Article Text

Download PDFPDF

Prenatal exercise is not associated with fetal mortality: a systematic review and meta-analysis
  1. Margie H Davenport1,
  2. Amariah J Kathol1,
  3. Michelle F Mottola2,
  4. Rachel J Skow1,
  5. Victoria L Meah3,
  6. Veronica J Poitras4,
  7. Alejandra Jaramillo Garcia4,
  8. Casey E Gray5,
  9. Nick Barrowman6,
  10. Laurel Riske1,
  11. Frances Sobierajski1,
  12. Marina James1,
  13. Taniya Nagpal2,
  14. Andree-Anne Marchand7,
  15. Linda G Slater8,
  16. Kristi B Adamo9,
  17. Gregory A Davies10,
  18. Ruben Barakat11,
  19. Stephanie-May Ruchat12
  1. 1 Program for Pregnancy and Postpartum Health, Physical Activity and Diabetes Laboratory, Faculty of Kinesiology, Sport, and Recreation, Women and Children’s Health Research Institute, Alberta Diabetes Institute, University of Alberta, Edmonton, Alberta, Canada
  2. 2 R Samuel McLaughlin Foundation, Exercise and Pregnancy Laboratory, School of Kinesiology, Faculty of Health Sciences, Department of Anatomy and Cell Biology, Schulich School of Medicine and Dentistry, Children’s Health Research Institute, The University of Western Ontario, London, Ontario, Canada
  3. 3 Cardiff School of Sport and Health Sciences, Cardiff Metropolitan University, Cardiff, UK
  4. 4 Independent Researcher, Ottawa, Ontario, Canada
  5. 5 Healthy Active Living and Obesity Research Group, Children’s Hospital of Eastern Ontario Research Institute, Ottawa, Ontario, Canada
  6. 6 Clinical Research Unit, Children’s Hospital of Eastern Ontario Research Institute, Ottawa, Ontario, Canada
  7. 7 Department of Anatomy, Université du Québec à Trois-Rivières, Trois-Rivières, Quebec, Canada
  8. 8 John W Scott Health Sciences Library, University of Alberta, Edmonton, Alberta, Canada
  9. 9 Faculty of Health Sciences, University of Ottawa, Ottawa, Ontario, Canada
  10. 10 Department of Obstetrics and Gynecology, Queen’s University, Kingston, Ontario, Canada
  11. 11 Facultad de Ciencias de la Actividad Física y del Deporte-INEF, Universidad Politécnica de Madrid, Madrid, Spain
  12. 12 Department of Human Kinetics, Université du Québec à Trois-Rivières, Trois-Rivières, Quebec, Canada
  1. Correspondence to Dr Margie H Davenport, Program for Pregnancy and Postpartum Health, Faculty of Kinesiology, Sport, and Recreation, University of Alberta, Edmonton, AB T6G 2E1, Canada; mdavenpo{at}ualberta.ca

Abstract

Objective To perform a systematic review of the relationship between prenatal exercise and fetal or newborn death.

Design Systematic review with random-effects meta-analysis and meta-regression.

Data sources Online databases were searched up to 6 January 2017.

Study eligibility criteria Studies of all designs were included (except case studies) if they were published in English, Spanish or French and contained information on the population (pregnant women without contraindication to exercise), intervention (subjective or objective measures of frequency, intensity, duration, volume or type of exercise, alone [“exercise-only”] or in combination with other intervention components [eg, dietary; “exercise + co-intervention”]), comparator (no exercise or different frequency, intensity, duration, volume and type of exercise) and outcome (miscarriage or perinatal mortality).

Results Forty-six studies (n=2 66 778) were included. There was ‘very low’ quality evidence suggesting no increased odds of miscarriage (23 studies, n=7125 women; OR 0.88, 95% CI 0.63 to 1.21, I2=0%) or perinatal mortality (13 studies, n=6837 women, OR 0.86, 95% CI 0.49 to 1.52, I2=0%) in pregnant women who exercised compared with those who did not. Stratification by subgroups did not affect odds of miscarriage or perinatal mortality. The meta-regressions identified no associations between volume, intensity or frequency of exercise and fetal or newborn death. As the majority of included studies examined the impact of moderate intensity exercise to a maximum duration of 60 min, we cannot comment on the effect of longer periods of exercise.

