- Email: [email protected]
Systematic reviews of medical evidence: The use of meta-analysis in obstetrics and gynecology
33. 0 1997 by The American Gynecologists.)
/
Review SYSTEMATIC
REVIEWS
EVIDENCE: ANALYSIS
OF MEDICAL
THE USE OF METAIN OBSTETRICS
AND GYNECOLOGY J@ey F. Peipert, MD, MPH, and Michael B. Bracken, PhD, MPH Objective: To review the technique of meta-analysis and its uses and limitations in obstetrics and gynecology. Data Sources: We reviewed four major journals in obstetrics and gynecology (American Journal of Obstetrics and Gynecology, Fertility and Sterility, Journal of Reproductive Medicine, and Obstetrics 8 Gynecology). Methods of Study Selection: Journals were reviewed to determine frequency of meta-analysis as a method of systematic review in obstetrics and gynecology. We also summarized objectives and scientific guidelines for performing a meta-analysis. Tabulation, Integration, and Results: Meta-analysis is used with increased frequency in obstetrics and gynecology as a way of systematically reviewing medical evidence. This technique is an attempt to improve on traditional methods of narrative review by an expert and as a framework for evidence-based medicine and developing practice guidelines. By combining data from replicate studies, a metaanalysis can increase statistical power, more precisely estimate the typical effect size of treatment or risk factor, and attempt to resolve controversies in the medical literature. Meta-analysis is a retrospective look at data already collected and is therefore subject to the biases of all retrospective studies. Conclusions: The technique of meta-analysis requires all the scientific rigor of a randomized clinical trial with careful attention to study design, including a formal protocol for literature search strategies, quality assessment of candidate studies, specific inclusion and exclusion criteria, issues of sampling and publication bias, statistical tests of homogeneity, and sensitivity analysis. (Obstet Gynecol 1997;89:628From the Department qf Obstetrics and Gynecology, Women and l@nts’ Hospital, Brown Uniuersity School of Medicine, Providence, Rhode Island; and the Department qf Epidemiology and Public Health, Yale University School @Medicine, New Haven, Connecticut.
628 0029-7844/97/$17.00 PI1 50029.7844(96)00490-5
College
of Obstetricians
and
Meta-analysis is used frequently to review and aggregate data systematically in clinical research’ and is seen more and more in the obstetrics and gynecology literature. In fact, obstetrics has led other medical specialties in the attempt to review systematically all randomized trials conducted in its discipline.2 Combining data using meta-analysis can increase statistical power and increase the ability to evaluate treatment effects and complications in clinical trials as well as associations between risk factors and diseasein etiologic research. In addition, by combining results from several studies generalizability may be increased. Meta-analysis is especially useful when sample sizes of individual clinical trials are too small to detect an effect or when a large trial is too costly and time consuming to perform.3
Traditional Medicine
Reviews and Evidence-Based
The traditional method of summarizing accumulated knowledge concerning a field of interest is the narrative review by an expert. An individual or group of individuals reviews the literature, chooses the articles it considers “best,” and reaches conclusions that may or may not be rooted objectively in the reviewed evidence. This narrative review format has several potential weaknesses.There is no systematic approach to obtain original or primary data or to integrate findings. Interpretation is dependent on the opinion of the reviewer. In addition, explicit standards may not exist to evaluate the quality of the studies under review. Further, the expert reviewer often makes no attempt to synthesize data quantitatively. In recent years, emphasis has been placed increasingly on evidence-based medicine.4,5 Evidence-based medicine de-emphasizes intuitive professorial edict, unsystematic clinical experience, and pathophysiologic rationale as sufficient grounds for medical decisionmaking and instead emphasizes the examination of empirical evidence from clinical research.5 The approach of evidence-based medicine is to review systematically the strength and quality of the evidence supporting (or refuting) a practice and to reach a consensus regarding practice guidelines based on this review. This shift in paradigm from the traditional approach to an evidence-based review requires that researchers and physicians acquire new skills to evaluate and summarize the existing medical literature and clinical evidence. These skills include efficient literature-searching
Obstetrics & Gynecology
for primary end points and, more arguably, within subgroups. A second purpose is to resolve uncertainty and controversy when research studies disagree and to identify gaps and problems in the primary research base. Third, by increasing sample size, meta-analysis may provide more precise estimates of effect size or the magnitude of the association, which may vary considerably from study to study. Fourth, meta-analysis may begin to address research questions not posed at the start of an individual treatment trial.
Scient$c Guidelines for Meta-Analysis pre-
1 se6
1986.88
1989-91
1992-94
Years Figure 1. Number of articles published using meta-analysis obstetrics and gynecology journals, ~~-1986 to 1994.
in four
and application of formal rules of evidence to evaluate the medical literature.’ To evaluate the use of meta-analysis in the reproductive health literature, we performed a MEDLINE search to establish the number of published studies in obstetrics and gynecology using meta-analysis. We limited our search to four peer-reviewed journals: Obstetrics 6 Gynecology, American Journal of Obstetrics and Gynecology, Journal qf Reproductive Medicine, and Fertility and Sterility. We evaluated the number of published studies during the period before 1986 and then within 3-year blocks beginning in 1986, noting an increase in the use of meta-analysis from zero in the pre-1986 and 19861988 periods to eight studies from 1989 to 1991 to 23 reports from 1992 to 1994 (Figure 1). In 1995, there were 15 articles using meta-analysis published in these four journals, an increase from an average of 5.2 per year during the previous 6 years. The clinician and researcher must have some basic knowledge about meta-analysis to evaluate studies using this technique and to determine whether the methodology was used appropriately and correctly. The purpose of our study was to review the technique of meta-analysis and its potential pitfalls and to illustrate to the clinician and researcher some scientific standards by which to judge an article using this methodology.
Purpose of Meta-Analysis There are several purposes of meta-analyses. Many individual trials and observational studies lack sufficient statistical power to detect important and clinically relevant differences. One purpose of meta-analysis is to increase statistical power to detect overall differences
VOL.
89, NO.
