| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
ORIGINAL RESEARCH |
From the Department of Obstetrics and Gynecology and the Center for Clinical Epidemiology and Biostatistics, University of Pennsylvania Health System, Philadelphia, Pennsylvania.
Address reprint requests to: George A. Macones, MD, Room 901, Blockley Hall, 423 Guardian Drive, Philadelphia, PA 19104-6021, E-mail: macones{at}cceb.med.upenn.edu
 |
Abstract |
|---|
Top
 Abstract Materials and MethodsResultsDiscussionReferences |
|---|
Methods: We performed a case-control study of patients treated with magnesium sulfate for idiopathic preterm labor. A case was defined as a patient who received magnesium sulfate tocolysis and delivered within 48 hours of admission. We selected a 48-hour delay to delivery as a clinically relevant endpoint for the maximization of steroid benefit. Controls were patients who received magnesium sulfate tocolysis and remained undelivered 48 hours after admission. Cases and controls were identified by merging a pharmacy billing database with International Classification of Disease codes for premature labor. Medical records were reviewed and risk factor information was obtained. We focused on risk factors within the first hour of admission, because our goal was to identify patients at high risk of delivery early in their hospital course. Backward stepwise logistic regression was used to develop explanatory and predictive models. The focus of the predictive model was to maximize the test’s sensitivity and negative predictive value.
Results: We identified 50 cases and 150 controls. The following six variables were included in the initial multivariable models based on bivariate analyses: white blood cell count at least 14.0 (1000/µL), cervical dilation at least 2 cm, bleeding, substance abuse, parity, and previous abortion. A two-variable model containing cervical dilation and bleeding had an overall accuracy of 73%, sensitivity of 62%, and specificity of 76%, and it was as sensitive and specific as more complex models.
Conclusion: Cervical dilation of at least 2 cm and bleeding on admission had an overall accuracy of 73% in predicting the likelihood of delivery within 48 hours in women receiving magnesium sulfate.
The prevalence of preterm deliveries in the United States is approximately 10–15%, half of which are attributable to idiopathic preterm labor.1,2 Given the enormous societal impact of preterm delivery, a large investigative effort has focused on the treatment of idiopathic preterm labor with tocolytic agents. Based on these studies, many obstetricians believe that parenteral agents can effectively delay delivery for 24–48 hours in up to 70–90% of patients treated, suggesting that 10–30% will fail tocolysis and deliver within 48 hours.3 Accurate prediction of which patients treated with standard tocolytic agents are likely to deliver before maximal steroid benefit is attained (ie, within 48 hours) could have important clinical influence.4 If we could predict who is likely to deliver within 48 hours of admission among women with preterm labor, we could target and test more aggressive treatments in this high-risk subgroup, thus improving neonatal outcome. Unfortunately, few data specifically address whether there are factors that can accurately identify those at high risk of delivery within 48 hours, among patients with preterm labor. The aim of our study was to determine whether individual factors or combinations of clinical factors could accurately predict delivery within 48 hours in patients receiving tocolysis for preterm labor.
|
Materials and Methods |
|---|
|
Top
AbstractMaterials and MethodsResultsDiscussionReferences |
|---|
Patients were excluded if amniocentesis found fetal lung maturity, if they did not meet our institutional definition of preterm labor (uterine contractions with documented cervical change or a cervical dilation of at least 2 cm in a patient without a previous examination), if there was clinically apparent intra-amniotic infection (diagnosed by the attending obstetrician) or placental abruption (the latter two groups are ineligible for tocolysis at our institution), or if there was rupture of membranes documented (those patients also are not eligible for tocolysis at our institution). All other patients were eligible.
As patient records were reviewed, it was decided whether they met the criteria for case or control. A case was defined as a patient who had received magnesium sulfate tocolysis but delivered within 48 hours of admission. We selected 48-hour delay to delivery as a clinically relevant endpoint for maximization of steroid benefit.5 Controls were patients who received magnesium sulfate tocolysis and did not deliver within 48 hours after admission.
From medical records we obtained information on clinical variables that we hypothesized could be predictors of rapid delivery, including patient demographics (age, race, and body mass index on admission), obstetric history (gravidity, parity, pregnancy history, and previous tocolysis during index pregnancy), medical conditions (diabetes, hypertension, or other conditions), and social history (cocaine, alcohol, tobacco use, and physical abuse during pregnancy), as well as admission physical examination results (cervical dilation, heart rate, and temperature) and admission laboratory tests (bacteruria, white blood cell count x 1000/µL, and hematocrit). Bleeding on admission was categorized as positive if the patient reported bleeding on admission or bleeding of any amount was noted by the admitting physician on physical examination. At our institution, the admitting physician can be a resident, fellow, or attending physician. We were unable to use information on cervical effacement in this analysis, because reporting of effacement is variable at our institution (ie, as a percentage or as length of the cervix in centimeters). We focused specifically on admission laboratory results and physical examination characteristics because we hoped to determine risk of delivery early in the course of a patient’s hospitalization (because we believe that future interventions in this high-risk group would be more effective if started early in the patient’s hospitalization).
