The need for this work

Regular physical exercise (PE) is associated with physical, psychological and social benefits in the general population.1 In recent years, a growing interest has emerged on the beneficial effects of PE during gestation.2–4 In fact, professional associations of obstetricians and gynaecologists5 and, more recently, international guidelines for physical activity1 6 endorse PE throughout the gestational period recommending pregnant women to accumulate at least 150 min of moderate-intensity aerobic activity per week, distributed over at least 3 days a week. However, being active every day is the most beneficial for maternal health.5 7

This evidence comes from several systematic reviews and meta-analyses supporting that prenatal exercise, besides benefiting newborn infants, could reduce the mothers’ risk for some disorders, such as gestational diabetes mellitus, excessive maternal weight gain, pre-eclampsia and hypertensive disorders of pregnancy, incontinence urinary or postpartum depression.3 4 8 9 Furthermore, maternal PE reduces caesarean delivery rates in healthy pregnant women.10 Despite this, <15% of pregnant women follow the physical activity recommendations.11 12

Moreover, although numerous systematic reviews and meta-analyses have addressed the effects of PE on maternal health, evidence regarding the risk of caesarean section, birth weight or Apgar score at delivery is still controversial. In this sense, a recent systematic review and meta-analysis13 did not find a significant association between prenatal exercise and the risk of caesarean section, while two meta-analyses indicated a decrease in the risk of caesarean delivery among pregnant women who exercised.10 14 Likewise, another meta-analysis concluded that prenatal PE was not associated with birth weight or Apgar score at delivery,15 whereas two previous meta-analyses supported that newborns of mothers who were active during pregnancy had a lower weight within the normal range and higher Apgar scores than their counterparts.16 17 However, despite these inconclusive findings, because the effects of exercise in women with excess of weight continue to be an unresolved question for both clinicians and pregnant women, it is worthwhile to conduct other broader research approaches, such as an umbrella review, which increases the likelihood of providing more consistent evidence on this issue.

If this happens when dealing with healthy pregnant women, it is not surprising that practitioners are reluctant to recommend exercise to pregnant women suffering from some disorders, such as hypertension or pre-eclampsia, since guidelines include these disorders as absolute or relative contraindications to exercising during pregnancy.5 Similarly, the PE recommendation for overweight pregnant women and pregnant women with obesity continues to be a debatable issue due to the low quality of evidence regarding its benefits.2

Finally, the safety or additional benefits of vigorous exercise for pregnant women and fetus health are widely debated. Although consistent evidence supports the beneficial effects of moderate PE in healthy pregnant women, the scarcity of studies addressing the risks and benefits of exercise at higher intensity prevents practitioners from recommending PE at higher dosages in terms of frequency, duration or intensity than recommended in clinical guidelines.2 6

Therefore, considering the myriad of systematic reviews and meta-analyses of randomised controlled trials (RCTs) addressing the impact of exercise during pregnancy on different maternal health outcomes, and the contradictory findings of these previous reviews, umbrella reviews represent an appropriate design to elucidate the reasons behind the conflicting findings of previous systematic reviews, and to provide clinicians and policymakers with an overall assessment of the evidence on this issue, which is necessary for both practitioners and pregnant women.18

Selection of studies and data extraction

First, record titles and abstracts will be independently evaluated to identify eligible studies according to the inclusion and exclusion criteria. Then, the full-texts of possible eligible studies will be comprehensively reviewed by two investigators (GS-M and RP-L). Disagreements will be solved by consensus between them, but if disagreements persist, a third investigator will solve the conflict (BN-P). The two investigators will extract data (authorship, date, study characteristics, type of exercise, main outcome and quality assessment tool) from each included study. Data extraction forms have been designed ad hoc (see online supplementary files 1 and 2). Corresponding authors will be contacted when there are missing data or to clarify unclear information.

Assessment of risk of bias and methodological quality of included studies

The methodological quality of the included systematic reviews and meta-analyses will be evaluated using the AMSTAR 2 tool,21 which was developed and validated to critically assess the quality of systematic reviews and meta-analyses. This instrument includes 16 criteria referring to relevant methodological aspects of studies. The quality of studies will be classified, according to the number of approved criteria, as follows: excellent, 15–16; very good, 12–14; good, 9–11; acceptable, 6–8 and deficient, 3–5.

