Early government policy responses to IPV in Australia, as elsewhere in Western countries, prioritised crisis intervention from police, legal and domestic violence services. More recently, a fervent debate about health care professional IPV screening and early intervention has included a focus on the lack of evidence for effective interventions[5]. While professional intervention has been recently examined,[6] less research attention has been paid to the potential contribution of non-professional support for primary care patients.

Pregnant women and those with young children frequently attend primary care [7] General practice (GP) and maternal and child health (MCH) clinics (free community-based centres for women with children from birth to six years of age) are universal, accessible and affordable primary care services in Melbourne, Australia. Thus, there are opportunities for prevention and early intervention strategies offered by the less stigmatising, confidential environment available in primary care. However, clinicians are often hampered by a range of individual and contextual barriers[8,9]. Inadequate training and support for identifying and managing IPV is complicated by limited evidence for what subsequent interventions to which they could refer, are beneficial for women from primary care settings[10,11].

Peer support and home visiting strategies have been shown to reduce maternal depression in the postnatal period,[12-14] but the picture is less clear when IPV is present[15]. In the most recent Cochrane review of IPV advocacy interventions, Ramsay et al (2009) argued for an increased range of intervention studies overall, but in particular for women remaining at home with perpetrators[6].

The MOSAIC (MOtherS' Advocates In the Community) model trialled in the current study combined evidence for the benefits of social support,[16] advocacy,[17] and antenatal mentoring [18] to reduce partner violence and improve women's mental and physical health. We located the study in primary care to contribute to the limited evidence about effective referral and intervention strategies in this setting and because mothers experiencing IPV are more prevalent in these populations. The development of the study is outlined in more detail in the published protocol[19].

In a pragmatic cluster randomised trial, we tested the effectiveness of the model to reduce partner abuse and depression and improve women's health and attachment to their children among pregnant or recent mothers identified in primary care.

All findings were consistently in favour of the intervention arm, and the lack of stronger evidence for the differences may be due in part to the smaller than anticipated numbers of participants, resulting in reduced power to detect small, but meaningful differences. Despite substantial efforts to enhance clinician IPV knowledge, skills and resources, and clinician agreement to the randomised trial design, MOSAIC suffered from low rates of identification and referral from both nurses and doctors, particularly in the trial's comparison arm. There is evidence from clinician interviews of two problems. The first is that despite considerable training, many clinicians continued to feel under-confident to ask about IPV. We conclude that major systemic challenges remain to be overcome before health care providers feel sufficiently supported and confident to identify and effectively care for the abused women in their respective populations. Second, there was clearly a reluctance to refer women in the comparison arm in comparison with the intervention arm, due to a perception of no benefit for this group of vulnerable women even after feedback of positive comments about participation from women recruited to the comparison arm. This may also contribute to bias.

Thus, the major threat to the validity of the study was clinician non-blinding[34]. This aspect of the design was thought to be unavoidable, as the mentor-mother program was designed to be integrated with primary care and women did talk about mentoring with their clinician. Low referrals overall and the 2:1 ratio of recruited women in the intervention versus comparison arms, introduced potential selection bias and had a substantial impact on our final sample size and our power to provide better evidence for intervention effects. Our approach to addressing selection bias involved adding PS to our analysis. While the PS analysis did result in altered effect sizes, it did not alter the direction of effects. The PS analysis did not alter the consistent trend towards more favorable outcomes in the intervention arm of the trial.

Could we have avoided the problem? One solution would have been to randomise individual women once they had been referred to the research team. With hindsight this has some merit, given the small number of women any individual clinician actually referred, which meant that contamination may have been unlikely. However, it is also unclear whether the trial would have experienced even lower numbers of referrals if clinicians had known that women referred would only have a 50:50 chance of receiving the offer of a mentor.

The self-report nature of the measures and the fact that we measured outcomes immediately after the intervention was completed, may be additional sources of bias. It remains unknown if any beneficial effect was sustained beyond 12 months.

In the only other mentoring study - Madres a Madres, a study among an abused pregnant population where the majority of mentoring was provided by telephone and only up until the time of birth-McFarlane et al found that the benefits of mentoring during pregnancy were not sustained to 18 months after birth[35,18]. Birth and infancy are periods of considerable strain among families where there is no IPV, so the additional stress for mothers experiencing IPV is self-evident, whenever the abuse commenced. The emotional and practical support from mentors during this infancy period, sustained for up to a year as in the current study, may explain why our findings suggest some benefit.

Our study findings are consistent with the growing number of studies providing evidence of the benefit of home-visiting for vulnerable mothers[36-38]. Yet, in the great majority of these, home visiting is provided by nurses and there is no evidence to date that nurse visits are effective when there is IPV present[15]. Nurses are often mandated to report child abuse and many women affected by IPV are fearful of losing their children or have many professionals already involved in their care. Given the findings of the current study we suggest there is a role for non-professional befriending models in the spectrum of professional and non-professional responses to IPV.

Angela Taft held a VicHealth Public Health Fellowship. Rhonda Small held a National Health and Medical Research Council (NH&MRC) Career Development Award. Lisa Gold is supported by an NH&MRC Capacity Building Award.

No-one from any funding body was involved in design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the manuscript.

Previous presentation of the information reported in the manuscript. Draft results of this study have been presented to FVPF conference, New Orleans and WONCAEurope conference, Basel, Switzerland, 2009 and the National Women's Health Conference, Melbourne May 2010.