Summary/conclusions Although the evidence in this field is of ‘very low’ quality, it suggests that prenatal exercise is not associated with increased odds of miscarriage or perinatal mortality. In plain terms, this suggests that generally speaking exercise is ‘safe’ with respect to miscarriage and perinatal mortality.

  • exercise
  • pregnancy
  • miscarriage
  • stillbirth
  • perinatal mortality

Statistics from Altmetric.com

Introduction

Miscarriage is a common adverse outcome of pregnancy before 20 weeks gestation.1 Approximately, 25% of women will experience at least one miscarriage in their lifetime.1 While a large proportion of miscarriages arise from chromosomal abnormalities that are not compatible with life, up to 50% have a normal karyotype and are due to other causes.2

Perinatal mortality includes stillbirth (20 weeks gestation or later) and infant mortality (early neonatal death up to 28 days of life).3 4 In 2015, the incidence of stillbirth in developed countries was approximately 3.4 per 1000 total births with higher rates in undeveloped countries (18.4 stillbirths per 1000 total births).5 Approximately, 10% of stillbirths worldwide are associated with obesity, diabetes and hypertension,3 disorders which may be improved with regular exercise. This is particularly relevant in high-income countries6 where physical activity rates are low and rates of such diseases are increasing. Therefore, understanding the impact of modifiable risk factors, such as physical activity, on the risk of fetal and newborn mortality is of critical importance.

Current national and international guidelines recommend that women without contraindications to exercise be physically active throughout pregnancy.7–9 Although prenatal exercise is strongly encouraged by health organisations around the world, 85% of pregnant women fail to meet current guidelines for prenatal exercise of 150 min per week of moderate intensity exercise.10 Of particular relevance to the first trimester, there is a societal fear that exercise may increase the risk of miscarriage—a perception that contributes to decreased participation in exercise early in pregnancy.11

Therefore, the purpose of this systematic review was to evaluate the relationship between prenatal exercise and fetal or newborn mortality (miscarriage and perinatal mortality). This paper forms part of a series of reviews which will inform the 2019 Canadian guideline for physical activity throughout pregnancy (herein referred to as Guideline).12

Methods

In October 2015, the Guidelines Consensus Panel including researchers, methodological experts, a fitness professional and representatives from the Society of Obstetricians and Gynaecologists of Canada (SOGC), Canadian Society for Exercise Physiology (CSEP), The College of Family Physicians of Canada, Canadian Association of Midwives, Canadian Academy of Sport and Exercise Medicine, Exercise is Medicine Canada and a representative health unit (the Middlesex-London Health Unit), assembled to identify priority outcomes for the Guideline update. The Guidelines Consensus Panel selected 20 ‘critical’ and 17 ‘important’ outcomes related to prenatal exercise and maternal/fetal health. One of the ‘critical’ outcomes (ie, fetal and newborn mortality) is examined in this review. This systematic review and meta-analysis was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines and the checklist was completed.13

Protocol and registration

Two systematic reviews examining the impact of prenatal exercise on fetal, newborn and maternal health outcomes were registered with PROSPERO, the international prospective register of systematic reviews (fetal health: Registration no. CRD42016029869; Available from: https://www.crd.york.ac.uk/PROSPERO/display_record.asp?ID=CRD42016029869; maternal health: Registration no. CRD42016032376; https://www.crd.york.ac.uk/PROSPERO/display_record.asp?ID=CRD42016032376). Because the relationships between prenatal exercise and fetal or newborn mortality are examined in studies related to fetal, newborn and maternal health, records retrieved from the searches for both of these reviews were considered for inclusion in the present study.

