Advocacy interventions to reduce or eliminate violence and promote the physical and psychosocial well-being of women who experience intimate partner abuse

Physical health of abused women

Intimate partner abuse is one of the most common causes of non-fatal injury in women. In the USA a review estimated that 50% of all acute injuries and 21% of all injuries in women requiring urgent surgery were the result of partner abuse (Guth 2000).

Abused women also experience many chronic health problems. The most consistent and largest physical health difference between abused and non-abused women is the experience of gynaecological problems (e.g. sexually-transmitted diseases, vaginal bleeding and infection, genital irritation, chronic pelvic pain, urinary-tract infections) (Campbell 2002). Population based studies from the United States show that the likelihood of abused women exhibiting these symptoms are three times greater than average (McCauley 1995). Other conditions include chronic pain (e.g. headaches, back pain) and central nervous system symptoms (e.g. fainting and seizures), (Campbell 2002; DiazOlavarrieta 1999), self-reported gastrointestinal symptoms (e.g. loss of appetite, eating disorders) and diagnosed functional gastrointestinal disorders (e.g. irritable bowel syndrome) (DiazOlavarrieta 1999; Coker 2000), and self-reported cardiovascular symptoms (e.g. hypertension, chest pain) (Tollestrup 1999).

Health of abused women during pregnancy

Research evidence shows that intimate partner abuse continues when a woman becomes pregnant - indeed, it may even escalate (Gazmararian 2000; Mezey 1997). A review by Campbell shows that prevalence rates of abuse during pregnancy are very similar in industrialised and non-industrialised countries (Campbell 2002). Most studies in the United States show that between 4%-8% of pregnant abused women are physically abused during pregnancy. This compares closely with 6%-8% during the past year in the United Kingdom, 6%-7% in Canada, at least 7% in South Africa, but is substantially lower than the 11%-21% in Sweden, and 13% in Nicaragua. The health risks for abused mothers and their unborn children are substantial. The most serious outcome is the death of the mother (Parsons 1999) or the foetus (McWilliams 1993; Jejeebhoy 1998). In well-designed studies, the outcome most associated with partner abuse is low birth weight, although there is also evidence for increased risk of miscarriage (Gazmararian 2000; Taft 2004) and foetal injury (Mezey 1997).

Psychosocial health of abused women

The most prevalent mental health sequelae of intimate partner abuse are depression and post-traumatic stress disorder (McCauley 1995; Ratner 1993; Golding 2002; Coid 2003). Women living in abusive relationships are three times more likely to be diagnosed depressed or psychotic (Stark 1996) and they often have feelings of low self-esteem and hopelessness (Kirkwood 1993). Among Australian women attending general practice who have been abused, depressed women are significantly more likely to have experienced combined physical, emotional and sexual abuse than are non-depressed abused women (Hegarty 2004). Living in a violent relationship may exacerbate a predisposition to depression; however, a woman's first exposure to abuse can also be a causal factor for subsequent depression (Campbell 1999; Silva 1997). Abused women are nearly four times as likely to suffer from post-traumatic stress disorder compared with non-abused women, and this can be directly related to experiencing intimate partner abuse (Golding 2002; Silva 1997). There is also evidence from the United States, Scandinavia and Papua New Guinea that increased suicidal tendencies are associated with abuse (Counts 1987; Golding 2002). Other signs of emotional distress associated with intimate partner abuse are self-harm and para-suicide (Stark 1996; Heath 2003), anxiety, insomnia and social dysfunction (Ratner 1993). A Nicaraguan study found that 70% of cases of emotional distress in women were a direct consequence of abuse (Ellsberg 1999).

In industrialised countries a further mental health problem associated with partner violence is the abuse of alcohol and drugs (McCauley 1995; Ratner 1993; Golding 2002). Substance abuse and intimate partner abuse often co-occur. Women who have experienced physical or psychological violence are fifteen times more likely to abuse alcohol and nine times more likely to abuse drugs than are non-abused women (Stark 1996). There is also evidence that alcohol and drug abuse for some women is directly attributable to intimate partner abuse (Stark 1996; Golding 2002).

Primary outcomes

Incidence of abuse

Forms of abuse included: (i)

physical

Abuse could be assessed using self-report measures (scales such as Index of Spouse Abuse, Women's Experience of Battering, Conflict Tactics Scale, or a single question about continuing abuse) or from the recording of abuse in medical or police records.

Psychosocial health (i)

quality of life (measures such as SF-36)

Secondary outcomes

Physical health (i)

deaths, all-cause and partner abuse-related (documented in medical/police records/regional and national databases)

Psychosocial health (i)

post-traumatic stress (measures such as Impact of Events scale)

Socio-economic outcome measures (i)

income

‘Proxy’ or intermediate outcome measures (including take-up of referrals to other agencies) (i)

the use of safety behaviours (e.g. use of coded telephone messages to a friend, keeping clothes at a friend's house, hiding emergency money)

We recognised that post-intervention changes in some of these proxy measures may be associated with both ‘positive’ and ‘negative’ health outcomes for abused women and require careful interpretation. For instance, increased refuge/shelter usage may reflect proactive behaviour on the behalf of abused women but it may also reflect an escalation of violence that has led to the women needing to seek safety. Where authors reported any adverse outcomes from interventions, such events were recorded and are discussed in the narrative summary.

Timing of outcome assessment

We documented the duration of follow-up in all included studies. We do not know the optimum period of follow-up. Thus, while an intervention may have some immediate positive effects on the health of an abused woman (such as a reduction in physical violence), other outcomes may not be so readily apparent. For example, even after leaving an abusive relationship, a woman may be traumatised for many months afterwards and any positive mental health effects may not be evident for some time. For purposes of this review, we defined short-term follow-up as up to and including 12 months, medium-term follow-up as from 12 to 24 months, and long-term follow-up as more than two years.

Handsearching

We handsearched the following journals from 1980 to September 2004: American Journal of Public Health, Australian and New Zealand Journal of Public Health, Journal of Family Violence, Medical Journal of Australia, Violence and Victims, and Women's Health. We did not find any additional papers and did not update hand-searches beyond September 2004.

Citation tracking

We examined the reference lists of acquired papers, and tracked citations forwards and backwards.

Other search strategies

In order to check for possible omissions, we emailed the first or correspondence authors of all the primary studies included in the review and also relevant researchers and members of intimate partner abuse groups and related organisations around the world. Efforts were made to make contacts in European countries where English is not the first language via the Domus Medicus organisation, and worldwide via the MRC Gender & Health Unit and the Department of Gender and Women's Health at WHO.

