Study flow diagram. Psych indicates psychological reasons; ETOH, alcohol use.
Rhodes KV, Drum M, Anliker E, Frankel RM, Howes DS, Levinson W. Lowering the Threshold for Discussions of Domestic ViolenceA Randomized Controlled Trial of Computer Screening. Arch Intern Med. 2006;166(10):1107-1114. doi:10.1001/archinte.166.10.1107
Women experiencing domestic violence (DV) frequent health care settings, but DV is rarely identified.
We conducted a randomized controlled trial to determine the effect of computer screening on health care provider–patient DV communication at 2 socioeconomically diverse emergency departments (EDs). Consenting nonemergent female patients, aged 18 to 65 years, were randomized to self-administered computer-based health risk assessment, with a prompt for the health care provider, or to “usual care”; all visits were audiotaped. Outcome measures were rates of DV discussion, disclosure, and services.
Of 2169 eligible patients, 1281 (59%) consented; 871 (68%) were successfully audiotaped, and 903 (71%) completed an exit questionnaire. Rates of current DV risk on exit questionnaire were 26% in the urban ED and 21% in the suburban ED. In the urban ED, the computer prompt increased rates of DV discussion (147/262 [56%] vs 123/275 [45%]; P = .004), disclosure (37/262 [14%] vs 23/275 [8%]; P = .07), and services provided (21 [8%] vs 10 [4%]; P = .04). Women at the suburban site and those with private insurance or higher education were much less likely to be asked about experiences with abuse. Only 48% of encounters with a health care provider prompt regarding potential DV risk led to discussions. Both inquiries about and disclosures of abuse were associated with higher patient satisfaction with care.
Computer screening for DV increased but did not guarantee that DV would be addressed during ED encounters. Nonetheless, it is likely that low-cost interventions that allow patients the opportunity to self-disclose can be used to improve detection of DV.
Domestic violence (DV) is a major cause of morbidity and mortality for women.1- 4 While evidence-based reviews do not support routine screening,5- 7 there is qualitative evidence from abuse survivors that identification by health care providers is both desirable and helpful.8- 11 Indeed, failure to recognize DV may result in missed diagnoses, unnecessary testing,12,13 and high rates of health care utilization.14- 16 However, these potential opportunities to intervene are frequently missed owing to health care provider reluctance to initiate discussions about sensitive issues.17- 19 Likewise, abused women are unlikely to volunteer the information unless specifically asked.20,21 The end result is silence between patients and health care providers.
Victims of DV often seek care in emergency departments (EDs),22- 24 but this setting is particularly challenging due to time pressures and the urgency of unscheduled care.25,26 Trained and dedicated staff can increase identification and referrals for DV, but these effects have not been sustainable.27,28 One well-designed system-change model improved health care provider attitude, knowledge, beliefs, and screening behavior without increasing the number of identified DV cases.29 Computer-based health risk assessment has the potential for patient self-disclosure, allowing the health care provider to focus on assessment and referral. Patients are more likely to disclose sensitive and illegal behaviors on computer-based surveys than on paper questionnaires or personal interviews.30- 34 Ideally, computer screening could take advantage of patients' waiting time to provide valuable information about a variety of health risks. The goals of this study were to assess rates of computer-based DV disclosure at 2 socioeconomically diverse sites and to determine the effect of computer screening on DV communication between patients and health care providers during the course of the ED visit.
Between June 2001 and December 2002, we conducted a randomized controlled trial of a self-administered computer-based health risk assessment tool (Promote Health Survey),35 which generates health recommendations for patients and alerts physicians to a variety of potential health risks, including DV. Our primary outcomes, assessed by audiotape analysis, were rates of discussion of DV, patient disclosure of DV to the health care provider, and evidence of DV services provided during the visit, defined as a safety assessment, counseling by the health care provider or social worker, or referrals to DV resources. Secondary outcomes were medical chart documentation of DV screening (positive or negative), DV “case finding” (chart documentation of current or past DV), and overall patient satisfaction assessed by exit questionnaire.
The study was conducted at 2 socioeconomically diverse EDs: an urban academic medical center that serves a predominately publicly insured inner-city African American population, and a suburban community hospital serving a privately-insured suburban white population. Both EDs are part of the same emergency medicine residency training program and participate in joint educational conferences, including annual training about DV. Health care providers were informed that the study objective was to study the effect of a computer prompt on DV communication and were encouraged to screen all women for abuse. Participating health care providers and patients signed written consent, and the institutional review boards of both hospitals approved the protocol.
