Interventions for promoting the initiation of breastfeeding

BACKGROUND: Despite the widely documented health advantages of breastfeeding over formula feeding, initiation rates remain relatively low in many high-income countries, particularly among women in lower income groups. OBJECTIVE : To evaluate the effectiveness of interventions which aim to encourage women to breastfeed in terms of changes in the number of women who start to breastfeed. METHODS : Search methods: We searched the Cochrane Pregnancy and Childbirth Group's Trials Register (July 2007), handsearched the Journal of Human Lactation, Health Promotion International and Health Education Quarterly from inception to 15 August 2007, and scanned reference lists of all articles obtained. Selection criteria: Randomized controlled trials, with or without blinding, of any breastfeeding promotion intervention in any population group except women and infants with a specific health problem. Data collection and analysis: One review author independently extracted data and assessed trial quality, checked by a second author. We contacted investigators to obtain missing information. MAIN RESULTS: Main results: Eleven trials were included. Statistical analyses were conducted on data from eight trials (1553 women). Five studies (582 women) on low incomes in the USA with typically low breastfeeding rates showed breastfeeding education had a significant effect on increasing initiation rates compared to standard care (risk ratio (RR) 1.57, 95% confidence interval (CI) 1.15 to 2.15, P = 0.005). Subgroup analyses showed that one-to-one, needs-based, informal repeat education sessions and generic, formal antenatal education sessions are effective in terms of an increase in breastfeeding rates among women on low incomes regardless of ethnicity and feeding intention. Needs-based, informal peer support in the antenatal and postnatal periods was also shown to be effective in one study conducted among Latina women who were considering breastfeeding in the USA (RR 4.02, 95% CI 2.63 to 6.14, P < 0.00001). AUTHORS' CONCLUSIONS: This review showed that health education and peer support interventions can result in some improvements in the number of women beginning to breastfeed. Findings from these studies suggest that larger increases are likely to result from needs-based, informal repeat education sessions than more generic, formal antenatal sessions. These findings are based only on studies conducted in the USA, among women on low incomes with varied ethnicity and feeding intention, and this raises some questions regarding generalisability to other settings.


B A C K G R O U N D
This review aims to assess ways of promoting breastfeeding. Fewer resources have been spent on this than on the promotion of formula feeding. Women in most countries encounter promotion of formula feeding in various forms, a factor which has been implicated in women choosing to feed their babies on formula (WHO Data Bank 1996).
There is extensive evidence for short-term and long-term health risks of formula feeding and the World Health Organisation recommends that all infants should be fed exclusively on breast milk from birth to six months of age (WHO 2002). Babies who are not breastfed exclusively for the first three to four months are more likely to suffer health problems such as gastroenteritis (Howie 1990;Ip 2007;Kramer 2001;Quigley 2006;Quigley 2007), respiratory infection (Ip 2007;Kramer 2001;Victora 1989;Wright 1989), otitis media (Aniansson 1994;Duncan 1993;Ip 2007), urinary tract infections (Marild 1990;Pisacane 1992), necrotising enterocolitis ( Ip 2007;Lucas 1990a), atopic disease if a family history of atopy is present (Burr 1989;Lucas 1990b;Saarinen 1995) and diabetes mellitus (Karjalainen 1992;Mayer 1988;Virtanen 1991). Research also indicates a positive relationship between having been breastfed and the bone health of the child (Lucas 1990a) and with improved cognitive development (Kramer 2008). In addition, breastfeeding is beneficial to the mother's health. Women who formula feed are significantly more likely to develop epithelial ovarian cancer (Gwinn 1990;Rosenblatt 1993) and are more likely to develop breast cancer (Ip 2007;Layde 1989;Newcomb 1994;UK Study Group 1993) than women who breastfeed.
Other social and practical benefits to the breastfeeding mother include the increased likelihood she will use up the body fat de-posited in pregnancy (Dewey 1993;Kramer 2002), substantive savings on the expenses associated with formula feeding (except in the case of mothers participating in welfare schemes and receiving subsidised formula milk powder), and the avoidance of effort and risks involved in preparing formula feeds (MIDIRS 1997;Bryant 1990).
Attempts have been made to quantify public cost benefits of breastfeeding. Recent National Institute for Health and Clinical Excellence (NICE), guidance on modelling the cost effectiveness of breastfeeding to the UK National Health Service (NHS) estimates that peer support which achieves an estimated 20 percentage point increase in breastfeeding initiation would save the NHS money over the long term (NICE 2007). The model suggests that such a scheme would avert 2.7 per 10,000 cases of pre-menopausal breast cancer in mothers and 285 per 10,000 cases of infections requiring hospitalisation in the first year of life. At the current NICE threshold for cost per quality-adjusted life year (QALY) gained (£20,000-£30,000), it estimates that expenditure on breastfeeding support would be justified in competition with other demands on NHS resources by an increase in initiation rates of about 15 percentage points. The authors note that evidence on other potential health benefits and savings, such as improved health of breastfed babies later in life and reduced spending on infant formula, is too complex to be accounted for by the model. Results will therefore tend to underestimate the cost-effectiveness of peer support interventions. The authors found the quality of evidence concerning the efficacy and cost of public health interventions designed to promote breastfeeding to be poor.
Despite the documented risks of formula feeding, many women choose to formula feed their babies. Given the wide variations in rates between countries and population groups (see below), the reasons for this are likely to be socio-cultural and include personal, social and structural biases against breastfeeding. These may include attitudes of family and close friends, attitudes to breastfeeding in public and employment practices (Renfrew 1998). The availability of subsidised infant formula milk through welfare food programmes, such as the UK based Healthy Start Scheme and the USA based Women, Infant and Children Supplemental Feeding Program, may be an economic factor which contributes unintentionally to women in low-income groups deciding to formula feed. The extent to which individual countries have adopted the World Health Organisation's International Code of Marketing of Breast Milk Substitutes (WHO 1981) may also be a contributing factor to the infant feeding decision, particularly for women in low-and middle-income countries. The WHO Code (WHO 1981) was adopted by the World Health Assembly in 1981 and is currently supported by over 20 countries with draft laws awaiting adoption in others. Increased support to adopt the WHO/UNICEF Code (WHO 1981) into national legislation would prohibit the provision and promotion of breast milk substitutes to pregnant women, new mothers or families.
International rates of initiation of breastfeeding are extremely variable between and within countries (see Appendix 1). Historically, breastfeeding rates have been high in many resource poor-countries despite variation in urban and rural areas. In Scandinavia and Eastern Europe, many countries have a high incidence of women starting to breastfeed. Other individual countries with breastfeeding initiation rates of over 90% include Japan, Switzerland, Luxembourg and Turkey. In central and southern Europe, initiation rates remain relatively high at around 70% to 80%, for example, in Italy, Spain and Greece.
North America and much of Western Europe have experienced lower rates of initiation of breastfeeding since the early 20th century. Significant increases have been experienced in some countries over the last decade. For example, recent UK data show that 78% of mothers initiate breastfeeding in England, 70% in Scotland, 67% in Wales and 63% in Northern Ireland (Bolling 2007) (see Appendix 2); and 85% and 74% of mothers start to breastfeed in Canada (Millar 2005) and the US (CDC 2004) respectively.
In all countries, breastfeeding initiation rates are closely related to social class, income and educational levels. In those high-income countries where breastfeeding rates are typically low, the lowest rates are found among women in low-income groups. In England and Wales, for example, only 65% of women classified as having 'never worked' or 65% of women in 'lower occupations' initiated breastfeeding in 2000 compared to 88% of women classified in 'higher occupations' (Bolling 2007). It is therefore important to examine which interventions might have an impact on rates in these groups.
The purpose of this review is to examine interventions which aim to encourage women to breastfeed, to evaluate their effectiveness in terms of changes in the number of women who initiate breastfeeding and to report any other effects (beneficial or adverse) of such interventions.

