. The high NPV, supported by this combination of results, suggests that fFN conveys an important benefit in terms of identifying women who are not an immediate risk of preterm birth and who may not require intervention.

The EVUSD literature presents special challenges to combining results in this fashion. Continuous predictors, such as cervical measurements, cannot be captured adequately from the level of detail permitted in publications. Appropriately, therefore, authors of the articles we reviewed examined their data and established categories of cervical change or dichotomous cutoff points for examining test characteristics or calculating likelihood ratios. As a result of their statistically driven analyses, the studies show greater inconsistency in the cutoff points used than was true for the fFN articles.

Of the five EVUSD studies in women with signs or symptoms of preterm labor, three evaluated the use of wedging or funneling to determine risk of preterm birth. ^{111 112 113} Combined, these investigators evaluated 314 women, of whom 98 delivered before 37 weeks' gestation. Of those who delivered preterm, 59 had funneling or wedging noted on ultrasound examination; 39 did not have visible funneling. In addition, 23 women had funneling and delivered at term (a total of 23 false-positives), and 171 women had normal ultrasounds and term deliveries. These figures are equivalent to the following rates: sensitivity, 60.2 percent; specificity, 79.2 percent; PPV, 56.7 percent; and NPV, 81.4 percent. The positive likelihood ratio for funneling is 2.8 (95 percent CI: 2.1, 3.9) and 0.50 (0.39, 0.65) for the absence of funneling.

Two studies ^{55 112} used the cervical length of 3 cm or less as a diagnostic cutoff point to assign higher risk of preterm birth. This threshold was associated with the following rates: sensitivity, 88.5 percent; specificity, 59.3 percent; PPV, 49.5 percent; NPV, 92.0 percent; and positive likelihood ratio of 2.2 (95 percent CI: 1.7, 2.7); negative likelihood ratio of 0.19 (0.10, 0.39).

Table 12. Predictive value of biologic markers in women (more...)

Table 12. Predictive value of biologic markers in women with symptoms of preterm labor

	Sensitivity	Specificity	Test Characteristics by Incidence of Preterm Birth In Women Presenting for Preterm Labor
			Included Studies			Low Risk Example Within 7 Days = 3% Before 37 Weeks = 10%		Medium Risk Example Within 7 Days = 10% Before 37 Weeks = 25%		High Risk Example Within 7 Days = 15% Before 37 Weeks = 45%
			Observed Incidence	PPV	NPV	PPV	NPV	PPV	NPV	PPV	NPV
Birth within 7 days
fFN	89.4	83.3	5.3	22.9	99.3	14.3	99.6	37.2	98.6	48.6	97.8
Birth before 37 weeks
fFN	54.7	85.6	27.3	58.8	83.4	29.7	94.5	55.9	85.0	75.7	69.8
EVUSD: Funneling/wedging	60.2	79.2	31.2	56.7	81.4	24.3	94.7	49.0	85.6	70.4	70.9
EVUSD: Cervical length < 3 cm	88.5	59.3	31.1	49.5	92.0	19.4	97.8	42.0	93.9	64.0	86.2

fFv = fetal fibronectin; EVUSD = endovaginal ultrasound; PPV = positive predictive value; NPV = negative predictive value

The test characteristics reviewed above are summarized in Table 12. The right portion of the table provides examples of the predictive values of these biologic markers in the studies included in this summary and for a range of hypothetical rates of preterm birth in women presenting with signs and symptoms of preterm labor. As shown in the table, as the proportion of women who gave birth preterm increases, the NPV falls, and the PPV improves. The aggregate predictive values reflect expected performance of the biologic markers in populations similar to that of the study participants with the observed incidence of preterm birth comparable to that summarized in the table.

Conclusions

Biologic markers supplement clinical judgment. Based on a patient's obstetric history, her description of current symptoms, and findings from physical examination and monitoring of contractions, care providers estimate the probability of preterm birth. An estimate of high probability results in further testing and/or treatment; an estimate of lower probability leads to continued observation. Thresholds for deciding whether to treat women in preterm labor or to observe them further are highly individualized and often not explicit. For example, physicians with aggressive treatment policies may favor intervention at any time that they believe that a woman has a greater than 10 percent chance that her preterm labor might result in preterm birth and will likely further adjust that judgment based on gestational age. A conservative philosophy might reserve treatment for cases in which the probability estimate of a preterm birth is greater than 30 percent, accepting that time and observation will further clarify the need to treat.

Regardless of where this threshold to treat lies for individual providers, there exists a range of probabilities between an estimated probability at which they "always" treat and an equivalent estimated probability at which they are comfortable sending a patient home with no further evaluation. When this range is relatively wide, the additional diagnostic or prognostic information provided by fFN or EVUSD testing may help the clinician to make a more sound decision.

Clinical applications of likelihood ratios

Imagine a provider whose threshold to treat is 20 percent and who would accept a 2 percent risk that the pregnancy would be delivered before term. Faced with a patient who is being evaluated for her second episode of preterm labor at 33 weeks' gestation and who has contractions every 8 minutes, no cervical change, and a history of a prior preterm birth at 36 weeks, the clinician may believe that the woman's current episode has a 10 percent chance of leading to progressive cervical change and delivery. This is within the clinician's range of clinical uncertainty; additional information, such as fFN or EVUSD data, can have direct decisionmaking utility.

The question is how to apply such information easily across a range of patient populations with varied prevalence of preterm birth. Likelihood ratios for categories of test results accomplish this; they can be used to modify the odds of a patient's having a specific outcome such as preterm birth or birth within the next week. Posttest odds are equal to the pretest odds multiplied by the likelihood ratio.

In this hypothetical example, the provider's pretest probability, based on clinical judgment, was 10 percent. This value can be converted to odds using this formula:

Pretest odds are 0.11:1 or 1:9.

From the combined analysis of fFN discussed above (with a likelihood ratio of 3.8 percent for birth before 37 weeks), if the woman's fFN test was positive, then her posttest odds would be given by the following equation:
Posttest odds = Pretest odds x Likelihood ratio = 0.11 x 3.8 = 0.42

To convert this back to a probability, we use this formula:

In short, with a positive fFN test, a woman's posttest odds are 0.42, which is equivalent to a probability of 30 percent. This is a value within the range of probabilities for which the hypothetical clinician would start treatment. Pocket nomograms are available to ease use of these conversions in clinical settings. ¹¹⁶

Figures 9 and 10 remove these intermediate calculations and allow direct inspection of the relationship between a clinician's initial assessment of risk of preterm birth (the X-axis) and the probability after adjustment for evaluation of biologic markers (the Y-axis). If the clinician is certain that a woman will deliver, then both probabilities are near 1.0 (in the upper right-hand corner of the figures), and the biologic marker contributes little. The lower range of clinical assessment of probability (toward the lower left-hand corner) is of greater interest. In this range, biologic markers offer additional information that can aid decisionmaking, prompting treatment or confirming the choice of expectant management.

An alternative approach to using these graphs is to acknowledge that clinical estimates of risk of spontaneous preterm birth for an individual are imprecise, whereas the prevalence of preterm birth from preterm labor in a specified population can be determined with certainty. This prevalence can serve as the prior probability. In this case, the prevalence is located on the X-axis, and the Y-axis reflects the posterior probability, in that population, adjusted for the biologic marker result. Whether interpreted as adjustment of individual risk or risk for a population, these biologic markers offer additional information that refines assessment of the probability that preterm birth will result from an episode of preterm labor.

Description of the Tocolytic Literature

First-Line Tocolytics: Beta-Mimetics

First-Line Tocolytics: Calcium Channel Blockers

First-Line Tocolytics: Magnesium Sulfate

First-Line Tocolytics: Nonsteroidal Anti-Inflammatory Disease Drugs

First-Line Tocolytics: Other Agents

Maintenance Tocolytics: Beta-Mimetics

Seven of the RCTs compared a treatment group to a control group. ^{44 172 175 175 176 177 178 179} All except three studies masked the treating clinician to the protocol. ^{173 175 176} Most of the RCTs ^{44 174 175 176 177 179 180 181} found no efficacy differences in treatment. In contrast, Brown and Tejani ¹⁷² in a small study found beta-mimetic treatment to show efficacy in terms of mean prolongation days and infant birth weight. In a more recent and larger study of more than 200 women, Rust, Bofill, Arriola, et al. ¹⁷⁸ found no differences between groups in delivery outcomes; however, the length of hospital stay for the infant was shorter in the control group than in the group treated with magnesium or terbutaline. This was the only study to include a pill count to evaluate compliance with treatment.

A small observational study ¹⁸² contained no statistical analysis. Another larger cohort study compared women with and without cervical change ¹⁷² and found delivery and infant outcomes to be generally superior in women without cervical change. One RCT compared two beta-mimetics, ritodrine and terbutaline and found no difference. ¹⁷³

Maintenance Tocolytics: Calcium Channel Blockers

Maintenance Tocolytics: Magnesium Sulfate

Maintenance Tocolytics: Non-Steroidal Anti Inflammatory Disease Drugs

Tocolytic Harms

Owing to the large numbers of categories and the multiple classes of agents assessed in this evidence report, we have chosen to summarize the harms data graphically in the form of "bubble charts." Tables specifying the individual rates of the harms in each study would be unwieldy, so we believe that these figures will convey the information more readily. The articles included in this analysis appear in the tocolytic evidence tables, but we also included additional articles that had not met our inclusion criteria for the benefits analysis.

The graphs give the rates of each of the categories of harm within each class of tocolytic medication: beta-mimetics, calcium channel blockers, NSAIDs, magnesium sulfate, ethanol, and atosiban. Each circle within the graph represents a single study. The size of the circle is proportional to the number of subjects receiving the medication (or placebo) in the study. A study with 100 subjects would have a circle twice as large as a study with 50 subjects, assisting the reader in giving more "weight" to larger studies.