4, APRIL
1997
Meta-analysis is a method of synthesizing data that incorporates a qualitative assessment of the methodology of reviewed studies and a quantitative method of combining and analyzing data. However, meta-analysis is a retrospective look at data already collected and, therefore, may be subject to many of the biases of retrospective studies. It is important to make the process of meta-analysis as scientific and rigorous as possible.6-8 Some authorities have suggested that a metaanalysis should adhere to scientific principles inherent in the model of the randomized clinical trial.6 Specifically, the methods should include an explicit protocol for the meta-analysis, an assessment of the combinability of data, including tests of homogeneity when appropriate, evaluation and measurement of potential biases, description of the statistical methods used, including a sensitivity analysis when appropriate, and a discussion regarding the applicability of the results. Each of these topics will be discussed. Protocol
and Study
Design
of the Meta-Analysis
As in any research endeavor, a protocol should be developed and used for the meta-analysis. Study objectives and questions to be answered should be prespecified clearly before analysis. As an example in the recent obstetrics and gynecology literature, Macones and colleagues’ clearly state their objective to evaluate the efficacy of oral beta-agonist maintenance therapy in delaying delivery and decreasing the incidence of preterm birth and its complications. Secondary objectives also can be proposed. For example, in the meta-analysis evaluating the safety of metronidazole in the first trimester of pregnancy, lo if dosage data were available it might be possible to determine whether varying doses of metronidazole affected teratogenicity. Details of the meta-analysis should be clearly stated, such as the methods for identifying studies. These should include how the literature search was carried out, identification of unpublished trials, search of texts, review of references of published trials, and consulta-
Peipert and Bracken
Meta-Analysis
629
tions with experts. Some overviews include every study; others may delete unpublished work, abstracts, dissertations, correspondence, and other less formal reports. Computer searches should specify data bases searched and key words used in the process. The report should include a list of trials reviewed and trials excluded and justification for the exclusions. Ideally, study inclusion criteria are delineated at the protocol stage and depend on the specific objectives of the study. Inclusion criteria may be based on study design, sample size, choice of experimental or control groups, dosage or magnitude of exposure, and outcomes of interest. Many overviews emphasize formal and explicit criteria for evaluating the quality of studies to be included in meta-ana1yses.i’ A description of the baseline characteristics of the patients studied should be provided, as well as the specific criteria for the diagnosis. Combinability
of Individual
Studies
A major challenge in meta-analysis is to determine whether results of separate trials can be combined meaningfully. Are the studies “similar enough” to be pooled? What criteria should be used to evaluate similarities and differences? Meta-analysis is analogous to pooling data from individual centers from a large multicenter trial when each center may have some variability in protocol compliance and study population. These problems are usually much greater for meta-analysis of different studies. The researcher should note these differences and discuss how they might affect the conclusions of the study. A major controversy in clinical research and epidemiology is whether meta-analysis can be performed appropriately using observational studies (cohort and case-control studies) or whether it should be limited to randomized clinical trials. In theory, randomization eliminates allocation bias and confounding. In nonexperimental studies, it is impossible to rule out bias and confounding as explanations for observed results. Even when a series of nonexperimental studies have results in the same direction, it is possible that they all suffer from the same bias or fail to control for the same confounder. The quality of information from a metaanalysis cannot transcend that of the individual studies.‘* For this reason, many epidemiologists believe that pooling observational studies may result in biased estimates of effect size.13 It is generally agreed that observational studies and randomized trials should be analyzed separately. However, with the inherent biases already in observational trials, some authorities suggest that nonexperimental studies cannot be combined and the practice of pooling
630
Peipert and Bracken
Meta-Andysis
nonexperimental data should be abandoned.r3 Others disagree and consider meta-analysis of nonexperimental studies to be in an early stage of development.i4
Control
of Bias
As in any study, bias should be controlled for as much as possible. Selection bias may occur when selecting studies for inclusion in the meta-analysis. To avoid the bias of selecting studies that are consistent with the beliefs of the meta-analyst, the study selection should be made by looking at the methods section, not at the results of the study. Bias may also be introduced when assessingstudy quality. If possible, a group of colleagues should develop a quality assessmentprotocol and data forms. Assessors should evaluate studies according to these criteria. Quality criteria for randomized trials have been proposedi and include concealment of treatment allocation schedule, generation of allocation sequences, inclusion in the analysis of all randomized participants, and double blinding. To avoid bias in quality assessment,reviewers should be blinded to the authors, the institution, and interventions. Quality scores can be constructed, in which case only studies meeting a predefined cutoff point might be included in the analysis. It may also be possible to assign weights to the studies based on the study scores, and study weight may become a covariate in the meta-analysis to determine how study quality influences results.3 Potential for error and bias in the quality assessment process is due greatly to the subjective nature of such assessments.Therefore, it may be helpful to have two or more reviewers perform assessments.Disagreements can be discussed at a consensus meeting, or a third party may assist in bringing resolution.
Statistical
Analysis
The statistical issue of heterogeneity relates to the problem of variability among studies and the appropriateness of combining them into one meta-analysis. Statisticians distinguish between two types of variability: within- and between-study variability. Statistical tests for homogeneity are available for evaluating and measuring variability. It may be helpful to examine a graphic display of study outcomes to form a subjective impression of study homogeneity. Therefore, a first step in a meta-analysis may be to display the estimated treatment effects for each study together with their respective confidence intervals. These graphic displays complement statistical testing. Critical readers should
Obstetrics 6 Gynecology
evaluate meta-analyses to determine whether the issue of homogeneity is discussed. When studies have a common measure of outcome, a summary measure can often be derived from pooling the results. The Mantel-Haenszel test or some modification thereof is used most commonly in published meta-analyses. The two general methods for producing a combined estimate of the effect size and its confidence interval are the “fixed-effects” method and the “random-effects” method.16 The specific statistical techniques and procedures selected are beyond the scope of this review but need careful consideration with respect to the choice of effect measures” and methods for calculating them.i8 As in other study types, reports should acknowledge an awareness of potential sources of error including Type I error (concluding that there is a difference when none exists) or Type II error (concluding no difference exists when one does). Estimates of typical effect sizes and 95% confidence intervals are more useful to readers than the results of tests of significance (P values). Sensitivity
89, NO.
4, APRIL
Kmhn. 1981 Same-Helm,
1981
Wemom.
1981
Hcirterbcrg,
I
t
1997
I I
=
1985A
I
I
He~swrberg, 19858 Darj, 1987
I
=-I
Heist&erg,
1988
I
=
I
=
Sorensen. 1992
I
=
2316
I
1987
1988
1547
I
=
Heist&erg,
Lewdlois.
1447
=
I
915
-
I
2434
I
2489
I
3563 3737
I
LaIrson, 1992
4115
Ncilscn.
