Effect of antenatal peer support on breastfeeding initiation: a systematic review

Literature search

The following bibliographic databases and resources were searched: British Nursing Index (1981–2008), the Cumulative Index to Nursing and Allied Health Literature (CINAHL) (1982–2008), the Cochrane Library, EMBASE (1980–2008), MEDLINE (1950–2008) and Current Controlled Trials. Reference lists of retrieved articles were manually searched. An updated search was carried out in January 2009 in MEDLINE (1950–2009) (Appendix 1, available at www.cmaj.ca/cgi/content/full/cmaj.091729/DC1).

Inclusion criteria

Citations and papers were selected using an inclusion–exclusion form (Appendix 2, available at www.cmaj.ca/cgi/content/full/cmaj.091729/DC1). We included studies in which the participants were pregnant women, peer-support intervention was provided in the antenatal period irrespective of whether it was also provided in the immediate postnatal period, any comparator was used, breastfeeding initiation was reported, and the study design was either an RCT, quasi-randomized or cohort study with concurrent control. Peer support was defined as support offered by women who had themselves breastfed, who were usually from the same socioeconomic background and locality as the women they were supporting and who had received appropriate training. Peer supporters could be either voluntary or in receipt of basic remuneration or money for expenses. 10 Universal peer support was described as that offered to all women, and targeted peer support as that offered only to women who were considering breastfeeding. For the purposes of this review, breastfeeding initiation was defined as any attempt to breastfeed, even if only once. Nonrandomized studies were included to explore the full spectrum of evidence, given that many studies on this subject are nonrandomized.

No language restrictions were applied. Potentially relevant citations were identified through a comprehensive electronic search. Inclusion and exclusion criteria were applied to all citations, and hard copies of potentially relevant papers were obtained and assessed for relevance (L.I., K.J.). When there was uncertainty, it was resolved by consulting coauthors (K.S.K. and C.M.).

Data extraction and assessment of study quality

Data were extracted on participants, intervention, type of peer support (universal or targeted), outcome, study type, methods, results and quality (L.I. and K.J.). A tool 12^,13 was adapted to assess the risk of bias in both experimental and observational studies (Appendix 3, available at www.cmaj.ca/cgi/content/full/cmaj.091729/DC1). The tool classified study quality into high, medium or low for selection, performance, measurement and attrition bias based on descriptions of the study design, execution and analysis. Studies rated as having the same level of quality in two or more of the four categories were considered to be of that quality overall.

Data synthesis

The data were tabulated and studies were categorized according to whether they had involved universal or targeted peer support. Risk ratios (RRs) with 95% confidence intervals (CIs) were used where available or were calculated from the other measures of effect reported. When pooling studies, it is important to choose a summary statistic that is likely to be constant across settings. Although trials typically report the proportions initiating breastfeeding, it is more likely that the relative risk of not initiating breastfeeding would be constant across settings where initiation rates vary (i.e., that an effective intervention would cause a greater number of women to breastfeed in a setting where rates are low than where few women do not breastfeed). Therefore, for the purposes of meta-analysis, we have pooled the RR of failure to initiate breastfeeding and have redefined it as the absolute number of additional women initiating breastfeeding.

Meta-analysis was only considered for studies without a high risk of bias. We derived the RRs of not initiating breast-feeding and 95% CIs. We explored heterogeneity among the included studies both qualitatively, by comparing their characteristics, and quantitatively, using the χ² test of heterogeneity and the I² statistic. Where appropriate, we combined results from included studies for each outcome to give an overall estimate of the treatment effect. For cluster trials, we computed the design effect from data presented in the reports (intraclass correlation coefficients and cluster-adjusted estimates) and adapted the standard errors of the relative risk to make appropriate allowance for clustering. 14 Where intraclass correlation coefficients were not reported, we computed a design effect using mean intraclass correlation coefficients from the trials in which they were available.

Identification of the literature

In the primary search, 371 citations were identified, of which 348 were excluded because of irrelevance or duplication. Of 23 studies assessed in full, 12 were excluded, leaving 11 for review (Figure 1). Of these studies, seven 15^–21 involved an intervention of universal peer support and four 22^–25 involved targeted peer support (Tables 1 and 2). All control groups received routine maternity care as practised in their respective regions.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1: Identification of relevant literature on antenatal peer support to improve rates of breastfeeding initiation.

Quality of the studies

The quality of the 11 included studies varied. Six 16^,18^,19^,22^–24 of the seven RCTs were classified as high-quality overall, with three involving targeted and three involving universal peer support. The remaining RCT 15^,17 was of medium to low quality, and involved universal peer support. The four remaining studies were observational and classified as being of medium to low quality, with one involving targeted and three involving universal peer support. 17^,20^,21^,25

Study settings

Six of the 11 studies were undertaken in the United States, 15^,20^,23^–25 one in Mexico 18 and four in the United Kingdom (two in Scotland 17^,19 and two in England 16^,24). The populations in all studies predominantly comprised low-income women.

Intensity of the interventions

All studies included both antenatal and postnatal peer support in their interventions. Differences in the intensity of the interventions are shown in Table 3, which lists the number and venue of intended antenatal support contacts and compliance in terms of actual contacts and population coverage. The two US-based RCTs of targeted peer support 22^,23 also included daily in-hospital peer support starting within 24 hours after birth, which may have had an additional effect on initiation rates.