We listed as "harms" clinical markers and events that the authors of the individual studies considered as adverse events or "side effects." In general, these adverse events represent process measures of somewhat unclear significance for the long-term health of the mother and infant. For example, the hypokalemia that is sometimes found after treatment with beta-mimetics may have no clinical consequences but may be associated with increased risk of cardiac arrhythmia. This level of detail generally is not available in this literature. Confidence intervals are generally not provided in the articles, and statistical interpretation is often difficult. Our summary of the literature, by definition, cannot be more detailed than the detail in the studies performed.

We were unable to form synthetic estimates across studies of the rates of treatment harms. The definitions of the clinical categories often were vague or not provided at all in the source articles, which would lead to problems if we assumed that the clinical categories actually were equivalent. Not all categories of medications were represented on every graph. For example, data on "metabolic harms" (e.g., elevated blood glucose, hypokalemia) were collected only for beta-mimetics.

Maternal Harms

Serious cardiovascular harms

Figure 11. Maternal harms - serious cardiovascular (more...)

   Figure 11. Maternal harms - serious cardiovascular

Cardiac side effects, including arrhythmias, heart failure, and chest pain, are well known to occur with the beta-mimetics, and these findings were confirmed by the graph in Figure 11. The estimates varied widely among the studies, and the range in the severity of the "serious harms" was substantial. Some of the variability is likely attributable to the assiduousness with which investigators assessed the cardiac effects of these agents.

Minor cardiovascular harms

Figure 12. Maternal harms - minor cardiovascular

   Figure 12. Maternal harms - minor cardiovascular

The pattern for minor cardiovascular harms was similar to that for the major harms, with higher rates in patients taking the beta-mimetic class of medications (Figure 12). Again, a great deal of variability occurred within categories such as the beta-mimetics. Although these types of events are uncomfortable, they were generally not considered to be of major health consequence.

Maternal metabolic harms

Figure 13. Maternal harms - metabolic

   Figure 13. Maternal harms - metabolic

As shown in Figure 13, a definite pattern of increased metabolic abnormality was seen for women who received beta-mimetic agents. The detail in the articles does not allow us to assess the clinical importance of the hyperglycemia, hypokalemia, or other adverse effects; but such abnormalities clearly were very common.

Gastrointestinal harms

Figure 14. Maternal harms - gastrointestinal

   Figure 14. Maternal harms - gastrointestinal

Many agents, including placebo, are associated with gastrointestinal harms (Figure 14). In the case of placebo, the "harm" is the association of symptoms with the medication when, in fact, they are not caused by the medication. Moderate rates of gastrointestinal harms were found for many of the agents. Interestingly, the few studies examining NSAIDs appeared to show relatively low rates of maternal gastrointestinal effects in this literature when compared with the other agents. This result may be because of the young age of the subjects and the relatively short duration of the NSAID exposure. Overall, there appeared to be little clear relationship among class of medication and gastrointestinal side effects.

Maternal psychologic harms

Figure 15. Maternal harms - psychologic

   Figure 15. Maternal harms - psychologic

The category of "psychologic" harms encompasses symptoms that are relatively nonspecific (Figure 15). The beta-mimetic studies assessed this category of harms most commonly and found them more frequently when compared with studies of other agents in which it was assessed.

Overall maternal results

The overall picture for maternal harms is one of more frequent occurrence of harms in users of beta-mimetic medications, especially for the cardiac, metabolic, and psychologic categories. We note, however, that these harms also were searched for more frequently in studies of these agents.

The articles were often unclear as to whether a single patient could appear more than once within a single category of harm. For example, if a single patient had tachycardia and palpitations, she could be counted in each category. Because we saw both of these effects as belonging to "minor maternal cardiovascular harms," the summary graphs may be subject to a degree of double counting that could bias the study outcome by raising somewhat the estimate of harms. Conversely, the reviewers for this report were impressed that many of the studies did not search adequately for maternal harms associated with tocolytic treatment, which would tend to bias the study outcome by lowering the estimate of the rate of harms.

Fetal/Neonatal Harms

Ductal constriction/tricuspid regurgitation

Figure 16. Fetal/newborn harms - ductal constriction/tricuspid (more...)

   Figure 16. Fetal/newborn harms - ductal constriction/tricuspid regurgitation

As shown in Figure 16, premature constriction of the ductus arteriosis would be expected to be of concern with the anti-inflammatory nonsteroidal class of medications. Two studies addressed this issue, and they did appear to find somewhat increased rates of ductal constriction. The study with the highest rates used routine prepartum echocardiograms specifically to estimate ductal constriction. The study did not address the clinical significance of this ductal constriction.

Intraventricular hemorrhage

Figure 17. Fetal/newborn harms - IVH

   Figure 17. Fetal/newborn harms - IVH

Intraventricular hemorrhage (IVH) is an uncommon but potentially serious neonatal complication, with increased risk among premature infants (Figure 17). No pattern of increased risk appeared for any pharmacologic intervention. The studies with the highest rate of IVH were case series of premature, low-birth-weight infants, in whom one would expect higher rates of IVH in both intervention and placebo groups.

Necrotizing enterocolitis

Figure 18. Fetal/newborn harms - NEC

   Figure 18. Fetal/newborn harms - NEC

We found little evidence of a difference among the treatment groups in the risk of necrotizing enterocolitis (NEC) (Figure 18). Prior research has shown that indomethacin may be associated with an increased risk of certain neonatal adverse outcomes, including intraventricular hemorrhage and NEC. ^{184 185} Moreover, indomethacin therapy in utero may cause constriction of the ductus arteriosis that may culminate in closure. ^{186 187} The drug is used in neonatology as a medical treatment to close persistently patent ductuses in newborns. Given that all of these complications also are associated with preterm birth, the magnitude of additional risk posed by indomethacin is unclear.

Sepsis

Figure 19. Fetal/newborn harms - sepsis

   Figure 19. Fetal/newborn harms - sepsis

Estimates of sepsis in the literature were hindered by definitional problems of "sepsis." Some studies required a positive blood culture, others defined sepsis clinically, still others had categories of "possible sepsis." The study with a category of "possible sepsis" did exhibit the highest rates of this harm. Overall, we found no pattern of differential neonatal harm among these agents (Figure 19).

Tachycardia

Figure 20. Fetal harms - tachycardia

   Figure 20. Fetal harms - tachycardia

The estimation of fetal/neonatal tachycardia is hampered by lack of a consistent definition of tachycardia, which likely contributed to the wide range of estimates. Some studies gave no specific heart rate but rather simply reported a percentage for tachycardia. As shown in Figure 20, not surprisingly the beta-mimetics appear to contribute to higher rates of rapid heart rate.

Overall fetal/neonatal results

The fetal/neonatal harms literature provided little consistent evidence of short-term harms to the infants of women receiving tocolytic agents. Insufficient evidence was available from which to draw conclusions regarding longer term effects of tocolytic medications on the health of children.

The analysis did not address the potential harms associated with the use of multiple classes of tocolytic medications during a single pregnancy. A few studies used such regimens, generating the challenge of differentiating the effects of single agents compared with those of multiple agents. One would expect that the harms associated with the use of multiple agents would be at least as great as those of both agents used alone.

Results of the Antibiotic Review

The articles reviewed consistently report three outcomes: (1) prolongation of pregnancy after enrollment, (2) mean gestational age at delivery, and (3) preterm delivery at a specific cutoff point, usually less than or equal to 37 weeks. Of the 15 RCTs reviewed, two consistently found improvement in all three of these outcome measures, ^{191 186} but five found no improvement in any of the three measures. ^{193 194 195 196 197} Five studies found a significant increase in pregnancy prolongation, ^{191 192 198 199 200} and three studies found a diminution in the number of preterm deliveries. ^{192 198 201}

Thus, the results of this research are mixed and definitive statements are hard to make about antibiotic efficacy in gravidas with preterm labor symptoms. This mixture of positive and negative results emphasizes the need for meta-analysis; Chapter 4 discusses the results of the meta-analysis on antibiotics conducted as part of this review.

Issues and Limitations in the Antibiotic Literature

Of the 15 RCTs reviewed, none used the same definition of symptoms to label a woman as having preterm labor. The array of criteria used included frequency of contractions, intensity of contractions, duration of contraction, and a wide range of measures of cervical dilation. Four trials used particularly broad definitions such as "painful" contractions, ¹⁹¹ dilation greater than or equal to 4 cm with no lower limit on contraction frequency, ²⁰¹ contractions less than 10 minutes apart for more than 60 minutes, ¹⁹² and "preterm labor." ²⁰² Lack of a standard, reproducible definition for preterm labor detracts from our ability to compare studies.

Study enrollment was quite varied in these 15 trials. Only one trial had more than 200 subjects (n=277), ¹⁹⁴ and five trials enrolled fewer than 100 women. ^{196 197 200 202 203} Smaller samples force a reader to look at surrogate outcomes (e.g., pregnancy prolongation or gestational age at delivery) because they lack the power to show statistically that antibiotic therapy improves or worsens perinatal outcome. Moreover, clinically significant findings may not be discernable when statistically insignificant results are reported from studies with small sample sizes. Romero, Mazor, Oyarzun, et al. ¹⁸⁸ calculated that their multicenter study of antibiotic efficacy, the largest study we reviewed, with 277 participants, had insufficient power to detect a difference in perinatal outcome. These investigators estimated that a sample size of 2,270 subjects would be required to detect a 33 percent reduction in the event rate between the placebo group and the antibiotic group.