O B J E C T I V E S
1. To identify and describe health promotion activity intended to increase the rate of initiation of breastfeeding.
2. To evaluate the effectiveness of different types of health promotion activity, in terms of changing the number of women who initiate breastfeeding.
3. To compare the effectiveness of different types of health promotion interventions as appropriate.
4. To assess the impact of these interventions on secondary outcomes, namely, duration of any or exclusive breastfeeding and any adverse outcomes as a result of the intervention.

Types of studies
Randomised controlled trials, with or without blinding. There was no limitation of study by country of origin or language.

Types of participants
All those exposed to interventions intended to promote breastfeeding. This includes pregnant women, mothers of newborn infants and women who may decide to breastfeed in the future. We also included population subgroups of women, such as women from low-income or ethnic groups. Women and infants with a specific health problem, e.g. mothers with HIV/AIDS or infants with cleft palate, or premature babies, are excluded from this review.

Types of interventions
Any intervention aiming to promote the initiation of breastfeeding, which takes place before the first breastfeed. Evaluations of interventions taking place after the first breastfeed or whose primary purpose is to affect the duration or exclusivity of breastfeeding are excluded from this review.

Types of outcome measures
Initiation and duration of any and exclusive breastfeeding.

Search methods for identification of studies Electronic searches
We searched the Cochrane Pregnancy and Childbirth Group's Trials Register by contacting the Trials Search Co-ordinator (July 2007). The Cochrane Pregnancy and Childbirth Group's Trials Register is maintained by the Trials Search Co-ordinator and contains trials identified from: 1. quarterly searches of the Cochrane Central Register of Controlled Trials (CENTRAL); 2. weekly searches of MEDLINE; 3. handsearches of 30 journals and the proceedings of major conferences; 4. weekly current awareness alerts for a further 44 journals plus monthly BioMed Central email alerts. Details of the search strategies for CENTRAL and MEDLINE, the list of handsearched journals and conference proceedings, and the list of journals reviewed via the current awareness service can be found in the 'Specialized Register' section within the editorial information about the Cochrane Pregnancy and Childbirth Group. Trials identified through the searching activities described above are each assigned to a review topic (or topics). The Trials Search Co-ordinator searches the register for each review using the topic list rather than keywords. Further details of searching carried out for the initial version of this review can be found in Appendix 3.

Searching other resources
We handsearched the Journal of Human Lactation, Health Promotion International and Health Education Quarterly from inception to August 2007. We scanned reference lists of all relevant papers retrieved. We did not apply any language restrictions.

Data collection and analysis
Two authors independently assessed over 1400 titles and abstracts of studies, identified from all sources, for relevance. Where no clear decision could be made on the basis of the title or abstract, we considered the studies relevant. This process identified 83 potentially relevant studies for which we retrieved full reports for more detailed consideration. One author used a prescreen form to systematically assess retrieved papers against the inclusion criteria and to classify included studies by the type of health promotion intervention. This included three papers which required translation into English before prescreening, namely, papers in Portuguese, Russian and Croatian. We contacted authors to clarify or obtain relevant details of individual studies, particularly to request details of their randomisation processes. A second author independently checked fifty papers that were classified as possible exclusions during the prescreening process. We excluded all. Two studies we previously identified as ongoing have been completed but not yet published and are awaiting assessment for this review. One author used data extraction and quality appraisal forms to extract data from the remaining eleven studies. A second author then checked the data. Any disagreements were settled through discussion between authors. This resulted in eleven studies being included in this review, eight of which contributed data to the meta-analyses. Studies were classified by type of intervention: health education of pregnant women, health education of significant others of pregnant women, peer support, health education of pregnant women and peer support, breastfeeding promotion packs, and early mother-infant contact. One author entered data into the Review Manager software (Revman 2008). We assessed the validity of each included study according to the criteria outlined in the Cochrane Handbook (Higgins 2008). We assessed selection bias on the basis of concealment of allocation: adequate; unclear; or inadequate. We rated performance bias, attrition bias and detection bias as: adequate; unclear or partially adequate; or inadequate. For eight studies, we considered meta-analysis appropriate to evaluate the effect of each type of intervention which measured the primary outcome of initiation of breastfeeding. The reasons for not including data from three studies (Caulfield 1998;Forster 2004;Wolfberg 2004) in the meta-analyses are given in Description of studies. Figures show calculated individual and pooled risk ratios with 95% confidence intervals for dichotomous data on initiation of breastfeeding from studies where data allow an estimation. We calculated the individual risk ratios on an intention-to-treat basis where the data from primary studies allow participants to be analysed according to the group to which they were initially allocated, regardless of whether or not they later withdrew or were lost to the study. Key differences in characteristics of participants or methods of implementation of the intervention, or both, are discussed in the interpretation of results (see Results). We undertook subgroup analysis to compare the differential effect of interventions on initiation rates for studies within the education and education or peer support categories. This was possible due to the similarities between subgroups of interventions and low levels of statistical heterogeneity . The nature of health promotion interventions to achieve a positive outcome in terms of an increase in the number of women starting to breastfeed warrants reversal of the traditional Cochrane Database of Systematic Reviews convention whereby a risk ratio of less than one indicates that the intervention is better than the control (Alderson 2004). For the purposes of this review therefore, a risk ratio of more than one indicates that the intervention has a more favourable effect on initiation rates than the control. This is displayed by the dot appearing to the right of the central vertical line that indicates no difference. To summarise results, we began by using a fixed-effect approach. This approach assumes differences among study results are due solely to the play of chance (Higgins 2006). If the forest plot showed substantial statistical heterogeneity (I 2 > 50%) using a fixed-effect approach, we then re-analysed the study results using a random-effects model to estimate the average treatment effect (Higgins 2006). No studies measuring the primary outcome of initiation rates of breastfeeding were excluded from the review and no sorting of studies for ordering of meta-analyses was considered necessary on the basis of methodological quality. Three studies included in the review did not contribute data to the meta-analyses, as detailed below.