Eligibility criteria

This study was guided by the participants, interventions, comparisons, outcomes and study design framework.14

Population

The population of interest was pregnant women without contraindication to exercise (according to the SOGC/CSEP and American College of Obstetricians and Gynaecologists (ACOG) guidelines).7 8 15 Absolute contraindications to exercise were defined as: ruptured membranes, premature labour, persistent second or third trimester bleeding, placenta previa, pre-eclampsia, gestational hypertension, incompetent cervix, intrauterine growth restriction, high order pregnancy, uncontrolled type 1 diabetes, hypertension or thyroid disease or other serious cardiovascular, respiratory or systemic disorders. Relative contraindications to exercise were defined as: a history of spontaneous abortion, premature labour mild/moderate cardiovascular or respiratory disease, anaemia or iron deficiency, malnutrition or eating disorder, twin pregnancy after 28 weeks or other significant medical conditions.7 8 15

Intervention (exposure)

The intervention/exposure was objective or subjective measures of frequency, intensity, duration, volume or type of exercise. Although exercise is a subtype of physical activity, for the purpose of this review, we used the terms interchangeably. Exercise was defined as any bodily movement generated by skeletal muscles that resulted in energy expenditure above resting levels.16 Prenatal exercise could be acute (ie, a single exercise session) or habitual (ie, usual activity). Interventions including exercise alone (termed ‘exercise-only’ interventions) or in combination with other interventions (such as diet; termed ‘exercise+co-interventions’) were considered. Exercise-only interventions could include standard care. Studies were not eligible if exercise began after the initiation of labour.

Comparison

Eligible comparators were: no exercise or different frequencies, intensities, durations, volumes and types of exercise or exercise in different trimesters.

Outcome

Relevant outcomes were miscarriage (fetal mortality <20 weeks) and perinatal mortality (stillbirth [20 weeks gestation or later] and infant mortality [early neonatal death up to 28 days of life]).

Study design

Studies of any design, except for case studies, were included. Studies were also excluded if they were not original research (narrative or systematic reviews and meta-analyses).

Information sources

A comprehensive search was created and run by a research librarian (LGS) in the following databases: MEDLINE, EMBASE, PsycINFO, Cochrane Database of Systematic Reviews, Cochrane Central Register of Controlled Trials, Scopus and Web of Science Core Collection, CINAHL Plus with Full-text, Child Development and Adolescent Studies, ERIC, Sport Discus, ClinicalTrials.gov and the Trip Database up to January 6, 2017 (see online Supplementary file 1 for complete search strategies).

Supplementary file 1

Study selection and data extraction

Records were imported into EndNote for de-duplication and subsequently reviewed for inclusion using EndNote (for maternal health search) or RefWorks (for fetal or newborn health search). The titles and abstracts of all retrieved articles were independently screened by two reviewers. Abstracts that were deemed to have met the initial screening criteria by at least one reviewer were automatically retrieved as full-text articles. Full-text articles were independently screened by two reviewers against study inclusion criteria prior to extraction. For studies where at least one reviewer recommended exclusion, further review was conducted by MHD and/or SMR for final decision on exclusion. In the event of a disagreement that could not be resolved through discussion, the study characteristics were brought to the Guidelines Steering Committee who oversaw the systematic reviews (MHD, MFM, SMR, CG, VP, AJG and NB) for final decision about inclusion/exclusion by consensus. Studies that were selected for inclusion from either the maternal or fetal search strategy were imported into DistillerSR for data extraction. Included studies from the maternal, fetal or newborn searches were de-duplicated against one another in DistillerSR and were considered as one review going forward. A staged approach was used to determine if there was sufficient evidence from randomised controlled trials (RCTs) for each outcome to inform the Guideline, or if it was necessary to consider other study designs. If fewer than 2000 women were included in the meta-analysis, the impact of prenatal exercise on the specific outcome was examined further utilising observational evidence (non-randomised interventions, cohort, cross-sectional and case–control studies).

Data extraction tables were created in DistillerSR in consultation with methodological experts and the Guidelines Steering Committee. Data from records that met the inclusion criteria were extracted by one person and independently verified by a content expert (MHD, MFM or SMR). For each unique study, the most recent or complete version (publication) was selected as the ‘parent’ paper; however, relevant data from all publications related to each unique study were extracted. Data extracted were study characteristics (ie, year, study design, country) and characteristics of the population (eg, number of participants, age, pre-pregnancy body mass index (BMI), parity and pregnancy complications including pre-eclampsia, gestational hypertension and gestational diabetes), intervention/exposure (prescribed and/or actual exercise frequency, intensity, time and type of exercise, duration of the exercise training, measure of physical activity) and outcomes (miscarriage and perinatal mortality). If data were unavailable for extraction, authors were contacted to request additional information.