Complicating the interpretation of this literature further is the array of cointerventions used in the management of women with symptoms of preterm labor. All but one study used tocolytics, ¹⁹⁷ and the majority of studies allowed clinicians to prescribe more than one tocolytic. Likewise, the use of corticosteroids and nonpharmacologic interventions (e.g., hydration or bed rest) was ubiquitous throughout these studies. In no instance were regression techniques included to control for the potential confounding that could be caused by the use of these cointerventions and to support the finding that the treatment effect could be attributed to antibiotic therapy.

In lieu of adequate numbers to assess perinatal outcome (which, given the rarity of perinatal death and serious morbidity, would require enrolling thousands of women), survival analysis would be an additional analytic technique for assessment of interventions in women with preterm labor. Despite the advantages of survival analysis, only three trials ^{193 194 200} employed the technique. One of these trials showed a treatment benefit; ²⁰⁰ the other two did not. ^{193 194}

A certain number of preterm births subsequent to assignment of the diagnosis of preterm labor in women with symptoms of labor will be medically indicated. Thus, treatment trials in women with preterm labor symptoms should exclude medically indicated preterm deliveries from their analysis or report on them separately. To our knowledge, none of these studies took this approach, and their failure to account for the etiologic heterogeneity of preterm birth may result in erroneous conclusions.

No two trials used the same antibiotic regimen. Beta-lactam, cephalosporin, macrolide, and metronidzole antibiotics were given to women with symptoms of preterm labor. Routes of administration and duration of use varied widely from study to study. Thus, there is insufficient consistency to determine whether any conclusions reached in relation to the results shown in this literature would apply to antibiotics in general or to the particularly convincing regimen used in an individual study.

In any study of a pharmaceutical intervention, bias is controlled by recording compliance with treatment in the study population and searching for potential confounding by recording differences in compliance. Our review of these 15 RCTs discloses only two studies that measured compliance. ^{194 203}

A limitation of all but two trials is gestational age at enrollment. The prevalence of positive amniotic fluid cultures-a direct measure of in utero infection-and histologic chorioamniotis-an indirect measure-vary inversely with gestational age. ^{194 204} Thus, gravidas at earlier gestational ages are more likely to have infection as an etiology and, in theory, should be more likely to benefit from antibiotic treatment. The converse should be true about women who present with preterm labor symptoms at later gestational ages.

The final limitation is generalizability of these findings to the universe of women with symptoms of preterm labor. Three studies have dropout rates of greater than 20 percent, ^{191 198 203} and two studies have insufficient data to assess dropout rates. ^{194 201} In addition, only five studies reported the number of women with preterm labor symptoms who were screened to find the enrolled population. ^{192 193 194 195} The rates of study participation range from 13 percent ¹⁹⁴ to 50 percent. ¹⁹⁹ Such low enrollment figures may result in participants' being unrepresentative of the population of women with symptoms of preterm labor.

Selection Criteria for Inclusion of Studies

We first reviewed the completed evidence tables for the various interventions to identify studies for inclusion in the meta-analyses. We included only RCTs. Results from prospective cohort studies, case series, retrospective analyses, and other nonrandomized trials were excluded from the analyses. For inclusion in the meta-analyses, the trials also had either to be restricted to women with intact membranes or to have reported the outcomes separately for this subgroup.

Furthermore, only RCTs making comparisons of interest were considered for inclusion. For adjunctive antibiotic and home uterine activity monitoring trials, the comparison of interest was between standard treatment with the intervention and standard treatment with either placebo or no intervention. All antibiotic and home uterine activity monitoring RCTs had at least one intervention arm and one arm without the intervention and, therefore, were potential candidates for the meta-analyses.

By contrast, the tocolytic trials frequently compared one tocolytic agent with another; only a small number of trials included a placebo or an arm with no tocolytic treatment. Because the various tocolytics work through different mechanisms, we were interested in comparisons among different classes of tocolytics; therefore, tocolytic trials were considered for the meta-analyses if they compared one class of tocolytics with another and/or to placebo or no tocolytic treatment. The tocolytic classes included were beta-mimetics, calcium channel blockers, magnesium sulfate, NSAIDs, and ethanol. Because trials comparing one beta-mimetic with another one calcium channel blocker with another added no additional information for the comparisons of interest, they were excluded from the meta-analyses unless they also had an arm with another class of tocolytic or a placebo or no tocolytic arm.

Of all the trials considered for these analyses, 23 trials of first-line tocolytics, 10 trials of maintenance tocolytics, 13 trials of adjunctive antibiotic treatment, and 3 trials of home uterine activity monitoring met the inclusion criteria. We then extracted the information on the outcomes of interest for each trial.

Although the three outcomes of interest were the most frequently reported, they were by no means universally or uniformly reported. For prolongation of pregnancy, some trials reported the mean number of days that pregnancy was prolonged as well as either the standard deviation, the standard error, or both. Others reported only the means. Others reported the number or percentage of pregnancies prolonged for various durations (e.g., <2 days, >7 days, <14 days). Similarly, some trials reported mean gestational age at delivery with or without standard deviations or standard errors; others reported the number or percentage of deliveries at or beyond different gestational age cutoffs (e.g.,>32 weeks, >36 weeks, >37 weeks). Still others reported mean birth weight with or without standard deviations or standard errors and a few reported the percentage of births greater than or equal to 2500 g. Some trials neglected to report any data on one or more of these outcomes altogether. Thus, the actual number of trial results that could be combined for any particular analysis was smaller than the number of trials meeting the inclusion criteria.

Methodology

Where possible, we used the continuous measures of the outcomes of interest and combined the mean differences, d, across studies:
d = M_t - M_c
where M _t and M _c are the sample means of the treated and control arms, respectively. Assuming a normal distribution for the individual observations with equal variances in each arm of the trial, the variance of the difference, S ², is computed:

where SEM _t ² is the standard error of the mean of the treated arm and SEM _c ² is the standard error of the mean of the control arm. Thus, to use this measure, the sample size, means, and standard errors (or standard deviations) must be available from the reported information. For analyses in which too few studies reported means with standard deviations or standard errors, we computed the odds ratio, ω, for selected cutoff values of the outcome measures. We could then use this value as the effect measure for combining results across studies:
w = P_t(1 - P_c)/P_c(1 - P_t)
where p _t is the proportion of the treated group that successfully reached the cutoff value and p _c is the proportion of the control group that successfully reached the cutoff value.

To manage the multiple treatment problem of first-line tocolytic treatments in which we had five classes of tocolytics to compare with placebo/no tocolytics, we used the generalization of the two-arm methods developed by Hasselblad.²⁰⁹ For all interventions and outcomes, we combined the results using a Bayes random effects model as described by Hedges and Olkin,²¹⁰ allowing the true treatment effect to vary across studies. For dichotomous outcomes, we estimated the model parameters with the EGRET software package; for continuous measures, we used a customized program developed by one of the authors, Victor Hasselblad, Ph.D.

In addition to the meta-analysis results on the three main questions noted earlier (e.g., estimates of prolongation of gestation), we report the results of tests of homogeneity for the four main sets of analyses (both types of tocolytics, antibiotics, and home uterine activity monitoring). These tests indicate the extent to which the trials are sufficiently alike to permit them to be combined with confidence, and the possibility that various unknown factors are present that have not been adequately controlled for.

First-Line Tocolytics

Table 13. Trial arms used in the meta-analysis of first-line (more...)

Table 13. Trial arms used in the meta-analysis of first-line tocolytics: Odds ratio for delivery at term (at or beyond 36 to 38 weeks gestation)

Study	Placebo	Beta- Mimetics	Calcium Channel Blockers	Magnesium Sulfate	NSAIDs	Ethanol
Merkatz, et al.,136 1980	Yes	Yes	No	No	No	No
Larsen, et al.,121 1986	Yes	Yes	No	No	No	No
Leveno, et al.,151 1986	Yes	Yes	No	No	No	No
Guinn, et al.,133 1997	Yes	Yes	No	No	No	No
Meyer, et al.,126 1990	No	Yes	Yes	No	No	No
Kupferminc, et al.,130 1993	No	Yes	Yes	No	No	No
Jannet, et al.,134 1997	No	Yes	Yes	No	No	No
Papatsonis, et al.,159 1997	No	Yes	Yes	No	No	No
Garcia-Velasco and Gonzalez,153 1998	No	Yes	Yes	No	No	No
Wilkins, et al.,122 1988	No	Yes	No	Yes	No	No
Kurki, et al.,129 1991	No	Yes	No	No	Yes	No
Castren, et al.,144 1975	Yes	Yes	No	No	No	Yes
Lauersen, et al.,117 1977	No	Yes	No	No	No	Yes
Caritis, et al.,148 1982	No	Yes	No	No	No	Yes
Glock and Morales,162 1993	No	No	Yes	Yes	No	No
Floyd, et al.,163 1995	No	No	Yes	Yes	No	No

Table 14. Effect of first-line tocolytics: Odds of (more...)

Table 14. Effect of first-line tocolytics: Odds of delivery at term (at or beyond 36 to 38 weeks gestation)

Treatment	Estimated Odds Ratio	Standard Error of Log	95% Confidence Interval
Beta-mimetics	1.622	0.147	1.216, 2.164
Calcium channel blockers	2.485	0.237	1.562, 3.954
Magnesium sulfate	1.867	0.231	1.187, 2.936
NSAIDs	4.948	0.562	1.645, 14.89
Ethanol	0.949	0.263	0.567, 1.589

Figure 21. Meta-analysis results: Effect of first-line (more...)

