In the roundtable that follows, clinicians discuss a study published in this issue of the Journal in light of its methodology, relevance to practice, and implications for future research. Article discussed:
Pamidi S, Pinto LM, Marc I, et al. Maternal sleep-disordered breathing and adverse pregnancy outcomes: a systematic review and metaanalysis. Am J Obstet Gynecol 2014;210:52.e1-14.
What was the primary aim of this study?
How do systemic review and metaanalysis differ?
What search strategy and inclusion criteria were used?
How were the data analyzed?
What were the main findings?
How might the results be incorporated into prenatal care?
For a summary and analysis of this discussion, see page 87
Sleep-disordered breathing (SDB) is marked by breathing pauses, microarousals, and hemodynamic changes. In the general population, the ailment is associated with multiple adverse cardiac and metabolic outcomes, including high blood pressure, cardiovascular disease, stroke, and altered glucose metabolism. Pregnant women frequently have SDB-related symptoms, which can be triggered by weight gain, airway edema, and hormonal changes. These tend to worsen as the pregnancy progresses. While studies have examined possible links between SDB and adverse pregnancy outcomes, the results have been conflicting. This month, Journal Club members discussed a new study that looked for a relationship between SDB in pregnancy and gestational hypertension/preeclampsia, gestational diabetes, and low birthweight.
Molly J. Stout, MD and George A. Macones, MD, MSCE, Associate Editor
Stout: What was the primary aim of this study?
Epplin: The goal was to determine whether pregnant women with SDB have a higher rate of adverse pregnancy outcomes, such as gestational hypertension/preeclampsia, gestational diabetes, and low-birthweight infants.
Stout: What is the difference between systematic review and metaanalysis?
Wood: A systematic review is a literature review based on a clearly formulated research question. The researchers attempt to identify all relevant work, assess the quality of those studies, and qualitatively summarize the results. In a meta-analysis, on the other hand, statistical methods are used to mathematically combine the results of studies identified in a systematic review. By combining the results of several studies, the power of the analysis is increased. This can be useful for examining rare outcomes. Additionally, by statistically combining the results of similar studies, researchers can assess whether treatment effects are similar or different in diverse clinical scenarios.
Stout: Can you describe the search strategy and inclusion criteria?
Epplin: The authors used 3 main scientific databases for their search. They used search terms related to SDB, such as “snoring” and “sleep apnea,” and combined those with terms related to adverse pregnancy outcomes. Conference abstracts, reviews, or case reports were excluded. Bibliographies of 3 review papers were evaluated for additional citations. To be included, the study had to address their primary question of the association of SDB with adverse pregnancy outcomes, and it had to include a comparison group.
Stout: How did the authors define SDB for these analyses? Why is this important?
Wood: The authors defined SDB according to either a polysomnographic (PSG) diagnosis of obstructive sleep apnea-hypopnea or a probable/presumed diagnosis of obstructive sleep apnea (OSA); the latter was based on clinical symptoms, which included snoring, nocturnal choking/gasping, and witnessed apneas or on simplified sleep recordings that showed evidence of upper airway obstruction (inspiratory flow limitation) and/or repetitive oxygen desaturations. This is important. If the authors only included studies with strict PSG diagnosis, they might have missed essential findings in studies of patients whose diagnoses were based on clinical symptoms—clinical symptom scoring could be the most practical method used in daily medicine. The downside of including clinical diagnosis is that some participants in individual studies might not have truly had OSA.
Stout: Please describe the analysis. Specifically, can you comment on the assessment of heterogeneity and the type of meta-analysis modeling used?
Trudell: The authors performed a systematic review of the literature and where possible, used the method of meta-analysis. Meta-analysis is a general term that refers to the combining of results from different studies to identify agreement, disagreement, or other patterns among all findings. This is a 2-step process. The first step involves computing summary statistics for each individual study; the type of summary statistic calculated depends on the type of data. In the present study, the authors used odds ratios for dichotomous (yes-no) data (presence of hypertensive disorders, gestational diabetes, low birthweight); and differences in means for continuous data (apnea hypopnea index, infant birthweight). The second step in meta-analysis is calculating a combined effect—in this case, the overall effect of SDB on adverse pregnancy outcomes.