Sequence generation

Description: the method used to generate the allocation sequence was described in sufficient detail so as to enable an assessment to be made as to whether it should have produced comparable groups. Review authors’ judgment: was the allocation concealment sequence adequately generated?

Ratings: ‘Yes’ (low risk of bias); ‘No’ (high risk of bias) and ‘Unclear’ (uncertain risk of bias)

Allocation concealment

Description: the method used to conceal allocation sequence was described in sufficient detail to assess whether intervention schedules could have been foreseen in advance of, or during, recruitment. Review authors’ judgment: was allocation adequately concealed?

Ratings: ‘Yes’ (low risk of bias); ‘No’ (high risk of bias) and ‘Unclear’ (uncertain risk of bias)

Blinding

Description: any measures used to blind outcome assessors were described in sufficient detail so as to assess possible knowledge of which intervention a given participant might have received. Review authors’ judgment: was knowledge of the allocated intervention adequately prevented during the study?

Ratings: ‘Yes’ (low risk of bias); ‘No’ (high risk of bias) and ‘Unclear’ (uncertain risk of bias)

Incomplete outcome data

Description: Data on attrition were reported as well the numbers involved (compared with total randomized), and reasons for attrition were reported or obtained from investigators. Review authors’ judgment: were incomplete data dealt with adequately by the reviewers? (See also ‘Dealing with missing data’, below). Ratings: ‘Yes’ (low risk of bias); ‘No’ (high risk of bias) and ‘Unclear’ (uncertain risk of bias)

Selective outcome reporting

Description: attempts were made to assess the possibility of selective outcome reporting by investigators. Review authors’ judgment: were reports of the study free of suggestion of selective outcome reporting?

Ratings: ‘Yes’ (low risk of bias); ‘No’ (high risk of bias) and ‘Unclear’ (uncertain risk of bias)

Other sources of bias

Was the study apparently free of other problems that could put it at a high risk of bias?

We specified a priori that the following three criteria would be assessed under the heading ‘Other sources of bias’:

Baseline measurement of outcome measures

Description: information on the comparability of the groups in terms of the primary outcome measures at baseline were described. Review authors’ judgements: were the primary outcomes comparable between groups at baseline? Ratings: ‘Yes’ (low risk of bias); ‘No’ (high risk of bias) and ‘Unclear’ (uncertain risk of bias)

Reliability of outcome measures

Description: the primary outcomes were assessed using reliable measures (Cronbach's alpha 0.6 or above). Review authors’ judgements: were the primary outcomes assessed using reliable measures?

Ratings: ‘Yes’ (low risk of bias); ‘No’ (high risk of bias) and ‘Unclear’ (uncertain risk of bias)

Protection against contamination

Description: any measures taken to prevent or minimise the possibility that women in the control arm might receive part or all of the intervention were described so as to assess possible contamination between groups. Review authors’ judgements: did the study investigators take steps to prevent or minimise contamination? Ratings: ‘Yes’ (low risk of bias); ‘No’ (high risk of bias) and ‘Unclear’ (uncertain risk of bias)

Binary outcomes

For binary outcomes, a standardised estimation of the Odds Ratio (OR) with a 95% confidence interval was calculated. If the data to calculate the odds ratio were not reported or not available from the authors of studies, we have provided the findings as stated by the study authors.

Continuous outcomes

Continuous data were analysed where means (or mean changes) and standard deviations were available or obtainable from the authors of studies. In those instances where means and standard deviations were not available, the findings as reported by the study authors are provided. Where mean changes and standard deviations were reported but the findings were skewed (most likely because of the small sample size), we assumed a normal distribution and analysed accordingly. The majority of the studies included in the review used a range of scales to measure similar outcomes. As such, the treatment effects for these outcomes were standardised by dividing the mean difference in post-intervention scores or change from baseline scores for the intervention and control groups by the pooled standard deviation to create the Standardised Mean Difference (SMD) with 95% confidence intervals. For the few outcomes which were assessed using the same scale, the Weighted Mean Difference (WMD) was calculated.

Timeframe

Review authors intend to update this review within three years.

Recruitment

Four of the studies recruited women in health care settings (Hyman 2001; McFarlane 2000; McFarlane 2006; Tiwari 2005), four in domestic violence shelters (Constantino 2005; Jouriles 2001; Sullivan 1991a; Sullivan 1992), one primarily from domestic violence shelters but also from social and family service agencies (Sullivan 2002), and one in a District Attorney's office (McFarlane 2002). It is possible that women recruited in the health care settings may differ from those recruited elsewhere as they may not yet be at a stage where they are ready to accept help (Zink 2004). One of the studies (Sullivan 1992) recruited more participants over time. At the outset of the study, 146 women were recruited from shelters but this rose to a total of 283 reported at the two-year follow-up. However, their report of the three-year follow-up data was based on the original sample only.

The severity of the abuse experienced by participants was not always stated explicitly but mostly could be implied from either the study setting or the mean baseline abuse scores. For the five trials recruiting women either exclusively or primarily from domestic violence shelters (Constantino 2005; Jouriles 2001; Sullivan 1991a; Sullivan 1992; Sullivan 2002) it is likely that relatively severe abuse was being experienced and mean baseline abuse scores available for three of the studies support this (Sullivan 1991a; Sullivan 1992; Jouriles 2001). For the five studies recruiting participants from settings other than domestic violence shelters, the severity of the abuse was more variable. One study (McFarlane 2002), which recruited participants seeking protection orders, did not report any data on abuse severity. However, in order to obtain the orders, the women either had to present with a physical injury or a police report of the assault. In a study set in a hospital emergency department (Hyman 2001), 90% of the participants reported having been physically or sexually assaulted, 93% were experiencing ongoing abuse and 68% had received previous medical treatment for injuries. The severity of abuse experienced by participants at study entry in the three remaining trials (based either in antenatal or public health clinics) was less severe. Mean baseline scores indicated that moderately severe physical and psychological abuse was experienced by women at study entry in two of the studies (McFarlane 2000; McFarlane 2006), while in the third trial (Tiwari 2005) participants reported moderately severe psychological abuse but less severe levels of physical and sexual abuse.