Description of studies
See: Characteristics of included studies; Characteristics of excluded studies; Characteristics of studies awaiting classification. See Characteristics of included studies table. This review has a total of eleven included studies and 70 excluded studies. Two studies, completed but not published when this update went to press, are awaiting assessment. Three studies (Caulfield 1998;Forster 2004;Wolfberg 2004) met the inclusion criteria for this review but due to methodological concerns have not contributed data to the analyses. Caulfield 1998 randomised four WIC clinics in one US city. One clinic received no intervention, one received a breastfeeding motivational video, another received a peer counselling intervention, and the fourth clinic received both the video and peer counselling interventions. The clinic populations were 90% to 96% African-American, with breastfeeding rates of 2% to 6% at one month. Outcome data are presented for 242/674 potentially eligible women (36%) for whom complete data were available. Among these 242 women, the overall breastfeeding initiation rate (defined as ever putting the infant to the breast) was 48%. In the control clinic (n = 57) 26% initiated breastfeeding; in the video clinic (n = 64) the rate was 50%; in the peer counsellor clinic (n = 55) it was 62%; and in the clinic that used both the interventions (n = 66) the breastfeeding initiation rate was 52%. We sought statistician's advice about whether we should include this study as a cluster-randomised trial. We were advised against inclusion because with only one clinic in each treatment group, it is not possible to calculate the standard error of difference. The paper reports that although clinic-level data had suggested the four clinics were comparable, characteristics of women who enrolled in the study were noted to vary by clinic. Therefore, although it appears that either intervention was better than none for promoting breastfeeding initiation, it is not possible to estimate their effects with any precision. Forster 2004 reported a randomised controlled trial that took place in Melbourne, Australia, in a hospital that had been Baby-Friendly accredited for four years before the trial began. Women booking to have their first infant at the hospital were eligible if they were public patients, 16 to 24 weeks' pregnant at recruitment, and able to speak and write in English. In addition to Baby-Friendly standard care (n = 327), there were two interventions; practical skills (n = 327) and attitudes (n = 327) (details of the interventions appear in the Characteristics of included studies table). Breastfeeding initiation rates were high in all three groups (97% for the practical skills intervention, 95% for the attitudes intervention, and 96% for Baby-Friendly standard care). The researchers conclude that in settings where breastfeeding initiation is high, neither of the two interventions can be recommended as effective. Results from this large, well-reported trial are not included in the meta-analyses for this review. We considered the control group who received standard care compliant with the Baby-Friendly initiative had in effect received a third intervention that meant we could not compare this control group with the control groups of other studies in the review. Wolfberg 2004 reported a randomised controlled trial with expectant fathers who were partners of women seeking prenatal care at one US hospital. Participants were predominantly black women receiving benefits under the federal Women and Infants Supplementary Feeding Program. The intervention group attended a twohour class on infant care and breastfeeding promotion, and the control group attended a class on infant care only. Both classes were facilitated by the same person, a black father who was knowledgeable and used a non-didactic and informal method of education. Women whose partners attended the intervention class had a 74% breastfeeding initiation rate (20/27) compared with 41% (13/32) whose partners attended the control class. The authors reported significant attrition. They approached 567 pregnant women and reported data from 57 couples (10%). The numbers randomised to each group are not reported therefore results from this trial have not been included in the analyses for this review. Further details of these studies appear in the Characteristics of included studies table.
Among the excluded studies are four studies which were excluded due to lack of information about outcome data or method of randomisation despite efforts to contact the authors (Chapman 1986;Grossman 1988;Mattar 2007;Winterburn 2003), and one study which is ongoing but would not be included on completion due to the aims of the study being outside the focus of this review (Chong 2007). Two studies are awaiting assessment (Hoddinott 2007;Wambach 2006). All eleven included studies evaluated the effect of the intervention in terms of the numbers of women who initiated breastfeeding (Brent 1995;Caulfield 1998;Chapman 2004;Coombs 1998;Forster 2004;Hill 1987;Howard 2000;Lindenberg 1990;Ryser 2004;Serwint 1996;Wolfberg 2004). Nine studies were conducted in the USA (Brent 1995;Caulfield 1998;Chapman 2004;Coombs 1998;Hill 1987;Howard 2000;Ryser 2004;Serwint 1996;Wolfberg 2004), one in Australia (Forster 2004) and one in Nicaragua, Central America (Lindenberg 1990). Nine of the studies have evaluated the impact of the intervention on both initiation and duration rates (Brent 1995;Caulfield 1998;Chapman 2004;Forster 2004;Hill 1987;Howard 2000;Lindenberg 1990;Serwint 1996;Wolfberg 2004). Nine studies have targeted participants on low incomes (Brent 1995;Caulfield 1998;Chapman 2004;Coombs 1998;Hill 1987;Lindenberg 1990;Ryser 2004;Serwint 1996;Wolfberg 2004). With the exception of the study conducted by Lindenberg (Lindenberg 1990) in Nicaragua, all of these interventions were implemented amongst low-income women in the US. Four of the US studies have evaluated the effect of an intervention amongst a low-income group belonging primarily to an ethnic minority group, namely African-American women (Caulfield 1998;Serwint 1996;Wolfberg 2004) and Latina women (Chapman 2004). The following types of breastfeeding promotion intervention were evaluated by the eight trials included in the analyses for this review (see Characteristics of included studies' for details):