   Figure 21. Meta-analysis results: Effect of first-line tocolytics on the odds of delivery at term (at or beyond 36 to 38 weeks gestation), by tocolytic class

Figure 21

We conducted an overall test for goodness of fit, which gave a chi-square of 11.476 for 13 degrees of freedom, p=0.571. Thus, there was no evidence of lack of fit. We also conducted a test to see if beta-mimetics, calcium channel blockers, and magnesium sulfate had the same effects. This calculation gave an overall chi-square test of 8.762 for 2 degrees of freedom, p=0.0125. Thus, there is evidence that these three tocolytics may have a differential effect from each other relative to placebo in terms of increasing the number of births at or beyond 36 to 38 weeks gestation.

Maintenance Tocolytics

Table 15. Studies of the effects of tocolytics as maintenance therapies (more...)

Table 15. Studies of the effects of tocolytics as maintenance therapies

Study	Placebo	Beta- Mimetics	Magnesium Sulfate
Creasy, et al.,44 1980	Yes	Yes	No
Brown and Tejani,172 1981	Yes	Yes	No
Ricci, et al.,180 1991	Yes	Yes	Yes
Parilla, et al.,175 1993	Yes	Yes	No
How, et al.,176 1994	Yes	Yes	No
Lewis, et al.,177 1996	Yes	Yes	No
Rust, et al.,178 1996	Yes	Yes	Yes
Guinn, et al.,179 1998	Yes	Yes	No

Table 16. Effect of maintenance tocolytics on pregnancy (more...)

Table 16. Effect of maintenance tocolytics on pregnancy outcomes

Treatment	Endpoint (units)	Estimated Difference	95% Confidence Interval
Beta-mimetics	Prolongation (days) Gestational age (weeks) Birth weight (grams)	-0.68 -0.46 -16.4	-4.68, 3,33 -1.03, 0.11 -189.0, 155.3
Magnesium sulfate	Prolongation (days) Gestational age (weeks) Birth weight (grams)	-3.31 -0.55 -202.8	-9.67, 3.06 -1.38, 0.28 -405.7, 0.1

Figure 22. Meta-analysis results: Effect of maintenance (more...)

   Figure 22. Meta-analysis results: Effect of maintenance beta-mimetics on mean gestational age relative to placebo

Figure 23. Meta-analysis results: Effect of maintenance (more...)

   Figure 23. Meta-analysis results: Effect of maintenance magnesium sulfate on mean gestational age relative to placebo

Figure 24. Meta-analysis: Effect of maintenance beta-mimetics (more...)

   Figure 24. Meta-analysis: Effect of maintenance beta-mimetics on prolongation of pregnancy relative to placebo

Figure 25. Meta-analysis results: Effect of maintenance (more...)

   Figure 25. Meta-analysis results: Effect of maintenance magnesium sulfate on prolongation of pregnancy relative to placebo

Figure 26. Meta-analysis results: Effect of maintenance beta-mimetics (more...)

   Figure 26. Meta-analysis results: Effect of maintenance beta-mimetics on birth weight relative to placebo

Figure 27. Meta-analysis results: Effect of maintenance (more...)

   Figure 27. Meta-analysis results: Effect of maintenance magnesium sulfate on birth weight relative to placebo

21

26

Table 17. Homogeneity of effects of maintenance tocolytics (more...)

Table 17. Homogeneity of effects of maintenance tocolytics on pregnancy outcomes

Treatment	Endpoint (Units)	Chi-square for Homogeneity	Degrees of Freedom	p-Value
Beta-mimetics	Prolongation (days) Gestational age (weeks) Birth weight (grams)	7.57 0.67 46.27	6 5 7	0.272 0.985 <0.001
Magnesium sulfate	Prolongation (days) Gestational age (weeks) Birth weight (grams)	0.43 0.58 0.07	1 1 1	0.512 0.445 0.790

Adjunctive Antibiotics

Table 18. Effect of antibiotics on prolongation of (more...)

Table 18. Effect of antibiotics on prolongation of pregnancy (measured in days)

Study	Estimated Prolongation (Days)	Standard Error	95% Confidence Interval
Morales, et al.,198 1988	13.55	3.39	6.898, 20.20
Nadiasauskiene, et al.,191 1996	10.50	6.91	-3.067, 24.07
McGregor, et al.,199 1991	9.90	4.44	1.174, 18.63
McCaul, et al.,203 1992	0.50	8.53	-16.34, 17.34
Newton, et al.,195 1989	0.10	4.73	-9.200, 9.400
Gordon, et al.,193 1995	-0.10	4.29	-8.527, 8.327
COMBINED	6.42	2.63	1.266, 11.56

Table 20. Effect of antibiotics on birth weight (measured (more...)

Table 20. Effect of antibiotics on birth weight (measured in kilograms)

Study	Increase in Birth Weight (kg)	Standard Error	95% Confidence Interval
Svare, et al.,192 1997	0.292	0.170	-0.04241, 0.6264
Morales, et al.,198 1988	0.288	0.128	0.0375, 0.5385
Norman, et al.,200 1994	0.225	0.145	-0.060, 0.5104
Cox, et al.,197 1996	0.169	0.176	-0.1739, 0.5119
McGregor, et al.,1991991	0.145	0.135	-0.1195, 0.4099
Newton, et al.,1961991	0.135	0.181	-0.2174, 0.4874
Nadisauskiene and Bergstrom201 1996	0.056	0.104	-0.1476, 0.2596
Gordon, et al.,1931995	0.026	0.124	-0.2177, 0.2697
Newton, et al.,1951989	0.008	0.134	-0.2547, 0.2707
McCaul, et al.,2031992	-0.059	0.247	-0.5481, 0.4295
Oyarzun, et al.,2111998	-0.063	0.108	-0.2748, 0.1488
Romero, et al.,1941993	-0.148	0.092	-0.3287, 0.0327
COMBINED	0.068	0.045	-0.01938, 0.1558

Figure 28. Meta-analysis: Effect of adjunctive antibiotics (more...)

   Figure 28. Meta-analysis: Effect of adjunctive antibiotics on prolongation of pregnancy relative to no antibiotics

Figure 29. Meta-analysis results: Effect of adjunctive (more...)

   Figure 29. Meta-analysis results: Effect of adjunctive antibiotics on estimated gestation age at birth relative to no antibiotics

Figure 30. Meta-analysis results: Effect of adjunctive antibiotics (more...)

   Figure 30. Meta-analysis results: Effect of adjunctive antibiotics on birth weight relative to no antibiotics

28

30

Table 19. Effect of antibiotics on estimated gestational (more...)

Table 19. Effect of antibiotics on estimated gestational age at birth (measured in weeks)

Study	Increase in Gestation Age (Weeks)	Standard Error	95% Confidence Interval
Nadiasaukiene, et al.,191 1996	2.80	1.02	0.7942, 4.806
Norman, et al.,200 1994	1.30	0.82	-0.3110, 2.911
Newton, et al.,196 1991	1.00	0.90	-0.7581, 2.758
Morales, et al.,198 1988	0.95	0.41	0.1380, 1.762
McGregor, et al.,199 1991	0.50	0.66	-0.8040, 1.804
Gordon, et al.,193 1995	0.10	0.55	-0.9718, 1.172
Cox, et al.,197 1996	0.10	0.95	-1.765, 1.965
Newton, et al.,195 1989	0.00	0.60	-1.174, 1.174
McCaul, et al.,203 1992	-1.10	1.01	-3.093, 0.8931
COMBINED	0.59	0.24	0.1198, 1.057

Home Uterine Activity Monitoring

Table 21. Effect of home uterine activity monitoring (more...)

Table 21. Effect of home uterine activity monitoring on pregnancy outcomes

Endpoint (Units)	Estimated Difference (treated - control)	95% Confidence Interval
Gestational age (weeks)	0.19	-0.37, 0.76
Survival to 37 weeks (odds ratio)	1.22	0.73, 2.05
Birth weight (grams)	-47.0	-201.1, 107.1

Figure 31. Meta-analysis results: Effect of home uterine (more...)

   Figure 31. Meta-analysis results: Effect of home uterine activity monitoring on the odds on the odds of delivery at 37 or more weeks gestation relative to no monitoring

Figure 32. Meta-analysis results: Effect of home uterine (more...)

   Figure 32. Meta-analysis results: Effect of home uterine activity monitoring on gestational age relative to no monitoring

Figure 33. Meta-analysis results: Effect of home uterine (more...)

   Figure 33. Meta-analysis results: Effect of home uterine activity monitoring on birth weight relative to monitoring

31

33

Table 22. Homogeneity of effects of home uterine activity (more...)

Table 22. Homogeneity of effects of home uterine activity monitoring on pregnancy outcomes

Endpoint (units)	Chi-square for Homogeneity	Degrees of Freedom	p-Value
Gestational age (weeks)	0.83	2	0.660
Survival to 37 weeks (odds ratio)	0.44	1	0.508
Birth weight (grams)	1.88	2	0.391

Benefits Data Grades

We assign four basic grades for benefits. The first three grades indicate that the quantity of data is sufficient to make a judgment and that the overall data are good, fair, or poor in terms of supporting the conclusion that the intervention (e.g., pharmaceutical agent or biologic marker) is better than an alternative intervention or placebo. The fourth grade indicates that the quantity and quality of data are insufficient to draw any conclusions, chiefly because of small sample sizes or poor methods in the studies reviewed. The specific category definitions are as follows:

Harms Data Grades

We assigned two categorical ratings to the overall harms data reported for tocolytics in Chapter 3, "low" or "high." The rating reflects our judgment of the probability of risk from the intervention in question. The specific category definitions are as follows:

Low probability of risk:

- The side effects of treatment with the specific tocolytic that we noted from our review of articles were short term and not life-threatening.