Participants

A total of 1526 participants took part in the trials. Most of these women (84%) were recruited on the basis that they were experiencing current (within the last 12 months) physical or sexual abuse. Only three studies extended inclusion beyond this to also include women experiencing emotional abuse (Constantino 2005; Hyman 2001; Tiwari 2005). The majority of the participants were married or still intimately involved with the perpetrator at study entry. However, there were exceptions to this. In one study 79% of the women were no longer involved with the abusive partner (Sullivan 2002) and in another study (Jouriles 2001) the women were only recruited if the perpetrator no longer lived in the residence. The age range of the women recruited extended from 15 to 61 years, although more usually the women were aged between 24 and 35 years. The ethnicity of the samples varied across trials, but most included a mix of whites, African Americans and Hispanics. Only two studies involved single ethnic groups, these being a study reporting on Hispanic women (McFarlane 2000) and the Hong Kong study where all the participants were Chinese (Tiwari 2005). The socioeconomic status of participants was mixed, but most of the women were on low incomes and about two-thirds to three-quarters had not attended college. Only in two studies did a reasonably high proportion of the women either have an income that was average or above average for the country (Tiwari 2005) or an education that included attending some college (Hyman 2001).

The interventions

The models of advocacy employed in the trials were heterogeneous.

Content

In the one study where abused women were seeking protection orders against their perpetrators, the main thrust of the intervention was to discuss safety issues with participants (McFarlane 2002). In the other nine trials, the content of the advocacy was more varied; in these studies the advocacy included advice on safety and accessing resources, as well as the provision of support. The theoretical framework underlying each intervention was not always stated explicitly. However, in general, the advocacy provided to the abused women participants was based on the concept of empowerment. All of the interventions were pragmatic, in that the help provided was tailored to meet the wants and needs of the individual women. One of the trials extended this by also including advocacy for the children of mothers who had been abused (Sullivan 2002), and a second study (Jouriles 2001) also provided support for the children and taught child-management skills to the mothers. One of the studies evaluated a group intervention that allowed for interaction between the participants, as well as interaction between the intervention participants and the advocate (Constantino 2005).

Advocates

One of the interventions was administered by therapists (Jouriles 2001). The remainder were administered by trained advocates, though there was some variability in their professional status. One of the interventions was led by a professional advocate (Hyman 2001), four were administered by research nurses (Constantino 2005; McFarlane 2002; McFarlane 2006; Tiwari 2005), and three of the interventions were led by trained para-professional students (Sullivan 1991a; Sullivan 1992; Sullivan 2002). The tenth study was somewhat different in that it evaluated the effectiveness of advocacy provided by a professional advocate, as well as this type of advocacy when supplemented by the services of a trained nonprofessional “mentor mother” who also offered advocacy support (McFarlane 2000). With one exception, all of the interventions assessed were new and non-established modes of care within the settings in which the studies were conducted. The exception to this was a trial in an emergency room where two modes of “interventions” already in regular use were evaluated (Hyman 2001).

Duration

The duration of the interventions varied considerably and, to a large extent, this was related to the setting in which they were conducted. In general, the studies conducted outside of health care were of longer duration, both in terms of the length of the sessions and the period over which the sessions were offered. Typically, these types of interventions took place over a period of months (ranging from 8 weeks to 8 months) and provided advocacy totalling 12 hours (Constantino 2005), 23 hours (Jouriles 2001), 42 hours and a further 50 hours provided to the children (Sullivan 2002), and 60-80 hours (Sullivan 1991a; Sullivan 1992). The one study where women were recruited within a District Attorney's office provided telephone advocacy for up to eight weeks but the calls were relatively brief and totalled a mean of 72 minutes (McFarlane 2002). All of the interventions in health care settings were of relatively brief total duration. In two of the trials, short sessions of advocacy were offered over a prolonged period. In the first of these (McFarlane 2000), women were recruited in antenatal clinics and offered advocacy (or advocacy plus the services of a “mentor mother”) until the time of their delivery. The authors report that most sessions lasted about 30 minutes with women having an average of four to five sessions, thereby totalling up to approximately 150 minutes (personal communication). In the second study (McFarlane 2006), a 20 minutes intervention was administered on five separate occasions, each six months apart, thereby totalling approximately 100 minutes. The remaining two interventions within health care settings were both brief one-off sessions and lasted between 20-30 minutes (Tiwari 2005), and 90 minutes (Hyman 2001).

Outcomes

In terms of our primary outcomes, eight of the ten studies measured some form of abuse, three assessed quality of life outcomes, three measured depression, and four studies measured anxiety or psychological distress. There was little consistency across the trials in relation to the scales used to measure any of these outcomes.

Six different scales (or subscales) and a single item question were employed to measure abuse. Physical violence was the form of abuse most commonly measured. When assessing this outcome, four of the trials used revised versions of the physical violence subscale of the Conflict Tactics Scale (CTS) (Jouriles 2001; Sullivan 1991a; Sullivan 1992; Tiwari 2005), two studies employed the physical abuse subscale of the Severity of Violence Against Women Scale (SVAWS) (McFarlane 2000; McFarlane 2006), and one study (Hyman 2001) used the physical abuse subscale of the Index of Spouse Abuse (ISA). Emotional abuse was assessed using two measures. Two trials (McFarlane 2000; McFarlane 2006) employed the threats of abuse subscale of the SVAWS, while the Index of Psychological Abuse (IPA) was utilized in a further two studies (Sullivan 1992; Tiwari 2005). A single question was used to ascertain if sexual abuse had occurred in the one study (Tiwari 2005) that measured this outcome. Overall abuse was assessed by two different methods: one trial (Hyman 2001) combined the physical and non-physical abuse subscales of the ISA, while a second (Sullivan 2002) measured overall abuse by using a composite of three different scales (modified versions of the IPA, the CTS Physical Violence subscale and an unspecified 12-item injury scale). Two other aspects of abuse also were measured in one of the trials (McFarlane 2006): in this study the investigators asked about homicide risk using the Danger Assessment Scale (DAS) and harassment at work using the Employment Harassment Scale (EHS).

Quality of life was measured using two scales. One trial (Tiwari 2005) asked women to complete the SF-36 Health Survey, while two studies (Sullivan 1992; Sullivan 2002) both used an adapted version of a scale by Andrews and Withey (Andrews 1976). In terms of depression, again two different scales were employed. One study (Tiwari 2005) used a Chinese translation of the Edinburgh Postnatal Depression Scale (EPDS), while two trials (Sullivan 1992; Sullivan 2002) each measured this outcome using the Centre for Epidemiological Studies Depression Scale (CES-D). Anxiety/psychological distress was assessed using three scales. In two studies (Hyman 2001; Jouriles 2001) the participants were asked to complete the Symptoms Checklist 90-Revised Global Severity Index (SCL-90-R GSI), a third study (Constantino 2005) employed the Brief Symptom Inventory (BSI), and a fourth (Sullivan 1992) used the Rape Aftermath Symptom Test (RAST).

