See the related articles on pp 774-81, 782-94, 800-2 and 804-5.

Little research has been done on the clinical characteristics of physically and sexually abused women who present to primary care practices [6]. Most information comes from studies done in specialized medical settings [7-17]; other information is obtained from studies of women living in shelters [18], student populations [19], or general population samples [20, 21]. In these studies, abuse was associated with depression, anxiety, and other psychological disorders [6-812, 15-20]; drug and alcohol abuse [7, 8, 15-17]; sexual dysfunction [8, 15, 17, 21]; functional gastrointestinal disorders [6, 10, 14, 17]; headaches [6, 9]; chronic pain [10, 11, 13, 15, 16]; and multiple somatic symptoms [10, 12, 16-19].

Despite the seriousness of domestic violence, women rarely volunteer a history of abuse when they see their physicians [22]. In addition, although some authors have suggested that physicians should question all female patients about physical and sexual abuse [23], few primary care physicians do so [22]. Although some women would probably disclose a history of current abuse if a physician directly inquired about it, other patients may be hesitant to disclose such information because of fear or shame. A physician may also be overwhelmed when a woman presents with multiple physical and psychological problems and may defer inquiry to address the immediate symptoms. If certain somatic and emotional problems are more common in patients experiencing current domestic violence who present to internal medicine practices, the presence of these symptoms might raise physician suspicion and lead to an initial inquiry. If the association between these symptoms and abuse is strong enough, physicians might inquire again about abuse on subsequent patient visits, even if the woman had initially denied the abuse.

Our primary objectives were to 1) determine the prevalence of domestic violence among female patients presenting to four community-based, primary care, adult medicine practices that serve patients of diverse socioeconomic background and 2) identify demographic and clinical differences between currently abused patients and patients not currently being abused. A secondary purpose was to use this information to develop a clinical model that succinctly explains our data.

Methods

Top Methods Results Discussion Author & Article Info References

Participants and Survey Administration

Between February and July 1993, we surveyed adult female patients for 1 to 2 months in each of four community-based, primary care internal medicine practices in the Baltimore area. All practices were affiliated with The Johns Hopkins Bayview Medical Center, a private university teaching hospital. These practices serve approximately 23 000 adult patients.

Office nurses determined patients' eligibility for participation. A patient was considered ineligible if she was blind or could not read because she had forgotten her glasses, was illiterate, did not speak English, was mentally retarded or demented, was too acutely ill to participate, or was accompanied by a male or female companion who refused to leave the examining room. To ensure that patients were surveyed only once, each patient indicated on a cover sheet whether she had already completed a survey. If so, her response was excluded.

If an adult female patient was eligible to participate, an office nurse explained the purpose of the study, reassured the patient that all responses would be kept anonymous, and asked for participation. To maximize patient privacy, recruitment was done in the examining rooms, and patients completed the survey while awaiting the physician. To preserve the anonymity of participants, no names or coding numbers appeared on the survey. Patients placed surveys in a box after completion so that their responses would not be read by medical personnel or family or friends who may have accompanied them to the office. Because the survey was anonymous, the Institutional Review Board of the Johns Hopkins Bayview Medical Center did not require informed consent from the participants.

During the study period, 3203 female patients presented to the offices. Of these, 283 (8.8%) were ineligible, and 528 (16.5%) were not approached because the nurses failed to recruit them. Of the 2392 remaining patients, 1952 (81.6% of those approached, 66.9% of those eligible, 60.9% of those presenting) completed the questionnaire; 398 (16.6% of those approached, 13.6% of those eligible, 12.4% of those presenting) refused to complete the questionnaire; and 42 (1.8% of those approached) did not answer any of the questions about violence.

We derived prevalence statistics from the total sample of 1952 patients. For the analysis of clinical and demographic differences between patients who were currently experiencing domestic violence and patients who were not, we included 1826 patient responses (93.5%). We did not include the remaining 126 patients for the following reasons: Twenty-six (1.3%) had experienced violence from a stranger within the last year; 23 (1.2%) did not identify the abuser and it was therefore unknown whether the abuse had been domestic; and 77 (3.8%) did not meet strict criteria for current abuse because they answered "no" to the question "Within the last year, have you been hit, slapped, kicked or otherwise physically hurt by someone?" but in subsequent questions indicated that they had been physically threatened, pushed, or grabbed in the last year. These patients, who reported some form of actual or threatened abuse within the last year but did not meet our definition for current domestic violence, had characteristics that differed from those of patients who were currently experiencing domestic violence and those of patients who were not. Such patients were therefore excluded from this analysis.