Quality of evidence assessment

The Grading of Recommendations Assessment, Development and Evaluation (GRADE) framework was used to assess the quality of evidence across studies for each study design and health outcome.

Evidence from RCTs began with a ‘high’ quality of evidence rating and was graded down if there was concern with risk of bias, indirectness, inconsistency, imprecision or risk of publication bias, because the presence of these factors reduces the level of confidence in the observed effects. Evidence from all non-randomised interventions and observational studies began with a ‘low’ quality rating and, if there was no cause to downgrade, was upgraded if applicable according to the GRADE criteria (eg, large magnitude of effect, evidence of dose response).

Specifically, the risk of bias in RCTs and non-randomised intervention studies was assessed following the Cochrane Handbook,15 and risk of bias in observational studies was assessed using the characteristics recommended by Guyatt et al 16, which has been used by other physical activity reviews.17 Studies of all designs were screened for potential sources of bias including selection bias (RCT/intervention: inadequate randomisation procedure; observational: inappropriate sampling), reporting bias (selective/incomplete outcome reporting), performance bias (RCT/intervention: compliance to the intervention; observational: flawed measurement of exposure), detection bias (flawed measurement of outcome), attrition bias (incomplete follow-up, high loss to follow-up) and ‘other’ sources of bias. Risk of bias across studies was rated as ‘serious’ when studies having the greatest influence on the pooled result (assessed using weight (%) given in forest plots or sample size in studies that were narratively synthesised) presented ‘high’ risk of bias. The greatest influence on the pooled result was determined as followed: the studies that had the greatest individual percentage contribution in the meta-analyses, when taken together, contribute to >50% of the weight of the pooled estimate. Serious risk of bias was considered when sample size of studies that were narratively synthesised was similar to the total sample size of studies contributing to >50% of the weight of the pooled estimate in the meta-analyses. Due to the nature of physical activity interventions, it is not possible to blind participants to group allocation. Therefore, in the case that the blinding of allocation was the only source of bias, the risk of selection bias was deemed to be ‘low’. Performance bias was rated as ‘high’ when <60% of participants performed 100% of prescribed exercise sessions or attended 100% of counselling sessions (defined as low compliance) or when compliance to the intervention was not reported. Attrition bias was rated as ‘high’ risk when >10% of data were missing at the end of the study and intention-to-treat analysis was not used. However, the Guidelines Steering Committee decided that attrition bias would not be downgraded for miscarriage as women would not be able to continue with the exercise programme following a miscarriage.

Inconsistency was considered serious when heterogeneity was high (I2 ≥50%) or when only one study was assessed (I2 unavailable). Indirectness was considered serious when exercise-only interventions and exercise+co-interventions were combined for analysis, or when the effect of exercise+co-intervention on odds of fetal or newborn mortality was assessed. Imprecision was considered serious when the 95% CI crossed the line of no effect and was wide, such that the interpretation of the data would be different if the true effect were at one end of the CI or the other. When only one study was assessed, imprecision was not considered serious because inconsistency was already considered serious for this reason. Finally, in order to assess publication bias, funnel plots were created if at least 10 studies were included in the forest plot (see online Supplementary figures 12 and 13). If there were fewer than 10 studies, publication bias was deemed non-estimable and not rated down. Due to time constraints and feasibility, one reviewer evaluated the quality of the evidence using the GRADE protocol and a second person checked the GRADE tables as a quality control measure.18 Quality of evidence assessment is presented in the online Supplementary tables 1 and 2.