Follow-up

The duration of the follow-up period varied considerably between studies. Subsequent to any assessment of outcomes at the completion of an intervention, one study conducted no further follow-up of the women (Constantino 2005), and three trials carried out assessments at 10 weeks (Sullivan 1991a), 3-4 months (Hyman 2001), and 4 months (Sullivan 2002). One study (Tiwari 2005) states that outcomes were assessed at 6 weeks post delivery, but this could represent a follow-up period of anywhere between 16-34 weeks. Two studies report data at various time points up to 18 months post-intervention (McFarlane 2000; McFarlane 2002), two have multiple follow-ups for up to two years (Jouriles 2001; McFarlane 2006), and one reports data at multiple time points for up to 3 years post-intervention (Sullivan 1992).

Some of the papers reported reasons why there was a loss to follow-up. The primary reason stated was that some women could not be located (Hyman 2001; Jouriles 2001; McFarlane 2006; Sullivan 2002; Tiwari 2005). Other reasons included refusals to carry on with the study (Hyman 2001; Sullivan 2002), women returned to their abusive partner (Sullivan 1991a), difficulties with transport, childcare or living conditions (Constantino 2005), and the death of one control group woman who committed suicide (McFarlane 2002).

Sequence generation

All of the ten studies described treatment allocation as random but only three trials provided sufficient information on sequence generation (Hyman 2001; McFarlane 2006; Tiwari 2005). One study reported using random numbers tables to randomise participants (Hyman 2001) and two trials stated that they used a computer-generated method (McFarlane 2006; Tiwari 2005). The remaining seven studies either made no reference to sequence generation or provided insuffcient details.

Allocation concealment

Only two trials provided sufficient information and fulfilled the requirement of allocation concealment (Sullivan 1992; Tiwari 2005). In one study (Tiwari 2005), it was reported that a researcher not involved in the study placed the allocations in consecutively numbered sealed envelopes, and in the second study (Sullivan 1992) again sealed envelopes were used to maintain allocation concealment. Three of the trials did not have allocation concealment. In these studies, women were assigned to the condition selected in the week (McFarlane 2002; McFarlane 2006) or month (McFarlane 2000) she attended. All five of the remaining studies made no reference to allocation concealment.

Blinding

Blinding of participants or of those delivering advocacy interventions is not feasible. Two of the trials provided details of the measures taken to ensure blinded assessment of the primary outcomes. One of these studies (Hyman 2001) stated that the researchers were not provided with any information about randomisation status, while the other study (Tiwari 2005) stated that the researchers did not become aware of the women's randomisation status until after all follow-up data had been collected and questions were being asked about the women's satisfaction with their care.

There was no blinded assessment in three of the trials. In the first of these (Sullivan 1991a) we are told that the follow-up interviews with women who received advocacy were longer as they included questions about the programme, thus the researchers must have been aware of the women's randomisation status. In the second trial (McFarlane 2002) the same researcher who administered the intervention also collected all follow-up data, and in the third trial (McFarlane 2006) we are told that budget limitations precluded the use of blinding.

The other five studies did not give any explicit information on the blinding of assessors.

Baseline measurement of outcomes

We assessed reporting of baseline measurement of outcomes in the reviewed studies. In four of these, the study investigators stated that there were no substantive between-group differences in outcome measures at baseline (Jouriles 2001; McFarlane 2000; Sullivan 1991a; Sullivan 2002), while a fifth study (Sullivan 1992) stated that there were no such differences for their larger sample but information was not provided in relation to their original sample. Five of the trials did not state specifically that there were no baseline differences between groups, but they did report means and standard deviations and these looked comparable (Constantino 2005; Hyman 2001; McFarlane 2002; McFarlane 2006; Tiwari 2005) and thus we have rated these as also fulfilling this criterion.

Reliable primary outcome measures

One study (McFarlane 2002) considered a single outcome and this was assessed using the Safety Behavior Checklist, a measure previously employed by the study authors (McFarlane 1994). They state that the measure had been tested and proven to be extremely effective, but no statistics were reported and therefore reliability was rated by us as “not clear”.

All other nine studies measured multiple outcomes and often the primary outcomes were not stated explicitly, although this could sometimes be implied from the sample size calculation (Hyman 2001; Tiwari 2005) or from the research question (McFarlane 2000; Sullivan 1992). Based on this information, it was judged that all nine trials employed reliable measures to evaluate their primary outcomes. One study (Tiwari 2005) employed well-established measures that had not been evaluated for the mode of administration used, but given the high reliability associated with these measures (Conflict Tactics Scale (Straus 1988; Tang 1994), SF-36 Health Survey (Ware 1993; Lam 1998), Edinburgh Postnatal Depression Scale (Cox 1987; Lee 1998)), we still rated this trial as meeting the criterion.

Protection against contamination

Seven studies randomised the women on an individual basis. One of these (Tiwari 2005) took steps to counter possible contamination: the researchers who collected and analysed the data were not involved in the design of the analysis and were unaware of the study hypotheses; opportunity for the participants to meet were minimised; there was no record of a woman's allocation status in her medical record to influence other staff. One study (Hyman 2001) reported that there was contamination and approximately 19% women in the intervention group received control group care, while about 23% of women allocated to the control arm received the intervention. In the other five studies, where women were randomised on an individual basis (Constantino 2005; Jouriles 2001; Sullivan 1991a; Sullivan 1992; Sullivan 2002), it is possible that communication might have occurred between women in different arms of the trial and therefore we rated these as “not clear”.

Three studies used a method where women were allocated according to the week (McFarlane 2002; McFarlane 2006) or month (McFarlane 2000) in which they attended either a health care facility or a District Attorney's office. In one of these (McFarlane 2006), the same health care professionals managed both groups and therefore there was potential for the control group also to receive elements of the intervention accidentally. The other two studies (McFarlane 2000, McFarlane 2002) had designs which may have minimised potential contamination in that women in the different trial arms were unlikely to come into personal contact with each other and the treatments were not provided by the same individuals.

Incidence of abuse

Physical abuse

Six studies measured physical abuse; three of these evaluated brief advocacy interventions (McFarlane 2000, McFarlane 2006, Tiwari 2005) while the other three appraised intensive interventions (Jouriles 2001, Sullivan 1991a, Sullivan 1992). One study measured both severe and minor physical violence separately (Tiwari 2005), and this study was also different from the others in that mean change from baseline was assessed. Additional statistical data were obtained directly from two of the study investigators (Sullivan 1992; Tiwari 2005).

