Effect of maternal positioning during cardiopulmonary resuscitation: a systematic review and meta-analyses

We searched the Cochrane Central Register of Controlled Trials (CENTRAL), MEDLINE, Embase and OpenGrey databases for relevant studies on 16 November 2019, and we updated them on 3 April 2021. We did not restrict the publication year. We also checked the reference lists of all included studies and relevant existing systematic reviews for additional studies. We used subject headings in combination with key words. We devised three sets of search terms: (i) population of interest (pregnant women), (ii) health condition of interest (cardiac arrest) and (iii) intervention (or exposure) evaluated (Table 1).

The study population included pregnant women who experienced cardiac arrest. We made no restrictions regarding maternal age, care settings or nationality. Regarding the intervention, we included studies that examined the effect of maternal positioning or methods to relieve aortocaval compression during CPR. We also included any type of study (randomised control trials [RCTs], nonrandomised clinical trials and observational studies). Because of the rarity of cardiac arrest during pregnancy, we included simulation-based studies using patient mannequins. We excluded reviews and commentaries as well as studies without English language abstracts (Table 2).

The primary outcomes of interest included the survival rate of mothers or fetuses/neonates with favourable neurologic outcomes and the return of spontaneous circulation following maternal cardiac arrest. The secondary outcomes of interest were the quality of CPR and any adverse events.

We imported identified studies into Covidence, a web-based tool for systematic reviews. Two review authors (NE and TY) independently screened the studies for relevance based on titles and abstracts; they then screened based on full texts. We resolved any discrepancies via discussions with the review team until we reached a consensus.

Using data extraction forms designed specifically for this review, two review authors (NE and MF) extracted data from the included studies. We contacted the authors of original studies to obtain missing information and unpublished data. Two review authors (MF and NE) independently assessed the risk of bias in the included studies using a revised Cochrane risk-of-bias tool (RoB2) developed specifically for crossover trials [29] because all studies included in this review applied a crossover design.

Findings from nonrandomised crossover studies are presented narratively. Whenever sufficient data were available from RCTs to estimate the effect size of the intervention, we conducted meta-analyses using Cochrane's Review Manager (RevMan) version 5.3 [30]. We calculated the weighted mean difference and 95% confidence interval (CI) for continuous outcomes. We performed the random effects meta-analyses, because we assumed that the impact of the maternal positioning during cardiopulmonary resuscitation varied from study to study [31]. We assessed clinical heterogeneity (e.g. variability in the interventions such as chest compression on different surfaces; the floor or on the bed) as well as methodological heterogeneity (e.g. variability in study design such as RCTs or nonrandomised studies) within each comparison. Where meta-analyses were performed, we assessed statistical heterogeneity with tau² in addition to visual inspection of the forest plots [32]. We assessed heterogeneity using tau², rather than I², as tau² is the appropriate measure for indicating the presence of clinically relevant heterogeneity while I² may be misleading as it depends on the sample size of studies [33]. Furthermore, I² is not an absolute measure of heterogeneity [34, 35]. We conducted subgroup analyses based on the clinical heterogeneity (chest compression delivery surfaces). If there was a concern to the robustness of the result caused by missing outcome data, sensitivity analysis would have been performed, by comparing results from different methods of dealing with missing data (e.g. available case analysis, imputed case analysis) [36, 37]

We used the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach [38] to assess the body of evidence for all the identified outcomes. We assigned one of four levels — high, moderate, low or very low — to each outcome by considering five domains, including the within-study risk of bias, inconsistency, indirectness, imprecision and publication bias [39]. If sufficient studies had been available (> = 10), then we would have constructed funnel plots to assess publication bias.

The databases we searched identified 1,836 articles, including 346 duplicates. We screened a total of 1,490 titles and abstracts and selected 79 articles for full-text evaluation. We identified no additional articles from the reference lists of the included studies or review articles, and of the 79 articles that underwent full-text evaluation, we excluded 71 for the reasons stated in the PRISMA flowchart (Figure 1). A total of eight studies met the inclusion criteria, including six crossover RCTs [40‐45] and two nonrandomised crossover studies [46, 47].

An overview of included studies is presented in Table 3. All the available studies used mannequins, and none involved living subjects. One crossover RCT [43] examined the effect of manual left uterine displacement in the supine position and compared the results to those in the left-lateral tilt position. Four crossover RCTs [40‐42, 44] and one nonrandomised crossover study [46] compared the quality of CPR on a mannequin lying supine (manual left uterine displacement) with that of the left-lateral tilt position. One crossover RCT examined the optimal methods for producing lateral tilt [45], and one nonrandomised crossover study [47] examined the effect of chest compression at various angles between 0° and 90° of inclination. All participants in the included studies were health professionals and performed two or more sequential interventions.