Health education of pregnant women -five trials
Brent 1995: white low-income women, unmarried and with an educational level of 12 years or below. Coombs 1998: formal education programme (self-help manual) for low-income pregnant women. Hill 1987: formal health education combined with written literature delivered by health professionals to mostly white, low-income women. Ryser 2004: Best Start health education program (repeated oneto-one, needs-based counselling) delivered to low-income women who intended to formula feed or were undecided. Serwint 1996: single formal health education session delivered in the antenatal period to low-income women who were mostly African-American.
(2) Peer support -one trial Chapman 2004: home and public health facility based peer support during prenatal and perinatal period to predominantly Latina women who were considering breastfeeding.

(3) Breastfeeding promotion packs -one trial
Howard 2000: breastfeeding promotion packs in a sample of white, well-educated, women in middle-or high-income groups.

(4) Early mother-infant contact -one trial
Lindenberg 1990: hospital based early mother-infant contact combined with minimal breastfeeding education intervention delivered to primiparous women living in poor urban areas of Managua with a mean age of 20 years. No trials of population-based programs to promote initiation of breastfeeding were found.

Risk of bias in included studies
Assessment of studies for potential sources of selection, performance, attrition and detection bias and overall risk of bias (as recommended by Higgins 2008) are detailed in Characteristics of included studies. Seven of the eight studies were assessed as having unclear or inadequate allocation concealment (Brent 1995;Coombs 1998;Hill 1987;Howard 2000;Lindenberg 1990;Ryser 2004;Serwint 1996), raising concerns regarding the effect of selection bias on study findings. The remaining study was assessed as having achieved adequate concealment (Chapman 2004). In regard to attrition bias, four of the eight studies included in meta-analysis reported breastfeeding initiation for all participants (Brent 1995;Chapman 2004;Hill 1987;Lindenberg 1990). The remaining four studies had losses to follow up between recruitment and breastfeeding initiation which were inadequately reported of around 8% (Ryser 2004;Serwint 1996), 19% (Howard 2000 and 25% (Coombs 1998). Given that there are genuine pragmatic considerations when delivering and evaluating breastfeeding promotion interventions, the ability to reduce performance bias is limited and this should be recognised as an inherent weakness of this particular type of evidence base rather than of the particular studies included in this review. The only study which was considered to have adequately addressed potential sources of performance bias was the evaluation of a breastfeeding promotion pack compared to a commercial formula pack (Howard 2000), which was able to maintain blinding of both participants and providers through the use of sealed, similarly designed, packs more comparable with the use of a placebo and treatment in a therapeutic trial. In the case of detection bias, the objective nature of the outcome being assessed, namely, whether a woman starts to breastfeed or not at a predefined timepoint, limits the scope for potential influence by the assessor, regardless of their being blind to the participant's group allocation.

Effects of interventions
Statistical analyses for the primary outcome of initiating breastfeeding are reported below for eight trials involving 1553 women (Brent 1995;Chapman 2004;Coombs 1998;Hill 1987;Howard 2000;Lindenberg 1990;Ryser 2004;Serwint 1996). Studies were analysed within four types of intervention: health education, peer support, breastfeeding promotion packs, and early mother-infant contact. Subgroup analyses have been conducted within the health education interventions to test the effect of particular types of education intervention on initiation rates.