Effect sizes were calculated for two of the three studies that offered brief advocacy but these were not pooled because of the one trial that reported mean change from baseline. As indicated in (see Analysis 1.1, and Data and analyses), there was no evidence that advocacy leads to a reduction in severe physical abuse in women recruited from public health and women-and-children clinics (McFarlane 2006), either in the short-term at 12 months follow-up (SMD 0.07, 95% CI -0.15 to 0.29) or in the medium term at 24 months follow-up (SMD -0.02, 95% CI -0.24 to 0.19) (Analysis 1.1). Findings were more equivocal in the study which measured severe and minor abuse in pregnant women attending for antenatal care (Tiwari 2005). In this good quality study, a brief 30 minutes session of advocacy was not associated with any reduction in severe abuse as measured at 16-34 weeks post-intervention (change-score SMD 0.09, 95% CI -0.29 to 0.46) (Analysis 1.2) but a reduction in minor abuse was observed (change-score SMD -0.45, 95% CI -0.83 to -0.07, Analysis 1.2). The third trial that offered brief advocacy did not report any standard deviations, nor were these (or alternative relevant data) available from the authors (McFarlane 2000). However, in their paper the study authors report finding no evidence to suggest that advocacy (or advocacy plus the services of a “mentor mother”) reduced physical abuse in women recruited from antenatal clinics, either in the short or medium-term.

The effectiveness of intensive advocacy was assessed in three studies that all recruited women exiting from domestic violence shelters. One study measured physical abuse using both continuous and dichotomous measures, while two trials assessed this outcome using a dichotomous measure, see (Analysis 1.3, Analysis 1.4). In the one study where a continuous measure of physical abuse was employed (Sullivan 1992), there was no evidence to confirm or refute the hypothesis that advocacy leads to a reduction in severe physical abuse at 12 months follow-up (SMD -0.11, 95% CI -0.35 to 0.13) or at 3 years follow-up (SMD 0.16, 95% CI -0.19 to 0.52). (However, at the two-year follow-up it was reported that women who received advocacy did experience significantly less physical abuse than women in the control arm (SMD -0.24, 95% CI -0.48 to 0.00)). A similar pattern of results also emerged with the dichotomous data. Three studies contributed to a meta-analysis of the 12 months follow-up data (Jouriles 2001; Sullivan 1991a; Sullivan 1992); the point estimate favoured advocacy but the confidence intervals just crossed the line of no effect (OR 0.62, 95% CI 0.35 to 1.09). Two studies provided data for a meta-analysis of the 12-24 months follow-up data (Jouriles 2001, Sullivan 1992) and this suggests that participants who received advocacy experienced significantly less physical abuse than women in the control arm (OR 0.43, 95% CI 0.23 to 0.80). However, by three-year follow-up - when data were only available for the study by Sullivan and her colleagues (Sullivan 1992) - no benefit of advocacy was observed (OR 1.07, 95% CI 0.52 to 2.23) (Analysis 1.4).

Best case and worst case scenario analyses estimated the effects of missing data on the results of the two meta-analyses. For the 12 months follow-up data, imputing an improved outcome for women who dropped out of the study produced larger effect sizes and confidence intervals excluding no effect. When missing women in the intervention group were assumed to be no longer abused and those in the control arm were assumed to be abused: OR 0.53, 95% CI 0.30 to 0.93 (Analysis 1.5). When missing women in both intervention and control groups were assumed to be no longer abused: OR 0.55, 95% CI 0.31 to 0.97. Imputing a poor outcome for women who dropped out of the study led to a reduction in the size of the effect and wider confidence intervals including no effect. When missing women in the intervention group were assumed to be abused but those in the control arm were assumed to be no longer abused: OR 0.85, 95% CI 0.50 to 1.46. When missing women in both intervention and control groups were assumed to be abused: OR 0.82, 95% CI 0.48 to 1.41 (Analysis 1.5). A similar pattern of results emerged when the effects of missing data at 12-24 months follow-up were estimated (Analysis 1.6).

For descriptive purposes only, a forest plot outlining evidence for the effectiveness of any type of advocacy (brief or intensive) on physical abuse (as measured using a continuous measure) is given in (Analysis 1.7).

There is insufficient evidence at present to draw any unequivocal conclusions about the benefits of advocacy, either within health care settings or within the community, to reduce severe physical abuse. It is possible that intensive advocacy can reduce physical abuse one to two years after the intervention but that this effect wanes over a more protracted length of time. The primary studies are underpowered and only three could be pooled for some outcomes. The meta-analyses were still underpowered, with relatively wide confidence intervals. One high quality study suggests that brief advocacy may benefit women suffering minor abuse, but most studies did not measure minor abuse.

Sexual abuse

One study (Tiwari 2005) based in an antenatal setting evaluated the effect of receiving advocacy on subsequent sexual abuse. (Additional statistical data to those reported in the paper were obtained directly from the study investigators). Sixteen to 34 weeks post-intervention, there were no mean change differences in reported levels of sexual abuse between women who received a single 30 minute session of advocacy and those who did not (change-score SMD -0.11, 95% CI -0.49 to 0.26) (Analysis 2.1).

Emotional abuse

Emotional abuse was measured in four studies; three studies evaluated the effectiveness of brief advocacy (McFarlane 2000; McFarlane 2006; Tiwari 2005) and a fourth evaluated an intensive intervention (Sullivan 1992). The Tiwari study again measured mean change from baseline and the additional statistical data we required were obtained directly from the study investigators.

We calculated effect sizes for two of the three studies that offered brief advocacy, but we were unable to pool the findings of these as one of the trials reported mean change from baseline and was analysed separately. A study set in public health and women-and-children clinics (McFarlane 2006), did not indicate any significant reduction in emotional abuse at 12 months post-intervention (SMD 0.09, 95% CI -0.13 to 0.31) or at 24 months follow-up (SMD -0.06 95% CI -0.28 to 0.16) (Analysis 3.1). In contrast, the antenatal-based study by Tiwari and colleagues (Tiwari 2005) reported a statistically significant reduction in emotional abuse at 16-34 weeks after the receipt of a brief one-off session of advocacy (change- score SMD -0.72, 95% CI -1.11 to -0.34), (Analysis 3.2). The third trial did not report any standard deviations, nor were these (or alternative relevant data) available from the authors (McFarlane 2000). In the published paper of this study, however, the authors report that they found no evidence that multiple brief sessions of advocacy (or advocacy plus the services of a “mentor mother”) reduced emotional abuse for women recruited in an antenatal setting in the short or medium-term.

The one study that provided intensive advocacy to women exiting shelters (Sullivan 1992) did not show any significant reduction in emotional abuse at 12 months post-intervention - either when this was assessed using a continuous measure (SMD -0.03, 95% CI -0.27 to 0.21) (Analysis 3.3) or when using a dichotomous measure (OR 0.58, 95% CI 0.30 to 1.13) (Analysis 3.4). Similarly, there was no evidence of a reduction of emotional abuse as a result of receiving advocacy at 24 months follow-up (SMD -0.07, 95% CI -0.31 to 0.18) (Analysis 3.3).

