The sensitivity and specificity of four questions (HARK) to identify intimate partner violence: a diagnostic accuracy study in general practice

Violence against women is a global issue affecting millions who experience it and have to live with its consequences [1]. Intimate partner violence (IPV) including physical, sexual and emotional abuse is a major public health problem.

The WHO Violence Against Women study [2] found that the prevalence of lifetime physical violence and sexual violence by an intimate partner, among ever-partnered women varied from 15 to 71% in urban and rural settings in 10 countries. The prevalence of IPV is higher among women seeking primary care than in community surveys of the same geographic populations [3].

In a study in 12 east London general practices it was found that 41% of women waiting to see their general practitioner (GP) or practice nurse had experienced physical violence from a partner or former partner. 17% had experienced it within the past year [4].

IPV causes short and long term health problems. From controlled studies in a wide range of settings, we know that these include injury, chronic pain, gastrointestinal and gynaecological conditions (including sexually transmitted diseases) [5]. Consequences of IPV extend to perinatal health with it being an independent risk factor for deficit in gestational weight gain during pregnancy [6] and strong evidence of an IPV association with low birth weight [7].

The psychological health problems associated with domestic violence are no less serious and have psychological parallels with the trauma of being taken hostage and subjected to torture [8]. The most prevalent mental health sequelae of IPV are depression and post-traumatic stress disorder [9].

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, and there is evidence that substance abuse is a consequence as well as a potential cause of IPV [10]. Children exposed to domestic violence also often experience emotional and behavioural problems [11]. In the developing world it has been shown that children exposed to severe and recurrent IPV are more likely to be admitted with severe acute malnutrition [12].

It is difficult to calculate the exact societal economic impact of IPV but the costs are high. In the United States annual costs of intimate partner rape, physical assault, and stalking exceed $5.8 billion, nearly $4.1 billion of which is for direct medical and mental health care services [13]. In the United Kingdom the annual cost to the national health service of physical assaults is £1.2 billion [14].

The Department of Health in England now recommends that "All trusts should be working towards routine enquiry" [15]. In the US, the Family Violence Prevention fund consensus guidelines recommend that all adolescent and adult patients should be routinely asked about domestic violence [16]. Although there is ongoing debate about the evidence for screening or routine enquiry [17], there is unquestionably a need for clinicians to ask about domestic violence more often than they currently do.

A study of women attending general practices in east London found that only 17% of women experiencing IPV reported that their doctor had asked them about domestic violence [4]. We know that women who are experiencing violence want to disclose this to trusted doctors and get support [18], but that a high proportion of women who are experiencing abuse do not disclose this spontaneously in clinical consultations [4].

Short questions that reliably identify women experiencing IPV who present in clinical settings are a pre-requisite for developing an appropriate response from health services to this substantial public health problem [19].

Many primary health care professionals, including general practitioners (GPs) and practice nurses, occasionally enquire about domestic violence. It has not been adequately determined whether their questions identify women experiencing IPV.

We conducted a cross-sectional survey of women in GP waiting rooms. The fifty-one general practices in Newham, a multi-ethnic inner city area of London, were stratified according to the number of doctors and the proportion of south Asian names on the practice register [35]. Equal numbers of practices were selected from each stratification group using a randomisation programme (SPSS version X). This was in an attempt to ensure that the practice population reflected the local area population.

Each practice was sent a recruitment letter with information about the study. If practices expressed an interest, a research team member met with the primary care team to answer any questions. We excluded practices that did not have a private room available, as then privacy for the survey could not be provided. If a practice decided not to take part or was excluded, the reason for this was documented and another practice was randomly selected from within the same stratification group.

We approached consecutive women in practice reception areas waiting to see a doctor or nurse. We included women aged more than 17 years who in the last year had been in an intimate relationship. We excluded women who were accompanied by children over four years of age or by another adult, were too unwell to complete the questionnaires, unable to understand English or unable to give informed consent. In the waiting room, women were asked to participate in a study designed to improve women's health care. We sought consent for the administration of the HARK and CAS questionnaires in a private room. All participants were given information on local domestic violence services. The East London and City ethics committee approved the study.

The number of potentially eligible subjects was recorded by the researcher in the waiting room. A record was made of the number of women who were excluded due to the exclusion criteria, those who the researchers were unable to approach at very busy times, women who were approached and agreed that they would be seen by the clinician first and then undertake the study but were not seen again ("did not come back"), those who refused participation in the waiting room and those who declined consent in the private room.

The HARK and CAS were self-administered. We expected to be able to recruit approximately 500 women. On the assumption that the prevalence of IPV in the past year was 20%, we calculated that there was a 90% chance of estimating sensitivity at 76% or above with this sample size.

We approached 24 practices and 12 agreed to participate; 11 declined and one was excluded as it had no private room. Two hundred and thirty two women were recruited from May to October 2003. Figure 1 shows recruitment of individual participants to the study. Seven hundred and thirty seven women did not meet the inclusion criteria. Fourteen women were not approached because there were too many women in the waiting room for all to be approached. Two hundred and three women "did not come back." One hundred and eighty six women declined participation in the waiting room. Eleven women declined consent in the private room. The response rate of 54% (232/(232 + 186 + 11) was adjusted for the women who "did not come back."

The average age of participants was 35 years (range 18–70 years). 51% were in a paid job and 53% owned a house or flat. 40% of participants described their ethnic origin as white British, 25% as black British, African or Caribbean and 18% as Indian, Pakistani or Bangladeshi.