(1) Health education interventions (comparison one)
Five studies (Brent 1995;Coombs 1998;Hill 1987;Ryser 2004;Serwint 1996) (including 582 women) evaluated the effect of health education on the initiation of breastfeeding. When all studies were combined for meta-analysis, a statistically significant increase in the number of women starting to breastfeed was demonstrated as a result of the health education interventions (risk ratio (RR) 1.57, 95% confidence interval (CI) 1.15 to 2.15). These interventions were all conducted among women on low incomes in one high-income country setting (USA). Substantial statistical heterogeneity between the six studies included in this analysis is suggested by the heterogeneity score above 50% (I 2 = 53.4%). Subgroup analysis of two studies (Brent 1995; Ryser 2004; including 162 women) evaluating the effect of repeat, informal breastfeeding education which was personalised to each woman's needs demonstrated a statistically significant increase in the number of women starting to breastfeed as a result of the intervention (RR 2.40, 95% CI 1.57 to 3.66). Statistical heterogeneity between these studies was small (I 2 = 7.0%). Both studies were conducted in the USA among women on low incomes from a range of ethnic groups. One intervention comprised repeat prenatal education sessions, daily perinatal input and repeat postnatal support on breastfeeding issues from a lactation counsellor based on the woman's needs and interests (Brent 1995). The other intervention comprised repeat prenatal education sessions from a breastfeeding researcher on breastfeeding issues based on the woman's concerns (Ryser 2004). Professional education was also directed at nursing and medical staff who interacted with the breastfeeding mother-infant pair. The Brent 1995 study included women regardless of feeding intention whereas the Ryser 2004 study included women who intended to formula feed or had not yet decided. Both education interventions were delivered by a breastfeeding professional or 'expert', namely, a lactation counsellor (Brent 1995) or a breastfeeding researcher (Ryser 2004). The type of routine care received by the control group has not been specified by Ryser 2004 although care was similar between the comparison groups except for the intervention component provided to the experimental group. Brent describes routine care as optional prenatal breastfeeding classes, postpartum instruction by nurses and doctors and outpatient follow-up in the paediatric ambulatory department (Brent 1995). The education intervention implemented in the prenatal period only (Ryser 2004) aimed to increase initiation rates and measured changes in breastfeeding at one week postpartum. The intervention which was implemented across the prenatal, perinatal and postnatal periods aimed to increase both the initiation and duration rates (Brent 1995). This study reported a significant increase in the median duration of breastfeeding as a result of this intervention (intervention: 84 days; control: 33 days). Subgroup analysis of three studies, including 420 women, ( Coombs 1998;Hill 1987;Serwint 1996) evaluating the effect of generic, formal, single breastfeeding education sessions on the initiation of breastfeeding found a positive, but non-statistically significant increase in the number of women starting to breastfeed as a result of this type of health education intervention RR 1.26, 95% CI 1.00 to 1.60). Statistical heterogeneity was not found between these studies (I 2 = 0%). One intervention comprised generic information on parenting, including the advantages of breastfeeding, to each woman during a prenatal visit in her third trimester (Serwint 1996). Routine care was described as similar except no prenatal visit was provided. The study by Hill (Hill 1987) evaluated a one-off, generic lecture and discussion about breastfeeding with a post-test questionnaire to assess change in knowledge. Routine care comprised optional breastfeeding classes and postnatal support for mothers who wish to breastfeed (Hill 1987). A generic self-help manual was provided to each woman seven weeks before her expected date of birth during a routine prenatal breastfeeding counselling session from a nutritionist (Coombs 1998). Women in both comparison groups received a total of two prenatal interviews and two postnatal interviews. Some caution is required in interpretation of these findings due to variation between studies; for example, definitions of routine care, the methods, content and duration of the health education interventions. However, all studies were conducted among lowincome women in the USA with a range of ethnic backgrounds and feeding intentions and the type of formal, generic, one-off education session was consistent across all three studies.
(2) Peer support for women considering breastfeeding (comparison two) A single study evaluating the effect of prenatal, perinatal and postnatal peer support services among a total of 165 participants (Chapman 2004) was shown to be effective at increasing initiation rates among predominantly Latina women who were considering breastfeeding in the USA (RR 4.02, 95% CI 2.63 to 6.14, P<0.00001). Authors describe many of the study population as feeling socially uncomfortable with breastfeeding in the USA. The personalised, problem-solving approach of the intervention had been developed for 10 years in collaboration with the study hospital. Peer counsellors were community women who have completed high school, breastfed for six months and received 30 hours of internationally recognised classroom training in breastfeeding management. Counsellors served as observers for three to six months with experienced peer counsellors, received a payment (US$12) and the potential for health benefits if working at least 20 hours per week. Counselling services included at least one prenatal home visit, daily postpartum visits during hospitalisation and at least three home visits following return from hospital. Routine care was patient-led comprising breastfeeding information in response to participant's questions and written materials available at the prenatal clinic. Perinatal care included handson assistance and education from maternity ward nurses. Written breastfeeding materials and access to a Lactation Consultant for breastfeeding problems were also available if requested as was a 'warm line' where nurses answer postpartum breastfeeding questions. A significant increase in duration rates of breastfeeding was not demonstrated at one or three months postpartum. Failure to adhere to protocol, particularly the delivery of half of postnatal home visits in the first month, was a study limitation due to staffing problems.

(3) Breastfeeding promotion packs (comparison three)
A single study (Howard 2000) involving 547 women reported on the outcome of initiation of breastfeeding. The provision of a non-commercial breastfeeding promotion pack compared to a formula company produced pack has been shown to have no effect (RR 0.93, 95% CI 0.80 to 1.08, P=0.34) on increasing initiation rates among women of middle-or higher-income groups in a highincome country setting (USA).The authors also reported no effect on rates of stopping breastfeeding up to two weeks (RR 1.58, 95% CI 0.97 to 2.56).

(4) Early mother-infant contact followed by separation (comparison four)
A single study (Lindenberg 1990) (including 259 women and their babies) reported on the outcome of initiation of breastfeeding. It must be noted that whilst this intervention evaluated early motherinfant contact immediately after birth, women and their babies were then separated for the rest of their stay. This minimal contact intervention was shown to have no effect (RR 1.05, 95% CI 0.94 to 1.17, P = 0.39) on increasing initiation rates among women living in poor urban areas in Nicaragua.

D I S C U S S I O N
The meta-analysis of the five studies evaluating the effectiveness of health education interventions for increasing initiation rates of breastfeeding showed the interventions were effective overall (risk ratio (RR) 1.57, 95% confidence interval (CI) 1.15 to 2.15). The studies were small, with the largest having 200 participants. All five evaluated programmes were delivered in the USA to low-income women with a range of feeding intentions and where baseline breastfeeding rates are typically low. Programme components varied; however, all forms of health education included in this review seem to have increased breastfeeding rates.
Subgroup analyses of different types of health education intervention showed that both one-to-one, needs-based, informal repeat sessions and generic, formal antenatal sessions favour the intervention in terms of an increase in breastfeeding rates among women on low incomes regardless of ethnicity and feeding intention. Findings from these studies do suggest however that larger increases in the numbers of women are likely to result from needsbased, informal repeat education sessions than more generic, formal antenatal sessions. These findings are based only on studies conducted in the USA, however, raising some questions regarding generalisability to other settings.
The single study evaluating needs-based, informal peer support in the antenatal and perinatal periods was shown to be effective at increasing initiation rates among predominantly Latina women who were considering breastfeeding in the USA (RR 4.02, 95% CI 2.63 to 6.14). These findings are consistent with those reported above for needs-based, informal health education sessions from a health professional.
The evaluation of hospital breastfeeding promotion packs compared to formula-company produced materials about infant feeding (Howard 2000) showed this intervention to be ineffective at increasing initiation rates of breastfeeding. This trial was of high quality with sample size sufficient to provide 80% power to detect a 15% difference in breastfeeding initiation between the groups. However, approximately 40% of women in both groups reported receipt of formula company promotion items from sources other than their obstetric provider. These findings suggest that this form of breastfeeding promotion intervention is not the most effective use of breastfeeding promotion resources in a context of widespread use of formula company promotion items. Future research to evaluate the effect of full implementation of the WHO Code (WHO 1981) and the use of breastfeeding promotion packs in an environment free of formula-company produced materials would provide a useful contribution to the evidence base for the promotion of breastfeeding.
The evaluation of early mother-infant contact immediately after birth prior to complete separation until discharge on breastfeeding initiation rates (Lindenberg 1990), which was the only study included in this review that was conducted in a low-to middleincome country setting (Nicaragua), found no effect. In this study the overall breastfeeding initiation rate was 87%, higher than in the other included studies. Standard care was complete separation of mother and infant throughout hospitalisation (normally 12 to 24 hours). No sample-size calculation is reported. This finding suggests that early mother-infant contact followed by complete separation did not increase or decrease breastfeeding initiation rates in a population of women with relatively high breastfeeding rates and living in poor urban areas in a low-to middle-income country. Generalisation of the result of this evaluation is not recommended due to the moderate quality and size of the study and to fundamental concerns regarding the practice of routine separation of mother and infant prior to hospital discharge. The World Health Organisation recommends mothers and infants should not be separated after birth unless there is an unavoidable medical reason (WHO 1998). In addition, the literature on the promotion of the duration of breastfeeding provides clear evidence of the benefits of ongoing mother and infant contact during the hospital stay to support the mother's ability to breastfeed (Bonnin 1989;Inch 1989;Perez-Escamilla 1994;Renfrew 2000).
The Australian study (Forster 2004) conducted in a hospital which had been accredited with Baby Friendly status (WHO 1998) for four years prior to trial implementation was also conducted among a population with high breastfeeding rates (96% for control group receiving routine Baby Friendly Initiative care). Whilst the authors of this good quality study conclude that the high breastfeeding rates may have been a factor affecting the lack of effectiveness of the education interventions (Forster 2004), the 'intervention' of Baby Friendly Initiative implemented in this hospital as routine care is also noteworthy. In the absence of a formal trial evaluating the effectiveness of the Baby Friendly Initiative intervention in this hospital or hospital baseline data on breastfeeding initiation rates prior to implementation of Baby Friendly Initiative, it is not possible to attribute the current high breastfeeding rates across the study population to the Baby Friendly Initiative (Forster 2004). Other good quality non-randomised evidence on the effectiveness of the Baby Friendly Initiative does, however, demonstrate significant increases in initiation rates among all women attending maternity services with the Baby Friendly Initiative Full Accreditation Award (Bartington 2005;Kramer 2001;Philipp 2001;Radford 2001;Tappin 2001). Indeed, the routine implementation of the UNICEF Baby Friendly Initiative to full accreditation level, or an externally evaluated structured programme using the Baby Friendly Initiative as a minimum standard, has been recommended as clinical guidance and effective action for UK maternity and community services by the National Centre for Health and Clinical Excellence (Demott 2006;Dyson 2006). This is consistent with UNICEF/WHO adoption of the Baby Friendly Hospital Initiative as a global programme to support successful breastfeeding for all mothers and babies (WHO/UNICEF 1989).