A forest plot showing the effectiveness of any type of advocacy (brief or intensive) on emotional abuse (as measured using a continuous measure) (Analysis 3.5). With the exception of the trial by Tiwari and colleagues, there is little evidence that advocacy reduces or leads to a cessation of emotional abuse. This conclusion holds for studies conducted both within and outside of health care settings, and for studies offering either short or more intensive sessions of advocacy.

Financial abuse

No studies reported data for this outcome.

Overall abuse

Two studies reported on overall abuse (physical and emotional combined); one evaluated a brief session of advocacy for abused women attending an accident and emergency department (Hyman 2001) and one provided intensive advocacy over a 16-week period for women recruited from domestic violence shelters and social and family service agencies (Sullivan 2002). Both studies supplied additional statistical data to enable us to calculate the effect sizes. Neither study found a significant benefit from advocacy. The point estimate in the Hyman study (Analysis 4.1) favoured the intervention group but the confidence intervals included zero (SMD -0.35, 95% CI -0.90 to 0.20), while the Sullivan trial indicated a trivial benefit to the control group of women (SMD 0.03, 95% CI -0.42 to 0.48) (Analysis 4.2). The effectiveness of any type of advocacy (brief or intensive) on overall abuse is given in Analysis 4.3.

Other abuse measures

Psychosocial health

Six of the ten studies evaluated the effects of advocacy on the primary outcomes measures assessing psychosocial health.

Quality of life

Quality of life was assessed in three of the studies. One of the studies evaluated a brief advocacy intervention and reported mean changes from baseline for each of the eight subscales comprising the SF-36 Health Survey measure of quality of life (Tiwari 2005), while the other two trials measured overall quality of life following an intensive advocacy intervention (Sullivan 1992; Sullivan 2002). Additional statistical data were obtained directly from the investigators of all three trials.

The Tiwari study (Tiwari 2005), where women recruited in an antenatal clinic received a one-off session of advocacy, had mixed results. Women who received advocacy reported significant improvements on three out of eight quality of life subscales as at 16-34 weeks post-intervention: physical functioning (change-score SMD 0.50, 95% CI 0.12 to 0.88), role physical (change-score SMD 0.41, 95% CI 0.03 to 0.78), and role emotional (change-score SMD 0.56, 95% CI 0.18 to 0.94). There were no between group differences on four subscales: general health (change-score SMD -0.09, 95% CI -0.47 to 0.28), vitality (change-score SMD 0.03, 95% CI -0.34 to 0.40), social functioning (change-score SMD 0.16, 95% CI -0.21 to 0.53), and mental health (change-score SMD 0.02, 95% CI -0.35 to 0.40). Women in the intervention arm fared significantly worse on the remaining bodily pain subscale (change-score SMD -0.46, 95% CI -0.83 to -0.08) (Analysis 7.1).’

The two studies conducted by Sullivan and colleagues (Sullivan 1992; Sullivan 2002) each evaluated the effectiveness of advocacy delivered over a period of 10-16 weeks for women exiting domestic violence shelters or referred from social and family service agencies. Both studies reported 12 months follow-up data and, when pooled, a higher quality of life was indicated for women in the intervention group; however, the confidence intervals included zero (WMD 0.23, 95% CI 0.00 to 0.46). Only one of these two studies followed up participants beyond 12 months (Sullivan 1992); data were obtained both at 2 years post-intervention, and again at 3 years but only for the original smaller sample of recruited women. Point estimates for effect on quality of life were positive and favoured the intervention, although again the confidence intervals spanned no difference at 2-year follow-up (WMD 0.25, 95% CI -0.02 to 0.52) and 3-year follow-up (WMD 0.30, 95% CI -0.07 to 0.67) (Analysis 7.2). NB., these findings differ from the significant effects reported by the study authors. We calculated weighted mean differences based on the reported raw means and standard deviations. The study authors reported findings based on multivariate analyses (2-year follow-up) and employed a p < 0.10 significance level to avoid type II errors as a result of the smaller sample size (3-year follow-up).

Overall, these studies are inconclusive. Only one study evaluated whether brief advocacy improves quality of life in women who have experienced partner abuse and this had mixed findings. For advocacy provided over a period of months, the point estimates suggest that such an intervention may help abused women to attain a better quality of life both in the short and longer term. The confidence intervals spanned no effect but primarily were indicative of improved quality of life following the receipt of advocacy.

Depression

Three studies assessed the effect of advocacy on depression; one evaluated the effect of a brief intervention on post-natal depression (Tiwari 2005), and two studies appraised advocacy interventions that were intensive (Sullivan 1992; Sullivan 2002). Additional statistical data from two of the study investigators (Sullivan 1992; Tiwari 2005) were obtained by us.

In the trial by Tiwari and colleagues (Tiwari 2005), the results (Analysis 8.1) show that fewer women developed post-natal depression if they received a brief advocacy intervention in the antenatal period (OR 0.23, 95% CI 0.10 to 0.57). For the two intensive advocacy studies (Sullivan 1992; Sullivan 2002), both assessed depression at 12 months follow-up. A meta-analysis of these data revealed no effect of intensive advocacy on depression for women exiting domestic violence shelters (WMD -0.05, 95% CI -0.19 to 0.09). NB., the authors of the Sullivan 2002 study reported this as a significant effect, based on a multivariate analysis. Only one of these two intensive advocacy studies measured depression over a longer time period (Sullivan 1992): at two years follow-up, there was no evidence that receiving advocacy was consistent with less depression among women in the intervention arm (WMD -0.08, 95% CI -0.24 to 0.08) (Analysis 8.2).

Overall, there is inconsistent evidence that advocacy has a beneficial impact on the degree of depression experienced by abused women. There is limited evidence that a one-off session of advocacy within healthcare may help pregnant abused women to suffer less post-natal depression in the months following the birth.

Anxiety/Psychological distress

Only one study measured anxiety as an outcome measure (Sullivan 1992). Three other studies measured psychological distress (Constantino 2005; Hyman 2001; Jouriles 2001) and, as this is a similar concept to anxiety, we discuss them together here. One of the four studies appraised a brief advocacy intervention, while the remaining three all evaluated the effectiveness of intensive advocacy. One of the latter studies (Constantino 2005) reported mean change from baseline and was analysed separately. Additional statistical data were obtained for two of the studies (Hyman 2001; Sullivan 1992).