5188
1993
! 0.1
Analysis
A sensitivity analysis is a process of addressing how resistant the results of a meta-analysis are to changes in the way the meta-analysis was conducted. For example, if one randomized trial in a meta-analysis is of borderline quality, the meta-analyst may want to conduct the analysis both with and without this trial to determine whether the overall results vary. In a published metaanalysis evaluating oral tocolytics,’ the researchers noted one study that yielded contradictory results. They performed their analyses both with and without this study to determine whether their results and conclusion would change. Sensitivity analysis demonstrates how meta-analysis results vary through a range of assumptions, tests, and inclusion criteria.7 In this way, sensitivity analysis may be used to investigate why two meta-analyses on a subject come to different conclusions based on the choice of studies included in the analysis.3 Publication bias also is a factor to be considered in meta-analysis. Negative studies, ie, studies that fail to yield statistically significant results, are lesslikely to be submitted and accepted for publication than positive findings.‘“-‘r As a result, there may be general bias in the medical literature toward studies that show statistically significant differences with a corresponding bias in meta-analyses that are based only on published studies. For example, Simesz2demonstrates that a conclusion based on a meta-analysis about the effect an alkylating agent alone compared to combination che-
VOL.
ENQuANTlBlOTlCS
I 5
0.5 Summay
relative risk estimates and 95% confidence
intervals
Figure 2. An example of cumulative meta-analysis with summary relative risks and 95% confidence intervals for the use of antibiotics to prevent infection after induced abortion.27 (Reprinted with permission from Sawaya GF, Grady D, Kerlikowske K, Grimes DA. Antibiotics at the time of induced abortion: The case for universal phrophylaxis based on a meta-analysis. Obstet Gynecol 1996;87:884-90.)
motherapy had on survival in patients with ovarian cancer varied depending on whether only published trials were included in the meta-analysis or an attempt was made to include unpublished trials. If only published trials were included, there was a statistically significant increase in median survival in patients treated with combination chemotherapy, but when other unpublished data were included, there was no significant advantage of combination chemotherapy. Some authorities advocate not making the effort to identify unpublished trials, because the data and research have not undergone peer review.23 However, excluding data from unpublished trials may lead to a loss in precision of the estimate of the effect size.24 Attempting to locate unpublished data can be daunting unlessa registry of ongoing and completed trials exists. One such early registry in perinatal care was the Oxford Databaseof Perinatal Trials, andz5now published aspart of the Cochrane Data Base of Systematic Reviews, UK Cochrane Center, Summertown Pavillion, Middleway,
Peipert and Bracken
Meta-Analysis
631
Table
1. Criteria
to Evaluate
the Quality
Limitations
of a Meta-Analysis
1. Is the purpose of the study (ie, hypothesis/objective) stated clearly? 2. Is there a working research protocol? 3. Are literature search strategies described in detail? 4. Are study inclusion and exclusion criteria specified? *Are included and excluded papers listed? *Are reasons for exclusions justified? -1s the issue of publication bias addressed? 5. Was a blinded and systematic assessment of study quality conducted? 6. Are visual displays and tests of homogeneity provided? 7. Are appropriate statistical analyses performed? 8. Are sensitivity analyses used? 9. Are conclusions drawn for treatment recommendations (beneficial, equivocal, or harmful)? 10. Were results reported in sufficient detail to enable replication results by the reviewer? 11. Do the authors mention directions for future research?
of
Oxford OX2 7LG, England; E-mail: [email protected]. AC.UK. This data base is updated every 6 months and contains published and unpublished trials in the field of perinatal medicine. At the least, investigators should evaluate the potential for publication bias to explain the meta-analysis findings8
Problems
of Applicability
Once the meta-analyst has derived typical effect sizes and confidence intervals, he or she should attempt to put the findings into clinical and public health perspective. Does the meta-analysis establish definitively that one agent is more effective than another, that they are clearly the same, or that the results and conclusions are still in question, suggesting the need for further study? A meta-analysis may point to specific avenues for further research, such as different drug doses and regimens and groups of patients who may show particular benefit from therapy. A relatively new application of meta-analysis is cumulative meta-analysis: the performance of an updated meta-analysis every time a new trial appears for the purpose of evaluating the results as a continuum.26 This technique, illustrated in Figure 2,27 allows us to pinpoint the first time a difference in outcome between treatment and control groups becomes statistically significant. These data provide insight into how rapidly or how slowly medical interventions or treatments are adopted after demonstration of proven benefit. Cumulative meta-analysis can supply practitioners and policy makers with up-to-date information on emerging and established therapeutic protocols.26
632
Peipert
and
Bracken
Mefa-Analysis
of Meta-Analysis
As with other retrospective studies, meta-analysis has its limitations, and epidemiologists have been quick to highlight them.‘3~28~29In spite of its limitations, however, meta-analysis fills a critical need in medicine, that is, the need to reconcile conflicting research studies. Increasingly, new research areas are being examined using meta-analytic techniques, such as those that evaluate diagnostic test accuracy.30 Epidemiologists and statisticians generally agree that each meta-analysis should be conducted like a scientific experiment. Petittii4 has pointed out that meta-analyses, like case-control studies, are easy to do poorly. Several authors have proposed checklists by which to evaluate the quality of a meta-analysis (Table 1).3,31 Clinicians and researchers evaluating these studies should consider these guidelines and quality indicators when reviewing meta-analyses and be aware of the uses and limitations of this technique.
References 1 Thacker SB. Meta-analysis: A quantitative approach to research integration. JAMA 1988;259:1685-9. 2. Chalmers I, Enkin M, Keith MJNC. Effective care in pregnancy and childbirth. Oxford: Oxford University Press, 1989. 3 L’abbe KA, Detsky AS, O’Rourke K. Meta-analysis in clinical research. Ann Intern Med 1987;107:224-33. 4. Grimes DA. Teaching evidence-based medicine in obstetrics and gynecology. Obstet Gynecol 1995;86:451-7. 5. Evidence-Based Medicine Working Group. Evidence-based medicine: A new approach to teaching the practice of medicine. JAMA 1992;268:2420-5. 6. Gerbarg ZB, Horwitz RI. Resolving conflicting clinical trials: Guidelines for meta-analysis. J Clin Epidemiol 1988;41:503-9. 7. Sacks HS, Berrier J, Reitman D, Ancona-Berk VA, Chalmers TC. Meta-analyses of randomized controlled trials. N Engl J Med 1987;316:450-5. 8. Petitti DB. Meta-analysis, decision analysis, and cost-effectiveness analysis: Methods for quantitative synthesis in medicine. New York: Oxford University Press, 1994;58:1-130. 9. Macones GA, Berlin M, Berlin JA. Efficacy of oral beta-agonist maintenance therapy in preterm labor: A meta-analysis. Obstet Gynecol 1995;85:313-7. 10. Burtin I’, Taddio A, Aribumu 0, Einarson TR, Koren G. Safety of metronidazole in pregnancy: A meta-analysis. Am J Obstet Gynecol 1995;172:525-9. 11. Sinclair JC. Assessing evidence concerning treatment and prevention of diseases of the newborn. In: Sinclair JC, Bracken MB, eds. Effective care of the newborn infant. Oxford: Oxford University Press, 1992. 12. Feinstein AR. Meta-analysis: Statistical alchemy for the 21st century. J Clin Epidemiol 1995;48:71-9. 13. Shapiro S. Meta-analysis/shmeta-analysis. Am J Epidemiol 1994; 9:771-S. 14. Petitti DB. Of babies and bathwater. Am J Epidemiol 1994;9:77982. 15. Schulz KF, Chalmers I, Hayes RJ, Altman DG. Empirical evidence
Obstetrics & Gynecology
16. 17.