Women aged 18 to 65 years and triaged as medically nonemergent were recruited sequentially from the ED waiting rooms for a study of physician-patient communication. Consenting patients were randomly assigned in a 1:1 ratio to usual care or to the DV Promote Health Survey. Treatment assignment was ascertained by the research assistant by opening sealed randomization envelopes in sequential order; the envelopes were prepared from a randomization list generated by computer in blocks of size 10 to ensure balance between groups over short time spans, such as shifts and days of the week, as well as over the entire course of the study. The Promote Health Survey solicited information about lifestyle and health risks and generated a 1-page summary of patient risks, which was stapled to the chart during the ED visit. All patients consented to audiotaping of their entire ED visit and to an exit questionnaire before leaving the ED. Digital audio recorders were attached to intravenous drip poles that traveled with the patient. Audiotapes were subsequently edited to remove all names and non–health care provider–patient conversations; a 7-hour ED visit might be edited to 20 minutes of actual health care provider–patient interaction.
The development and testing of the computer-based Promote Health Survey, which served as our intervention, has been described previously.35 Patients completed the survey on a touch-screen computer in a private room. If the patient answered “Yes” to any of the DV risk questions listed in Table 1, “Physician to Assess Risk” box at the top of the computer-generated report said “Possible partner violence: Assess for current abuse” and suggested referral options. The DV screen (also completed by both groups on the exit questionnaire) was derived from the previously validated questionnaires, the Abuse Assessment Screen36,37 and Partner Violence Screen,38 and modified after cognitive interviewing and field testing. A woman was considered to be at potential risk for current DV if the abuse happened within the last year or involved a current partner. Because there is no gold standard for DV, health care providers were told that positive responses were risk factors for DV and should prompt an assessment.
A structured audiotape coding scheme was developed by 4 of us by independently listening to 30 tapes, comparing results, and reaching consensus in an iterative fashion, a well-accepted qualitative research technique.39 Research assistants, blinded to group assignment, were trained using a coding manual, sample cases, and self-testing until 100% agreement was reached with the investigators. To monitor coding quality, approximately 10% of all recordings were double coded. Agreement was 87% for mention of DV and 95% for patient disclosure of DV.
Charts of all enrolled patients were coded using a structured chart abstraction form to assess evidence of DV documentation. To monitor chart abstraction quality, 10% of all charts were doubly abstracted. Agreement was 85% for DV screening (“positive,” “negative,” or “not screened”). Case finding was defined as chart documentation of a positive screen finding.
The primary analysis evaluated the effect of the randomized intervention, the Promote Health Survey, within each site, on audio-coded rates of DV discussion, disclosure, and services, using mixed logistic regression models40,41 with intervention as the fixed effect of interest and primary provider as a random effect, entered to adjust standard errors of the intervention effect for correlation due to clustering of patients within providers. If this correlation were 0, the mixed logistic model would be equivalent to a univariate logistic model or a χ2 test. A subgroup analysis of the intervention effect was conducted on patients reporting DV risk on the exit questionnaire. All tests of intervention effect were conducted as intent-to-treat analyses. Multivariate logistic regression models were used to evaluate simultaneous effects of the Promote Health Survey intervention, ED site, and patient characteristics on the DV audio outcomes and on chart documentation of DV screening. Two-sample t tests and χ2 tests were used to compare patient characteristics and rates of patient disclo sure of DV on exit questionnaire between sites, as well as associations between patient satisfaction (dichotomized as “very satisfied” vs lower ratings) and DV discussion and disclosure. All analyses were conducted using STATA SE, version 8.2 (StataCorp, College Station, Tex).
We enrolled 1281 (59%) of 2165 eligible female patients who presented during data collection periods; 871 (68%) of these were successfully audiotaped; 903 (70%) completed an exit questionnaire, and 1178 charts were reviewed (92%). The Figure is a flowchart of all potentially eligible and approached patients and reasons for refusal or noncompletion. Generation of the health care provider prompt required completion of the computer survey, but 115 (18%) of the patients randomized to the computer program had the process interrupted, most frequently by being called for treatment. As a result, the full realization of our intervention was only 82%. For our primary outcomes, the analysis was based on the 871 complete audible audiotapes from 331 suburban and 540 urban encounters.
Table 2 lists the patient characteristics for the urban and suburban hospital EDs. Since there were no significant demographic differences by group assignment, we focus on differences by site. As expected, women patients at the urban ED were younger (mean age, 32 years), predominantly African American (86%), unmarried (64%), and more likely to be living in poverty. Women at the suburban site had a mean age of 36 years, were predominantly white (80%), more likely to be married (43%), and to have higher levels of income, education, and private insurance.
