Lifestyle interventions for overweight and obese pregnant women to improve pregnancy outcome: systematic review and meta-analysis

The eligible studies included RCTs and non-RCTs that evaluated antenatal dietary and lifestyle interventions in obese and overweight pregnant women whose outcome measures included quantitative maternal and fetal health outcomes. Systematic reviews and trials of women with existing gestational diabetes, or trials of pre-conception or postpartum interventions, were not included. Inclusion of trials was not restricted by language, publication date or country. Systematic reviews and observational studies were excluded.

Literature searches were performed using five mainstream electronic databases (Cochrane Library, MEDLINE, EMBASE, CINAHL, Maternity and Infant care), and eight other databases (PsyclINFO via OVID SP, PyscLNFO via OVID SP, Science Citation Index via Web of Science, Social Science Citation Index via Web of Science, Global Health, Popline, Medcarib, Nutrition database).

The following MeSH terms, words and combinations of words, were used in constructing the systematic search: overweight OR obesity; pregnancy OR pregnancy complications OR pregnancy outcome OR prenatal care, prenatal, antenatal, intervention, randomised controlled trial, life style, "early intervention (education)", health education, education, patient education handout, patient education, exercise, exercise therapy, health promotion, diet, carbohydrate-restricted, diet, fat-restricted, diet, reducing, diet therapy, weight loss. Full details of the search strategy are shown in Table 1. The searches were unlimited by time up to January 2012 and limited to human studies and clinical trials. The systematic search was undertaken in the mainstream databases and targeted searches were conducted in the other databases.

Electronic literature searches, study selection, methodology, appropriateness for inclusion and quality appraisal were performed independently and in duplicate by two authors (E-ON and RV). Disagreements between reviewers were resolved by consensus. Included studies were divided into two groups (RCTs and non-RCTs) and separately meta-analysed.

Two independent reviewers extracted the data. As a first step, each paper was screened using the title and the abstract. In the next round, studies were assessed for methodological quality and appropriateness for inclusion by two reviewers working independently from the full text of the manuscript. This was done without consideration of the results.

For each included trial, data was extracted on maternal gestational weight gain; gestational diabetes; Caesarean section; large for gestational age baby (> 4 kg); and birth weight. The included studies have been summarised in Tables 2 and 3.

The quality of studies was assessed based on how the studies had minimised bias and error in their methods. We categorised the studies according to criteria based on PRISMA guidelines [49] and the Cochrane Library [50]. For example, high quality trials reported study aims; control comparison similar to the intervention group; relevant population demographics pre- and post-intervention; and data on each outcome. These study characteristics are tabulated in Tables 4 and 5. A final assessment categorised the studies as high, medium or low quality.

This is shown in Figure 1 and in a tabulated format contained within Table 2 and 3.

The main measure of effect of the meta-analysis was the odds ratio or standardised mean difference. The data syntheses were conducted according to the Cochrane methodology [50]. First, we used statistical meta-analysis techniques to assess the efficacy of the interventions of controlled trials. Chi-square statistics tests were used to test for heterogeneity (Q statistics) between controlled trials. When there was no significant heterogeneity, we combined effect sizes in a fixed effect statistical meta-analysis using Review Manager (RevMan; Version 5.0, Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2008). The meta-analyses were performed by calculating the odds ratios (for proportion data) or standardised mean differences (for scale data) using a fixed effects model. Quantitative analysis was performed on an intention-to-treat basis focused on data derived from the period of follow-up. There was heterogeneity between studies because of the smaller sample size of some of the studies (poor quality), variation of the study population and the intensity and duration of the interventional strategies being evaluated. A random effects model was used to adjust for heterogeneity.

The review process is outlined in Figure 1 and the selected papers summarised in Tables 2 and 3.

Fifteen trials met the inclusion criteria: 13 RCTs [26‐34] and six non-RCTs [35, 36, 38, 39, 51, 52]. All 19 trials were performed in developed countries: five in the USA, three in Canada, three in Australia, two in Finland and one in Denmark, Netherlands, Sweden, Spain, Brazil and Belgium (Tables 2 and 3). Five RCTs were judged to be of medium quality [27, 29, 34]. The rest were deemed low quality (Tables 4 and 5).

The pooled RCTs included a total of 1,228 participants and the pooled non-RCTs included 1,534 participants. Participants were predominantly white except in the studies by Asbee et al. [27], Gray-Donald et al. [38] and Hui et al. [33]. In the Asbee et al. study, the majority were described as being of Hispanic ethnicity [27].

For all included RCTs, the control group received no intervention or standard care. In the non-RCTs, most used non-parallel controls [35, 38, 39, 51] or controls from another centre [36]. The outcomes investigated in the trials were gestational weight gain, gestational diabetes, Caesarean section delivery, large for gestational age baby and birth weight.

Of the 19 controlled trials, 16 measured gestational weight gain (10 randomised, 6 non-randomised); 8 recorded gestational diabetes (6 randomised, 2 non-randomised); 10 recorded Caesarean delivery (6 randomised, 4 non-randomised); 10 measured large for gestational age (6 randomised, 4 non-randomised); and 7 measured birth weight (7 randomised). Meta-analyses for the different outcomes are shown in Tables 6 and 7, and Figures 2, 3, 4, 5, 6, 7, 8, 9 and 10.