Bias due to randomisation: Of the randomised crossover trials included in this review [40‐45], none except one [45] reported the processes used to generate the random allocation sequence and/or allocation concealment. Bias due to deviations from intended interventions: Given the nature of the interventions, participants (rescuers) in all studies were aware of their assigned intervention (e.g. chest compression in the supine or lateral tilting positions) during each period of the trial. Four studies [40, 41, 44, 45] ensured a washout period to minimise the carryover effect (after 2 minutes of chest compressions in the first assigned position, the participants rested for 10 minutes to minimise rescuer fatigue), whereas no information was available to assess the carryover effect in the remaining studies [42, 43, 46, 47]. Bias due to missing outcome data: There were no missing outcomes [40‐44, 46, 47], or the proportion of missing outcomes was small [45]. Bias due to outcome measurement: The outcomes were assessed using the PC SkillReporting software system, which was connected to the patient mannequin (Laerdal Resusci Anne®) in all the included studies [40‐46] except one [47]. Where the participants could not see the monitor screen displaying the outcomes during the chest compressions, the risk of bias was considered low [41, 44]; however, where information regarding blinding of the outcomes was not provided, the risk of bias was rated as of some concern by taking into account the possibility that knowing the outcomes altered the participants’ performance [40, 42, 43, 45‐47]. Bias due to selection of the reported result: No studies provided a trial protocol. Overall: Six RCTs [40‐45] were rated as having some concern for risk bias, whereas the two nonrandomised crossover studies [46, 47] were considered of high risk for bias because at least one domain had a high risk of bias (Table 4).

For all the outcomes included in this review, the quality of evidence was rated as very low using the GRADE criteria. The certainty of evidence was downgraded for risk of bias, indirectness, inconsistency and imprecision of results. More specifically, we downgraded one level for a potential risk of bias due to the randomisation process, insufficient washout period between phases and/or unblinding of participants for all the outcomes measuring for quality of CPR. We also downgraded two levels for indirectness of evidence because all the outcome was assessed with simulation-based studies using hypothetical cardiac arrest maternal patient mannequins. We further downgraded one level for serious unexplained inconsistency (heterogeneity) for quality of chest compression as measured with correct chest compression depth rate and another one for serious imprecision (wide confidence intervals) of the mean effect for correct chest compression depth rate and correct hand position rate.

Study effect estimates for quality of chest compression varied between RCTs and nonrandomised study with conflicting results; RCTs favouring the spine position and nonrandomised study [46] favouring the left-lateral tilt position. The nonrandomised study was published in 1992, whereas RCTs were published more recently, in the 2010s.Some of this variation is likely to be caused by differences in the definition used for measuring quality of chest compression that reflect changes in the clinical guidelines’ recommendations for CPR in past decades, which we discuss further in the Comparison with Existing Guidelines and Reviews section below. There is a lack of clarity of the definition of high-quality chest compression and high risk of selection bias due to a lack of randomisation in the non-randomised study. Therefore, we only included the results from RCTs in the meta-analyses.

Heterogeneity is not a reason for downgrading the evidence for quality of CPR apart from quality of chest compression as measured using the percentage of correct chest compression depth (Tau² = 48.95, I² = 47%). The percentage of correct chest compression depth was consistently lower in the left-lateral tilt position than the supine position. However, the effect of size (mean difference in the percentage of correct chest compression depth) varied from study to study: three appear to have large effects (15.9–40.5%) and one much smaller effect (5.7%). There are many probable causes of heterogeneity, which cannot be explained by a subgroup analysis by chest compression delivery surfaces (floor or bed) or study population (experienced or inexperienced rescuers). The estimated effect of maternal positioning is larger, 57% [42], when a chest compression was performed from patients’ right side, showing lower correct percentage (27%) in the left-lateral tilt position, compared with the spine position (86%). We need more studies to gain a reliable estimate of heterogeneity and reasons for it.

The 2020 American Heart Association (AHA) Guidelines [48] recommend that “priorities for the pregnant woman in cardiac arrest should include provision of high-quality CPR and relief of aortocaval compression through left-lateral uterine displacement” (Supplement, p. 454). This recommendation is based primarily on the physiology of pregnancy, extrapolations from the non-arrest pregnancy states [49, 50] and non-randomised simulation-based studies [46, 47]. However, the interpretation of the recommendation is not straightforward because the recommendation was based on inconsistent results, including the non-randomised simulation-based studies [46, 47] conducted 30 to 40 years ago.