- Considering the body of evidence, the frequency of occurrence was not substantially different from the frequency for an alternative tocolytic treatment or no treatment.

"High" probability of risk:

- The side effects of treatment with the specific tocolytic that we noted from our review of articles were life-threatening.

- Considering the body of evidence, the frequency of occurrence was substantially different from the frequency for an alternative tocolytic treatment or no treatment.

First-Line Tocolytic Therapy

The largest body of literature examined for this review concerned treatment with tocolytic agents for acute episodes of preterm labor: The tocolytics discussed included beta-mimetics, calcium channel blockers, magnesium sulfate, NSAIDs, and ethanol. We included both efficacy and effectiveness studies; research compared treatment with a particular class of tocolytic with a control group as well as with a different class of tocolytic drug. Overall, we assigned a quality grade to the first-line tocolytic literature of Fair (B-1), having determined that the data were sufficient for evaluating the quality of the findings and that the evidence comes from at least one well-designed RCT.

Literature comparing an intervention group of women receiving first-line treatment with tocolytics relative to a control group were available only for beta-mimetics, magnesium sulfate, and ethanol. Results were mixed: Some evidence of efficacy was shown in terms of estimated gestational age at delivery and prolongation of pregnancy with beta-mimetic or magnesium sulfate treatment. Only one beta-mimetic study ¹³⁶ found a relationship between treatment and improved infant outcomes, showing significantly higher birth weight in the treatment arm. Data from the ethanol study ¹⁴⁴ were insufficient to evaluate efficacy.

The largest body of literature reviewed compared the outcomes of treating women with beta-mimetics with the outcomes of treatment with other classes of tocolytics. Results again were mixed: Beta-mimetics were shown to have efficacy only relative to ethanol treatment in relation to prolonging pregnancy. ¹⁴⁸ Other RCTs showed that all other classes of tocolytics had greater or not significantly different effects relative to beta-mimetics on birth outcomes. We also found mixed results for infant outcomes, but no study reported that beta-mimetics were a superior treatment. No other differences in efficacy were found between classes of tocolytics, and data from observational studies did not contradict the results of the RCTs.

Supplementing our qualitative review, some of the RCT data available from the individual studies were combined using meta-analytic techniques (Chapter 4). This methodology allowed us to compare all classes of tocolytics (i.e., beta-mimetics, calcium channel blockers, magnesium sulfate, NSAIDs) with a no-treatment group. The data suggest that all tocolytics, with the exception of ethanol, were effective in extending births to term (at or beyond 36 to 38 weeks gestation) compared with results in a no-treatment group. Evidence shows that there may be effect differences among beta-mimetics, calcium channel blockers, and magnesium sulfate; but more studies are needed for conclusive findings.

In sum, although inconsistencies exist in these data, overall the evidence supports the notion that treatment offers small improvements in pregnancy prolongation. If necessary, this may provide adequate time for administering corticosteroids for fetal lung development. Sufficient data were not available to determine directly whether this treatment has a directly beneficial effect on neonatal morbidity and mortality. Data concerning relative efficacy between drug classes are generally inconclusive, but they clearly support the conclusion that ethanol is less efficacious than other tocolytic options.

Maintenance Tocolytic Therapy

With the exception of one small study, ¹⁷² the efficacy studies of maintenance tocolytic literature showed no difference between treatment and control arms in managing women who had recently experienced an episode of preterm labor. These results were confirmed through our meta-analytic synthesis of results (Chapter 4 ). In terms of gestational age at birth, prolongation of pregnancy, or birth weight, no benefits from maintenance treatment were uncovered.

We graded this literature as Poor (C-1) in terms of showing efficacy or effectiveness of maintenance tocolytics. The data were sufficient for evaluating the quality of the literature, and no evidence supported the use of this therapeutic modality relative to no treatment.

Potential Harms From Tocolytic Therapy

The harms analysis examined tocolytic treatments for different classes of pharmaceuticals, but we did not distinguish between first-line and maintenance therapies. We assessed adverse events for both the mother and the fetus or neonate.

We graded beta-mimetics as "high" in probability of maternal risk. These drugs were shown to pose a risk to the mother of serious cardiovascular harms, minor cardiovascular harms, metabolic harms, and psychologic harms. In comparison, all other classes of tocolytic treatments were graded as "low" in relation to maternal risk. Significant levels of serious maternal complications were not shown.

We graded all classes of tocolytics as "low" risk in relation to fetal or neonatal harms. Evidence of short-term harms was inconsistent, and evidence to evaluate longer term problems was insufficient.

Additional Comments Concerning Tocolytic Therapy

The evidence from our analysis supports the generally held clinical belief that ethanol is an inappropriate treatment for women with preterm labor symptoms. Other treatments offer greater efficacy at less risk. Similar concerns would seem to be warranted with regard to treatment with beta-mimetics. Although this judgment may be an artifact of the relative wealth of data on this class of tocolytics relative to others, the benefits of beta-mimetics were never found to exceed other options, and the harms were shown to be potentially more severe.

Definitional Issues

Depending on the topic and the release date of the pharmaceuticals, we reviewed a literature base that could date back more than three decades. In that time, the standards for research, including the conduct of RCTs as well as case control, longitudinal cohort, and similar effectiveness studies, have risen appreciably.

Nonetheless, in our view, a good deal of work needs to be done to shore up critical building blocks of research in this area. In particular, researchers need to attend to three issues:

• 1

  Define preterm labor explicitly and carefully.

• 2

  Be clear about which cointerventions were used and when were they administered during the study.

• 3

  Separately analyze outcomes for those women with medically indicated preterm births and compare the results with those for the larger group of women with preterm labor.

Outcome Measures

Perhaps the most important methodologic issue is that of measuring outcomes in trials or other studies of interventions for women in preterm labor. Our work has documented the extent to which researchers differ in their definitions of even standard outcomes. This matter is especially important if regulators (e.g., the FDA) or health plan administrators insist that an intervention must improve perinatal outcomes before it can be approved or adopted (as the FDA did with the oxytocin-agonist atosiban). However, any trial using perinatal endpoints would have to be quite large and expensive because progress in the care of tiny newborns has resulted in a relative rarity of poor perinatal outcomes; indeed, the numbers and expense of such trials are likely to be prohibitive.

Insistence on improvements in perinatal outcome is epidemiologically sound. Therefore, investigators and policymakers are likely to continue to rely on surrogates relating to delivery such as pregnancy prolongation or the number of preterm births before an arbitrary cutoff.

One way to address this problem is to improve the analysis of available data. As an answer to this problem, we have already noted the utility of survival analysis, which could provide continuous delivery outcomes stratified by gestational age at enrollment in the study. Such an approach is clinically and biologically relevant. It allows both clinicians and the public to decide whether an intervention's effects were relevant to an individual's particular situation, meaning that researchers can begin to clarify whether (and if so, to what extent) even a small delay in delivery at a particular gestational age has the potential to be significant.

Our recommendation, therefore, is that investigators, to the extent possible, supplement the use of categorical outcomes measures (e.g., preterm or term birth) and variables involving specific cutoff points for gestational age by including concepts such as the length of time a pregnancy was extended-prolongation-as a form of survival of the pregnancy. That is, we advocate for the assessment of the impact of an intervention on pregnancy prolongation through the use of survival analysis stratified by gestational age at the onset of preterm labor.

Implicit in this recommendation is an assumption that prolongation will improve perinatal outcome. This assumption has intuitive appeal given that the correlation between gestational age at birth and improved perinatal mortality and morbidity is strongly positive; however, no one knows the advantages or disadvantages of delaying delivery for an individual once preterm labor is established. Perhaps the in utero environment of the mother who is experiencing preterm labor is hostile and prolongation does more harm than good. Thus, the ultimate test for any therapeutic intervention will always be improvements in perinatal outcome.

Given the rarity of poor perinatal outcomes in an age of effective neonatal care, the need exists for composite perinatal outcomes that reflect the morbidity and mortality of premature birth. Potential measures that could be combined include perinatal death, neurologic deficit, and oxygen requirement at hospital discharge. Perinatologists could advance the field in important ways by constructing composite markers that accurately and consistently reflect the burden of being born before term. Such an approach has been useful in other fields such as cardiology. A valid composite measure could then become the outcome of choice in intervention trials.

Moreover, long-term followup of neonates is needed. Certain conditions that could be a direct harm from an intervention or a byproduct of pregnancy prolongation in a hostile environment may not become obvious until an infant matures. Followup of particular importance regarding developmental and neurologic outcomes as the fetal and neonatal brain has yet to complete development at the time of birth and hospital discharge.

Epidemiology

First, the reader of this report will appreciate the need for basic and epidemiologic research in preterm labor. In our judgment, far more work needs to be done in this area just to lay a better framework for future intervention-specific efficacy and effectiveness research. Said another way: a field of knowledge so primitive that the patient's complaints (i.e., preterm labor symptoms) are used as the diagnostic criterion for the condition possessing the same name must invest in basic and epidemiologic research.

We lack basic knowledge regarding the biologic mechanisms resulting in birth before term. Moreover, there is a dearth of knowledge regarding the incidence and prevalence of preterm symptoms and modifiable risk factors for preterm birth. The issue of etiologic heterogeneity versus homogeneity for preterm birth is especially poorly understood.

We also draw attention to the need for better understanding of these basic facts in ethnic minorities, for at least two reasons. First, the demographics of this nation are changing rapidly, including among women of child-bearing ages. Second, clinicians should have better information at hand to address the needs of women who may have low English literacy skills, and who may thus be at a disadvantage in terms of communicating their symptoms or understanding the medical management regimens being recommended to them.