Almost all eligible health care providers (101/104) gave consent to participate in the study; 80 of the consenting health care providers (30/39 attending physicians, 46/58 residents, and 4/4 nurse practitioners) had audiotapes used in the analysis. Participating health care providers were 74% white, 14% African American, and 12% Asian or other. The average ages of attending physicians, nurse practitioners, and residents were 44, 45, and 29 years, respectively.
Table 1 reports frequencies of positive responses to DV items on the exit questionnaire by site. Within sites, the intervention and control groups did not differ significantly in rates of current or lifetime DV disclosure. Of the 903 women completing the exit questionnaire, 151 (26%) of 578 at the urban ED and 67 (21%) of 325 at the suburban ED disclosed risk of current DV (current threat, control, isolation, insulted or put down, afraid to disagree, physically hurt, or forced into sex by partner). Lifetime prevalence of DV exposure was higher among women at the urban site (39% vs 31%). If a woman disclosed DV to the computer, 90% of the time she also disclosed DV on the exit questionnaire.
Our primary results, summarized in Table 3, show the effect of the computer intervention on rates of DV discussion, patient disclosure of DV risk to the health care provider, and receipt of DV services by site: overall and for the subset who disclosed DV on the exit questionnaire. At the urban ED, rates of DV discussion were higher in the intervention group (147/262 [56%]) than in the usual care group (123/278 [45%]) (P = .004); rates of patient disclosure of DV to the health care provider also tended to be higher in the intervention group (37/262 [14%] vs 23/278 [8%]) (P = .07). The number of patients who received DV services was small but significantly higher in the intervention group: 57% (n = 21) of the 37 patients who disclosed DV in the intervention group compared with only 43% (n = 10) of the 23 patients in the usual care group were given any counseling or verbal referral by the provider. At the suburban ED, the rate of DV discussion and thus disclosure was uniformly low in both groups (DV discussion, intervention vs control, 11% vs 9%; DV disclosure, 5% in both groups).
Overall, only 48% of audiotaped encounters during which patients disclosed DV risk to the computer led to a DV discussion. Only 17% of suburban compared with 61% of urban women disclosing DV on the computer had any mention of the topic during their visit. Among patients disclosing DV on exit questionnaire, those in the intervention group were both significantly more likely to have DV discussions and twice as likely to disclose DV to the health care provider. At the urban ED, the presence of a DV discussion was associated with patients being more likely to say they were “very satisfied” with their ED visit: 62% of patients who had any discussion of DV reported very high satisfaction compared with 50% of those for whom the topic did not arise (P = .01). This association was not observed at the suburban ED.
Chart documentation of DV screening and documentation of risk factors varied considerably between sites, but the effect of the intervention on these outcomes was not significant at either site.
The effects of patient characteristics, site, and the Promote Health Survey intervention on audio-coded DV discussion, DV disclosure to the health care provider, DV services, and chart documentation of DV screening are summarized in Table 4. In multivariate analysis, effects of the Promote Health Survey intervention are unmodified by patient characteristics and remain essentially the same as the unadjusted effects in Table 3. Rates of all outcomes were substantially and significantly higher at the urban ED. At both sites, higher rates of DV discussion were associated with less education, younger age, and presentation with gynecologic and/or urinary tract complaints. Rates of DV disclosure were lower at both sites when health care providers were caring for women with private insurance. Higher rates of disclosure were associated with younger age at the suburban site and lower education at the urban site. At the urban site, patients without insurance were more likely to receive DV services. Chart documentation of DV screening was strongly associated with being at the urban site, younger age, public or no insurance, and presenting with gynecologic and/or urinary complaints. Multivariate models for DV discussion and disclosure retained most factors that were significant in the univariate analysis. By contrast, age was the only significant factor in the multivariate model for chart documentation of DV screening.
We found that women at both urban and suburban sites were willing to disclose DV risks to a computer. However, computer screening increased rates of health care provider–patient DV communication at the urban ED only. This was accompanied by increases in disclosureto the health care provider and counseling and/or referrals, thus underscoring the potential of computer screening to have a positive effect on outcomes of care for abused women. To our knowledge, this is the first study to use audiotaping, as opposed to retrospective reports or chart documentation, to assess the rates and content of health care provider–patient DV interactions.