One of the studies often utilised in clinical practice guidelines is Goodwin’s non-randomised simulation-based study published in 1992 [46]. Goodwin found that chest compression quality was more reduced in the supine position than in the wedged position, using the human wedge manoeuvre. According to Goodwin, the common reason for inaccuracy is “compression of too great a force” ([46], p. 434), but the correct definition of cardiac compression and compression were not provided. Our systematic review revealed that findings from recent RCTs contradict Goodwin’s findings, possibly because of changes in CPR recommendations in past decades [51]. For example, the current CPR guidelines recommend a target chest compression depth of 5–6 cm, whereas it was 4–5 cm (AHA Guidelines 2005) in the past. Even further back, it was defined as the difference in the height of a rescuer’s shoulder, not in a victim’s chest, using 2.5–5 cm in the 1992 AHA Guidelines [52].

Although our review included only indirect evidence from simulation-based studies, the trials included had more sophisticated studies that overcome the methodological limitations commonly observed in previous studies (such as lack of randomisation, the potential risk of carryover effect and the inaccuracy of measuring outcomes). Our results showed that resuscitation in the supine position enhances the quality of the resuscitation activity. Together with evidence from previous systematic reviews on the non-arrest pregnant population [17, 53] that shows that manual left uterine displacement effectively relieves aortocaval pressure in pregnant women with hypotension, it is reasonable to conclude that manual left uterine displacement in the supine position is more effective than a left-lateral tilt position to increase the chest compression quality during resuscitation. This can, in turn, contribute to increased maternal and foetal survival rates following maternal cardiac arrest.

Given its systematic and comprehensive literature search, our review enhanced evidence regarding the effect of maternal positioning during maternal CPR, particularly on the quality of chest compressions. Where we found information in the included studies to be insufficient, we contacted the original researcher, if doing so was possible. However, the quality of evidence produced by our systematic review was still poor, mainly because of indirect evidence from the mannequin studies.

We did not identify any study that evaluated the effect of maternal positioning using real patients. Therefore, there were no data on survival rates and the return of spontaneous circulation following maternal cardiac arrest. Foetal/neonatal outcomes were also unavailable. Thus, the only outcomes available constituted indirect evidence of the quality of CPR, which was obtained from simulation-based studies using hypothetical cardiac arrest maternal patient mannequins. Therefore, there are serious limitations regarding the applicability and transferability of the findings of our systematic review to real maternal patients.

From the study design point of view, all studies included in our review were crossover trials in which each healthcare professional involved performed chest compressions on a mannequin in two or more maternal positions in random order. Because each participant acted as their own control, this design allowed them to express the difficulty with chest compressions that they experienced during a particular maternal position. The crossover trials could have provided more precise effect size estimates than parallel-group trials if appropriate statistical analyses (paired analyses) had been applied [54, 55]. However, this was not the case in some of our included studies. Data on within-subjects correlation were unavailable, so this advantage of a crossover design could not be utilised. Our meta-analysis estimated the average effect of an intervention (maternal positioning), but given the small number of studies to be synthesised for each outcome, the statistical model used for the meta-analysis (random effects model) could not estimate the between-study variance (the extent of variation among the effects observed in different studies).

Knowledge gaps still exist concerning the effect and efficacy of CPR with manual uterine displacement versus tilt positioning on clinical outcomes following maternal cardiac arrest. Simulation-based RCTs specifically designed to evaluate the favourable or unfavourable effects of manual left uterine displacement should be carried out to assess the quality of CPR, including delay and interruption of CPR in relation to performing manual left uterine displacement. Further studies also must focus on establishing what could be the best strategies (including for manual left uterine displacement) for high-quality CPR. This is important because there are various recommendations regarding manual left uterine displacement, possibly referencing the situations or settings wherein maternal cardiac arrest occurs. For example, guidelines recommend ‘placing a hand below the uterus on the maternal right and pushing the uterus slightly upwards and to the left’ ([50], p.29), which can be done with one-hand or two-hand techniques and from the left or right side of the patient [24, 48, 56]. There is, however, a lack of evidence about whether and how these different strategies affect the quality of CPR. Because maternal cardiac arrest is rare and RCTs evaluating the effects of maternal position with real patients would be unrealistic, the development of a nationwide database that collects data concerning both in-hospital and out-of-hospital maternal cardiac arrest patients would be beneficial. Such a database would be critical to predict the clinical outcomes of such cases, including the survival rates of mothers and babies with favourable neurologic outcomes, after cardiac arrest vis-à-vis the strategies used for relieving aortocaval compression during maternal resuscitation.