Recent survey data suggests that 47% of women in the US have experienced intimate partner violence (IPV) - an abuse or aggression committed by a current or former partner that includes physical violence, sexual violence, stalking, psychological violence, and reproductive coercion.1 About one in three women reported severe physical violence perpetrated by an intimate partner over their lifetime.1 Notably, these statistics from the Centers for Disease Control and Prevention (CDC) precede the COVID-19 pandemic, and the rate of IPV has since increased due to the widespread adoption of remote jobs globally.2,3 Alarmingly, it is estimated that approximately half of female homicide victims in the United States and internationally are related to IPV.4–8Patients experiencing IPV turn to healthcare providers for associated injuries and psychosomatic issues, often before seeking help from law enforcement or social services.1,9,10 Therefore, initiating discussions about IPV in healthcare settings is crucial and potentially a matter of life and death. Early interventions can break escalating cycles of abuse and promote the safety of patients experiencing it.11,12 Today, IPV detection relies primarily on self-reported screening surveys. However, IPV often remains undetected due to both patient and provider factors including reluctance to disclose abuse, limited training of healthcare workers to recognize IPV warning signs, and time constraints on clinical encounters.11,13,14 These issues are exaggerated in emergency or urgent care settings due to the rapid pace of care delivery and lack of longitudinal relationships between patients and providers.Objective and systematic analysis of radiologic studies provides an opportunity for earlier IPV detection in patients who are not yet ready to disclose.15,16 While radiologists are required by law to report suspicion of non-accidental trauma in children, a similar identification and reporting system does not exist for adult patients experiencing IPV in most states.17 However, it is increasingly appreciated that by recognizing imaging patterns due to IPV related target and defensive injuries, frequency and combination of injuries, imaging utilization, and injuries inconsistent to the provided mechanism, radiologists can raise the suspicion for IPV.18Previous investigations have explored injury patterns in patients experiencing IPV and reported a higher number of radiological studies performed in this patient population. However, many of these studies were descriptive, focused on single organ systems or imaging modalities, and studied relatively small patient cohorts from ED settings.19–21 The aim of this study was to further delineate imaging utilization and injury patterns associated with IPV. By comparing radiologically evident injuries, imaging utilization patterns, and screening usage and cancellation rates between patients experiencing IPV and age- and race-matched controls, we establish more systematic radiological criteria for raising suspicion for IPV.
Methods
This HIPAA compliant retrospective study with a case-control design was performed with IRB approval (2016P002096) at a tertiary Level I Trauma center in an integrated healthcare system in the Northeast United States. Informed consent was waived due to difficulty in locating individuals and the impact on scientific validity if we could use data only from individuals from whom we were able to obtain consent. The case population consisted of women who reported physical IPV for the first time at 18 years of age or older and were referred to our institution’s domestic violence and intervention program between 2013–2018 and underwent at least one radiology study (N = 265). We excluded males (N = 23) because of their distinct patterns of IPV-associated injury including higher rates of lacerations and cutting injuries.22 Using our institution’s Research Patient Data Registry (RPDR), we attempted to perform 1-to-4 matching on age and race to build the control group with female patients who had at least one radiology study during the same period but never reported IPV to the violence prevention program. Age matching was performed using patients’ dates of birth with 10-year age buckets. For example, a case who was 25 years old at the time of matching could be matched to a control who was 15 to 35 years old at the same time. Controls who were less than 18 years old at the time of matching were excluded. Race was collected from the electronic health record (EHR) and was considered because IPV prevalence varies between racial groups.1,23 If a patient’s race was not available in the database, their race was categorized as “Unknown”, which was treated as a category akin to known races for the purpose of matching. Several iterations were performed to build an optimal control group, resulting in variations in the data collection periods between cases (data collected January 2021) and controls (data collected in January 2023). From the control group, we excluded 185 patients who were less than 18 years old or had a noted history of IPV, yielding a final cohort of 875 controls (Figure 1). Of note, it was difficult to obtain adequate matching with respect to race due to the disparate underlying racial compositions of the case group and eligible pool of control patients, particularly after the exclusions of potential control patients as described above.Final signed reports summarizing all radiologic studies available in the EHR for cases and controls at our integrated health care system were obtained from RPDR. The imaging modality and organ system(s) imaged were extracted for each study. Reports from cases were divided among four Emergency Radiology (ER) fellows for manual review to determine whether one or more injuries were documented in each study. Each case was reviewed by one reader. Each injury was mapped to a single anatomic site and injury type (Supplemental Table 1) and was counted only once at the time when it was