18.
19. 20. 21.
22. 23.
24.
25.
of bias: Dimensions of methodologic quality associated with estimates of treatment effects in controlled trials. JAMA 1995;273: 408-12. Spector TD, Thompson SG. The potential and limitations of meta-analysis. J Epidemiol Community Health 1991;45:89-92. Sinclair JC, Bracken MB. Clinically useful measures of effect in binary analyses of randomized trials. J Clin Epidemiol 1994;47: 881-9. Bracken MB. Statistical methods for analysis of effects of treatment in overviews of randomized trials. In: Sinclair JC, Bracken MB, eds. Effective care of the newborn infant. Oxford: Oxford University Press, 1992. Easterbrook I’J, Berlin JA, Gopalan R, Mathews DR. Publication bias in clinical research. Lancet 1991;337:867-72, Rosenthal R. The “file drawer problem” and tolerance for null results. Psycho1 Bull 1979;86:638-41. Dickersin K, Min YI, Meinert CL. Factors influencing publication of research results: Follow-up of applications submitted to two institutional review boards. JAMA 1992;263:374-8. Simes RJ. Publication bias: The case for an international registry of trials. J Clin Oncol 1986;4:1529-41. Chalmers TC, Berrier J, Sacks HS, Levin H, Reitman D, Nagalingam I’. Meta-analysis of clinical trials as a scientific discipline. II. Replicate variability and comparison of studies that agree and disagree. Stat Med 1987;6:733-44. Dickersin K, Scherer R, Lefebvre C. Identifying relevant studies for systematic reviews. In: Chalmers I, Altman DG, eds. Systematic reviews. London: BMJ Publishing Group, 1995:32. Chalmers I, Hetherington J, Newdick M, Mutch L, Grant A, Enkin M, et al. The Oxford Database of Perinatal Trials: Developing a register of published reports of controlled trials. Control Clin Trials 1986;7:306-24.
MISOPROSTOL RIPENING
FOR CERVICAL
AND LABOR INDUCTION:
A META-ANALYSIS his Sanchez-Ramos, MD, Andrew M. Kaunitz, MD, Robert L. Wears, MD, Isaac Delke, MD, and Francisco L. Gaudier, MD
Objective: To analyze published randomized trials assessing the safety and efficacy of misoprostol for cervical ripening and labor induction. Data Sources: We supplemented a search of entries in electronic data bases with references cited in original studies and review articles to identify randomized trials of misoprostol for cervical ripening and labor induction. Methods of Study Selection: Two blinded investigators performed independent trial quality evaluation and data abstraction of randomized clinical trials assessing the effi-
From the Department of Obstetrics and Gynecology, the Division qf Mnternal-Fetal Medicine, and the Department sf Emergency Medicine, University 4 Florida Health Science Center, Jacksonville, Florida.
VOL.
89, NO.
4, APRIL
1997
26. Lau J, Schmid CH, Chalmers TC. Cumulative meta-analysis of clinical trials builds evidence for exemplary medical care. J Clin Epidemiol 1995;48:45-57. 27. Sawaya G, Grady D, Kerlikowske K, Grimes DA. Antibiotics at the time of induced abortion: The case for universal prophylaxis based on a meta-analysis. Obstet Gynecol 1996;87:884-90. 28. Olkin I. Statistical and theoretical considerations in meta-analysis. J Clin Epidemiol 1995;48:133-46. 29. Greenland S. Can meta-analysis be salvaged? Am J Epidemiol 1994;9:783-7. 30. Irwig L, MaCaskill I’, Glaszion P, Fahey M. Meta-analytic methods for diagnostic test accuracy. J Clin Epidemiol 1995;48:119-30. 31. Thacker SB, Peterson HB, Stroup DF. Meta-analysis for the obstetrician-gynecologist. Am J Obstet Gynecol 1996;174:1403-7.
Address reprint requests to: Jefrey F. Peipert, MD, MPH Department qf Obstetrics and Gynecology Women and 1urJant.s’ Hospital 201 Dudley Street Providence, RI 02905
Received lune 24, 1996. Received in revised October Accepted October 28, 2996.
form
22, 1996
Copyright 0 1997 by The American College of Obstetricians Gynecologists. Published by Elsevier Science Inc.
and
cacy of misoprostol as a cervical ripening and labor-inducing agent. Tabulation, Integration, and Results: We calculated an estimate of the odds ratio (OR) and risk difference for dichotomous outcomes, using both a random- and fixedeffects model. Continuous outcomes were pooled using a variance-weighted average of the within-study difference in means. Of 16 studies identified, eight met our criteria for meta-analysis. These eight trials included 966 patients (488 received misoprostol and 478 were controls). Women who received misoprostol for cervical ripening and labor induction had a significantly lower overall cesarean rate (OR 0.67, 95% confidence interval ICI] 0.48, 0.93) and a higher incidence of vaginal delivery within 24 hours of misoprostol application (OR 2.64, 95% CI 1.87, 3.71). Use of misoprostol was associated with a higher incidence of tachysystole (OR 2.70, 95% CI 1.80, 4.04) but not hyperstimulation (OR 1.91, 95% CI 0.98, 3.73). The incidences of abnormal 5-minute Apgar scores and admissions to the neonatal intensive care unit were similar in the misoprostol and control groups. The pooled estimate of the mean interval from start of induction to delivery was 4.6 hours fewer (95% CI -3.5, -5.7) in the misoprostol group. Conclusion: Published data confirm the safety and efficacy of intravaginal misoprostol as an agent for cervical ripening and labor induction. (Obstet Gynecol 1997;89:633-42.