Evidence synthesis: statistical analysis and narrative synthesis

Statistical analyses were conducted using Review Manager V.5.3. (Cochrane Collaboration, Copenhagen, Denmark). Significance was set at p<0.05. For all dichotomous outcomes, ORs were calculated. Inverse-variance weighting was applied to obtain OR using a random-effects model. For RCTs and non-randomised interventions, sensitivity analyses were performed to evaluate whether the effects were different when examining the relationships between exercise-only interventions versus exercise+co-interventions and the outcomes of interest. When possible, the following a priori determined subgroup analyses were conducted for exercise-only interventions and observational studies. These subgroups included: (1) women diagnosed with diabetes (gestational, type 1 or type 2) compared with the general population (which may have included unreported cases of diabetes); (2) women with overweight/obesity (BMI >25.0 kg/m2) prior to pregnancy compared with the general population (women who were of various BMI; mean BMI <25 kg/m2 but possibly with some individuals with BMI >25.0 kg/m2); (3) women >35 years of age compared with women who were <35 years; (4) women who were previously inactive compared with those who were previously active (as defined by individual study authors); (5) gestational age that the intervention was initiated (<12 weeks compared with >12 weeks) and (6) type of exercise intervention. If a study did not provide sufficient detail to allow it to be grouped into the a priori subgroups, then a third group called ‘unspecified’ was created. Tests for subgroup differences were conducted, with statistical significance set at p<0.05. Subgroup differences were interpreted only when statistically significant differences were found. The percent of total variability that was attributable to between-study heterogeneity (ie, not to chance) was expressed using the I2 statistic. In the case of I2 >50%, heterogeneity was explored further with additional subgroup analyses and the overall result was presented using the random-effects model. In studies where there were no observed events in the intervention or control group, data were entered into forest plots, but were considered ‘not estimable’ and excluded from the pooled analysis as per the recommendation in the Cochrane Handbook.19

Dose–response metaregression20–22 as carried out by weighted no-intercept regression of log ORs with a random effects for study, using the metafor23 package in R24 V.3.4.1. Models did not include an intercept term since the log OR is assumed to be zero when the exercise dose is zero. Restricted cubic splines with knots at the 10th, 50th and 90th percentiles of the explanatory variable25 were used to investigate whether there was evidence for a non-linear relationship. Fitting was performed by maximum likelihood, and non-linearity was assessed using a likelihood ratio test. When the model was statistically significant at p<0.05, the minimum exercise dose to obtain a clinically significant benefit as determined by the Guidelines Steering Committee was estimated by the minimum value of the explanatory variable at which the estimated OR was less than 0.75.

For outcomes where a meta-analysis was not possible, results were presented as a narrative synthesis, structured around each outcome. Within each outcome, results were organised by study design. Unless otherwise specified, studies were not included in meta-analyses if data were reported incompletely (SD, SE or number of cases/controls not provided), if data were adjusted for confounding factors or if the study did not include a non-exercising control group. In studies where data were included in the meta-analysis but additional information was available that could not be meta-analysed, the studies were included in both the meta-analysis and narrative synthesis.

Results

Study selection

Although all languages were included in the initial search, for feasibility reasons studies that were published in languages other than English, Spanish or French were excluded. A PRISMA diagram of the search results, including reasons for exclusion, is shown in figure 1. A comprehensive list of excluded studies is presented in the online Supplementary file 1.

Figure 1

Flow diagram of studies selected for the present study. *Six papers included in qualitative synthesis but one was from the same cohort study and counted as one unique study. #Two studies were included in both the qualitative and quantitative synthesis.

Although a staged approach was planned, after considering the evidence from RCTs, the Guidelines Steering Committee decided to include all study designs in the analysis.

Study characteristics

Overall, 46 unique studies (n=266 778 women) from 20 countries and six continents were included. The studies included 30 RCTs, eight non-randomised interventions, four cohort, one cross-sectional and three case–control studies. Among the included exercise interventions, the frequency of exercise ranged from 1 to 7 days per week, the intensity was low to vigorous, the duration of exercise ranged from 10 to 95 min per session and the types of exercise included aerobic exercise, yoga, resistance training and pelvic floor muscle training. Exercise was initiated in the first, second or third trimester (up to 31 weeks gestation). Additional details regarding the studies can be found in the online Supplementary file 1 (see the section Study Characteristics below and online Supplementary table 3).

Quality of evidence

Overall, the quality of evidence ranged from ‘very low’ to ‘moderate’ (see online Supplementary tables 1 and 2). The most common reasons for downgrading the quality of evidence were (1) serious risk of bias and (2) serious imprecision. Common sources of bias included poor or unreported compliance and inappropriate treatment of missing data when attrition rate was high, as well as unknown validity of the physical activity measures. Where possible, publication bias was assessed using funnel plots; in these cases, publication bias did not appear to be an issue.