The four HARK questions accurately identify women experiencing IPV in the past year and may help women disclose IPV in general practice. The estimated specificity (95%, 95% C.I. 91% to 98%) of the HARK score of ≥1 was higher than the sensitivity (81%, 95% C.I. 69% to 90%). The PPV (post-test probabilities) of HARK, which increase as the HARK score increases, also provide evidence that HARK is an effective short tool for identifying IPV.

The most straightforward way of using the HARK is as a simple test with a cut off of ≥1. Therefore if a clinician asks these four questions and their patient scores ≥1, this will identify 81% of women affected by IPV (as judged by the CAS). This is assuming that the tool performs in the same way that it did when a researcher administered it. There is an 83% probability that a woman with this score has experienced IPV in the past year (positive predictive value); and she is 16 times more likely to have been affected by IPV in the last year than some-one with a HARK score of 0 (likelihood ratio of a positive result).

The multilevel LRs and corresponding post-test odds make more use of the data from the test as it avoids dichotomising the HARK score into IPV present or not present [39]. When a woman is asked the four HARK questions she does not actually have a positive or negative score for IPV; instead she may score 0, 1, 2, 3 or 4 and each score has a different meaning (i.e. different likelihood ratio and post-test odds for IPV – see table 4). When an individual answers "no" to all of the HARK questions the likelihood ratio and post test odds (0.2 and 0.1 respectively) suggest that IPV is probably not present; whereas answering "yes" to three or four HARK questions produces a specificity of 100%, meaning that IPV is present. Answering "yes" to one or two of the HARK questions is less specific.

The majority of women who are experiencing IPV do not spontaneously disclose to clinicians. HARK can potentially accurately and quickly identify a high proportion of these women. This is a pre-requisite for effective intervention allowing the successful management of IPV in general practice. It has been shown that women want to disclose IPV to health care professionals, particularly primary care clinicians [18].

The high pre-test probability (prevalence) of IPV (23%) is consistent with other prevalence studies in primary health care settings [3].

To increase the external validity of the study, we recruited a wide range of practices, including small single handed ones with less than 3,000 patients which are common in inner city areas in the United Kingdom. However small practices had fewer patients in the waiting room available for recruitment than had been anticipated; with the recruitment of participants taking longer than planned. Consequently we were only able to recruit 46% of our target sample size within the timeframe of the study, resulting in less precise estimates of test accuracy, reflected in wider confidence intervals. Nevertheless our study is larger than some other validation studies of short instruments and our estimates of test characteristics are relatively precise.

Eighty two percent of women who did not fulfil the inclusion criteria were accompanied. The ethics committee that approved our proposal specified that potential participants should only be approached if they were unaccompanied in order to decrease the likelihood of an abusive partner discovering that the participant had completed a questionnaire on domestic violence. We did include women who were accompanied by children under the age of five years, as it was felt that a child this young was unlikely to jeopardise a participant's safety.

Overall women were enthusiastic about participation once they found out that the study was about domestic violence: only eleven women declined consent in the private room. One hundred and eighty six women declined participation in the waiting room but these women did not know that the study was specifically about domestic violence.

The National Census 2001 figures allowed us to compare our study population to the local population in the borough of Newham. The average age of the study population was 3 years older than the average age in the local population (32 years). The percentage of the study population in a paid job was 12% higher and the percentage that owned a house or flat was 9% higher than that in the local population (39% and 44% respectively). The percentage of the study population that described their ethnic origin as white British was 6% higher than that in the local population (34%) whilst the percentage that described their ethnic origin as Indian, Pakistani or Bangladeshi was 11% lower than that in the local population (29%). This comparison shows that despite our attempts, the study population were not totally representative of the local population. We believe that the higher socio-economic status of our study sample (as reflected by the higher percentage in a paid job and owning a house or flat) compared to the local population may reflect a response bias meaning that perhaps those women with lower socio-economic status and at greater risk of IPV were less likely to have taken part in this study. This may have affected the calculation of the prevalence, PPV and NPV of HARK. However there is no reason why this would necessarily affect the sensitivity/specificity calculations unless the 46% of women who did not take part in the study answered differently with regards to only one of the instruments (the HARK or the CAS). This is unlikely.

The strengths of this study are that it tested a short tool that can be used in routine general practice, against an abuse measure validated in primary care. Additionally, HARK's external validity has been increased by being conducted in a range of practices with a study population of varied ethnicity.

Limitations included the response rate of 54%, decreasing the external validity of the study. Although we consider the CAS to be the best research measure for IPV in a health care setting, we cannot exclude the possibility that not all women who were found to be positive for IPV with the HARK but negative with the CAS were false positives. Other investigators have found that when using two sets of validated questions each may identify some women as abused that would have been missed by the other tool [40].

The HARK questions could be incorporated into electronic medical records in primary care to prompt clinicians to ask about recent intimate partner violence and to encourage disclosure by patients. Future research should test the effectiveness of HARK in clinical consultations as part of system level interventions to improve the response of primary care to IPV.

Intimate partner violence against women is common and causes short and long-term ill health. Previously questions about intimate partner violence to elicit disclosure have been insufficiently validated for use in general practice or family medicine populations, particularly outside the US. The four short HARK questions accurately identify women experiencing intimate partner violence in the past year.