0029-7844/97/$17.00 PI1 SOO29-7844(96)00374-Z
633
/
Review SYSTEMATIC
REVIEWS
EVIDENCE: ANALYSIS
OF MEDICAL
THE USE OF METAIN OBSTETRICS
AND GYNECOLOGY J@ey F. Peipert, MD, MPH, and Michael B. Bracken, PhD, MPH Objective: To review the technique of meta-analysis and its uses and limitations in obstetrics and gynecology. Data Sources: We reviewed four major journals in obstetrics and gynecology (American Journal of Obstetrics and Gynecology, Fertility and Sterility, Journal of Reproductive Medicine, and Obstetrics 8 Gynecology). Methods of Study Selection: Journals were reviewed to determine frequency of meta-analysis as a method of systematic review in obstetrics and gynecology. We also summarized objectives and scientific guidelines for performing a meta-analysis. Tabulation, Integration, and Results: Meta-analysis is used with increased frequency in obstetrics and gynecology as a way of systematically reviewing medical evidence. This technique is an attempt to improve on traditional methods of narrative review by an expert and as a framework for evidence-based medicine and developing practice guidelines. By combining data from replicate studies, a metaanalysis can increase statistical power, more precisely estimate the typical effect size of treatment or risk factor, and attempt to resolve controversies in the medical literature. Meta-analysis is a retrospective look at data already collected and is therefore subject to the biases of all retrospective studies. Conclusions: The technique of meta-analysis requires all the scientific rigor of a randomized clinical trial with careful attention to study design, including a formal protocol for literature search strategies, quality assessment of candidate studies, specific inclusion and exclusion criteria, issues of sampling and publication bias, statistical tests of homogeneity, and sensitivity analysis. (Obstet Gynecol 1997;89:628From the Department qf Obstetrics and Gynecology, Women and l@nts’ Hospital, Brown Uniuersity School of Medicine, Providence, Rhode Island; and the Department qf Epidemiology and Public Health, Yale University School @Medicine, New Haven, Connecticut.
628 0029-7844/97/$17.00 PI1 50029.7844(96)00490-5
College
of Obstetricians
and
Meta-analysis is used frequently to review and aggregate data systematically in clinical research’ and is seen more and more in the obstetrics and gynecology literature. In fact, obstetrics has led other medical specialties in the attempt to review systematically all randomized trials conducted in its discipline.2 Combining data using meta-analysis can increase statistical power and increase the ability to evaluate treatment effects and complications in clinical trials as well as associations between risk factors and diseasein etiologic research. In addition, by combining results from several studies generalizability may be increased. Meta-analysis is especially useful when sample sizes of individual clinical trials are too small to detect an effect or when a large trial is too costly and time consuming to perform.3
Traditional Medicine
Reviews and Evidence-Based
The traditional method of summarizing accumulated knowledge concerning a field of interest is the narrative review by an expert. An individual or group of individuals reviews the literature, chooses the articles it considers “best,” and reaches conclusions that may or may not be rooted objectively in the reviewed evidence. This narrative review format has several potential weaknesses.There is no systematic approach to obtain original or primary data or to integrate findings. Interpretation is dependent on the opinion of the reviewer. In addition, explicit standards may not exist to evaluate the quality of the studies under review. Further, the expert reviewer often makes no attempt to synthesize data quantitatively. In recent years, emphasis has been placed increasingly on evidence-based medicine.4,5 Evidence-based medicine de-emphasizes intuitive professorial edict, unsystematic clinical experience, and pathophysiologic rationale as sufficient grounds for medical decisionmaking and instead emphasizes the examination of empirical evidence from clinical research.5 The approach of evidence-based medicine is to review systematically the strength and quality of the evidence supporting (or refuting) a practice and to reach a consensus regarding practice guidelines based on this review. This shift in paradigm from the traditional approach to an evidence-based review requires that researchers and physicians acquire new skills to evaluate and summarize the existing medical literature and clinical evidence. These skills include efficient literature-searching
Obstetrics & Gynecology
for primary end points and, more arguably, within subgroups. A second purpose is to resolve uncertainty and controversy when research studies disagree and to identify gaps and problems in the primary research base. Third, by increasing sample size, meta-analysis may provide more precise estimates of effect size or the magnitude of the association, which may vary considerably from study to study. Fourth, meta-analysis may begin to address research questions not posed at the start of an individual treatment trial.
Scient$c Guidelines for Meta-Analysis pre-
1 se6
1986.88
1989-91
1992-94
Years Figure 1. Number of articles published using meta-analysis obstetrics and gynecology journals, ~~-1986 to 1994.
in four
and application of formal rules of evidence to evaluate the medical literature.’ To evaluate the use of meta-analysis in the reproductive health literature, we performed a MEDLINE search to establish the number of published studies in obstetrics and gynecology using meta-analysis. We limited our search to four peer-reviewed journals: Obstetrics 6 Gynecology, American Journal of Obstetrics and Gynecology, Journal qf Reproductive Medicine, and Fertility and Sterility. We evaluated the number of published studies during the period before 1986 and then within 3-year blocks beginning in 1986, noting an increase in the use of meta-analysis from zero in the pre-1986 and 19861988 periods to eight studies from 1989 to 1991 to 23 reports from 1992 to 1994 (Figure 1). In 1995, there were 15 articles using meta-analysis published in these four journals, an increase from an average of 5.2 per year during the previous 6 years. The clinician and researcher must have some basic knowledge about meta-analysis to evaluate studies using this technique and to determine whether the methodology was used appropriately and correctly. The purpose of our study was to review the technique of meta-analysis and its potential pitfalls and to illustrate to the clinician and researcher some scientific standards by which to judge an article using this methodology.
Purpose of Meta-Analysis There are several purposes of meta-analyses. Many individual trials and observational studies lack sufficient statistical power to detect important and clinically relevant differences. One purpose of meta-analysis is to increase statistical power to detect overall differences
VOL.
89, NO.