Synthesis of data

Miscarriage

Overall, there was ‘very low’ quality evidence from 23 RCTs (n=7125 women) regarding the association between prenatal exercise (in all studies exercise was initiated after 8 weeks gestation) and miscarriage. The quality of evidence was downgraded from ‘high’ to ‘very low’ because of serious risk of bias, serious indirectness of the intervention and serious imprecision. Overall, prenatal exercise was not associated with increased odds of miscarriage compared with no exercise (OR 0.88, 95% CI 0.63 to 1.21, I2=0%, see figure 2).26–46

Figure 2

Effects of prenatal exercise compared with control on odds of miscarriage (randomised controlled trials). Sensitivity analyses were conducted with studies including exercise-only interventions and those including exercise+co-interventions. Analyses conducted with a random-effects model. Active: women who were previously active arm of the intervention; exercise: exercise arm of the intervention; diet+exercise: diet+exercise arm of the intervention. Note: studies with zero events in both arms are included in the forest plot but for such studies the OR is not estimable and hence cannot be included in the pooled analysis. M-H, Mantel-Haenszel method.

Sensitivity analysis

The pooled estimate for the exercise-only interventions was not significantly different than the pooled estimate for the exercise+co-intervention subgroups (p=0.32). There was ‘low’ (downgraded due to serious risk of bias and serious imprecision) and ‘very low’ (downgraded due to serious risk of bias, serious indirectness and imprecision) quality evidence showing no association between exercise-only interventions (10 RCTs, n=2248 women)40–46 and exercise+co-interventions (14 RCTs, n=4877 women),26–39 respectively, and odds of miscarriage compared with no exercise (figure 2).

Subgroup analysis

The tests for subgroup differences performed for exercise-only interventions were not statistically significant (see online Supplementary figures 1–4).

Meta-regressions

Meta-regression analyses did not identify a dose–response relationship between frequency, intensity, duration or volume of exercise and risk of miscarriage (online Supplementary figures 16–19). Linear models are presented unless the fit of the spline was significantly better (p<0.05).

Other study designs

The findings from non-randomised interventions and cohort studies were consistent with findings from RCTs, as follows.

There was ‘very low’ quality evidence (downgraded due to serious risk of bias, serious indirectness and serious imprecision) from three non-randomised exercise interventions (exercise only and exercise+co intervention) showing no increased odds of miscarriage with exercise during pregnancy compared with no exercise (n=487; OR 0.50, 95% CI 0.23 to 1.09, I2=0%; online Supplementary figure 5).47–49 Sensitivity analyses did not differentiate the effect of exercise-only intervention from that of exercise+co-intervention on the odds of miscarriage (online Supplementary figure 5).

Similarly, there was ‘very low’ quality evidence (downgraded due to serious risk of bias and serious imprecision) from three cohort studies (n=93 536; pooled estimate based on two studies (n=865))50 showing no increased odds of miscarriage with exercise during pregnancy (online Supplementary figure 6). One study that interviewed women prior to and after suffering a miscarriage regarding prenatal exercise habits (n=92 671 women) could not be included in the meta-analysis. This study demonstrated a progressive increase in the odds of miscarriage as volume of exercise increased.51 Indeed, the HR for risk of miscarriage in women exercising more than 7 hours per week before 18 weeks gestation was 3.7 (95% CI 2.9 to 4.7).51 However, approximately two-thirds of the cohort were interviewed following a miscarriage; the association was not significant in secondary analyses including only women who were interviewed about exercise habits prior to suffering a miscarriage.52

Finally, there was ‘very low’ quality evidence from two case–control studies that could not be included in the meta-analysis. In contrast to findings from RCTs, non-randomised intervention and cohort studies, the case–control study by Zhang showed a protective dose–response effect of exercise on the odds of miscarriage.53 While exercise performed once per week did not affect the odds of miscarriage compared with no exercise (OR 0.83, 95% CI 0.50 to 1.4; adjusted for maternal age, ventilation, folic acid supplements and supervisor support), a protective effect of exercise was found by increasing the frequency to 2–3 times per week compared with no exercise (OR 0.43, 95% CI 0.21 to 0.88) with further protective effect with exercise >3 times per week (OR 0.27 95% CI 0.11 to 0.68).53 Similar to the other study designs, the study by Maconochie (2007) found no association between strenuous exercise (once/week; 2–3/week; >4/week) compared with rare or no exercise and odds of miscarriage (cases, n=603; controls, n=6116; OR 1.00–1.31, 95% CI 0.75 to 2.00, adjusted for year of conception, maternal age, previous miscarriage and previous live birth).54

Perinatal mortality

Overall, there was ‘very low’ quality evidence from 13 RCTs (n=6837 women)28 29 31 39 44 45 55–61 indicating no association between prenatal exercise and odds of perinatal mortality (OR 0.86, 95% CI 0.49 to 1.52, I2=0%, see figure 3). The quality of evidence was downgraded from ‘high’ to ‘very low’ because of serious risk of bias, serious indirectness of the interventions and serious imprecision.