Finally, African American women and their offspring face profound disadvantages in terms of preterm birth. The discrepancy between perinatal outcomes in black and white women and the mechanisms contributing to such disparities must be investigated, so that clinicians and the public gain a better understanding of the epidemiologic, social, and other causal and contributing factors.

Also, the best type of clinician and optimal setting for care of the woman with preterm labor remains to be answered. Given the ubiquitous nature of these symptoms and our inability to assess risks, the minimum level of expertise and capacity needed to triage these patients effectively is not known.

Preterm labor with subsequent preterm birth has remained an enigma to clinicians and investigators. The condition has proven remarkably resistant to the inquiry of modern science. No satisfactory "meta-narrative" or paradigm has been constructed that reconciles all the observed phenomena. From the perspective of the authors of the evidence report, the failure to solve the muddled picture is because of faulty science and inadequate resources for investigation. By implication, better science and more resources to invest in efficacy and effectiveness studies will help to end the enigma.

Conversely, preterm labor and subsequent preterm birth may be the culmination of a complex, biopsychosocial process such as poverty or obesity. A narrow scientific focus on the clinical endpoint of preterm labor and relentless pursuit of a meta-narrative threaten to direct attention and resources from the changes needed in society that might ultimately result in substantial reductions in preterm birth rates.

Risk Factors and Biologic Markers

Additional study regarding risk factors is important, and this issue bridges the epidemiologic and clinical realms of research. Clinicians and researchers need to be able to assess accurately the risk of preterm birth in women with symptoms of preterm labor and to exclude women at low risk of early birth from intervention trials. Failure to assess risk will not only dilute the effects of the intervention in question, but also expose numerous women who are at low risk to unnecessary and potentially harmful intervention. Some investigators have tried to assess risk by constructing strict definitions of preterm labor (e.g., by incorporating a measure of contraction frequency and duration with measures obtained from digital examination of the cervix).

Biologic markers may enhance this goal because they are objective and reproducible and are superior to purely clinical assessment or prognostication of preterm birth. At least two of the biologic markers we reviewed-fFN and EVUSD-can serve in this role. Thus, we recommend that investigators studying tocolytics and antibiotics (in their capacity as part of a medical regimen for women in or suspected of preterm labor) use biologic marker negativity as an exclusion criterion.

Moreover, we advise that further clinical trials be conducted to demonstrate whether the enhanced knowledge and precision provided by biologic markers with respect to preterm birth risk actually can improve perinatal outcomes. Trials should be conducted in which clinicians are provided with biologic marker information on a subset of participants and not on others, so as to clarify the level of efficacy and usefulness of this information in "ideal" settings. This step is fundamental before the use of biologic markers is incorporated into routine care in "average" settings, so as to avoid dilemmas in which better measurement precision does not improve outcomes (as has been seen with electronic fetal monitoring).

Pharmaceutical Agents

Our meta-analytic results for the first-line tocolytics and for antibiotics (reported in Chapter 4) suggest that further trials of these modalities are needed. Use of antibiotics is almost universally recommended in women with preterm labor symptoms as a form of prophylaxis to prevent early-onset group B streptococcal sepsis. By contrast, the use of acute tocolysis has been disparaged by many. Well-designed trials of adequate size with appropriate outcome measures of acute tocolysis to prevent preterm birth in women with symptoms of preterm labor would be helpful and reasonable. We would call attention, in particular, to the likely benefits of a focus on beta-mimetics, calcium channel blockers, and magnesium sulfate.

Review of our evidence tables highlights not only the numerous agents used for tocolysis, but also the wide range of dosages prescribed. Clinicians and patients could potentially benefit from pharmacologic studies to assess optimal dosing and duration of therapy. Overdosing could lead to excessive harms from tocolysis, whereas underdosing could result in an inability to optimize outcomes. The same arguments and limitations apply to the duration of tocolytic dosing. Although we did not find evidence to support the use of so-called maintenance tocolysis once the acute preterm labor phase is over, the optimal duration for acute tocolysis is unknown.

Our report emphasizes the small, potential benefits of acute tocolysis-in terms of pregnancy prolongation-in the face of substantial potential maternal harms. This marginal benefits-to-risk ratio argues for continued development and testing of new tocolytic drugs. Clinicians and patients would benefit from agents that are selective for uterine smooth muscle quiescence. The oxytocin-agonist atosiban is an example of a novel tocolytic drug that met utero-selective criteria. Its failure to improve perinatal outcome kept the FDA from approving it as a tocolytic, but many authorities predict approval within the European community because of its ability to prolong pregnancy and lack of side effects. Should the European experience with atosiban be favorable, perhaps approval of this agent as a tocolytic should be reconsidered in the United States. Moreover, other oxytocin antagonists and novel classes of compounds with the ability to quiet the uterus should be studied and tested.

The studies we reviewed provided virtually no information on the quality of the mother's life while receiving treatment-other than in relation to physical harm. Data on lost wages or lost days of work would support cost-benefit analyses of treatment effectiveness. In addition, information on the psychological impacts of treatment and the treatment process could also provide valuable insights.

The best type of clinician and the optimal care setting of a woman with preterm labor remains to be answered. Given the ubiquitous nature of these symptoms and the inability to adequately assess risks, clinicians and institutions caring for these women need to be able to triage these patients. However, whether this needs to be at a tertiary perinatal center or a lower level of care is a topic for future study.

If the optimal class of drug, dose, and duration of tocolytic therapy is obfuscated, the evidence to guide clinicians' use of antibiotics is even more obscure. Given the array of microbicidal and static agents in the full evidence tables, we cannot delineate a rational approach to the selection and duration of such therapies. At present, it seems prudent in light of the marginal benefits of antibiotics to follow protocols for prophylaxis against early-onset Group B streptococcal sepsis with the expectation that such interventions will not only accomplish that goal, but also prolong pregnancy. Investigators should focus their research on determining the optimal agent, dose, duration, and possible combinations for antibiotics in women presenting with preterm labor.

Areas Not Warranting Further Research

Our review casts doubts on the wisdom of future research in three areas. The first is maintenance tocolysis, where we saw from both the synthesis of the literature and the meta-analyses that tocolytics used in a maintenance modality (as contrasted with first-line interventions) offered no demonstrable benefit in preventing preterm birth. Second, it seems clear that no further work need ever be done on ethanol as either a first-line or maintenance tocolytic. The third area is home uterine activity monitoring, where we were unable to find even indirect evidence that using this device alone is useful in preventing preterm birth.

Technical Expert Advisory Group

We gratefully acknowledge the substantial involvement of and assistance from the Technical Expert Advisory Group (TEAG). TEAG members are listed at the end of this section. The TEAG was meant in part to contribute to (a) advancing AHRQ's broader goals of creating and maintaining "science partnerships" and "public-private partnerships" and (b) meeting the needs of a broad array of potential customers and users of its products. Thus, it was both a substantive resource and a sounding board throughout the study, and it is the body from whom "expert inputs" were formally sought at several points through the project.

We constituted our TEAG from three types of technical experts and other partners. These types are (1) technical/clinical experts; (2) patients or representatives of organizations whose mission concerns the interests and perspectives of patients and consumers; and (3) potential users of the final evidence report or other materials, including explicitly a representative of the organization that nominated the topic (in this case, the American College of Obstetrics and Gynecology. All in all, we had six clinical/technical experts, one individual representing patient populations, and one individual representing potential user groups, for a total of eight on the TEAG.

The final decision about TEAG membership was based on candidates' availability for scheduled conference calls and other input, willingness to review materials and provide advice and assistance within a short turnaround time, and approval by the AHRQ Task Order Officer.

The RTI-UNC Center team solicited the views of TEAG members from the start of the project. Among other issues, TEAG members provided insights and reactions to key clinical questions, input to the literature review process by ensuring that we included all known published research meeting our inclusion criteria, review of our data extraction forms and view of our approach to meta-analysis. TEAG members have also provided valuable input concerning problems of focusing a literature search on treatments, and specifying appropriate outcomes, for a clinical topic as difficult to conceptualize as preterm labor.

In keeping with AHRQ's standards for employing a multidisciplinary approach to the development of evidence reports, we called on our TEAG for inputs at two key points during this task. First, the group was asked to comment on the literature synthesis and to give us feedback on our overall plans at that stage of the analysis, which included approaches to developing evidence tables and to summarizing information about patient outcomes associated with the treatment options being studied for the management of preterm labor.

Peer Reviewers

An important first step in the identification of potential peer reviewers was to determine the appropriate constituencies from which our reviewers should be approached. The categories we finally settled on, and the number of reviewers asked to participate in this effort, are shown in Table A. Although the categories are fairly self-explanatory, we note the following details for clarification. Individual experts engaged in related research (as contrasted with medical practice per se) were included in Category I (clinical practice and health care delivery) because they are based in health care delivery organizations and, we judged, likely to be involved to some extent in patient care. We assigned organizations that appear to be engaged chiefly in information dissemination (clearinghouses and the like) to Category IV (quality-of-care and consumer groups) on the grounds that much quality of care activity today assumes some effort to provide consumer information. We believe that these four categories represent the full range of health care experts, users, and patient groups that should be involved in reviewing this particular evidence report on the management of preterm labor; the specific peer reviewers are listed at the end of this appendix.