Each variable that was recorded was analyzed as a risk factor for delivery (within 48 hours of admission) by comparing cases with controls (Stata Statistical Software, College Station, TX). Unadjusted odds ratios (OR) and 95% confidence intervals (CI) were calculated for dichotomous variables, and t tests or Mann-Whitney U tests were used for comparing continuous variables. Since we wanted to develop a clinically useful prediction rule, we dichotomized the continuous variables and chose the optimum cut-point by creating receiver-operator characteristic curves. Multiple logistic regression was used to generate two separate models, an explanatory model and a predictive model.6 The purpose of the explanatory model was to assess the individual association between risk factors and the likelihood of delivery within 48 hours, whereas the purpose of the predictive model was to include variables that contributed significantly to the sensitivity of the model. We decided a priori to maximize sensitivity of the prediction rule, because we wanted to identify as large a proportion of patients likely to deliver within 48 hours as possible (because more aggressive treatment strategies could be tested in patients at high risk). For etiologic and predictive models, variables were initially entered into the model based on the level of significance in the unadjusted analyses, such that variables with P < .10 were eligible for inclusion. Variables were then removed using backward elimination. For the predictive model, differences in sensitivity and negative predictive value of the model (before and after elimination) were tested with McNemar’s test.7,8 The clinical prediction rule was internally validated using a bootstrap approach, which involves taking many samples from the original data and testing the predictive accuracy of the rule on each sample.9
Our sample size calculation was based on that for a case-control study. Specifically, to detect an OR of 2.75 for exposures with a prevalence of 0.20, the study would require 50 cases of delivery within 48 hours (assuming alpha = 0.05, 80% power, and a 3:1 control-to-case ratio). This sample size would allow a CI of 15% around our sensitivity estimate (assuming a sensitivity of 75%).
|
Results |
|---|
|
Top
AbstractMaterials and MethodsResultsDiscussionReferences |
|---|
Unadjusted comparisons of potential predictors between cases and controls are presented in Table 1
. For continuous and ordinal variables, we present bivariate results and analyzed them as either continuous or dichotomous (after an optimal cut-point for the variable has been established with receiver-operator characteristic analysis. Data for receiver-operator characteristic analysis are not presented but are available on request). For example, results for parity both evaluate it as an ordinal measurement and dichotomize it into less than two or at least two. Regarding obstetric history, cases had higher parity but had similar histories of preterm delivery and abortion. In terms of admission characteristics, cases had higher frequency of bleeding on admission, more advanced cervical dilation, and higher mean white blood cell counts. Cigarette smoking also tended to be more common in cases than controls, and there was a trend toward a higher frequency of cocaine use in the cases as well. Because of collinearity between alcohol, tobacco, and cocaine use, we combined these into a single category of substance abuse (patients were considered as positive for substance abuse if they used any substance in any amount).
|
). Substance abuse was not independently associated with an increased risk of delivery within 48 hours.
|
, a two-variable model consisting of cervical dilation and bleeding was as sensitive and specific as any of the more complex models. In this final model, a patient is classified at risk if either or both factors are positive.
|
|
Discussion |
|---|
|
Top
AbstractMaterials and MethodsResultsDiscussionReferences |
|---|
Other studies evaluated risk factors for failed tocolysis in idiopathic preterm labor, mainly in the form of secondary analyses, suggesting that cervical dilation on admission is a good discriminator of tocolytic efficacy. For example, Spisso et al10 retrospectively reviewed the records of patients with preterm labor who received magnesium sulfate for tocolysis.10 Eighty-four of the 119 patients with intact membranes (70.6%) had delivery delayed for at least 48 hours. In that study, cervical dilation greater than 2 cm was reported to be a risk factor for failed tocolysis and delivery. Miller et al11 found similar results in a clinical trial comparing magnesium sulfate to intravenous terbutaline. In that study, failure of magnesium sulfate tocolysis occurred more commonly in women with initial cervical dilation exceeding 3 cm. Chau et al12 noted that greater degrees of cervical dilation at admission were associated with higher rates of magnesium sulfate failure and delivery within 48 hours. Thus, the existing literature and everyday clinical experience support the fact that one of the strongest risk factors for failed tocolysis is the degree of cervical dilation on admission. The combination of two variables in our model, cervical dilation and bleeding, allowed a significantly greater degree of discrimination than either variable alone. If one only used cervical dilation on admission, then sensitivity for predicting delivery within 48 hours drops to 40% (compared with 62% in the two-variable model).