Survey Instrument

The survey instrument, called The Women's Health Questionnaire, was developed by a team consisting of a psychologist-researcher, a biostatistician, and seven primary care internists, one of whom is a health services researcher and three of whom had previously developed an educational program on domestic abuse.

The self-administered survey included approximately 85 questions and could be completed in 5 to 7 minutes. The two questions that defined the presence or absence of current domestic violence (see below) were taken from the three-question Abuse Assessment Screen developed by McFarlane and colleagues [24]. The third question, about abuse since the onset of pregnancy, was dropped because it was irrelevant to most of our respondents. The Abuse Assessment Screen has been shown to discriminate between abused and nonabused women better than two longer instruments whose reliability and validity have been measured—the Conflicts Tactics Scale [25] and the Index of Spouse Abuse [26]. Our survey also included questions on the frequency and severity of current abuse and on whether the respondent had ever been physically or sexually abused as a child or adult.

We used the CAGE questions to assess current and past alcohol abuse. Previous research has indicated that a score of 2 or more on the questions has a sensitivity of 74% to 76% and a specificity of 91% to 94% for the diagnosis of alcohol dependence or abuse (area under the receiver-operator characteristic curve, 0.89) [27, 28].

We used a 22-item symptom inventory, the Symptom Checklist-22 (SCL-22) Table 1, to measure anxiety, depression, somatization, and interpersonal sensitivity (low self-esteem). The 22 items were taken directly from the 58-item Hopkins Symptom Checklist, which has proven reliability and validity for measuring these variables [29]. Items were chosen on the basis of the size of their factor-analysis loadings on the relevant dimension. Internal consistency measures, or Cronbach measures, for the SCL-22 were the following: 0.79 for anxiety, 0.86 for depression, 0.75 for somatization, 0.82 for interpersonal sensitivity, and 0.92 for the total inventory.

The questionnaire also contained questions on demographic characteristics, physical symptoms, psychiatric history, use of street drugs, use of current medication, and medical history. The final questionnaire was ranked as being at a seventh-grade reading level according to the Flesch Readability Statistics (available in Microsoft Word, version 2.0). Before final revisions were made, a pilot survey was administered to 130 patients seen at an obstetrics and gynecology practice.

Definition

A patient was considered to be a current victim of domestic violence [currently abused] if 1) she answered yes to either of the following questions: "Within the last year, have you been hit, slapped, kicked, or otherwise physically hurt by someone?" or "Within the last year, has anyone forced you to have sexual activities?" and 2) she identified the abuser as a husband, ex-husband, boyfriend, or relative. Patients were defined as not currently abused if they answered "no" to the defining questions, even if they had experienced violence in previous years of adulthood or had been physically or sexually abused as a child.

Statistical Analysis

In the descriptive analysis we examined the distributions of all study variables. We examined continuous variables for evidence of non-normality, skewness, and outliers. Mean substitution for missing data was used for the SCL-22 if 75% or more of the scale items were completed.

In bivariate analyses, we compared the two groups of women using chi-square statistics for categorical variables and two-tailed t-tests for continuous data. We derived crude and adjusted prevalence ratios with 95% CIs using the ratio of current abuse prevalence in one category of a variable compared with the ratio in a reference category (for example, prevalence in high-income women compared with prevalence in low-income women) [30]. We chose prevalence ratios in preference to odds ratios because odds ratios can produce slightly inflated estimates of the effects of factors.

The associations described between independent variables and current abuse could be partially attributed to abuse before age 18 years. Fifty-three (49%) women who were currently being abused had also been abused before age 18 years. Statistically controlling for abuse before age 18 years did not significantly alter the relations reported for currently abused patients.