This study adds to the literature about interventions to increase DV screening.42- 44 While system changes such as chart prompts, staff training, incentives, and availability of advocacy services have been associated with increased DV documentation of screening,25,27,29 the rates have not been sustained over time, nor have they necessarily resulted in more DV case identification or DV services.29,45 In our study, women disclosed DV risk to a computer at rates comparable to those found in studies using dedicated screeners,22,46,47 which supports the notion that computer health risk assessment has the potential to be a useful tool for achieving routine screening for DV.
Computer screening resulted in an overall 75% increase in the odds of DV discussion. This effect size is comparable in magnitude to that of other successful public health interventions, such as those prompting health care provider interventions for depression, alcohol abuse, and smoking.48- 50 It is important to note that even small effect sizes can have large impacts if the prevalence rate of a condition is high in the population being served, as is the case with DV.51 Nonetheless, DV was still grossly underaddressed in the demanding pace of the ED visit; in fewer than half of encounters in which there was a prompt to the health care provider of potential DV risk was there any mention of the topic.
While our study design does not allow us to identify the specific reasons why health care providers, particularly suburban health care providers, failed to address DV when prompted to do so, we can speculate that issues like patient profiling along with competing demands, limited time, personal discomfort, lack of privacy, and the complexity of dealing with a multifactorial problem like DV might all contribute to failing to implement computer-prompted recommendations for DV assessment. More research focused on understanding and reducing the gap between information and health care provider action will be necessary to address the patient-identified need for DV risk assessment and counseling.
Women at the suburban site disclosed only slightly less current DV risk on exit questionnaire than at the urban site, 21% vs 26%. However, ED health care providers were 70% less likely to ask women at the suburban site about experiences with abuse than their urban counterparts, even controlling for all observable patient characteristics. While we cannot separate out effects of site, race, and socioeconomic status because they were all highly correlated, this result still raises the possibility of a health care provider bias against addressing DV with more affluent white women. Concern for health care provider bias has been raised in both the DV and ED literature.17,52 While biases may result from clinical experience, screening biases can become self-fulfilling.53 Lane et al54 found that minority children with accidental injuries were 3 times more likely to be evaluated for suspected abuse, resulting in higher rates of diagnosed fractures due to abuse. Sugg and Inui17 found that physicians who came from white middle-class backgrounds assumed that patients with similar sociodemographic characteristics as themselves were unlikely to be at risk for DV. Our study was not designed to evaluate health care provider bias, but our results are consistent with a different approach to urban vs suburban women. The high prevalence of DV risk at both sites raises a general concern since the rates of DV discussion for all women were grossly inadequate.
It is of interest that patients with any DV inquiry or discussion had higher rates of satisfaction with their ED visit. Prior ED studies have found lower satisfaction with care among abused women.55,56 The US Preventive Service Task Force raised the question of whether health care provider screening for DV could cause potential harm to the health care provider–patient relationship.57 Our study would suggest that patient satisfaction might be improved by DV screening, even in the absence of DV disclosure to the health care provider.
This study has a number of important limitations. Audiotaping had the advantage of providing direct evidence, as opposed to indirect evidence available from chart review, but it may have created a participation bias; patients who were uncomfortable discussing sensitive issues might have declined to have their visit audiotaped. Audiotaping also may have caused a Hawthorne effect owing to the nonblinding of the intervention and, along with the potential for treatment diffusion, may have decreased our ability to detect differences between groups. Domestic violence conversations may have occurred that were not captured because we had loss of data from inaudible recordings. Finally, there may have been important system issues at each site that were not captured in our study but influenced the effect of our intervention.
Domestic violence is a highly prevalent condition, but detection in the ED remains elusive. We found that female patients will disclose their DV risk to a computer. Our study both supports the potential for computer screening to increase identification and referral for DV and raises the concern that ED physicians, particularly suburban physicians, may need additional training to adequately recognize and respond to chronic, complex psychosocial issues.
Correspondence: Karin V. Rhodes, MD, Department of Emergency Medicine, University of Pennsylvania Hospital, 3000 Spruce St, Ground Ravdin, Philadelphia, PA 10104 (firstname.lastname@example.org).
Accepted for Publication: December 22, 2005.
Financial Disclosure: None.
Funding/Support: This project was funded by grant RO1 HS 11096-03 from the Agency for Healthcare Research and Quality. Dr Rhodes is also supported by grant K23/MH64572 from the National Institute of Mental Health, Bethesda, Md.
Acknowledgment: We wish to acknowledge the many people who have contributed to this project, particularly Michael Roizen, MD, Diane Lauderdale, PhD, Marshall Chin, MD, Jon Olsen, MD, Mary Guenther, MD, and the faculty, residents, staff, and patients of the University of Chicago Emergency Medicine Program, Chicago, Ill.