4, APRIL
1997
Meta-analysis is a method of synthesizing data that incorporates a qualitative assessment of the methodology of reviewed studies and a quantitative method of combining and analyzing data. However, meta-analysis is a retrospective look at data already collected and, therefore, may be subject to many of the biases of retrospective studies. It is important to make the process of meta-analysis as scientific and rigorous as possible.6-8 Some authorities have suggested that a metaanalysis should adhere to scientific principles inherent in the model of the randomized clinical trial.6 Specifically, the methods should include an explicit protocol for the meta-analysis, an assessment of the combinability of data, including tests of homogeneity when appropriate, evaluation and measurement of potential biases, description of the statistical methods used, including a sensitivity analysis when appropriate, and a discussion regarding the applicability of the results. Each of these topics will be discussed. Protocol
and Study
Design
of the Meta-Analysis
As in any research endeavor, a protocol should be developed and used for the meta-analysis. Study objectives and questions to be answered should be prespecified clearly before analysis. As an example in the recent obstetrics and gynecology literature, Macones and colleagues’ clearly state their objective to evaluate the efficacy of oral beta-agonist maintenance therapy in delaying delivery and decreasing the incidence of preterm birth and its complications. Secondary objectives also can be proposed. For example, in the meta-analysis evaluating the safety of metronidazole in the first trimester of pregnancy, lo if dosage data were available it might be possible to determine whether varying doses of metronidazole affected teratogenicity. Details of the meta-analysis should be clearly stated, such as the methods for identifying studies. These should include how the literature search was carried out, identification of unpublished trials, search of texts, review of references of published trials, and consulta-
Peipert and Bracken
Meta-Analysis
629
tions with experts. Some overviews include every study; others may delete unpublished work, abstracts, dissertations, correspondence, and other less formal reports. Computer searches should specify data bases searched and key words used in the process. The report should include a list of trials reviewed and trials excluded and justification for the exclusions. Ideally, study inclusion criteria are delineated at the protocol stage and depend on the specific objectives of the study. Inclusion criteria may be based on study design, sample size, choice of experimental or control groups, dosage or magnitude of exposure, and outcomes of interest. Many overviews emphasize formal and explicit criteria for evaluating the quality of studies to be included in meta-ana1yses.i’ A description of the baseline characteristics of the patients studied should be provided, as well as the specific criteria for the diagnosis. Combinability
of Individual
Studies
A major challenge in meta-analysis is to determine whether results of separate trials can be combined meaningfully. Are the studies “similar enough” to be pooled? What criteria should be used to evaluate similarities and differences? Meta-analysis is analogous to pooling data from individual centers from a large multicenter trial when each center may have some variability in protocol compliance and study population. These problems are usually much greater for meta-analysis of different studies. The researcher should note these differences and discuss how they might affect the conclusions of the study. A major controversy in clinical research and epidemiology is whether meta-analysis can be performed appropriately using observational studies (cohort and case-control studies) or whether it should be limited to randomized clinical trials. In theory, randomization eliminates allocation bias and confounding. In nonexperimental studies, it is impossible to rule out bias and confounding as explanations for observed results. Even when a series of nonexperimental studies have results in the same direction, it is possible that they all suffer from the same bias or fail to control for the same confounder. The quality of information from a metaanalysis cannot transcend that of the individual studies.‘* For this reason, many epidemiologists believe that pooling observational studies may result in biased estimates of effect size.13 It is generally agreed that observational studies and randomized trials should be analyzed separately. However, with the inherent biases already in observational trials, some authorities suggest that nonexperimental studies cannot be combined and the practice of pooling
630
Peipert and Bracken
Meta-Andysis
nonexperimental data should be abandoned.r3 Others disagree and consider meta-analysis of nonexperimental studies to be in an early stage of development.i4
Control
of Bias
As in any study, bias should be controlled for as much as possible. Selection bias may occur when selecting studies for inclusion in the meta-analysis. To avoid the bias of selecting studies that are consistent with the beliefs of the meta-analyst, the study selection should be made by looking at the methods section, not at the results of the study. Bias may also be introduced when assessingstudy quality. If possible, a group of colleagues should develop a quality assessmentprotocol and data forms. Assessors should evaluate studies according to these criteria. Quality criteria for randomized trials have been proposedi and include concealment of treatment allocation schedule, generation of allocation sequences, inclusion in the analysis of all randomized participants, and double blinding. To avoid bias in quality assessment,reviewers should be blinded to the authors, the institution, and interventions. Quality scores can be constructed, in which case only studies meeting a predefined cutoff point might be included in the analysis. It may also be possible to assign weights to the studies based on the study scores, and study weight may become a covariate in the meta-analysis to determine how study quality influences results.3 Potential for error and bias in the quality assessment process is due greatly to the subjective nature of such assessments.Therefore, it may be helpful to have two or more reviewers perform assessments.Disagreements can be discussed at a consensus meeting, or a third party may assist in bringing resolution.
Statistical
Analysis
The statistical issue of heterogeneity relates to the problem of variability among studies and the appropriateness of combining them into one meta-analysis. Statisticians distinguish between two types of variability: within- and between-study variability. Statistical tests for homogeneity are available for evaluating and measuring variability. It may be helpful to examine a graphic display of study outcomes to form a subjective impression of study homogeneity. Therefore, a first step in a meta-analysis may be to display the estimated treatment effects for each study together with their respective confidence intervals. These graphic displays complement statistical testing. Critical readers should
Obstetrics 6 Gynecology
evaluate meta-analyses to determine whether the issue of homogeneity is discussed. When studies have a common measure of outcome, a summary measure can often be derived from pooling the results. The Mantel-Haenszel test or some modification thereof is used most commonly in published meta-analyses. The two general methods for producing a combined estimate of the effect size and its confidence interval are the “fixed-effects” method and the “random-effects” method.16 The specific statistical techniques and procedures selected are beyond the scope of this review but need careful consideration with respect to the choice of effect measures” and methods for calculating them.i8 As in other study types, reports should acknowledge an awareness of potential sources of error including Type I error (concluding that there is a difference when none exists) or Type II error (concluding no difference exists when one does). Estimates of typical effect sizes and 95% confidence intervals are more useful to readers than the results of tests of significance (P values). Sensitivity
89, NO.
4, APRIL
Kmhn. 1981 Same-Helm,
1981
Wemom.
1981
Hcirterbcrg,
I
t
1997
I I
=
1985A
I
I
He~swrberg, 19858 Darj, 1987
I
=-I
Heist&erg,
1988
I
=
I
=
Sorensen. 1992
I
=
2316
I
1987
1988
1547
I
=
Heist&erg,
Lewdlois.
1447
=
I
915
-
I
2434
I
2489
I
3563 3737
I
LaIrson, 1992
4115
Ncilscn.