We used logistic regression to asses which combination of variables was most associated with current abuse. To maintain consistency with our use of prevalence ratios in the bivariate analyses, we then used Poisson regression to derive adjusted prevalence ratios [31], which controlled for all other variables contained in the model. The SAS procedure, PROC LOGISTIC (SAS Institute, Cary, North Carolina), was used to develop a model of factors independently associated with abuse. We present adjusted prevalence ratios only for variables used in the final clinical model. Because of the many significant variables, the regressions proceeded in several stages. Only variables that were significant in the bivariate analyses were considered for analysis. We first constructed a model containing the specific physical symptom variables. We next built a model containing demographic and all other characteristics (for example, SCL-22 score and history of substance abuse). We then combined these models to produce the most parsimonious model; the change in the chi-square statistic was used as the criterion. We also applied the Hosmer and Lemeshow goodness-of-fit method using deciles of risk [32].

We used the SAS statistical package (SAS Institute) for all analyses. All statistical tests were done at the two-sided 5% level of significance.

Results

Top Methods Results Discussion Author & Article Info References

Prevalence

Of the total sample of 1952 respondents, 108 women (5.5%) had experienced domestic violence in the previous year. Four hundred eighteen women (21.4%) had been physically or sexually abused at some time in their adult life; 429 (22.0%) had been physically or sexually abused before age 18 years; and 639 (32.7%) had been physically or sexually abused as an adult or child. Of the 429 patients who were abused as children or adolescents, 308 (15.8% of 1952) had been sexually abused.

Of the 108 currently abused women, 77 (71.3%) answered the question on the number of abusive episodes in the previous year. About one third of these women (n = 26 [33.8%]) had been abused once; 39% (n = 30) had been abused two to three times; and 27.3% (n = 21) had been abused four or more times in the past year.

Of 108 women who met the criteria for current domestic violence, 103 (95.4%) identified the type of abuse. We classified 49% of these 103 currently abused women as experiencing "high-severity" abuse (threatened or hurt with weapon; burned; choked; or hit, kicked, or hurt with resulting broken bones or head or internal injuries) and 51% as experiencing "low-severity" abuse (slapped, hit, or kicked with or without bruises, cuts, or sprains; pushed or grabbed; or threatened). All 27 respondents who reported current sexual abuse also reported being physically abused. A similar pattern was seen between the clinical characteristics of sexually and physically abused patients and those of patients who were only physically abused.

Demographic Characteristics

As described in the Methods section, we included 1826 respondents for further analysis. Demographic characteristics of the total sample and of the patients who reported being currently abused are shown in Table 2. The following demographic characteristics were associated with the presence of current domestic violence: younger age; single, separated, or divorced status compared with married status; living with a male friend or family members other than husband; receiving medical assistance or no health insurance; and annual family income less than $10 000. Race and education level were not significantly associated with current abuse.

Emotional Symptoms

Currently abused patients had significantly higher mean SCL-22 subscores for anxiety (P < 0.001), depression (P < 0.001), somatization (P < 0.001), and interpersonal sensitivity (low self-esteem) (P < 0.001) than patients who were not currently abused. As an alternate way to express the relations between abuse and the SCL-22 scores, we divided the subscale scores into approximate thirds (low, medium, and high scores). As shown in Table 3, patients with high scores for anxiety, depression, somatization, and interpersonal sensitivity (low self-esteem) had significantly higher risks for current abuse than patients with low scores. Currently abused patients were not significantly more likely to be receiving antidepressant or antianxiolytic medications.

Substance Abuse

As shown in Table 3, currently abused patients were much more likely than patients not currently being abused to report that their partner drank too much alcohol or used street drugs. They were also more likely to report a substance abuse problem themselves.

Other Medical History

As shown in Table 3, currently abused patients were more likely than patients not currently being abused to report having ever attempted suicide and to report visiting the emergency department at least once in the past 6 months. They were not significantly more likely to have been hospitalized for an emotional mental problem or to have had operations during their lifetime. The miscarriage rate was slightly higher in the abused group. When we controlled for age, current medication use did not differ between groups. Currently abused patients were much more likely to report that they had been abused before age 18 years.