For the brief intervention set in an accident and emergency department (Hyman 2001) there was evidence, at three to four months follow-up, that advocacy reduced psychological distress (SMD -0.62, 95% CI -1.18 to -0.06), see Analysis 9.1.

All three intensive interventions recruited women in domestic violence shelters. However, we were unable to pool the findings from the one trial that reported mean change from baseline. As indicated in Analysis 9.2, a meta-analysis of the pooled data from two studies (Jouriles 2001; Sullivan 1992) showed at up to 12 months follow-up there was no reduction of anxiety/psychological distress associated with the receipt of intensive advocacy (SMD -0.16, 95% CI -0.39 to 0.06). Sullivan and her colleagues also collected data at two further time-points. On neither occasion was any statistically significant benefit evident: 24 months follow-up (SMD -0.01, 95% CI -0.25 to 0.23), 36 months (SMD -0.13, 95% CI -0.49 to 0.22). In the one study where change-from-baseline data were reported (Constantino 2005), the women were only followed up immediately post-intervention. The point estimate was consistent with decreased psychological distress in participants, but the confidence intervals spanned no effect (change-score SMD -0.69, 95% CI -1.52 to 0.14) (Analysis 9.3). NB., in the Constantino trial, the authors reported a statistically significant effect (p = 0.013) after using a non-parametric test to analyse the change from baseline data.

For descriptive purposes only, a forest plot showing the effectiveness of any type of advocacy (brief or intensive) on anxiety/psychological distress is given in Analysis 9.4.

These studies suggest that brief advocacy offered within health care settings may reduce abused women's feelings of anxiety and psychological distress in the short-term but it is not known if such a benefit would persist over time. The benefits of offering intensive advocacy to abused women in the community are unclear at present.

Physical health

Outcomes listed in the protocol were not assessed by any of the included studies: deaths, physical injuries, chronic health disorders, general measures of physical health, and pre-term birth.

Psychosocial health

Post traumatic stress disorder

Women who received advocacy in one trial (Hyman 2001) did not significantly differ from women in the control groups on a measure of post traumatic stress disorder (SMD -0.45, 95% CI -1.00 to 0.11), see Analysis 10.1. These data were requested from the study investigator.

Perceived stress

Data from one study (Hyman 2001) suggests that women who received advocacy had lower levels of perceived stress as compared with women in the control condition (SMD -0.62, 95% CI -1.18 to -0.06), see Analysis 11.1. These data were requested from the study investigator.

Self efficacy

This outcome was measured in one study that evaluated the effectiveness of intensive advocacy for women exiting domestic violence shelters (Sullivan 1992). Additional statistical data relating to this outcome were obtained from the study invesigators. At all time points, there was evidence that women who received advocacy reported higher levels of self efficacy than women in the control arm; however, the confidence intervals were wide and included the line of no effect. The effect sizes were: 12 months follow-up SMD 0.07, 95% CI -0.17 to 0.31), 2 years SMD 0.15, 95% CI -0.09 to 0.39), and 3 years SMD 0.13, 95% CI -0.22 to 0.49) (Analysis 12.1).

Self esteem

One study (Sullivan 2002) reported a benefit in improved self esteem following advocacy (using a multivariate analysis), although our calculated effect size shows that this was a statistically non-significant improvement (SMD 0.39, 95% CI -0.06 to 0.85) (Analysis 13.1). These data were requested from the study investigator.

Social support

Social support was assessed in three of the studies and each evaluated the effect of intensive advocacy offered over a prolonged period to women recruited in or exiting domestic violence shelters or referred from social and family service agencies (Constantino 2005; Sullivan 1992; Sullivan 2002). All of these studies assessed social support using an overall measure, but one trial (Constantino 2005) also assessed four separate components of social support. This study was also different from the others in that the authors assessed mean change from baseline. Additional statistical data were obtained for all three of the studies.

Two of the trials (Sullivan 1992, Sullivan 2002) reported 12 months follow-up data that could be pooled. As indicated in the forest plot, there was no evidence to suggest that advocacy had a positive effect on social support in the short-term (WMD 0.08, 95% CI -0.16 to 0.33). Further follow-up data were only available from one of these studies (Sullivan 1992). There was a small, statistically non-significant improvement in the advocacy arm at 2 years (WMD 0.13, 95% CI -0.12 to 0.38), but by 3 years (based on the smaller original sample recruited in to the study) there was a clear beneficial effect on social support (WMD 0.39, 95% CI 0.04 to 0.74) (Analysis 14.1). Constantino and colleagues (Constantino 2005) reported finding no effect on overall social support following advocacy (using a non-parametric test), but our analysis indicates that there was a positive effect (change-score SMD 1.60, 95% CI 0.66 to 2.55) when the women were followed-up immediately post-intervention. Similarly, while the authors reported an improvement only on the “belonging” subscale, we found significant effects for all four sub-scales: “tangible” (change-score SMD 1.24, 95% CI 0.35 to 2.13), “appraisal” (change-score SMD 0.92, 95% CI 0.07 to 1.77), “self esteem” (change-score SMD 1.18, 95% CI 0.30 to 2.06), and “belonging” (change-score SMD 1.06, 95% CI 0.19 to 1.93) (see Analysis 14.2).

There is inconsistent evidence that advocacy improves social support in the short term when offered to abused women who have actively sought help. In the longer term, there may be a positive effect but this is based on the findings from one small study.

Alcohol / drug abuse, attempted suicide, self-harm

No study reported data on any of these outcomes.

Impact on relationships

Independence from the assailant

Three studies measured independence from assailant (Jouriles 2001; Sullivan 1991a; Sullivan 1992). Each of these trials recruited women in domestic violence shelters and provided weekly sessions of advocacy over a period of at least 10 weeks up to eight months. Additional statistical data were obtained for one of the studies (Sullivan 1992).