Implications for practice
Health education and peer support interventions included in this review are likely to result in some improvements in initiation rates among low-income women in the USA where baseline breastfeeding rates are typically low. The small number of trials and participants, all of which were conducted in the USA, warrant caution for generalisability of findings to other country settings, particularly those where breastfeeding rates are typically high.
The type of education or support intervention which may be most likely to increase initiation rates among low-income women appears to be needs-based, one-to-one, informal sessions delivered in the antenatal or antenatal or perinatal period by a trained breastfeeding professional or peer counsellor. As above, generalisability of findings into practice to settings other than the USA warrants some caution.
Early mother-infant contact followed by complete separation until hospital discharge was not effective when targeting women with high breastfeeding rates living in poor urban areas in a low-and middle-income country setting. Enabling mothers and infants to remain together for 24 hours a day, 'rooming-in', is one of the Ten Steps of the UNICEF/WHO Baby Friendly Hospital Initiative adopted as a global programme to support successful breastfeeding and demonstrated to increase initiation rates for all women in all settings.
Breastfeeding promotion packs, in contexts where formula feeding packs are widely distributed, may be an inappropriate use of valuable breastfeeding promotion resources that could be more effectively used for population-appropriate breastfeeding education. However, increased support to adopt the WHO/UNICEF Code (WHO 1981) into national legislation would prohibit the provision and promotion of breastmilk substitutes to pregnant women, new mothers or families, potentially increasing the impact of breastfeeding promotion.

Implications for research
The effectiveness of interventions reviewed here needs to be assessed in diverse countries and settings, particularly outside the USA, in studies that are adequately powered, have adequate methods of randomisation, adequate reporting of losses to follow-up and utilise intention-to-treat analysis.
Publication of evaluations of effectiveness should detail the content and method of the intervention delivered, the people (e.g. peer or professional) who delivered it, the training and experience these people had, baseline breastfeeding rates for the studysite population and feeding intention for participants within each comparison group.
Future research should aim to evaluate the effectiveness of the intervention to improve both the initiation and duration of any and/or exclusive breastfeeding at least up to six months to enable appropriate planning and implementation of interventions during pregnancy and the postnatal period.
Further research to evaluate interventions that combine health education or support before the birth with support during the days immediately after the birth should be evaluated and compared with those that offer education alone.
Research to evaluate the impact of adopting and/or implementing the WHO Code on Marketing of Breast-milk Substitutes is urgently needed as well as the use of breastfeeding promotion packs in an environment free of formula-company produced materials. Further research into early mother-infant contact followed by rooming-in until hospital discharge may be effective at increasing breastfeeding initiation rates among various population groups.
Good quality research to evaluate the effectiveness of breastfeeding promotion and support on breastfeeding rates among maternity and community services who achieve fully accredited Baby Friendly status would further inform policy and practice.

Methods
Randomisation by permuted block. Analysis was by intention to treat.
Participants 108 English speaking, nulliparous, pregnant women attending a prenatal clinic, regardless of infant feeding preference were recruited into study. Participants stratified by age into 3 groups (less than 20, 20-29, or at least 30 years) Interventions Experimental group: (N = 51). Bf education and support provided throughout the prenatal and postpartum periods and into the first year of the child's life. Education consisted of 2-4 individual 10-15 minute sessions with a lactation consultant discussing the benefits and practice of bf. Content of sessions was based on the women's needs and interests. After delivery, mothers were followed up with daily inpatient rounds by the lactation consultant. Further follow up consisted of a telephone call 48 hours after discharge, a visit to the lactation clinic at 1 week and lactation consultation present at each health supervision visit until weaning or when the infant was 1 year of age, whichever came first. Professional education was directed at nursing and medical staff who interacted with the bf dyad Control group: (N = 57). Routine care, consisting of optional prenatal bf classes; postpartum bf instruction by nurses and doctors; outpatient follow up in the paediatric ambulatory department Outcomes Incidence of breastfeeding in hospital.
Incidence of breastfeeding at 2 weeks. Incidence of breastfeeding at 2 months. Incidence of breastfeeding at 6 months. Median duration of breastfeeding. Subgroup analysis for women who indicated at the first prenatal visit that they planned to formula feed or were undecided

Notes
To determine if a comprehensive breastfeeding promotion programme increased the incidence and duration of breastfeeding in a low-income population

Item Authors' judgement Description
Allocation concealment? Unclear Unclear whether allocation concealment was adequate.
Blinding? Women and staff No Inadequate due to non-blinded study.