Physical Symptoms

Currently abused women reported more physical symptoms than did patients not currently being abused (mean ± SE, 7.3 ± 0.38 symptoms compared with 4.6 ± 0.08 symptoms [P < 0.001]). Collapsing the distribution of the total number of symptoms into approximate thirds provides an alternate way to view the association. As shown in Table 4, patients reporting 6 or more symptoms were nearly five times as likely to report abuse as those with 0 to 2 symptoms. The following symptoms were associated with current abuse: loss of appetite, frequent or serious bruises, nightmares, vaginal discharge, eating binges or self-induced vomiting, diarrhea, broken bones, sprains or serious cuts, pain in the pelvic or genital area, fainting or passing out, abdominal or stomach pain, breast pain, frequent or severe headaches, difficulty in passing urine, chest pain, problem with sleeping, shortness of breath, and constipation. Back pain, frequent tiredness, face pain, choking sensation, and falls were not associated with current abuse, even when we controlled for the patient's age.

Multivariate Analyses

In regression analysis, we found that the presence of three sociodemographic features (age less than 36 years, separated or divorced status, and coverage by medical assistance or no insurance), three psychosocial variables (high rating on any emotional symptom score, drug or alcohol abuse, and suicide attempt), and three physical symptoms (broken bones, sprains, or serious cuts; diarrhea; and vaginal discharge) were most associated with abuse (P < 0.001). According to the Hosmer and Lemeshow good-ness-of-fit method [32], based on deciles of risk source, the model fit well to the data (P = 0.923). As shown in Table 5, the risk for abuse increased as the number of risk factors increased.

Although a partner's substance abuse remained highly significant when added to the model, many respondents did not answer the question on this topic (n = 358). Because of this reduced sample size, we did not include this variable in the final model.

We did not include the total number of physical symptoms for the following reasons: 1) It was not significant when included in the model with the three specific symptom variables; 2) replacing the three specific symptom variables with the total number of physical symptoms did not substantially change the explanatory power of the model; and 3) the total number of physical symptoms overlaps conceptually with somatization and was highly associated with the specific emotional symptoms, especially somatization, that are already in the model.

Discussion

Top Methods Results Discussion Author & Article Info References

Ours is the largest study of domestic violence in a primary care adult patient population ever attempted. We studied patients in community-based, primary care practices who were diverse in terms of age, race, education, marital status, and family income. A review of 1992 U.S. census information [33] suggests that this sample resembles the U.S. population in terms of the percentage of married women, the percentage of women having private insurance, and the percentage of women in each income range (except the highest, which was somewhat under-represented). This sample consisted of fewer high school graduates and more older women than the U.S. population in general. Although the Hispanic population was underestimated, the percentage of African-Americans surveyed approached the national average (12.4% nationally compared with 16.7% in our study). Many previous studies have been done in patients seen in emergency departments, subspecialty clinics, and hospitals, where clinical presentation may differ. Previous studies have also focused primarily on women with lower socioeconomic status. To our knowledge, no study has used validated questions for domestic violence, alcohol abuse, and emotional distress or systematically reviewed physical symptoms in a primary care population. In addition, no previous studies in an internal medicine setting have developed a model that might help identify women at high risk for current abuse.

Many of our results were similar to results of studies done in other settings and further support the existence of a "battering syndrome." Compared with patients who were not currently being abused, currently abused patients had more physical symptoms; higher levels of depression, anxiety, and somatization; and lower levels of self-esteem. Currently abused patients were also more likely to be abusing alcohol or street drugs, to have attempted suicide, to have visited an emergency department in the 6 months before presentation, and to have had a partner who abused drugs or alcohol.

However, some of our findings differ from those of previous studies. The prevalence of current abuse in our study is lower than that found in a study done in a primary care setting (5.5% compared with 14%) [2].

One possible explanation for this difference is the methods for measurement. We used a relatively strict definition of abuse that required physical violence or coerced sexual activities and excluded pure verbal abuse; however, other studies have used a similarly strict definition [2, 24]. Investigators of one study have claimed that a questionnaire is less likely to detect abuse than a personal interview by a nurse [34]. That study, however, compared reported rates from questionnaires that were completed by patients in a public waiting room and that became part of their medical records with reported rates from personal, confidential interviews that were conducted in a private setting. In fact, the previously cited study in a primary care setting [2], which reported a higher prevalence rate than we found, also used anonymously completed questionnaires.