The risk of bias (quality) for the RCTs selected for the updated systematic review and meta-analysis will be assessed following the Cochrane Collaboration’s methodology. This tool is based on eight potential sources of bias: random sequence generation; allocation concealment; blinding of participants, of the evaluator, of the outcome assessment; incomplete outcome data; missing data and other.

Finally, the certainty of evidence and strength of recommendations for meta-analyses will be assessed using the Grading of Recommendations Assessment, Development and Evaluation (GRADE). This tool provides a rating of ‘high’, ‘moderate’, ‘low’ or ‘very low’ quality, and will provide a ‘weak’ or ‘strong’ recommendation. This will be accomplished using the GRADEpro software, and output tables will be added.

Data analysis

Tables will be designed to summarise the key characteristics of the included studies. Additionally, forest plots will be used to show results extracted from each meta-analysis.

Assessment of summary effects and heterogeneity

For each meta-analysis, the summary meta-analytic estimates and corresponding 95% CI will be calculated using both fixed-effects and random-effects models.22 23 The 95% prediction intervals will also be estimated for the summary random-effects estimates, which will account for the between-study heterogeneity and as well as explaining the uncertainty for the effect that could be expected if a new study examines the same association.24–26 Thus, this 95% prediction interval indicates the range where the true effect is expected for 95% of studies from the population of the included studies in the meta-synthesis or similar studies potentially conducted in the future. Additionally, for the largest RCT of each meta-analysis, the SD of the effect size will be calculated and scrutinised if the SD is <0.10.27 28 Since a higher accuracy on detecting publication bias has been empirically demonstrated using 0.1 as the threshold for significance in the most well-known publication bias tests,29 in our study the significance p values for Egger’s test is setup at 0.1. When meta-analyses have continuous data, the estimated effect will be converted to their equivalent ORs using accepted calculation strategies. For other measures, such as the mean difference or the risk difference, a few general estimations will be needed, such as, Glass’Δ or RR, respectively.30 31

Among-study heterogeneity will be assessed using the I² statistic.32 Usually, I² ranges between values of <25%, 25%–50%, 50%–75% and >75% which represent small, medium, large or very large amounts of heterogeneity, respectively.25 The corresponding p values will also be considered. Studies with insufficient data to perform the analyses will be omitted from the data synthesis. When substantial heterogeneity will prevent the calculation of pooled estimates of outcomes, a systematic review or narrative synthesis will be undertaken.

Whenever possible, a meta-analysis will be conducted including the most recent RCTs on this issue not included in previous meta-analyses, and pooled effect size estimates were calculated with their 95% CIs. Moreover, when several studies have been published after the latest meta-analysis, we will first conduct an additional meta-analysis including only the most recent studies, and then we will carry out the umbrella review that will include the newest one. Additionally, for outcomes of studies where a meta-analysis will not be possible, a narrative synthesis of the results will be presented.

Subgroup analysis and meta-regression

In the new meta-analysis, we will carry out subgroup and meta-regression analyses to examine influence of potential mediators such as gestational weight gain on the main outcome. We will also conduct meta-regression analysis on some intervention (length of intervention, duration of intervention sessions) and women (BMI, gestational age at delivery or birth weight) related variables. Finally, subgroup analyses by categorical variables as, type and intensity of exercise or weight status will be conducted.

Small studies effect assessment

Small study effects usually indicate publication or other reporting biases, although these effects may also reflect chance, genuine heterogeneity or other differences between large and small studies.33 The existence of a potential small study effect will be assessed, if small studies tend to show larger estimates of effect size in contrast to larger studies, using the regression asymmetry Egger’s test for continuous outcomes, and Harbord’s test for dichotomous ones. A p value <0.10 will be used to show evidence of small-study effects.34

Excess of significance evaluation

The excess significance test will be used to evaluate whether the observed number of studies (O) included in each meta-analysis with statistically significant results (positive studies, p<0.05) is different than the expected number of studies with significant results (E).35 The effect size of the largest study (smallest SE) in a meta-analysis will be used to calculate the statistical power of each component study.36 Furthermore, the largest study effect will be assumed to be the true effect. A two-sided p<0.10 will be considered statistically significant.

Then, the comparison between the observed and the expected number of studies will be done separately for each meta-analysis, and it will be amplified to groups including many meta-analyses when the observed and the expected values from each meta-analysis will be summed.

All statistical analyses and power calculations will be performed using STATA V.15.1 software (StataCorp, College Station, Texas, USA).