first observed (“unique injury”) even if it was seen again on subsequent studies. We included obstetric and gynecologic (OB-GYN) studies because of established associations with IPV.27 Reports from controls were manually reviewed by three ER fellows and one ER fellowship-trained attending radiologist. As our initial objective was to explore the imaging patterns related to IPV with controls added at a later stage, the case-control status was not blinded during the manual review. Radiologists reviewing reports from imaging studies of cases were blinded to the date of IPV diagnosis.We compared imaging utilization between cases and controls by organ system, imaging modality, study setting, and the number of completed and canceled screening studies. Organ systems included neurologic (craniofacial, neck and spine), thoracic, abdominal, musculoskeletal, OB-GYN, breast, cardiovascular, and other. Modalities included radiographs, CT, MRI, ultrasound, fluoroscopy, nuclear imaging, and interventions. Study settings were categorized as ED and non-ED. Screening studies included screening mammograms, bone density scans, and lung cancer CT screening.We analyzed temporal patterns of imaging utilization at three levels: time-of-day, day-of-week, and season. Daytime was defined as 8:00 AM through 8:00 PM. Weekend was defined as Friday 8:00 PM through Monday 8:00 AM. Calendar dates were converted to season using the time2season function from the R package hydroTSM (version 0.6–0).We compared injuries between cases and controls by anatomic location, injury type, and synchronicity of injuries. Table S1 summarizes injuries considered for each anatomic location. Synchronous injuries were defined as multiple unique injuries involving different anatomic regions detected in a single clinical encounter. Asynchronous injuries were defined as multiple unique injuries detected in separate clinical encounters.Statistical analyses were performed in SAS Analytics Software (version 9.4). Plots were generated using the ggplot2 package (version 3.4.3) in R (version 4.3.2).To account for differences in the time of first presentation to our institution and variability in the data collection period, we compared imaging utilization and injury patterns using person-time and incidence rates. Person-time was defined as the amount of time between a patient’s first encounter at our institution and the end of the data collection period (January 2021 for cases, January 2023 for controls). At the time of data collection, no patients in the case or control cohort were marked as deceased in RPDR, although it is possible that there were some deaths not recorded within our health system database. Also, as is the case for most retrospective analyses, it is difficult to determine if a patient has been lost to follow-up, so the calculated person-time estimates could be inflated due to the conservative method used here. That said, the same method was applied to both the case and control cohorts, and there is not an evident reason why their person-time estimates would be impacted differently by this approach. Incidence rates were calculated for each population as the number of studies or injuries divided by the total person-years contributed. Adjusted incidence rate ratios (aIRRs) and their corresponding 95% confidence intervals and p-values were calculated using Poisson regression with the number of studies or injuries as the outcome and the log(person-years) as the offset, with adjustment for age (category) and race/ethnicity.For binary variables (e.g., study setting, temporal analyses, presence of synchronous injuries), we used generalized linear models with a logistic link and a binomial distribution to calculate adjusted odds ratios (aORs). Covariates included age (category) and race/ethnicity. We excluded the insurance status from our analysis as it fluctuated over time for several patients. For nominal categorical variables with more than two categories, a reference category was chosen and each category was compared to the reference. When calculating crude or adjusted ORs, we controlled for correlation between multiple imaging studies per patient by using generalized estimating equations, implemented with PROC GENMOD in SAS. We excluded studies for which setting, date, or time were not available from the respective analyses.To compare the cumulative incidence of asynchronous injury in patients who suffered at least one injury, we performed a time-to-event analysis using Cox proportional hazards regression. The time variable was defined as the time from the first injury to the second encounter with unique injury (among subjects who had at least two encounters with unique injuries) or as the time from the first injury to the end of the follow-up period (among subjects with only a single encounter with injury). The predictor variables were case-control status, age (category), and race/ethnicity. The Hazard Ratio (HR) estimates the relative incidence of asynchronous injury at any given time for cases and controls among subjects who did not have a second encounter with injury as of that time.As described above, cases and controls were effectively matched on age but not race. Most of the statistical analyses utilized unconditional regressions with adjustment for age and race/ethnicity, which do not account for the matched study design. For each analysis involving adjustment for age, we considered the patient’s age at the time of data collection. For each analysis involving adjustment for race/ethnicity, these were treated as a single variable. Patients who identified as Hispanic ethnicity were categorized as Hispanic. Patients who identified as non-Hispanic ethnicity or without a documented ethnicity were categorized based on their recorded race. Most analyses involved testing of multiple covariates against each outcome, and a threshold of statistical significance was set at a two-sided alpha of 0.05 without adjustment for the number of comparisons.