No
Outcome assessors were not blinded to group allocations.
Incomplete outcome data addressed? All outcomes Yes Adequate, breastfeeding initiation reported for all 108 women in the study

Caulfield 1998
Methods Method of allocation of the four clinics: 4 slips of paper labelled with 1 of 4 clinics randomly selected from pot for centralised allocation to a pre-ordered list of comparison groups Analysis was not by intention to treat.
Participants 4 clinics administered through the Johns Hopkins University WIC program, that had similar rates for ethnicity (90.4-96.1% African-American) and breastfeeding rates at 1 month (2.0-5.9% in 1991). Women were recruited between April 1992 and January 1994 as they registered for prenatal care at the 4 clinics. 674 women were eligible. 242 had complete data (36%) and only these were included in the results. Differences were noted by clinic in parity, education and employment status before and during pregnancy of the included women Interventions 2 x 2 factorial design. Control (N = 57). Routine WIC services and nutrition education. Video intervention (N = 64). Breastfeeding motivational video, based on Best Start video, consisted of 8 trigger vignettes 2-5 mins in length, about benefits of and major benefits to breastfeeding, played continuously in the waiting area without staff supervision. Discussion with service provider following video was encouraged. Posters displayed in clinic areas and relevant sites Peer counselling intervention (N = 55). Women interested in breastfeeding received personalised information and support on breastfeeding issues of concern specific to each participant. Women received counselling 3 times during pregnancy. WIC counsellors were former WIC clients, had successfully breastfed and completed 5-week training program.
Video and peer counsellor (N = 66). All the components described above

Outcomes
Breastfeeding initiation. Breastfeeding initiation by infant feeding intention at enrolment. Breastfeeding at 7-10 days for those who initiated.

Notes
Not included in the meta-analysis on statistician's advice, because with only 1 clinic in each group, it is not possible to calculate the standard error of difference
Blinding? Women and staff Unclear Partially adequate. Randomisation at clinic level minimised crossover and contamination between groups compared to randomisation at level of individual. Unlikely that providers of intervention were blind Incomplete outcome data addressed? All outcomes Unclear Partially adequate.

Methods
Recruiter not the same as peer counsellors. Computerised random allocation of weekly cases: appears to be on-site but not stated. Data entry of cases likely to be Research Assistant who recruited but not likely to know how case would be allocated as SPSS random selection Analysis was by intention to treat.

Participants
Pregnant women attending Hartford Hospital, Connecticut, USA, on 1 of 3 days a week when recruitment conducted between July 2000 and August 2002. Inclusion (prenatal) over 18 years old, considering breastfeeding, low income. Inclusion (postpartum) healthy full-term singleton infant, no maternal history of HIV. Exclusion: infants admitted to Special Care.

Interventions
Control group (N = 75). Routine prenatal breastfeeding education consisted of individualised breastfeeding information offered in response to women's questions, and written breastfeeding materials from the prenatal clinic. Routine perinatal breastfeeding education included hands-on assistance and education from maternity ward nurses, written breastfeeding materials and access to an IBCLC for breastfeeding problems.
In addition to routine care as for control group, prenatal, perinatal (and postnatal) peer support services, consisting of at least 1 prenatal home visit to review benefits of breastfeeding, screen for inverted nipples, discuss breastfeeding myths, positioning and anticipatory guidance.

Coombs 1998
Methods Allocation method was an opaque container filled with 100 tags (50 -experimental group; 50 -control group). Following greater selection of women to the control group, a statistician calculated the number of C tags to be removed to bias further selection in favour of I tag until groups were balanced Analysis was not by intention to treat.
Participants 200 pregnant women, age 18 years or more, literate, no medical conditions likely to make bf difficult, willing to consider using the manual and to undertake interview about bf Those who agreed to participate after the interview differed significantly from those who declined in terms of parity, bf knowledge, attitudes, confidence and intention to bf Interventions Experimental group (N = 104).
Received the self-help manual 7 weeks before delivery during standard prenatal breastfeeding counselling from nutritionist. The manual was modelled on successful self-help smoking cessation interventions to reduce cigarette smoking among low-income pregnant women using cognitive behavioural theory. Received a total of 2 prenatal interviews and 2 postnatal interviews. In addition to BFHI accredited standard care, received the offer of a single session of 1.5 hours focusing on practical breastfeeding skills. 'Latching on' was explained and demonstrated using dolls and knitted 'breasts'. Breastfeeding complications and management were discussed. Partners were not present.
Attitudes intervention (N = 327). In addition to BFHI accredited standard care, received the offer of 2 1-hour sessions focusing on changing attitudes to breastfeeding. Partners/significant others were encouraged to attend. The first class included information about the advantages of breastfeeding and explored participants' views of breastfeeding and their perceptions of the attitudes of others. Between classes participants were encouraged to interview their own and their partner's mother. The second class included a group discussion based on these interviews, and discussion of resources for breastfeeding women. Women were encouraged to write a breastfeeding plan Outcomes Breastfeeding initiation. Breastfeeding and exclusive breastfeeding at 6 months.

Forster 2004 (Continued)
Notes Authors concluded that in settings where breastfeeding initiation is high, neither of the interventions could be recommended as effective.
Results not included in the meta-analysis because we considered the control group, BFHI standard care, had received an intervention that meant we could not compare this control group with the control groups of other studies in the review

Hill 1987
Methods Table of random numbers was used to allocate women of different parity to intervention or control groups Analysis was by intention to treat.
Participants 64 women intending to give birth at the study hospital and keep their infant, and who gave birth to a healthy infant, and had a telephone or agreed to return the Telephone Interview Survey by post 95% of the total sample were white women.

