Approaches to Screening for Intimate Partner Violence in Health Care Settings:  A Randomized Trial FREE

As intimate partner violence (IPV) has gained recognition as a major public health problem,^1- 2 research efforts have focused on the development of universal screening instruments and protocols for use in health care settings to identify women exposed to IPV.³ Many national medical organizations, governmental agencies, and advocacy groups have recommended universal or routine IPV screening,⁴ although there is a lack of research examining its effectiveness on health outcomes for women.^5- 7 An ongoing question in the field is whether health care professionals should routinely screen their female patients for exposure to IPV.⁸

Previous studies have demonstrated that women will disclose experiences of violence in response to screening^3- 7; however, few studies have compared methods of administration. In a review of IPV screening in the primary care setting, Chuang and Liebschutz⁹ identified 2 main approaches to screening: (1) verbal methods (questions asked by a clinician) and (2) self-administered methods, including written or computer-based questionnaires. The limited amount of research to date has generally compared IPV prevalence on face-to-face questioning with a written self-completed questionnaire. Anderst and colleagues¹⁰ compared verbal screening with a written self-administered questionnaire among women accompanying children to pediatric visits. Each woman received either verbal or written questions. A much smaller proportion of women responded to the written screening questions, and of those who did, none reported IPV exposure, leading the investigators to conclude that verbal screening was superior. However, the screens did not include the same questions, and the sampling methods varied between groups. A study by McFarlane et al¹¹ compared written self-report with a nurse interview questionnaire using the same 4 IPV questions in a sample of women attending a Planned Parenthood clinic. Higher prevalence estimates of abuse were recorded during nurse interviews. The methods were not randomly assigned, and it is not clear that both groups of women underwent the same study procedures. In contrast, a study of new obstetric patients showed higher disclosures with a written questionnaire compared with verbal questioning. However, this was based on a review of medical records comparing responses from patients who received both a self-report written instrument that included questions about exposure to IPV and a set of different verbal questions on the topic asked by a midwife.¹² In the largest of the studies comparing screening methods, more than 4600 women presenting in 11 emergency departments in Pennsylvania and California completed an identical screening questionnaire either in a written self-administered format or administered by a nurse. Those using the written approach were significantly more likely to disclose physical or sexual abuse in the past year.¹³

None of the studies comparing verbal questioning with written questionnaires involved random assignment of participants to receive alternate methods of IPV screening. Furthermore, IPV prevalence appeared to be the single criterion of effectiveness in each of these studies, even though patient acceptability and other feasibility issues are also important.

Rhodes and colleagues^14- 15 evaluated the feasibility and effect on IPV disclosure of using computers in an emergency department. They concluded that use of the computer led to more IPV disclosures and patient-practitioner discussion of IPV compared with a usual care group that did not specifically receive screening. These studies do not provide any information about the appropriateness of computer-based screening compared with verbal or written screening, but the findings suggest that this approach warrants further study.

The current study was undertaken to contrast 2 screening instruments with the goal of determining an optimal method (computer, written, face-to-face) of screening for IPV in health care settings, based on 3 criteria: (1) 12-month prevalence, (2) extent of missing data, and (3) participant preference.

The study was conducted in Ontario, Canada, from May 2004 to January 2005. Participants were recruited from primary, acute, and specialty health care settings: 2 family practices, 2 emergency departments, and 2 women's health clinics. All women who presented for an appointment at a participating site were approached. Women were eligible for participation if they were: (1) 18 to 64 years old, (2) at the site for their own health care visit, (3) able to separate themselves from individuals who accompanied them, (4) able to speak and read English, (5) not too ill to participate, and (6) able to provide informed consent. The trial flow diagram is shown in the Figure.

All participants provided written informed consent prior to enrollment. For safety reasons, no reference to “abuse” or “violence” was made until women were taken to a private room where the informed consent process occurred. Women were told that their health care provider (a physician or nurse) would not be informed of their responses to the screens, but was available to discuss any concerns they might have. In the face-to-face arm of the trial, however, health care providers would necessarily be aware of women's responses. All participants were provided with information about resources in the community and the option to shred any study material they felt might put them at risk. Health care providers received specialized training in responding to IPV.

The study was approved by the research ethics boards of McMaster University/Hamilton Health Sciences, The University of Western Ontario/London Health Sciences Centre, Cambridge Memorial Hospital, and Norfolk General Hospital's Medical Advisory Committee.

Sample size was calculated based on the null hypothesis of no differences in 12-month IPV prevalence across methods, with α set at .05 (2-tailed test) and power set at 1 − β=80%. Based on the literature,^16- 17 we expected an overall prevalence of IPV of 15%. It was hypothesized that prevalence across methods of administration would be: face-to-face, 10%; written, 15%; and computer, 20%; this required a sample size of 246 per group per care type.

