List of Abbreviations
Chapter 3 Systematic review and meta-analysis 3.1 Introduction 3.1 Introduction
3.4 Discussion .1 Limitations .1 Limitations
3.4.2 Key findings
The aim of this review was to evaluate the findings of published studies examining the association between sleep and pregnancy outcomes. It found that only a very limited number of studies had examined the association between unfavourable sleep events and pregnancy outcomes; and that not all unfavourable sleep characteristics and pregnancy outcomes had received an equal level of attention in the literature. Indeed, some sleep events (i.e. latency, disturbance and daytime sleepiness) and some pregnancy outcomes (i.e. Apgar score, birth weight and late maternal pregnancy events) have not been examined at all to-date. Nonetheless, despite the relatively few sleep characteristics and pregnancy outcomes examined, there was still evidence of a general tendency towards positive associations between unfavorable sleep and poor pregnancy outcomes in which unfavorable sleep characteristics were associated with an increased risk of poor pregnancy outcomes. However, much of this evidence had multiple methodological flaws, and there was some evidence of publication bias. These methodological flaws decreased the certainty offered by the current evidence available, and undermined the level of precision available – so that the summarized findings cannot be interpreted with substantial confidence and should be applied with circumspection and a degree of scepticism.
The review also found that the magnitude of the increased risk of poor pregnancy outcomes associated with unfavourable sleep characteristics varied according to the pregnancy outcome and/or sleep characteristic examined. This also varied in relation to a range of specific study design criteria, including: the design of the primary studies; the inclusion and exclusion criteria used in the recruitment of study participants; the gestational age at which sleep was measured; the choice of covariates included in any covariate adjustment sets; the appropriateness of any adjustment achieved; and the use of various reference points for what were considered ‘normal’ sleep characteristics as well as variation in the measurement tools used – all of which appeared to have affected the magnitude of the estimated risks, causing either an increase or decrease in the estimated ORs, and (in a few instances) appeared to have reversed the direction of the association observed.
Although our assumptions regarding the possible impact(s) of these aspects of study design and analysis on the magnitude and direction of the estimated ORs might be true, the lack of the precisely estimated ORs made it difficult to assess or confirm with any degree of confidence whether these aspects were actually the reasons why substantial differences were evident amongst the estimated results.
However, considering these aspects of study design, data measurement and analysis, and their likely impact on the estimation of any casual pathway between sleep and pregnancy outcomes (while also considering the potential bidirectional nature of any relationship between sleep and the majority of covariates included in the regression models examined in this review), it is sensible to consider each of the following issues (each of which were discussed previously in the methodology chapter [Chapter 2, Section 2.4, page 60] when the original ‘hypothesized’ DAG was first presented and explained):
I. First, the importance of considering each study’s design when specifying the relationship between sleep and pregnancy outcomes (particularly whether the design is retrospective or prospective).
II. Second, the importance of the temporal sequence of events, since this can be crucial when specifying the direction of any causal arc when dealing with potentially bidirectional relationships.
III. Third, the importance of reporting/knowing the time (i.e. the gestational age) at which sleep and other covariates were measured (to help avoid the inclusion of mediators in the covariate adjustment sets used).
IV. Fourth, the importance of carefully considering the timing of measurements when repeated measurements of sleep and/or other covariates are available (again, to help avoid the inclusion of mediators in the covariate adjustment sets used).
In addition to the covariates included in the hypothesized DAG described earlier in the present thesis, this was further elaborated by the inclusion of variables that had been adjusted for by one or more of the studies included in this chapter’s review–
each of which were carefully categorised as: sociodemographic indicators;
behavioral risk indicators; current (and/or previous) maternal health indicators;
current (and/or previous) obstetric-related variables; psychological indicators; and fetal health indicators. However, some of these (additional) variables were also considered likely to have had bidirectional relationships with sleep, and as such might have been likely to behave as mediators (rather than confounders) according to their position in the DAG.
3.4.3 Conclusion
In conclusion, this review of previous studies examining the association between (a range of) sleep (characteristics) and (a range of) pregnancy outcomes found some evidence that such associations do indeed exist. However, this evidence is limited in scope since not all measurable sleep characteristic and not all available pregnancy outcomes were examined by these studies; while sleep was never
‘holistically’ defined (an important consideration given its complex nature and substantial variability between each of the different trimesters of pregnancy).
Additionally, the evidence summarized and appraised in the present review is highly variable in quality both in terms of data (quality) and the presence of a number of potential biases (most importantly selection bias and confounding bias).
Finally, many of the evidence summarized herein came from studies that appear severely underpowered.
Meanwhile, the possibility of publication bias may indicate that much of the currently published evidence is prone to type I and type II error – not least because the majority of published results are nominally ‘statistically significant’ even though many of the studies involved appear underpowered. As such, it seems likely that the currently available/published evidence overestimates the strength of the relationship between sleep and pregnancy outcomes.
3.4.4 Recommendations
Further research is needed in this area with: greater consistency in the measurement of sleep and pregnancy outcomes (particularly with regard to the reference point used, the measurement tool used, and the gestational age when sleep was measured); more evidence from less commonly examined sleep characteristics and pregnancy outcomes (particularly sleep latency and macrosomia, respectively); and more robust study designs and analytical techniques (particularly as regards sampling, power estimation and adjustment for confounding). There is also scope for extending the research undertaken to include more participants with additional risks of poor pregnancy outcome (such as those with gestational diabetes or those at risk of diabetes), who are often excluded from existing study populations. Improving the conceptualization of sleep (beyond the definition and measurement of its many disparate, yet inherently related, characteristics and components), would also be worth pursuing (using, for example, latent class analysis).