5188
1993
! 0.1
Analysis
A sensitivity analysis is a process of addressing how resistant the results of a meta-analysis are to changes in the way the meta-analysis was conducted. For example, if one randomized trial in a meta-analysis is of borderline quality, the meta-analyst may want to conduct the analysis both with and without this trial to determine whether the overall results vary. In a published metaanalysis evaluating oral tocolytics,’ the researchers noted one study that yielded contradictory results. They performed their analyses both with and without this study to determine whether their results and conclusion would change. Sensitivity analysis demonstrates how meta-analysis results vary through a range of assumptions, tests, and inclusion criteria.7 In this way, sensitivity analysis may be used to investigate why two meta-analyses on a subject come to different conclusions based on the choice of studies included in the analysis.3 Publication bias also is a factor to be considered in meta-analysis. Negative studies, ie, studies that fail to yield statistically significant results, are lesslikely to be submitted and accepted for publication than positive findings.‘“-‘r As a result, there may be general bias in the medical literature toward studies that show statistically significant differences with a corresponding bias in meta-analyses that are based only on published studies. For example, Simesz2demonstrates that a conclusion based on a meta-analysis about the effect an alkylating agent alone compared to combination che-
VOL.
ENQuANTlBlOTlCS
I 5
0.5 Summay
relative risk estimates and 95% confidence
intervals
Figure 2. An example of cumulative meta-analysis with summary relative risks and 95% confidence intervals for the use of antibiotics to prevent infection after induced abortion.27 (Reprinted with permission from Sawaya GF, Grady D, Kerlikowske K, Grimes DA. Antibiotics at the time of induced abortion: The case for universal phrophylaxis based on a meta-analysis. Obstet Gynecol 1996;87:884-90.)
motherapy had on survival in patients with ovarian cancer varied depending on whether only published trials were included in the meta-analysis or an attempt was made to include unpublished trials. If only published trials were included, there was a statistically significant increase in median survival in patients treated with combination chemotherapy, but when other unpublished data were included, there was no significant advantage of combination chemotherapy. Some authorities advocate not making the effort to identify unpublished trials, because the data and research have not undergone peer review.23 However, excluding data from unpublished trials may lead to a loss in precision of the estimate of the effect size.24 Attempting to locate unpublished data can be daunting unlessa registry of ongoing and completed trials exists. One such early registry in perinatal care was the Oxford Databaseof Perinatal Trials, andz5now published aspart of the Cochrane Data Base of Systematic Reviews, UK Cochrane Center, Summertown Pavillion, Middleway,
Peipert and Bracken
Meta-Analysis
631
Table
1. Criteria
to Evaluate
the Quality
Limitations
of a Meta-Analysis
1. Is the purpose of the study (ie, hypothesis/objective) stated clearly? 2. Is there a working research protocol? 3. Are literature search strategies described in detail? 4. Are study inclusion and exclusion criteria specified? *Are included and excluded papers listed? *Are reasons for exclusions justified? -1s the issue of publication bias addressed? 5. Was a blinded and systematic assessment of study quality conducted? 6. Are visual displays and tests of homogeneity provided? 7. Are appropriate statistical analyses performed? 8. Are sensitivity analyses used? 9. Are conclusions drawn for treatment recommendations (beneficial, equivocal, or harmful)? 10. Were results reported in sufficient detail to enable replication results by the reviewer? 11. Do the authors mention directions for future research?
of
Oxford OX2 7LG, England; E-mail: [email protected]. AC.UK. This data base is updated every 6 months and contains published and unpublished trials in the field of perinatal medicine. At the least, investigators should evaluate the potential for publication bias to explain the meta-analysis findings8
Problems
of Applicability
Once the meta-analyst has derived typical effect sizes and confidence intervals, he or she should attempt to put the findings into clinical and public health perspective. Does the meta-analysis establish definitively that one agent is more effective than another, that they are clearly the same, or that the results and conclusions are still in question, suggesting the need for further study? A meta-analysis may point to specific avenues for further research, such as different drug doses and regimens and groups of patients who may show particular benefit from therapy. A relatively new application of meta-analysis is cumulative meta-analysis: the performance of an updated meta-analysis every time a new trial appears for the purpose of evaluating the results as a continuum.26 This technique, illustrated in Figure 2,27 allows us to pinpoint the first time a difference in outcome between treatment and control groups becomes statistically significant. These data provide insight into how rapidly or how slowly medical interventions or treatments are adopted after demonstration of proven benefit. Cumulative meta-analysis can supply practitioners and policy makers with up-to-date information on emerging and established therapeutic protocols.26
632
Peipert
and
Bracken
Mefa-Analysis
of Meta-Analysis
As with other retrospective studies, meta-analysis has its limitations, and epidemiologists have been quick to highlight them.‘3~28~29In spite of its limitations, however, meta-analysis fills a critical need in medicine, that is, the need to reconcile conflicting research studies. Increasingly, new research areas are being examined using meta-analytic techniques, such as those that evaluate diagnostic test accuracy.30 Epidemiologists and statisticians generally agree that each meta-analysis should be conducted like a scientific experiment. Petittii4 has pointed out that meta-analyses, like case-control studies, are easy to do poorly. Several authors have proposed checklists by which to evaluate the quality of a meta-analysis (Table 1).3,31 Clinicians and researchers evaluating these studies should consider these guidelines and quality indicators when reviewing meta-analyses and be aware of the uses and limitations of this technique.
References 1 Thacker SB. Meta-analysis: A quantitative approach to research integration. JAMA 1988;259:1685-9. 2. Chalmers I, Enkin M, Keith MJNC. Effective care in pregnancy and childbirth. Oxford: Oxford University Press, 1989. 3 L’abbe KA, Detsky AS, O’Rourke K. Meta-analysis in clinical research. Ann Intern Med 1987;107:224-33. 4. Grimes DA. Teaching evidence-based medicine in obstetrics and gynecology. Obstet Gynecol 1995;86:451-7. 5. Evidence-Based Medicine Working Group. Evidence-based medicine: A new approach to teaching the practice of medicine. JAMA 1992;268:2420-5. 6. Gerbarg ZB, Horwitz RI. Resolving conflicting clinical trials: Guidelines for meta-analysis. J Clin Epidemiol 1988;41:503-9. 7. Sacks HS, Berrier J, Reitman D, Ancona-Berk VA, Chalmers TC. Meta-analyses of randomized controlled trials. N Engl J Med 1987;316:450-5. 8. Petitti DB. Meta-analysis, decision analysis, and cost-effectiveness analysis: Methods for quantitative synthesis in medicine. New York: Oxford University Press, 1994;58:1-130. 9. Macones GA, Berlin M, Berlin JA. Efficacy of oral beta-agonist maintenance therapy in preterm labor: A meta-analysis. Obstet Gynecol 1995;85:313-7. 10. Burtin I’, Taddio A, Aribumu 0, Einarson TR, Koren G. Safety of metronidazole in pregnancy: A meta-analysis. Am J Obstet Gynecol 1995;172:525-9. 11. Sinclair JC. Assessing evidence concerning treatment and prevention of diseases of the newborn. In: Sinclair JC, Bracken MB, eds. Effective care of the newborn infant. Oxford: Oxford University Press, 1992. 12. Feinstein AR. Meta-analysis: Statistical alchemy for the 21st century. J Clin Epidemiol 1995;48:71-9. 13. Shapiro S. Meta-analysis/shmeta-analysis. Am J Epidemiol 1994; 9:771-S. 14. Petitti DB. Of babies and bathwater. Am J Epidemiol 1994;9:77982. 15. Schulz KF, Chalmers I, Hayes RJ, Altman DG. Empirical evidence
Obstetrics & Gynecology
16. 17.