Universal peer support

Nonrandomized controlled studies

In a quasi-randomized study examining universal peer support in Scotland, McInnes and colleagues 17 observed that women in the intervention group were more likely to initiate breastfeeding than those in the control group (23% v. 20%) and that this finding was significant only when adjusted for confounding variables that included socio-economic deprivation. Two nonrandomized studies, both US-based, 20^,21 showed a difference in breastfeeding initiation rates with universal peer support. Schafer and colleagues 20 reported that 82% of women initiated breastfeeding in the intervention group versus 31% in the control group (RR 0.26, 95% CI 0.16–0.44). Shaw and colleagues 21 reported that women in the intervention group were significantly more likely to initiate breast-feeding (53%) than those in the control group (33%) (RR 0.71, 95% CI 0.58–0.78).

The results of all of the included studies of universal peer support are shown in Figure 2.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2: Forest plot of the effect of antenatal peer support on noninitiation of breastfeeding. Values less than 1.0 favour intervention. CI = confidence interval, RR = risk ratio.

Targeted peer support

Explanation and comparison with other studies

Our findings may have been influenced by the intensity of the interventions, underlying breastfeeding rates and breastfeeding information provided to control groups. The interventions were generally similar, although in one trial, 17 community breastfeeding awareness events also took place, and one trial used only telephone peer support. 25

All study designs included additional postnatal peer support for breastfeeding women, so it was not possible to determine the sole effect of antenatal peer support on the duration or exclusivity of breastfeeding. The designs of two RCTs included daily peer support while in hospital after birth, 22^,23 which may have contributed to the significant results reported by both. It is therefore difficult to identify accurately whether antenatal peer support alone can improve initiation or whether a combination of antenatal and very early postnatal support is required.

The studies in the meta-analysis of targeted peer support had significant heterogeneity. This heterogeneity may have resulted from provision in the two US-based trials of in-hospital postnatal peer support, which may have affected initiation rates. 22^,23 The study by Graffy and colleagues 24 was located in the United Kingdom, where baseline rates of breastfeeding initiation are not high. At recruitment, however, all but 2% of women in the intervention group and 4% of those in the control group had already decided to breastfeed; therefore, “considering breast-feeding” may have been interpreted differently in this study. The primary outcome, which was any breastfeeding at six weeks, showed no difference between trial groups. Only the trial by Chapman and colleagues 23 stated breastfeeding initiation as a primary outcome; thus, the focus of the interventions in the other two trials 22^,24 may have been more on exclusivity of breast-feeding.

Unlike the RCTs, all of the nonrandomized trials reported a significant effect. Except for that of the trial by Caulfield and colleagues, 15 the interventions used in the nonrandomized studies do not appear to have been more intensive than those in the RCTs. The much lower underlying rate of breastfeeding initiation among participants in the non-RCTs may explain this finding, although this explanation was not true for the trial by Kistin and colleagues. 25 The effect is probably better explained by higher levels of selection and attrition biases.

Limitations

This review has some limitations. The quality of the included studies varied. All but one of the RCTs were rated as high-quality overall, but the nonrandomized studies were of lower quality. This difference was taken into account by inclusion of only the high-quality trials in the meta-analyses. Data for the outcome measure of breastfeeding initiation were collected using different methods. The majority of the studies relied on self-reporting by participants (either by completion of a questionnaire 17^,19^–21^,24 or by interview 15^,18^,22^,23^,25). Only one study used hospital data to report rates of initiation. 16 The largest trial was in the United Kingdom 16 and reported data on 2398 women. By comparison, the other RCTs reported data for between 127 18 and 644 24 women. The nonrandomized studies included between 207 20 and 926 17 women, and most were subject to high rates of loss to follow-up.

The intensity of the interventions varied, both in planned and actual contacts between peer supporters and participants, varying from one (in most) to three. 22 Coverage ranged from 53% 23 to 100%, 20 but this was not always reported. 15^,18^,19^,25 There was insufficient information given to determine whether a relation existed between intensity of antenatal peer support and initiation of breastfeeding. Routine advice about breastfeeding provided antenatally to controls was generally poorly described, but it appeared to have been mainly clinic-based and written information.

Three RCTs, although not designating breastfeeding initiation as a primary or secondary outcome, 18^,19^,22^,24 reported this measure. Exclusion of these results would not have changed the interpretation of findings.

In two RCTs, 18^,24 control rates of breastfeeding initiation were very high, allowing little scope for improvement from intervention. One of the two trials, which evaluated universal peer support, took place in Mexico, 18 where almost all women initiate breastfeeding (hence the higher baseline rate). There, continuation and exclusivity are problems, and these out-comes did improve with intervention. The other trial, which evaluated targeted peer support, was in the United Kingdom, 24 where the rate of breastfeeding initiation is under 70%. In the trial population, however, this rate was over 95%; thus, only women who had already nearly decided to breastfeed were recruited. This study’s primary outcome of breastfeeding at six weeks, however, showed no effect.

Conclusion

Universal antenatal peer support appears to be ineffective at increasing rates of breastfeeding initiation when provided in the form of one or two instances of contact between peer supporter and pregnant woman, with strong evidence of this conclusion from the United Kingdom. 16^,19 There may be a significant increase in rates of initiation of breastfeeding when antenatal peer support is targeted at women who are considering breastfeeding, but evidence of this effect was found only among low-income Hispanic women in the United States. Owing to differences in community rates of breastfeeding and levels of breastfeeding support included in routine care in the included studies, the findings of this review may have limited generalizability. When peer support is introduced as an intervention to improve rates of breastfeeding initiation, there should be concurrent, high-quality evaluation to determine its effectiveness. Future research might focus on more intensive interventions and on the combination of antenatal and immediate postpartum breastfeeding support.