Two of the trials measured this outcome at up to 12 months follow-up (Sullivan 1991a; Sullivan 1992), and thus we were able to pool the data (Analysis 15.1). No positive short-term effect on this outcome was found (OR 1.22, 95% CI 0.74 to 2.00). A similar size of effect was also evident in the two studies that measured independence from the assailant at 12-24 months follow-up (Jouriles 2001; Sullivan 1992). Specifically, the point estimate was consistent with increased independence and thus favoured the intervention but the confidence intervals crossed the line of no effect (OR 1.40, 95% CI 0.88 to 2.24). One of the studies (Sullivan 1992) also reported independence at three years, again there was no difference between the intervention and control groups (OR 0.97, 95% CI 0.47 to 1.98) (Analysis 15.1). The effects of missing data on the results of the two meta-analyses were estimated via best case and worst case scenario analyses. For the 12 months follow-up data, a best case scenario (imputing an improved outcome for women in the intervention group and a lack of improvement in the control arm) produced a larger effect size but the confidence intervals still included no effect. When missing women in the intervention group were assumed to be independent and those in the control arm were assumed to be non-independent: OR 1.58, 95% CI -.98 to 2.55. When missing women in both intervention and control groups were assumed to be independent: OR 1.13, 95% CI 0.71 to 1.80. When missing women in the intervention group were assumed to be non-independent but those in the control arm were assumed to be independent: OR 0.82, 95% CI 0.52 to 1.31. When missing women in both intervention and control groups were assumed to be non-independent: OR 1.13, 95% CI 0.70 to 1.82 (Analysis 15.2). A similar pattern of results emerged when the effects of missing data at 12-24 months follow-up were estimated, but on this occasion the best case scenario produced an even larger effect size and the confidence intervals excluded no effect: OR 1.81, 95% CI 1.16 to 2.84 (Analysis 15.3).

These results suggest that there is no clear evidence that the provision of intensive advocacy for women who have sought refuge in domestic violence shelters may help them eventually to become independent of their assailants.

Emotional attachment to the abuser

One study considered whether intensive advocacy offered to women exiting from domestic violence shelters had a positive effect on their emotional attachment to the perpetrator of the abuse (Sullivan 1992). Again, additional statistical data were supplied by the study investigators. As indicated in the forest plot, there was no evidence that abused women who received advocacy were any less emotionally attached than those in the control arm at either 12 months follow-up (SMD -0.08, 95% CI -0.32 to 0.16), or at 2 years (SMD 0.00, 95% CI -0.24 to 0.24). By 3 years, the point estimate was consistent with women in the intervention arm being less emotionally attached to the abuser, but the confidence intervals included zero (SMD -0.35, 95% CI -0.71 to 0.00) (Analysis 16.1).

Socio-economic measures

No study provided data on socio-economic outcome measures including including income, housing, participation in education or participation in work.

Safety behaviours

Two studies investigated the use of safety behaviours. These were both brief interventions (less than three hours total duration) but the settings were different. The first recruited women attending a district attorney's office to obtain a protection order (McFarlane 2002), and the second from health care settings (McFarlane 2006). Both reported data at 12 months post-intervention. We found a benefit associated with the receipt of advocacy (WMD 0.60, 95% CI 0.14 to 1.06) with women in the intervention arms of the trials employing more safety behaviours. This benefit was also apparent up to two years post-intervention (WMD 0.48, 95% CI 0.04 to 0.92) (Analysis 17.1). With only two studies measuring safety behaviours, we could not perform any sub-group meta-analyses based on interventions set in health service settings versus non-health service settings. Nonetheless, as the forest plots indicate, there is some suggestion that a brief advocacy intervention may only lead to a substantial and long-lasting improvement in the use of safety behaviours when the women are in fear for their safety (and so seeking protection orders). It is less clear whether the receipt of brief advocacy improves safety behaviours in abused women who are not actively seeking legal protection.

Resources

Use of resources

At protocol stage we planned to collect data on participants’ use of refuges/shelters, use of counselling, calls to police, police reports filed and protection orders sought, as individual items. Data in reported papers was however assessed in the form of a composite measure (the Community Resources Checklist) which did not permit us to report as planned. The checklist assesses if the following resources have been accessed: alcohol/drug treatments, battered women's groups, chuch/clergy, health care, legal services, police, shelters, social services, education services, and employment services. The last two of these resources were subsequently dropped from the measure by the study invesigators; therefore the findings reported below are based on two related but not identical measures.’

The use of resources was measured by two studies, one employing a dichotomous measure of the 10 items (McFarlane 2000) and one using a continuous measure of the revised 8-item version (McFarlane 2006). Both studies recruited women from health care settings and provided multiple brief sessions of advocacy over a period of months. Both studies conducted short-term (up to 12 months) and medium-term (12-24 months) follow-up. The effect sizes for the trial using a continuous measure (McFarlane 2006) are given in Analysis 18.1. There was a point estimate consistent with accessing more resources at both time points although the confidence intervals included zero: at 12 months follow-up SMD 0.22, 95% CI 0.00 to 0.44, and at 12-24 months follow-up SMD 0.15, 95% CI -0.07 to 0.37. The study employing a dichotomous measure (McFarlane 2000) did not find that advocacy increased the likelihood of resource use over and above that found in the control group of women at any follow-up point, either when advocacy was offered on its own or in combination with the services of a “mentor mother”. In fact, the reverse occurred and the likelihood of accessing resources dropped below baseline levels in all groups; particularly in the control (from 32% to 17%) and advocacy (from 33% to 17%) groups, but also for the group with access to a “mentor mother” (from 23% to 21%). We were unable to calculate the effect sizes for this trial because of the baseline differences between the three groups. The results from these two trials are contradictory and we cannot draw any conclusion about the effects of advocacy in health care settings on individual resource use.

Difficulty in accessing resources

This outcome was concerned with measuring dificulties in accessing resources and was measured using a composite 11-item measure (housing, material goods and services, education, employment, health care, child care, transportation, social support, legal assistance, financial issues, and issues regarding the children). Where participants had not directly tried to access a resource, they were asked to answer how difficult they expected it would be.

This outcome was assessed by two studies (Sullivan 1991a; Sullivan 1992), the latter study providing data additional to that reported in their papers. Both studies recruited women exiting a domestic violence shelter and provided intensive weekly advocacy sessions over a period of 10 weeks. However, Sullivan 1991a did not report any means or standard deviations and these were not available from the study authors. Findings from Sullivan 1992 were mixed. There was no evidence to suggest that abused women who received advocacy experienced fewer difficulties in accessing resources when compared with women in the control arm, either at 12 months follow-up (SMD -0.02, 95% CI -0.26 to 0.22) or at 3-year follow-up (SMD -0.20, 95% CI -0.55 to 0.15). However, at the 2-year follow-up, there was a benefit associated with receiving advocacy (SMD -0.28, 95% CI -0.52 to -0.04) (Analysis 19.1). As such, based on the findings of this one trial, at present there is insufficient evidence to indicate whether intensive advocacy does or does not decrease difficulties in accessing resources.

Maintenance of family ties

No study included in this review reported data on maintenance of family ties (i.e. children staying with mother).

Adverse events

Two participants died before completion of the studies into which they were recruited. In a study where women seeking protection orders were recruited, one woman in the control group committed suicide three weeks into the trial (McFarlane 2002) and in a study recruiting women within domestic violence shelters (Sullivan 1992), one woman in the intervention arm was murdered by her partner two weeks into the trial.