We selected 19 organizations or independent peer reviewers from the four categories noted in Table A. The individuals included the eight members of the TEAG because they played a major role throughout the project in conceptualizing the work and reviewing materials; moreover, as active professionals in the field, the RTI/UNC EPC believed that their comments at this stage would be very valuable. The remainder of the peer reviewer group were identified by issuing an invitation to the organization's executive officer/director (e.g., President, CEO) or to a public sector agency head asking them to nominate a peer reviewer or by soliciting nominations from the TEAG. A preliminary (and longer) list of organizations, agencies, or individuals was submitted to the AHRQ Task Order Officer for this project for review, comment, and approval. We then contacted all potential peer reviewers to determine their willingness to serve as peer reviewers, alerting them to the fact that this service would require them to prepare formal written reviews according to the checklist developed for this evidence report. Their comments and suggestions formed the basis of our revisions to the evidence report.

The peer reviewers who were selected for this Evidence Report are predominantly involved in clinical practice or research, professional associations, and public sector quality of care organizations.

Article Identification, Eligibility Screen & Abstract:BIOMARKERS FOR PRETERM LABOR

Initials of Reviewer __ __ __ Date__ __/__ __ (Mo/Day) Unique Article Identifier __ __ __

Screen

(If any response is "No" then STOP. If any response is "Can't Tell," bring to the attention of the Research Coordinator.

Preterm labor, preterm birth or time to delivery an outcome?	Yes	No	Can't Tell
40 or more subjects at completion?	Yes	No	Can't Tell
Population of women with signs/symptoms of PTL?	Yes	No	Can't Tell

Abstract

(Note that numeric values in many instances need to be preceded by <, > , < or >.)

1. Main Objectives (described by author, typically in the introduction).
_____________________________________________________________________________
_____________________________________________________________________________

2. Study Design. (Record as stated by the author. Typically found in the abstract, end of the introduction or beginning of the methods section. If you think the author has misstated the study design, record the discrepancy in the comments section.)

Diagnostic;

Randomized controlled trial

Non-randomized controlled trial

Prospective cohort study

Retrospective cohort study

Case-control study

Cross-over study

Case series

Cross-sectional survey

Cost-benefit/cost-effectiveness study

Other (describe in space to right) _________________________________

3. Subanalysis. Are there any subgroup or stratified analyses in this article?

Yes No

4. Country in which study was conducted: ______________________________
STUDY POPULATION

5. Inclusion Criteria

Gestational age: __________ to __________ completed weeks

Intact membranes

Fetal weight estimate: __________

Maternal age: __________ (lower limit) __________ (upper limit)

Singleton gestation

Multiple gestation

Signs and symptoms of active labor

High risk of preterm labor/delivery

If yes, how was "high risk" defined? _____________________________________________________

_________________________________________________________________________________

Other: ______________________________

6. Exclusion Criteria

Abnormal fetus	Maternal age __________ (upper limit)
	Maternal age __________ (lower limit)
Antibiotic therapy, current or recent	Maternal fever/Pyrexia: T _____
Cerclage/Need for cerclage	Maternal White Blood Cells (WBC) _____
Chorioamnionitis	Multiple gestation
Chronic disease	No prior prenatal care
Current bleeding/hemorrhage	Placenta previa
Current illegal substance abuse	Rupture of membranes
Dilation __________cm	Singleton gestation
Fetal distress	Uterine anomaly
Failure to give informed consent	Other maternal/fetal indication for delivery
Group B Strep (GBS) Positive	Other Positive Cultures ( ______________ )
Hypertensive disorders (incl. Pre-eclampsia)	Signs/Symptoms of labor
High risk of preterm delivery
If yes, how was "high risk" defined? ________________________________________________
_____________________________________________________________________________
Other: ________________________________________________________________________
Other: ________________________________________________________________________
Other: ________________________________________________________________________
Other: ________________________________________________________________________

7. Gestational age assignment. (Circle either and/or as appropriate.)

Not determined/not measured/can't tell

Last Menstrual Period (LMP)

And/or clinical sizing, including serial fundal heights

And/or ultrasound When was ultrasound conducted? _________________________________

8. Population Size:

Screened __________ Cannot Determine

Eligible __________ Cannot Determine

Enrolled __________ Cannot Determine

Completed __________ Cannot Determine

Included in analysis __________ Cannot Determine

8a. Reasons for exclusion from analysis:

Loss to follow-up (n= )

Other ____________________ (n= )

Other ____________________ (n= )

Other ____________________ (n= )

Other ____________________ (n= )

9. MASKING LEVEL

Were treating physicians masked to the results of the diagnostic tests?	Yes	No	Can't Tell
Were patients masked to the results of the diagnostic tests?	Yes	No	Can't Tell

Definition of Diagnostic Test

10. What screening test or tests was/were used?

10a. How are the test result groups compared in the analysis defined (e.g., positive fFN test; positive fFN test + some cervical length)?

11. Diagnostic testing

11a. Diagnostic Test 1
When initiated:
Frequency Performed:
Did the result of the test influence the frequency with which it was subsequently performed?
Yes No Can't Tell

Diagnostic Test 2
When initiated:
Frequency Performed:
Did the result of the test influence the frequency with which it was subsequently performed?
Yes No Can't Tell

12. If cervical length was measured, how was it measured?

13. Was wedging or funneling or breaking considered?

14. How were they defined?
Wedging: ________________________________________________
Funneling: _______________________________________________
Breaking: ________________________________________________

15. Non-Pharmaceutical Co-Interventions.

15a. Under what circumstances were non-pharmaceutical co-interventions administered?

15b. Which ones were used?

Bedrest	Routine Observation
Fetal Monitoring	Uterine Monitoring
IV Fluids
Other:______________________________________________________________________________

15c. Were non-pharmaceutical co-interventions administered in both the test positive and test negative group?
Yes No
If not, in which group were they administered? ___________________________________

16. Pharmaceutical Co-Interventions.

16a. Under what circumstances were co-interventions administered?

16b. What pharmaceutical co-interventions were administered?
tocolysis
specify drug _______________
antibiotics
specify drug _______________
other _______________________

16c. Were pharmaceutical co-interventions administered in the same fashion in the test positive and test negative groups?
Yes No

17. What evaluations other than the test being studied (i.e., fFN or E3) were performed?
a. Periodic pelvic examination _____ frequency
specify by speculum or digital cervical evaluation __________________
b. Periodic ultrasound _____ frequency
c. Cultures for infection
d. HUAM _____ frequency
e. Others _____________ _____ frequency
_____________ _____ frequency

18. Participant Characteristics. (Record as number and/or percent as presented in article. Follow percent with %.)

19. OUTCOME MEASURES.

19a. Definition of Preterm Labor. (Record each definition separately. Provide the letters corresponding to each symptom. Fill in any necessary data describing a particular symptom.)
Can't tell
Def. 1: ____________________
Def. 2: ____________________
Def. 3: ____________________
a. Contractions: __________/__________
b. History of contractions
c. Dilation: __________ cm
d. Change in dilation from baseline
e. Effacement: __________%
f. Cervical change
g. Length of cervix: __________ cm
h. Contractions unresolved with: __________
i. Other: ______________________________
j. Other: ______________________________

19b. Rates of outcomes
Specify mean or median. Numeric values may be preceded by <, >, <, or >. Record as number and/or percent as presented in article. Follow percent with "%". In the last column, if a difference between the groups is statistically significant, record the relevant p value. If it is not significant, state, "n.s." If statistical adjustments were performed (i.e., stratified analyses or multivariate modeling) report the adjusted results, the strata, and any variable used for adjustment in #.

20. Sensitvity/Specificity Analysis for Test ____________ for Outcome: _________________________
NOTE: You may use as many of these pages as is necessary to capture the information provided by test/outcome. The table below is designed to capture rates associated with a given test and outcome by the groups into which participants were divided for analysis (i.e., those defined in question 11). If results are provided for different outcomes, use a different page for each outcome.

20b. Were these results based on one-time testing or repeated testing?

20c. Numbers provided for sensitivity/specificity calculations:
Overall:

Group A:

Group B:

Group C:

Group D:

21. Were ROC curves presented? Yes No If so, on what page? _____

22. Describe any multivariate analysis performed. Include statistical technique, model description, and result highlights.

23. Was cost data collected?
Yes No

24. Was information on harms collected?
Yes No

25. Describe any additional analyses and/or subgroup analyses:

26. Limitations. (Distinguish between limitations stated by the author and those found by the reviewer.)
a) State limitations noted by the author.
b) State any other limitations you found in the article.

27. Conclusions:
Test effective at predicting the following outcome: _________________________________________
_________________________________________________________________________________
Test effective at predicting relevant outcomes in the following subgroup(s)
_________________________________________________________________________________
_________________________________________________________________________________

28. Comments. (Provide any additional information that you believe needs to be captured concerning this study.)

Article Identification, Eligibility Screen & Abstract: TOCOLYTIC USE FOR PRETERM LABOR

Initials of Reviewer __ __ __ Date __ __/__ __ Unique Article Identifier __ __ __

SCREEN

(If any response is "No" then STOP. If any response is "Can't Tell," bring to the attention of the Research Coordinator.

Preterm birth an outcome	Yes	No	Can't Tell
40 or more subjects at completion?	Yes	No	Can't Tell
Intervention used as part of treating subjects with signs/symptoms of preterm labor (not a preventive intervention used among asymptomatic patients)	Yes	No	Can't Tell

ABSTRACT

(Note that numeric values in many instances need to be preceded by <, >, < or >.)

1. Main Objectives (described by author, typically in the introduction).
_____________________________________________________________________________
_____________________________________________________________________________
_____________________________________________________________________________

2. Study Design. (Record as stated by the author. Typically found in the abstract, end of the introduction or beginning of the methods section. If you think the author has misstated the study design, record the discrepancy in the comments section.)