Interventions
Experimental group (N = 31). Attended a 40 minute lecture including 5-10 minutes for questions and answers; received a pamphlet with information that reinforced lecture content Control group (N = 33).
Routine breastfeeding classes to all women attending antenatal clinic with no lecture, discussion, pamphlet or post-test Outcomes Bf knowledge scores. Bf outcomes: no bf, any bf, bf less than 6 weeks, bf more than 6 weeks Notes To determine the effects of a breastfeeding education programme among low-income pregnant women in Chicago

Howard 2000
Methods Randomisation using computer-generated random number lists. Potential participants were identified by regularly reviewing first prenatal appointments scheduled at each of the 6 clinical sites. Randomly sized blocks of pregnant women were stratified further by obstetric practice before assigning to study group. 56% of the intervention (research pack) group reported prenatal receipt of formula company promotion items from sources other than their obstetric provider.
Authors stated that attrition from the study did not vary significantly by study group. Of the 547 women randomised, breastfeeding initiation data were not reported for 103 (18.9%) Analysis undertaken by authors for this review was by intention to treat based on data reported by study authors Participants 547 pregnant women attending prenatal visits at any one of 6 obstetric outpatient settings in Rochester, New York Participants were largely white (94.4%) and privately insured (96.8%) and most had plans to return to work within 6 months (60%)

Interventions
Experimental group (N = 270). At the first prenatal visit, participants received a research pack containing a generic diaper bag, non-commercial educational materials on pregnancy, infant feeding and infant growth and development, a coupon redeemable for $5 worth of infant items at a local store, and a package of electrical socket outlet covers.
Control group (N = 277). At the first prenatal visit, participants received a commercial pack containing a formula company diaper bag, formula company produced educational materials on pregnancy, infant feeding and infant growth and development, a can of powdered formula, a business reply card to join a 'baby club' redeemable for a case of infant formula, and several infant formula discount coupons

Howard 2000 (Continued)
Outcomes Any bf at delivery. Mean duration (days) of any bf. Cessation of breastfeeding during hospital stay. Cessation of breastfeeding in relation to breastfeeding goals. Risks for cessation of breastfeeding at 2 weeks.

Notes
To compare the effect of formula company-produced materials about infant feeding, to bf promotion materials without formula advertising, on breastfeeding initiation and duration

Lindenberg 1990
Methods Randomisation using a table of random numbers for the first 3 months. In the fourth month, a third group were assigned consecutively (due to ethical and organisational limitations) to a second intervention group. Results from this group have been excluded from this study due to the lack of randomisation for allocation.
Analysis was not by intention to treat.
Participants 259 women experiencing a normal, vaginal delivery with no complications and living in poor urban areas of Managua, Nicaragua

Interventions
Experimental group (N = 136). First 3 months of study: 45 minutes of mother-infant contact immediately after birth with standardised (uniform) breastfeeding promotion followed by complete separation until discharge. Standardised breastfeeding promotion consisted of a series of specific breastfeeding promotional messages.
Control group (N = 123). First 3 months of study: complete separation throughout hospitalisation with usual (ad Lindenberg 1990 (Continued) hoc) breastfeeding promotion. Ad hoc breastfeeding promotion consisted of the routine infant feeding information a mother might receive which, given the large volume of deliveries and short hospital stay, was usually very scant to non-existent Outcomes Any bf at 1 week. Exclusive bf at 1 week. Any bf at 4 months. Exclusive bf at 4 months.

Notes
To examine the effects of early postpartum mother-infant contact, followed by separation until discharge, on the incidence and continuation of breastfeeding

Ryser 2004
Methods Random assignment by participants selecting a sealed envelope (not sequentially numbered, opacity not specified) to determine assignment to intervention or control group Analysis was by intention to treat.
Participants 54 English speaking pregnant women of 18 years or more, literate, eligible for Medicaid, access to telephone and stated feeding intention of 'bottle (formula) feed' or 'undecided' Marital status and intention to formula feed differed significantly between comparison groups Interventions Experimental group (N = 26).
Received the Best Start Program (Bryant 1990), presented as a breastfeeding promotion campaign that aims to allow health professionals to examine women's misconceptions and educate them about their specific concerns. It has been marketed since 1992 and its materials have been used by various programs, including the SNPWIC Program. In this study, the researcher used the 'Best Start' videotapes, training manuals and handouts to implement the educational program during 4 prenatal visits (2 more than control group

Ryser 2004 (Continued)
as visits also included data collection phase).
Control group (N = 28). No exposure to Best Start Program. No details of routine breastfeeding promotion activities at the physician's office were provided Outcomes Any bf at 1 week postpartum. Attitudes to breastfeeding. Social and professional support.

Notes
To evaluate the effect of the 'Best Start' program on breastfeeding attitudes, intention and initiation in low-income women Methods Random number table with blocks of 10 to assign participants. Allocation of women to a paediatrician was not completely random as based on paediatrician availability according to mother's due date Analysis undertaken by authors for this review was by intention to treat based on data reported by study authors Participants 156 nulliparous women, > 18 years, between 8 and 28 weeks' gestation, who had not yet selected a paediatrician or wanted their infant to receive paediatric care at the hospitalbased paediatric clinic Both experimental and control groups comprised 91% of African-American women

Interventions
Experimental group (N = 81). In addition to routine care, received a scheduled prenatal visit between 32 and 36 weeks' gestation at a hospital-based clinic with the infant's future paediatrician. The clinic was in an urban academic medical centre where mothers received their obstetric care.

Serwint 1996 (Continued)
Prior to visits, paediatricians received training in counselling parents of newborn infants and bf techniques/promotion. During visits, paediatricians recorded data on timing of pregnancy, preparation for the infant, involvement of father, social support and maternal medical history. Parents-to-be were counselled on feeding options, advantages of bf, infant car safety, circumcision and access to paediatric healthcare.
Control group (N = 75). Similar management except no prenatal paediatric visits.

Outcomes
Bf intent before prenatal visit. Bf initiation at birth. Bf at 30 days postpartum. Bf at 60 days postpartum. Mothers who changed their mind in favour of bf after enrolment. Parent-physician relationship.

Notes
To assess the impact of prenatal paediatrician visits on breastfeeding decisions of lowincome mothers Inclusion: consent of woman and partner. All but 1 of the women who completed the study was receiving public assistance or WIC. Most women and partners were black and had more than high school education