Meta-analysis of RCTs showed that combined antenatal lifestyle, dietary and activity intervention restricts gestational weight gain (Table 6 and Figure 2) and there was a trend towards reduction in the prevalence of gestational diabetes in overweight and obese women (Table 6 and Figure 3). However, meta-analysis of non-RCTs only showed weak evidence that lifestyle intervention reduces gestational weight gain (Table 7 and Figure 7) and there was no evidence for a reduction in prevalence of gestational diabetes (Table 7 and Figure 8). There was no robust evidence that lifestyle intervention is associated with a lower prevalence of Caesarean delivery or macrosomia or any alteration in birth weight (Tables 6 and 7, Figures 4,5, 6, 8, 9 and 10).

The nature of the interventions varied widely between studies and some of the key features of the interventions are outlined in Tables 2 and 3. In summary, for the six non-RCTs, three of the interventions comprised individual and group or seminar components [36, 38, 39, 53], two were individual [35, 52] and one was unclear [51]. Of the 13 RCTs, one comprised individual and group components [33], eight were individual [27‐30, 32, 34] and three were group-based [26]. Where there were individual and group components, the latter were usually physical activity sessions. All of the non-RCTs included dietary and physical activity guidance, as did the majority of the randomised studies. Exceptions were two studies which included only nutritional guidance [29, 30] and one which included guidelines about weight gain and weight monitoring only [28]. The majority of studies included dietary or physical activity guidance, with one of the non-RCTs [35] and three of the RCTs [29, 32‐34] specifying that guidance was personalised.

Antenatal lifestyle, dietary and activity advice for overweight and obese pregnant women restricts maternal weight gain during pregnancy and lowers the prevalence of gestational diabetes in women who are overweight or obese. However, the quality of the study designs was generally poor. The reduction in gestational weight gain was observed to be statistically significant in the meta-analysis of randomised trials (10 RCTs; n = 1,228; -2.21 kg (95% CI, -2.86 to -1.57 kg)) but non-significant in the meta-analysis of non-randomised trials (six non-RCTs; n = 1,534). No effects of antenatal lifestyle interventions were identified in obese and overweight pregnant women in relation to Caesarean delivery, large for gestational age, birth weight and macrosomia (> 4 kg).

There is evidence to suggest antenatal lifestyle interventions may restrict gestational weight gain and a trend towards a reduced prevalence of gestational diabetes, but there was no statistical effect on other important clinical outcomes, possibly due to inadequate power of the combined sample size. The effect on restricted weight gain and gestational diabetes was not consistent across all the trial populations and therefore cannot be generalised. There was also wide variation in the types of interventions evaluated in the studies. The majority were individual-based and most provided generic guidance comprising mainly dietary and physical activity information, with few tailoring guidelines. There was considerable heterogeneity in intervention design and no obvious patterns between intervention type and study outcomes. For the gestational weight gain and gestational diabetes outcomes, both the successful and non-successful studies included those which were personalised, combined physical activity and dietary guidance and were individual-based. Moreover, degrees of weight gain restriction achieved were modest overall. It is even harder to make conclusions regarding the specific behaviour change strategies included (for example, monitoring and goal setting) or theoretical basis of interventions since these were typically poorly reported.

Identifying specific components of successful interventions aids understanding of how interventions are having an effect and clear reporting of intervention design allows for easier replication [54]. Previous reviews have attempted to make conclusions regarding specific effective components of interventions. Suggestions that weight monitoring and setting weight goals could be useful [46] and also monitoring along with education counselling and physical activity sessions [51, 55] have been made. Another review suggested that interventions should be based on the Theory of Planned Behaviour, but the rationale for using this model over others in this population was unclear [56]. None of these reviews examined intervention components systematically. A more recent review by Gardner et al. assessed interventions targeting gestational weight gain from a psychological perspective and specifically examined intervention content and delivery methods [57]. This review comprised 10 controlled trials, all included in the current review; only two of the studies reported basing interventions on theory and the studies used, on average, five behaviour change strategies (self-monitoring, feedback provision and setting behavioural goals were the most common), but no conclusions could be drawn as to their contribution to study outcomes. Broadly consistent with this were the four studies in the current review which were not included in the review by Gardner et al. [57]. Their review questioned the evidence supporting the benefits of weight monitoring, but tentatively suggested that information provision had been underused and that it might be of benefit to have a narrower focus of intervention targets [57].

Our study adds to a growing body of evidence that aims to evaluate lifestyle intervention as a means to minimise the adverse outcome associated with obesity in pregnancy. In comparison to other published reviews [45, 46, 56], we have adopted an original approach by broadening the literature source (multiple data sources, no language restriction), focusing on relevant clinical outcomes (such as Caesarean section, gestational diabetes, macrosomia), and improving our sensitivity by meta-analysing both RCTs and non-RCTs. Furthermore, to minimise bias, the review methodology was registered a priori (Prospero number CRD420111122 http://​www.​crd.​york.​ac.​uk/​PROSPERO). We therefore believe our review provides a comprehensive and reliable analysis of the current evidence and for the first time highlights that lifestyle intervention in pregnancy may reduce the prevalence of gestational diabetes.

The evidence summarised in this work comes from available studies of which most are of low quality, with only four studies fulfilling a medium quality score. Hence, the evidence base is weak and calls for more robust studies. Our trial population is relatively small, the intensity and duration of the interventions of trials varied and trials were predominantly USA in origin; a phenomenon common to many public health reviews, especially on obesity. Although our focus was on antenatal lifestyle intervention for obese and overweight pregnant women, our search yielded some studies that contained a mixed group of obese and normal weight women and we excluded all the non-obese participants from our analysis. Still, this may lead to inconsistencies in measuring the effect of the intervention as well as under- or overestimating the treatment effect. Furthermore, even though our search was systematic and rigorous, we could have missed eligible studies inadvertently.