Because each individual study included in the overall analysis has different patients in different clinical situations with varied inclusion/exclusion criteria, etc., it would be inappropriate to assume that the results of each study contribute to the overall disease effect equally. Therefore, we test for heterogeneity. If the studies are too different, that is, if there is too much heterogeneity, it is inappropriate to pool them together for a summary statistic (use them in a meta-analysis). The authors of this study used the I2 statistic to quantify heterogeneity, and where significant heterogeneity existed (I2 >75%), they did not obtain pooled estimates.
Two main modeling techniques are used in meta-analysis: fixed effects and random effects (DerSimonian and Laird random effects model). Fixed effects modeling assumes that a single true disease effect size exists in all studies. On the other hand, a random effects model assumes that the true effect size may vary from study to study. A random effects model will give a more conservative pooled effect size and will “collapse” into a fixed effects model if, in fact, no significant difference in the true effect size can be found between studies. There is no single right answer about which modeling technique is better. However, planning to use the random effects model provides the benefit of the relaxed assumption of a common disease effect and a more conservative estimate of effect.
Stout: Describe the main findings for each of the obstetric outcomes examined: gestational hypertensive disease, gestational diabetes, and low birthweight.
Epplin: Thirteen of 18 studies looking for a link between maternal SDB and gestational hypertension/preeclampsia found a positive association. This effect was present whether the study used a clinical or PSG diagnosis of SDB, and it remained positive in the group of studies that adjusted for confounders (pooled adjusted odds ratio [aOR], 2.34; 95% confidence interval [CI], 1.60–3.09).
Six studies examined gestational diabetes as an outcome; 4 used a symptom-based assessment of OSA and 2 used a PSG assessment. An adjusted analysis showed an increased risk for gestational diabetes with SDB (pooled OR, 1.86; 95% CI, 1.30–2.42), which was slightly lower than the unadjusted pooled OR of 2.11 (95% CI, 1.38–3.23).
Several studies reported mean birthweights of infants born to mothers with and without SDB. Two of these reported significantly lower mean birthweight in SDB mothers. There was significant heterogeneity between these studies, and thus, the data could not be reliably pooled for a summary effect size.
Stout: How might the varied diagnostic techniques cause bias in observational studies included in this meta-analysis? In what direction might these bias the findings?
Trudell: The varied diagnostic criteria for identifying SDB add to between-study heterogeneity. If a study included only patients with a PSG diagnosis, it may represent a more severely affected population, whereas studies that used both PSG diagnoses and clinical symptom scoring might have a population with more mixed disease severity. If patients' diagnoses are based only on symptom scoring, some of those included in a study might not truly have OSA or SDB. This type of misclassification, for example including patients who did not truly have OSA, could lead to type II error, or bias the findings of that individual study toward the null. As a result, the chance of missing a real association would be increased. For this reason, metaanalyses are important. Hopefully, by pooling the results of a number of studies in a metaanalysis, the effect of misclassification is overcome and a “true” pooled effect size can be obtained.
Stout: If we accept the authors' conclusion that a significant association exists between SDB and adverse outcomes, how you think this finding could be incorporated into prenatal care? What are the pros and cons of symptom-based vs polysomnographic diagnosis?
Wood: If we accept the study conclusions, we could consider conducting an evaluation for signs or symptoms of sleep-disordered breathing in the routine prenatal visit. Importantly, many women may not have medical care except during pregnancy, and in this group, prenatal care serves as a window of opportunity for health-care screening that might not otherwise happen. Women with a positive screen could be referred onward for further evaluation, including possible sleep studies or PSG. The assessment could be incorporated into prenatal care relatively simply. For example, a questionnaire that reviews symptoms of SDB could be given at the start of prenatal care and once again at each trimester, since a patient's symptoms can change throughout pregnancy. Some of the benefits of symptom-based diagnosis include the low cost and ease of identifying patients at risk.
However, interventions such as continuous positive airway pressure might require a PSG diagnosis. Furthermore, it is not known, based on this study, whether interventions to overcome sleep-disordered breathing during pregnancy would, in fact, improve outcomes. Further research should investigate whether interventions such as continuous positive airway pressure actually lead to a decrease in adverse outcomes in these patients. This would be essential to determining the utility of screening for SDB in pregnancy.
Published online: November 21, 2013
© 2014 Published by Elsevier Inc.