18.
19. 20. 21.
22. 23.
24.
25.
of bias: Dimensions of methodologic quality associated with estimates of treatment effects in controlled trials. JAMA 1995;273: 408-12. Spector TD, Thompson SG. The potential and limitations of meta-analysis. J Epidemiol Community Health 1991;45:89-92. Sinclair JC, Bracken MB. Clinically useful measures of effect in binary analyses of randomized trials. J Clin Epidemiol 1994;47: 881-9. Bracken MB. Statistical methods for analysis of effects of treatment in overviews of randomized trials. In: Sinclair JC, Bracken MB, eds. Effective care of the newborn infant. Oxford: Oxford University Press, 1992. Easterbrook I’J, Berlin JA, Gopalan R, Mathews DR. Publication bias in clinical research. Lancet 1991;337:867-72, Rosenthal R. The “file drawer problem” and tolerance for null results. Psycho1 Bull 1979;86:638-41. Dickersin K, Min YI, Meinert CL. Factors influencing publication of research results: Follow-up of applications submitted to two institutional review boards. JAMA 1992;263:374-8. Simes RJ. Publication bias: The case for an international registry of trials. J Clin Oncol 1986;4:1529-41. Chalmers TC, Berrier J, Sacks HS, Levin H, Reitman D, Nagalingam I’. Meta-analysis of clinical trials as a scientific discipline. II. Replicate variability and comparison of studies that agree and disagree. Stat Med 1987;6:733-44. Dickersin K, Scherer R, Lefebvre C. Identifying relevant studies for systematic reviews. In: Chalmers I, Altman DG, eds. Systematic reviews. London: BMJ Publishing Group, 1995:32. Chalmers I, Hetherington J, Newdick M, Mutch L, Grant A, Enkin M, et al. The Oxford Database of Perinatal Trials: Developing a register of published reports of controlled trials. Control Clin Trials 1986;7:306-24.
MISOPROSTOL RIPENING
FOR CERVICAL
AND LABOR INDUCTION:
A META-ANALYSIS his Sanchez-Ramos, MD, Andrew M. Kaunitz, MD, Robert L. Wears, MD, Isaac Delke, MD, and Francisco L. Gaudier, MD
Objective: To analyze published randomized trials assessing the safety and efficacy of misoprostol for cervical ripening and labor induction. Data Sources: We supplemented a search of entries in electronic data bases with references cited in original studies and review articles to identify randomized trials of misoprostol for cervical ripening and labor induction. Methods of Study Selection: Two blinded investigators performed independent trial quality evaluation and data abstraction of randomized clinical trials assessing the effi-
From the Department of Obstetrics and Gynecology, the Division qf Mnternal-Fetal Medicine, and the Department sf Emergency Medicine, University 4 Florida Health Science Center, Jacksonville, Florida.
VOL.
89, NO.
4, APRIL
1997
26. Lau J, Schmid CH, Chalmers TC. Cumulative meta-analysis of clinical trials builds evidence for exemplary medical care. J Clin Epidemiol 1995;48:45-57. 27. Sawaya G, Grady D, Kerlikowske K, Grimes DA. Antibiotics at the time of induced abortion: The case for universal prophylaxis based on a meta-analysis. Obstet Gynecol 1996;87:884-90. 28. Olkin I. Statistical and theoretical considerations in meta-analysis. J Clin Epidemiol 1995;48:133-46. 29. Greenland S. Can meta-analysis be salvaged? Am J Epidemiol 1994;9:783-7. 30. Irwig L, MaCaskill I’, Glaszion P, Fahey M. Meta-analytic methods for diagnostic test accuracy. J Clin Epidemiol 1995;48:119-30. 31. Thacker SB, Peterson HB, Stroup DF. Meta-analysis for the obstetrician-gynecologist. Am J Obstet Gynecol 1996;174:1403-7.
Address reprint requests to: Jefrey F. Peipert, MD, MPH Department qf Obstetrics and Gynecology Women and 1urJant.s’ Hospital 201 Dudley Street Providence, RI 02905
Received lune 24, 1996. Received in revised October Accepted October 28, 2996.
form
22, 1996
Copyright 0 1997 by The American College of Obstetricians Gynecologists. Published by Elsevier Science Inc.
and
cacy of misoprostol as a cervical ripening and labor-inducing agent. Tabulation, Integration, and Results: We calculated an estimate of the odds ratio (OR) and risk difference for dichotomous outcomes, using both a random- and fixedeffects model. Continuous outcomes were pooled using a variance-weighted average of the within-study difference in means. Of 16 studies identified, eight met our criteria for meta-analysis. These eight trials included 966 patients (488 received misoprostol and 478 were controls). Women who received misoprostol for cervical ripening and labor induction had a significantly lower overall cesarean rate (OR 0.67, 95% confidence interval ICI] 0.48, 0.93) and a higher incidence of vaginal delivery within 24 hours of misoprostol application (OR 2.64, 95% CI 1.87, 3.71). Use of misoprostol was associated with a higher incidence of tachysystole (OR 2.70, 95% CI 1.80, 4.04) but not hyperstimulation (OR 1.91, 95% CI 0.98, 3.73). The incidences of abnormal 5-minute Apgar scores and admissions to the neonatal intensive care unit were similar in the misoprostol and control groups. The pooled estimate of the mean interval from start of induction to delivery was 4.6 hours fewer (95% CI -3.5, -5.7) in the misoprostol group. Conclusion: Published data confirm the safety and efficacy of intravaginal misoprostol as an agent for cervical ripening and labor induction. (Obstet Gynecol 1997;89:633-42.
0029-7844/97/$17.00 PI1 SOO29-7844(96)00374-Z
633