Randomization was by day (or shift for sites with regular hours longer than 8 hours) in 6-week (for sites with no shifts or 2 shifts) or 9-week (for sites with 3 shifts) periods. A table for each day of the week was created, and a random number table was used to assign clinic shifts to 1 of 3 methods. For example, the table for a Sunday in an emergency department with two 12-hour shifts would have 3 columns (computerized, written, and face-to-face) and 2 rows (day, night). The random number table determined the order of the numbers 1 through 6 in the cells. So, for example, the Sunday of week 1 was allocated to written, night shift; the Sunday of week 2 was allocated to computerized, day shift, and so on, for the 6-week period. This ensured balance across shifts and days of the week. The research coordinator created calendars that informed site coordinators of the assignments. The order in which the screening instruments were completed was also randomly varied.

In addition to standard demographic questions, participants completed the Partner Violence Screen (PVS)¹⁸ and the Woman Abuse Screening Tool (WAST).¹⁹ These 2 measures were selected following a systematic review of screening instruments based on their psychometric properties and use in settings comparable with those in this study.³ The PVS (3 items) addresses physical abuse and feelings of safety; when compared with the Conflict Tactics Scales (CTS), the sensitivity was 71.4% and the specificity was 84.4%.¹⁸ The WAST (8 items) includes multiple forms of abuse (physical, sexual, and emotional) and has good internal consistency (Cronbach α coefficient of 0.75), and more than 90% of women reported being “comfortable” or “very comfortable” when administered the WAST in a previous study.¹⁹ Both screening instruments ask about experiences within the last 12 months. The instruments and information about scoring are shown in the Box.

Partner Violence Screen (PVS)¹⁸†

Woman Abuse Screening Tool (WAST)¹⁹‡

*Both the PVS and the WAST had preambles that indicated the questions applied to the last 12 months.

†Answering yes to question 1 (and indicating it was by a person in a current or previous relationship) or question 3, or no to question 2 met the criteria for intimate partner violence exposure.

‡Endorsing either question 1 (“a lot of tension”) or question 2 (“great difficulty”) met the criteria for intimate partner violence exposure; questions 3 to 8 were not used in this determination.

The Composite Abuse Scale (CAS),^20- 21 a 30-item validated research instrument, was selected as the criterion standard for its comprehensiveness and strong psychometric properties: the Cronbach α for each of 4 subscales is greater than 0.85, and they correlate highly with corresponding subscales of the CTS. The CAS was administered to determine the agreement of the WAST and PVS with this instrument. It was scored as recommended by summing the frequency scores for the 30 items; a score of 7 or more was the criterion for exposure to IPV.

To evaluate participant preference of screening approach, women were asked 3 questions about their method of screening: (1) Was it “easy”?; (2) Did you “like answering” [in that way]?; and (3) Was it “private enough”? The responses were scaled 1 (“not at all”) to 5 (“very easy” or “a lot”).

After obtaining consent, the on-site study recruiter provided participants with 1 of 3 methods, according to the randomization schedule.

Computer-Based Self-Completed Method. The participant was given a tablet computer and asked to complete the screening instruments (PVS and WAST, randomly ordered), followed by the evaluation questions. If a participant did not respond to a question, a reminder window appeared; she could then answer the question or continue without answering. Once done, the participant exited the questionnaire program and returned the tablet to the study recruiter. She then completed the demographic questions and CAS on paper.

Written Self-Completed Method. The participant was given a paper version of the demographic questionnaire and screening instruments (PVS and WAST, randomly ordered), followed by the evaluation questions; the questionnaire closed with the CAS. Completed questionnaires were returned to the recruiter in a sealed envelope.

Face-to-Face Method With Verbal Questioning by the Health Care Provider. After obtaining consent, the recruiter informed the health care provider of the patient's participation by inserting a pink slip of paper into the patient's chart. Participants were verbally screened by their health care provider with one of the 2 screening instruments, randomly determined. Due to the nature of the screening method, disclosures became part of the clinical encounter; women who disclosed abuse were offered the usual care provided by that site. Following the screen, the participant completed a written version of the demographic questions, the evaluation items, and the CAS.

Descriptive statistics were run for sample characteristics by group. Because the participant evaluation items showed a skewed distribution, items were dichotomized as less than 5 or 5.

Data were considered missing if abuse status could not be determined. For the abuse instruments, if a participant provided sufficient data to score positive, she was deemed “positive,” regardless of the number of missing items. If a participant scored negative on all completed items, with 1 or more incomplete items, her status was “missing.”

Multilevel logistic regression and the statistical software MLwiN²² were used to model disclosure, missing data, and participant evaluations of the screening methods. Evidence of clustering indicated that the analysis of prevalence and participant evaluations required the use of a 3-level model (i binary responses, nested within j women, nested within k time blocks), while the analysis of missing data required a 2-level model (no clustering between time blocks). The large-sample χ² test statistic was used to assess the statistical significance of model parameters.

Table 1 shows the characteristics of the sample by group. Almost 56% of the women were married, and just fewer than half had 1 or 2 children living at home. More than 50% were well-educated; just fewer than half were working outside the home.