Randomized controlled trial

Non-randomized controlled trial

Prospective cohort study

Retrospective cohort study

Case-control study

Cross-over study

Case series

Cross-sectional survey

Cost-benefit/cost-effectiveness study

Other (describe in space to right) _________________________________

3. PPROM. Are women with preterm premature rupture of membrane included in this analysis as an entry criteria?
Yes No

4. Subanalysis. Are there any subgroup analyses in this article?
Yes No

5. Country in which study was conducted: ______________________________

6. Definition of Preterm Labor. (Record each definition separately. Provide the letters corresponding to each symptom. Fill in any necessary data describing a particular symptom.)
Can't tell
Def. 1: ____________________
Def. 2: ____________________
Def. 3: ____________________
c. Bio-marker(s) used (list: ______________________________)
d. Contractions: __________/__________
e. History of contractions
f. Dilation: __________ cm
g. Change in dilation from baseline
h. Effacement: __________%
i. Cervical change
j. Length of cervix: __________ cm
k. Contractions unresolved with: __________
l. Other: ______________________________
m. Other: ______________________________

7. Inclusion Criteria (Record criteria other than was stated within definition of preterm labor)
Gestational age: __________ to __________ completed weeks
Intact membranes
Fetal weight estimate: __________
Maternal age: __________ (lower limit)
__________ (upper limit)
Singleton gestation
Other: ______________________________
Other: ______________________________

8. Exclusion Criteria

Abnormal fetus	Maternal age __________ (upper limit)
Allergy to protocol drug	Maternal age __________ (lower limit)
Antibiotic therapy, current or recent	Maternal fever/Pyrexia: T _____
Cerclage/Need for cerclage	Maternal White Blood Cells (WBC) _____
Chorioamnionitis	Multiple gestation
Chronic disease	No prior prenatal care
Current bleeding/hemorrhage	Placenta previa
Current illegal substance abuse	Rupture of membranes
Dilation __________cm	Singleton gestation
Fetal distress	Uterine anomaly
Failure to give informed consent	Other maternal/fetal indication for delivery
Group B Strep (GBS) Positive	Other Positive Cultures ( ______________ )
Hypertensive disorders (incl. Pre-eclampsia)
Other: ________________________________________________________________________
Other: ________________________________________________________________________
Other: ________________________________________________________________________
Other: ________________________________________________________________________

9. Gestational age assignment. (Circle either and/or as appropriate.)
Not determined/not measured/can't tell
Last Menstrual Period (LMP)
And/or clinical sizing, including serial fundal heights
And/or ultrasound When was ultrasound conducted? _________________________________

10. Population Size:
Screened __________ Cannot Determine
Eligible __________ Cannot Determine
Enrolled __________ Cannot Determine
Completed __________ Cannot Determine

11. MASKING LEVEL

Subject masked to treatment assignment	Yes	No	Can't Tell
Treating physician(s) masked to treatment assignment	Yes	No	Can't Tell
Other direct care providers masked to treatment assignment	Yes	No	Can't Tell	N/A
Pharmacist masked to treatment assignment	Yes	No	Can't Tell	N/A
Pathologist masked to treatment assignment	Yes	No	Can't Tell	N/A

12. Costs and cost effectiveness. Is any cost information included in this study?
Yes No

13. Participant Characteristics. (Record as number and/or percent as presented in article. Follow percent with %.)

	Treatment Group A	Treatment Group B	Comparison Group	Total
Age (±SD)
Race:
White
Black
Hispanic
Other
High School Grads
Married

14. Dominant Mechanisms(s) of Loss. (Provide the reason for discontinuation and if available, the number of patients).
None lost
Can't tell
Loss to follow-up __________
Side effects of treatment __________
Other: _________________________________________
Other: _________________________________________
Other: _________________________________________

15. Timing of intervention:
Within _____ hrs of evaluation/admission
Other: ____________________________
Cannot tell

16. Intervention Compliance Assessment. (State what action(s) was taken to ensure that subject completed treatment.)
None reported Mechanism: ______________________________

17. Intention to treat analysis provided?
Yes, list exclusions below and if available, number of subjects.
No
Developed allergy or did not tolerate intervention drug: __________
Positive cultures - urine: __________
Positive cultures - group B strep: __________
Positive cultures - amniotic fluid: __________
Positive cultures - other cervico-vaginal: __________
Did not comply with protocol for use of intervention drug: __________
Did not comply with follow-up: __________
Other: _____________________________________________
Other: _____________________________________________
Other: _____________________________________________
Were exclusions from analysis after Treatment Assignment? Yes No Can't Tell

18. Care Setting(s). List all settings where treatment was provided.
Hospital
Perinatal center
Home
Other: _______________________________
Other: _______________________________

19. Were harms or side effects of treatment included in the study? Yes, _____ page # (s) No

20. Is successful tocolysis defined? Yes No
If yes, define: __________________________________________________________________________
21. Definition of Treatment and Comparison Groups. How are the groups defined?
Treatment Group A: _____________________________________________________________________
Treatment Group B: _____________________________________________________________________
Comparison Group: _____________________________________________________________________

22. TOCOLYTIC DRUG REGIMEN BY GROUPS (If treatment includes a separate loading and maintenance dose of the same drug, i.e., the route and/or duration of treatment changes, list these as different drugs.)

	Treatment Group A	Treatment Group B	Comparison Group
Number of Subjects per group	N =	N =	N =
DRUG 1
Dose
Route
Commencement of Treatment
Duration of Treatment
DRUG 2
Dose
Route
Commencement of Treatment
Duration of Treatment
DRUG 3
Dose
Route
Commencement of Treatment
Duration of Treatment
DRUG 4
Dose
Route
Commencement of Treatment
Duration of Treatment

23. Non-Pharmaceutical Co-Interventions. Did comparison group receive co-interventions? Yes No

Bedrest	Routine Observation
Fetal Monitoring	Uterine Monitoring
IV Fluids
Other: ______________________________________________________________________________

24. CO-INTERVENTION DRUG REGIMEN BY GROUPS

	Treatment Group A	Treatment Group B	Comparison Group
Number of Subjects per group	N =	N =	N =
DRUG 1
Dose
Route
Aim of Treatment
Commencement of Treatment
Duration of Treatment
DRUG 2
Dose
Route
Aim of Treatment
Commencement of Treatment
Duration of Treatment
DRUG 3
Dose
Route
Aim of Treatment
Commencement of Treatment
Duration of Treatment
DRUG 4
Dose
Route
Aim of Treatment
Commencement of Treatment
Duration of Treatment

25a). OUTCOME MEASURES. Specify mean or median. Numeric values may be preceded by <, >, <, or >. Record as number and/or percent as presented in article. Follow percent with "%". In the last column, if a difference between the groups is statistically significant, record the relevant p value. If it is not significant, state, "n.s." If statistical adjustments were performed (i.e., stratified analyses or multivariate modeling) report the adjusted results, the strata, and any variable used for adjustment in #24.

Outcome	Intervention Group A (N = )	Intervention Group B (N = )	Comparison Group (N = )	Sig. Level
DELIVERY
Mean/Median Est. Gestation Age (EGA) at Delivery (SD)
Mean/Median Prolongation (SD) Days
PTL Only Group
Delivered at: __________ weeks
Delivered at: __________ weeks
Delivered at: __________ weeks
Delivered at: __________ weeks
Combined PPROM + PTL
Delivered at: __________ weeks
Delivered at: __________ weeks
Delivered at: __________ weeks
Delivered at: __________ weeks
Delivery at Term
PROM OR PPROM Outcome
MATERNAL OUTCOMES
Maternal Length of Stay (SD)
Chorioamnionitis
Endometritis/Post-partum
Adverse Drug Reactions
INFANT OUTCOMES
Mean/Median Birthweight (SD)
_____ gms
_____ gms
_____ gms
Apgar ______ at ______ min
Apgar ______ at ______ min
Mean/Neonatal Hospital Days
Mean/MedianNeonatal Deaths
Fetal Mortality
Neonatal Respiratory Distress Syndrome (NRDS), Hyaline Membrane Disease (HMD)
Bronchio Pulmonary Dysplasia (BPD)
Intraventricular Hemorrhage
Neonatal Sepsis
Necrotizing Enterocolitis
Neonatal Morbidity
Pneumonia
Neonatal Infection
OTHER

26. Describe any additional analyses and/or subgroup analyses:

27. Limitations. (Distinguish between limitations stated by the author and those found by the reviewer.)

28. Conclusions:
Intervention ineffective in altering outcomes detailed above.
Intervention effectively alters the following outcomes: ______________________________________
________________________________________________________________________________
Intervention effectively alters the following outcomes in the following subgroup(s)
________________________________________________________________________________
________________________________________________________________________________

29. Comments. (Provide any additional information that you believe needs to be captured concerning this study.)
Unique Article Identifier:
Study Type:
Sample Size:

	Maternal Harms	Treatment Grp1	Treatment Grp2	Control Grp
	Death
	Heart Failure
	Arrhythmia
	Abnormal glucose tolerance
	Elevated liver functions/hepatitis
	Renal failure
	Headache
	Tremor
	Hypotension
	Skin rash
	Stress, emotional distress or other mental health issues
	Discontinuations due to side effects
	Other:
	Other:
	Other
	Death
	Intraventricular hemorrhage
	Necrotizing enterocolitis
	Ductal constriction
	Bardycardia
	Cerebral Palsy
	Other neurologic effects
	Other:
	Other:
	Other:
	Other:
* Statistical significance: if nonsignificant note NS, otherwise provide p number.

Method of Ascertaining Harms:
Interviewer-administered side-effect checklist
Self-administered side-effect checklist
Open-ended question asked at each visit
Investigator observed at each visit
Spontaneous reporting by patient
Not otherwise specified