The relatively low prevalence rate found in our study is most likely due to the older age and higher socioeconomic status of the patients in our study; these characteristics were associated with a lower prevalence of abuse in our study and in some previous studies [2, 4]. In fact, 14% of women aged 18 to 35 years, 14% with annual family incomes less than $10 000, and 22% receiving medical assistance or no health insurance reported current abuse in our study.

In our study, the presence of many specific physical symptoms was associated with an increased risk for current abuse. However, some symptoms (face pain, choking sensation, falls, and back pain) were not associated with a higher prevalence of abuse; this is surprising because the face, neck, and torso are considered to be frequently involved in abuse [4].

Although our study confirmed an increased prevalence of depression, anxiety, and previous suicide attempts in currently abused patients, the use of antidepressive or anxiolytic medication and the rate of inpatient psychiatric admissions did not increase; this finding differs from those of previous reports from different settings [35, 36]. One possible explanation for these findings might be that many women experiencing domestic violence have unrecognized and untreated emotional distress.

Our clinical model, which included nine sociodemographic, psychological, and physical characteristics, showed that the prevalence of current abuse increased steadily as the number of these characteristics increased. This model, if validated in other populations, could help practicing physicians identify patients who are currently experiencing domestic violence. This is especially important because only 15.7% of patients in our study reported discussing their history of violence with a physician.

Our study has some limitations. There is currently no gold standard for the measurement of domestic violence. Because our study and most previous studies on the subject are based on self-report, true prevalence may be over- or underestimated. Although our study had a good response rate (81.6% of those approached), 8.8% of patients were ineligible and 16.5% were not approached because the nurses failed to recruit participants on busy office days. We polled the nurses on days when recruitment was low, and the nurses consistently stated they had forgotten to recruit patients when the offices were unusually busy; no other reason was given for the failure to approach patients on some days. Of patients approached, 16.6% refused to participate, and 1.8% could not be included because of incomplete responses to critical questions. Because no descriptive information is available on nonrespondents, unrecognized selection bias may have occurred. We also recruited a convenience sample of patients from four practices in one section of a city and its surrounding suburbs. Even though the study patients were demographically and socioeconomically diverse, the results may not be generalizable to all primary care adult patients. Finally, our study is cross-sectional, and causal relations cannot be determined.

Notwithstanding these limitations, our data suggest that, in a large, diverse, community-based population of primary care patients, 1 of every 20 women had experienced domestic violence in the year before presentation, 1 of every 5 women had experienced violence in her adult life, and 1 of every 3 women had experienced violence as either a child or an adult. Our study suggests that current domestic violence is associated with various factors that include age; marital status; socioeconomic status; certain physical symptoms; total number of physical symptoms; and an increased likelihood of emotional distress, substance abuse, and suicide attempts among patients presenting to primary care internal medicine practices. The magnitude of these associations support the idea that domestic violence is a significant medical public health problem and that although physicians should ask all female patients about domestic violence, they should be especially motivated to inquire about the possibility of current domestic violence when a woman presents with multiple somatic symptoms or emotional distress. Detection of domestic violence by physicians or other health care professionals might alter both diagnostic and treatment plans for these troubled women.

Dr. Kolodner: 3806 Fenchurch Road, Baltimore, MD 21218.

Dr. Dill: Montgomery County Health Department, 515 North Jefferson Davis Highway, Montgomery, AL 36104.

Dr. Schroeder: Middlesex Health Center, 1245 Eastern Boulevard, Baltimore, MD 21221.

Dr. DeChant: 2233 Wisconsin Avenue, Suite 525, Washington, DC 20007.

Dr. Ryden: NorthPoint Medical Center, 1005 NorthPoint Boulevard, Suite 700, Baltimore, MD 21224.

Dr. Bass: Johns Hopkins University, Division of General Internal Medicine, 1830 East Monument Drive, Eighth Floor, Baltimore, MD 21205.

Dr. Derogatis: 100 West Pennsylvania Avenue, Suite 302, Towson, MD 21204.

For copies of the study questionnaire, write to Dr. Kern at the above address.