Figure 3

Effects of prenatal exercise compared with control on odds of perinatal mortality (randomised controlled trials). Sensitivity analyses were conducted with studies including exercise-only interventions and those including exercise+co-interventions. Analyses conducted with a random-effects model. Exercise: exercise arm of the intervention;diet+exercise: diet+exercise arm of the intervention. Note: studies with zero events in both arms are included in the forest plot but for such studies the OR is not estimable and hence cannot be included in the pooled analysis. M-H, Mantel-Haenszel method.

Sensitivity analysis

The pooled estimate for the exercise-only interventions was not significantly different than the pooled estimate for the exercise+co-interventions (p=0.85). There was ‘low’ (downgraded due to serious risk of bias and serious imprecision) and ‘very low’ (downgraded due to serious risk of bias, serious indirectness and serious imprecision) quality evidence showing no association between exercise-only interventions (six RCTs, n=1651 women)29 44 45 58–60 or exercise+co-interventions (eight RCTs, n=5186 women),28 29 31 39 55–57 61 respectively, and odds of perinatal mortality compared with no exercise (figure 3).

Subgroup analysis

The tests for subgroup differences performed for exercise-only interventions were not statistically significant (see online Supplementary figures 7–10).

Other study designs

Findings from five non-randomised exercise interventions (n=660; online Supplementary figure 11),62–66 two of three cohort studies (n=60 322 women, online Supplementary figure 12)50 67 and one cross-sectional study (n=2557 women, online Supplementary figure 13)68 were in agreement with the findings from RCTs.

In contrast, there was ‘very low’ quality evidence (downgraded due to serious risk of bias and serious imprecision) from one case–control study (620 cases and 1240 controls) that could not be included in the meta-analysis reporting a protective effect of exercising 30 min ≥2 times/week on the odds of perinatal mortality compared with not exercising 30 min ≥2 times/week (OR 0.72, 95% CI 0.51 to 0.88; adjusted for history of miscarriage, previous induced abortion, frequency of night shift, frequent staying up late, regular physical exercise, smoking and alcohol consumption).69

Discussion

In this study of the relationship between exercise and fetal or newborn mortality, we found ‘very low’ quality evidence from RCTs (miscarriage: 23 trials, n=7125 women; perinatal mortality: 13 trials, n=6837 women) that there were no increased odds of miscarriage, stillbirth or newborn death in women who were active during pregnancy. This finding was supported by ‘very low’ quality evidence from non-randomised interventions, cohort, cross-sectional and case–control studies. Findings from the meta-regressions identified no associations between exercise volume, intensity or frequency and the odds of fetal or newborn mortality (see online Supplementary figures 13–16).

Several studies have indicated that fear of miscarriage was a barrier to exercise in early pregnancy resulting in a reduction or cessation of physical activity.11 70 71 However, the actual impact of early prenatal exercise on early pregnancy loss is not well understood. A recent systematic review of five observational studies suggested there was insufficient evidence to determine a relationship between early pregnancy exercise and miscarriage.72 The present review provides ‘very low’ quality evidence that prenatal exercise does not increase the odds of miscarriage; however, the majority of studies were initiated in the late first/early second trimester (>8 weeks gestation) when the risk of miscarriage is already declining. More studies initiated in the preconception period and continuing during pregnancy are required.