Twelve-month prevalence ranged from 4.1% to 17.7% depending on method, instrument, and setting (Table 2). Prevalence was significantly lower in family practices (χ²₁ = 18.5; P<.001) and women's health clinics (χ²₁ = 29.4; P<.001) vs emergency departments. There was also a statistically significant interaction between method and instrument: for the written method, prevalence was lower on the WAST than on the PVS (χ²₁ = 5.5; P=.02).

Proportions of missing data differed by instrument and method (Table 2). Lower levels of missing data occurred for the WAST vs PVS (χ²₁ = 9.2; P=.002) and for the written method vs the face-to-face and computer-based methods combined (χ²₁ = 11.9; P<.001). In the analysis of missing data, there was no variability associated with time blocks, no statistically significant interactions between instrument and method, and no statistically significant differences between settings in levels of missing data.

On all 3 evaluation indicators, women chose computerized and written methods over face-to-face questioning (ease: computerized [χ²₁ = 21.5; P<.001] and written [χ²₁ = 92.1; P<.001]; preference: computerized [χ²₁ = 121.1; P<.001] and written [χ²₁ = 107.0; P<.001]; and privacy: computerized [χ²₁ = 36.7; P<.001] and written [χ²₁ = 46.4; P<.001]). There were no statistically significant interactions between instrument and method or between setting and method.

The estimated test characteristics of the PVS and WAST screens were compared with the CAS. The sensitivities (PVS, 49.2%; WAST, 47.0%) as well as the specificities (PVS, 93.7%; WAST, 95.6%) were very similar. The positive predictive value of the WAST (55.3%) was minimally higher than for the PVS (47.0%), and the negative predictive values were almost the same (PVS, 94.2%; WAST, 94.0%), leading to very similar accuracies (PVS, 89.2%; WAST, 90.6%).

This randomized trial compared 3 methods of IPV screening using 2 instruments on IPV detection, extent of missing data, and acceptability of screening approach, yielding some interesting findings. Although some literature suggests that use of computer-based questionnaires may lead to higher disclosures of sensitive issues than other approaches,^24- 25 we did not find that computer-based screening increased the detection of IPV relative to other screening methods.

We found that there was an interaction between method and instrument, with the written WAST having the lowest prevalence. We were surprised by this finding, although the evidence regarding verbal vs written disclosure is mixed.^10- 13 It is noteworthy that lower written disclosure was specific to the WAST.

Use of written questionnaires led to significantly fewer missing data, in contrast with the findings of Anderst et al.¹⁰

On all 3 measures of acceptability (ease of responding, likeability, and privacy), the face-to-face method was least preferred by participants. These findings have some recent support in the literature. A study comparing audiotaped screening with written screening in a pediatric emergency department²⁶ found no statistically significant difference in IPV disclosures between the 2 methods but several patterns in women's preferences. Specifically, women found the audiotaped method to be less risky and more private than the written approach, and among both the entire sample and the subgroup of women disclosing abuse, the written and audiotaped methods were significantly preferred to the idea of disclosing IPV directly to a health care provider. Coupled with our findings, and those of Glass et al,¹³ there seems to be emerging evidence that direct questioning by clinicians is less favored by women compared with self-report versions, whether delivered by computer, audiotape, or written questionnaire.

The estimated sensitivities and specificities of both instruments in relation to the CAS were remarkably similar; the low sensitivity means that a sizeable proportion of women who disclosed exposure to IPV on the CAS were not identified on either the WAST or the PVS. This is likely because the CAS includes many more questions covering a broad range of abusive behaviors in several domains, including harassment.

This study has limitations that need to be considered in interpreting the results. First, review of the sample characteristics by method shows that the women completing the computer-based screen had a higher proportion of participants in the lowest income quintile compared with those administered the other 2 approaches. Women of lower socioeconomic status might respond differently to a computer-based approach; however, there was no difference in other related variables, including education and work status, reducing the likelihood of a bias regarding acceptability of the computer screen.

Second, although we attempted to keep all other aspects of the protocol consistent across methods, the responses of women who underwent the face-to-face approach were known to the health care providers asking them. This aspect could have influenced women's willingness to disclose, although interestingly, there were no consistent patterns in disclosure by method. In addition, they were asked questions from only 1 instrument, either the WAST or the PVS, to reduce burden on both the clinician in having to administer more than 1 questionnaire verbally and the participant in having to verbally respond to 2 instruments with similar questions.

Last, although the CAS is a useful standard for comparison, it is not free of error: this error, in conjunction with the error associated with the screening tests, will serve to attenuate the estimation of sensitivities and specificities. The CAS was chosen because it was the most comprehensive measure of the IPV experience.

In summary, the findings from this study examining 3 approaches to IPV screening in health care settings suggest that the face-to-face approach is the least preferred by women, irrespective of instrument. With regard to selection of method based on prevalence, however, there was an interaction between method and instrument: it appears that the written format of the WAST may lead to some underestimation of disclosure. In theory, sensitivity of the WAST could be improved by changing the scoring criteria to include more items (see Box). The result of least missing data by written self-completed questionnaire is worth noting, especially for research applications, but also when considering clinical policies for IPV detection and intervention. Prevalence, missing data, and preference are all important considerations for both clinical and research efforts in IPV screening.