The results of the present study must be interpreted with the following limitations in mind. First, our explanatory and prediction rules were generated mainly from a population of inner city, low socioeconomic status women. Thus, it is uncertain whether we can generalize these results to other populations. Although we achieved fairly high sensitivities and specificities with the prediction rule, it is common for the test characteristics to decline when applied to other populations. We hope that investigators who serve other populations will try to validate our results in other groups of pregnant women. Second, it is important to realize that the predictive values we found are a function of the prevalence of delivery within 48 hours in this study. Because this is a case-control study, our prevalence of delivery within 48 hours was determined by our a priori sample size calculation and control-to-case ratio (in this study, three controls per case, for a prevalence of delivery within 48 hours of 25% or 50 in 200). If the prevalence of delivery within 48 hours is different than 25%, the predictive values we report will not be applicable. We believe that 25% is a reasonable estimate of the prevalence of delivery within 48 hours for women receiving tocolysis. This prevalence of delivery within 48 hours is supported by a recent meta-analysis of magnesium sulfate for preterm labor.3 We believe that the predictive values we report are reasonable estimates of what would be expected clinically. Third, recent evidence suggested that biochemical markers, such as cervicovaginal fibronectin, can be used as diagnostic tests in patients with symptoms of preterm labor (who did not necessarily receive tocolysis).13 At our institution, we do not routinely use this marker. However, even without the use of fibronectin, we were able to attain reasonable sensitivity and specificity (for delivery within 48 hours) using clinical variables alone. We believe that our study population is different from those in many of the fibronectin studies, in that all of our patients received parenteral tocolysis. Many of the patients in the present study would not have been eligible for fibronectin testing because their cervical dilation was over 3 cm (where the positivity rate of fibronectin is quite high). We acknowledge that the addition of this marker might further enhance the test characteristics in the subgroup of patients eligible for fibronectin testing.
This study used clinical factors to predict failure of magnesium sulfate tocolysis and delivery within 48 hours. Determining which women are at highest risk for tocolytic failure early in their hospitalization can identify an appropriate population for future studies of second-line treatment strategies to prolong gestation and gain the maximum benefit from corticosteroid therapy. Explanatory and predictive models showed that cervical dilation at least 2 cm and bleeding on admission are strongly associated with rapid delivery. The combination of these variables correctly identified over 60% of women who delivered within 48 hours, with a negative predictive value of over 80%. Based on the sensitivity and high negative predictive value of this model, we believe that this simple rule, if validated prospectively and in other populations, will be useful in targeting and implementing future research in women receiving parenteral tocolysis, who are at highest risk of delivery.
|
Footnotes |
|---|
Received June 22, 1998. Received in revised form August 14, 1998. Accepted August 27, 1998.
|
References |
|---|
|
Top
AbstractMaterials and MethodsResultsDiscussionReferences |
|---|
2. Arias F, Tomich P. Etiology and outcome of low birth weight of preterm infants. Obstet Gynecol 1982;60:277–81.
3. Macones GA, Sedev HM, Berlin M, Morgan MA, Berlin JA. Evidence for magnesium sulfate as a tocolytic agent. Obstet Gynecol Surv 1997;52:652–8.[Medline]
4. Collaborative Group on Antenatal Steroid Therapy. Effect of antenatal dexamethasone on the prevention of respiratory distress syndrome. Am J Obstet Gynecol 1981;141:276–87.[Medline]
5. Crowley P. Update of the antenatal steroid meta-analysis: Current knowledge and future research needs. U.S. Department of Health and Human Services, Public Health Service. National Institutes of Health. National Institute of Child Health and Human Development. NIH Pub No. 95-3784, November 1994.
6. Hosmer DW, Lemeshow D. Applied logistic regression. New York: Wiley, 1989.
7. Feinstein AR, Wells CK, Walter SD. A comparison of multivariate mathematical methods for predicting survival—I. Introduction, rationale, and general strategy. J Clin Epidemiol 1990;43:339–47.[Medline]
8. Wells CK, Feinstein AR, Walter SD. A comparison of multivariate mathematical methods for predicting survival—III. Accuracy of predictions in generating and challenge sets. J Clin Epidemiol 1990;43:361–72.[Medline]
9. Efron B, Gong G. A leisurely look at the bootstrap, the jackknife, and cross-validation. Am Statistician 1983;37:36–48.
10. Spisso KR, Harbert GM, Thiagarajah S. The use of magnesium sulfate as the primary tocolytic agent to prevent premature delivery. Am J Obstet Gynecol 1982;142:840–5.[Medline]
11. Miller JM, Keane MWD, Horger EO. A comparison of magnesium sulfate and terbutaline for the arrest of premature labor. J Reprod Med 1982;27:348–52.[Medline]
12. Chau AC, Gabert HA, Miller JM. A prospective comparison of terbutaline and magnesium for tocolysis. Obstet Gynecol 1982;80: 847–51.
13. Iams JD, Casal D, McGregor JA, Goodwin TM, Kreaden US, Lowensohn R, et al. Fetal fibronectin improves the accuracy of diagnosis of preterm labor. Am J Obstet Gynecol 1995;173:141–5.[Medline]
This article has been cited by other articles:
|
|
 |
|
  J. D. Iams Prediction and Early Detection of Preterm Labor Obstet. Gynecol., February 1, 2003; 101(2): 402 - 412. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 D. F Colombo Predicting spontaneous preterm birth BMJ, August 10, 2002; 325(7359): 289 - 290. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