The impact of prenatal exercise on perinatal mortality (20 weeks gestation until 28 days of life) has not been extensively studied. Known risk factors for stillbirth include smoking, obesity, pre-eclampsia and intrauterine infection.3 However, nearly two-thirds of all stillbirths occur in apparently healthy pregnancies and are largely unexplained.6 Decreased venous return and depressed blood flow to the uterus, such as that experienced during supine sleep, have been postulated to increase the risk of stillbirth.73 74 As exercise has also been suggested to result in reduced uterine blood flow, a comprehensive examination of exercise and stillbirth is warranted.74 Previous meta-analyses have included perinatal mortality as a secondary outcome and identified only one study (n=220275) or two studies (n=6176) for inclusion; however, to our knowledge, a comprehensive systematic review examining the impact of prenatal exercise on odds of perinatal mortality was not previously conducted. The present systematic review and meta-analysis provides ‘very low’ quality evidence that prenatal exercise does not increase the odds of perinatal mortality. In addition, one case–control study reported a protective effect of prenatal exercise against stillbirth. It is important to note that women with obesity are at elevated risk for miscarriage and stillbirth.52 We attempted to examine the impact of prenatal exercise in women with obesity on risk of miscarriage and stillbirth; however, only a small proportion of women with obesity were included in these analyses (overweight or obesity, n=141/2248 for miscarriage, n=287/1651 for stillbirth). Additional studies are required to investigate the influence of prenatal exercise in women with obesity on miscarriage and perinatal mortality.

Strengths of our study include the use of rigorous methodological standards (GRADE) to guide the process, our inclusion of a variety of study designs as well as our considering the grey literature and articles in English, French and Spanish. Where it was not possible to include reported results in a meta-analysis, results were narratively synthesised.

We acknowledge that the quality of evidence from RCTs was rated down because of serious risk of bias (low compliance to the intervention) and serious imprecision (95% CI crossed the line of no effect and was wide). In future studies, strategies to improve intervention adherence and compliance should be used, and monitoring and reporting of compliance needs to be improved. More exercise-only interventions are needed to adequately power subgroup analyses. Moreover, the majority of included studies examined the impact of moderate intensity exercise with a maximum duration of 60 min per session. This prevented the identification of whether an exercise threshold (frequency, intensity, type or duration) for increased/decreased odds of perinatal mortality exists. A final limitation is that it was not possible to examine the effect of different types of exercise in this review because all studies included aerobic exercise.

Conclusion

‘Very low’ to ‘moderate’ quality evidence showed no increased odds of miscarriage or perinatal mortality with prenatal exercise. High-quality RCTs initiated prior to 8 weeks gestation are needed to further clarify the relationship between prenatal exercise and risk of miscarriage.

What is already known

  • Approximately, 25% of women will experience at least one miscarriage in their lifetime.

  • In 2015, the incidence of stillbirth was approximately 3.4 per 1000 total births in developed countries and 18.4 per 1000 total births in undeveloped countries.

What are the new findings

  • Data from randomised controlled trials showed no increased odds of miscarriage or of perinatal mortality in pregnant women who exercised compared with those who did not exercise.

  • Similarly, observational studies found no association of miscarriage or perinatal mortality with maternal exercise.

  • There was no association between volume, intensity or frequency of exercise and fetal mortality.

Acknowledgments

We would like to thank Bailey Shandro (UAlberta) and Anne Courbalay (UQTR). The authors wish to acknowledge Mary Duggan from the Canadian Society for Exercise Physiology which is one of the primary knowledge users.

References

Footnotes

  • Contributors MHD, MFM, KBA, GAD and S-MR contributed to the conception of the study. MHD, MFM, VJP, AJG, CEG, NB, LGS, KBA, GAD, RB and S-MR contributed to the design of the study and development of the search strategy. LGS conducted the systematic search. AK, RS, VLM, LR, FS, MJ, TN and A-AM completed the acquisition of data. MHD, NB and MN performed the data analysis. MHD and AK were the principal writers of the manuscript. All authors assisted with the interpretation; contributed to the drafting and revision of the final article and approved the final submitted version of the manuscript.

  • Funding This project was funded by a Canadian Institute of Health Research Knowledge Synthesis Grant. MHD is funded by an Advancing Women’s Heart Health Initiative New Investigator Award supported by Health Canada and the Heart and Stroke Foundation of Canada. RS is funded by a Canadian Institutes for Health Research Doctoral Research Award. AAM is funded by a Fonds de Recherche en Santé du Québec Doctoral Research Award.

  • Competing interests None declared.

  • Patient consent Not required.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.