Results
We compared 265 female IPV cases to 875 matched female controls with at least one radiologic study (Figure 1, Table 1). Cases contributed 3013 person-years, and controls contributed 8878 person-years (median 8.3 vs. 8.7 person-years per patient, p = 0.017). Age distributions were similar for cases and controls (p = 0.79). Hispanic ethnicity was more prevalent among cases (21% vs. 4%, p < 0.0001). Racial breakdowns of cases and controls were not significantly different among non-Hispanic patients (p = 0.05) (Table 1).Cases and controls contributed 6401 and 13462 total imaging studies, respectively (median 13 vs. 4 studies per patient, p < 0.0001) (Table 1). Cases underwent imaging studies at a higher rate than controls (2125 vs. 1516 studies per 1000 person-years; aIRR [1.5–1.6], p < 0.0001) (Table 2).The rates of abdominal, thoracic, musculoskeletal, neurologic (craniofacial, neck, and spine), and OB-GYN imaging were higher in cases (Table 2). Differences were most pronounced for neurologic (aIRR: 2.1 [2.0–2.3], p < 0.0001), thoracic (aIRR: 1.8 [1.6–1.9], p < 0.0001), and musculoskeletal (aIRR: 1.7 [1.6–1.8], p < 0.0001) imaging. Similar utilization rates were observed for breast imaging (aIRR: 0.9 [0.8–1.0], p = 0.21). Cardiovascular imaging was performed at a modestly higher rate among controls (aIRR: 0.7 [0.5–0.9], p = 0.012) (Table 2).Multiple modalities were used more frequently for cases than controls, with the strongest trends observed for CT (aIRR: 2.1 [1.9–2.2], p < 0.0001), radiographs (aIRR: 1.5 [1.4–1.6], p < 0.0001), and ultrasound (aIRR: 1.4 [1.3–1.5], p < 0.0001) (Table S2). Weaker trends in the same direction were observed for interventional studies (aIRR: 1.5 [1.1–2.0], p = 0.0037) and MRI (aIRR: 1.2 [1.1–1.3], p = 0.0042). Cases and controls had similar utilization rates for fluoroscopy (aIRR: 1.0 [0.7–1.4], p = 0.95) and nuclear medicine (aIRR: 0.9 [0.6–1.2], p = 0.36). No modalities were utilized at significantly higher rates in controls (Table S2).Cases and controls underwent screening imaging studies at similar rates (aIRR: 0.9 [0.8–1.1], p = 0.29) (Table S3). Cases had significantly higher cancellation rates across all studies (11.5% vs. 7.2%; aOR: 1.8 [1.5–2.3], p < 0.0001) including screening mammograms (40% vs. 18%; aOR: 3.4 [1.9–6.0], p < 0.0001) (Table S3).Among all imaging exams for which study setting was available, 34.9% and 13.1% were performed in the ED for cases and controls, respectively (aOR: 3.8 [2.9–5.0], p < 0.0001). Among the studies which demonstrated injury, 62.1% were performed in the ED for cases versus 40.0% for controls (aOR: 2.1 [0.8–5.0], p = 0.11) (Table S4).Cases were more likely to undergo imaging studies overnight (aOR: 1.6 [1.3–1.9], p < 0.0001) and on weekends (aOR: 1.3 [1.1–1.5], p = 0.0055) (Table S5). The distribution of studies across seasons was similar for cases and controls (Table S5). When considering only studies with unique injury, cases had a modestly higher proportion of studies performed overnight (aOR: 1.8 [1.0–3.2], p = 0.06) though not on weekends (aOR: 1.20 [0.6–2.4], p = 0.55) (Table S6).There were 131 (49.4%) cases with at least one imaging study showing evidence of unique injury, compared to 100 (11.4 %) controls with at least one injury. These 131 cases contributed 314 unique injuries across 288 radiological studies from 243 clinical encounters, and the 100 controls contributed 119 unique injuries seen on 116 radiological studies from 112 clinical encounters (Table S7). The overall incidence rates of injury for cases and controls were 104 and 13 injuries per 1000 person-years, respectively (aIRR: 15.1 [12.5–18.1], p < 0.0001) (Table 3). Among individuals with at least one injury observed on a radiological study, the number of unique injuries per patient ranged from 1 to 9 for cases and 1 to 6 for controls (Table S7). The mean and median number of unique injuries per patient were significantly higher for cases than controls (mean [1.88] vs. 1.19 [0.62], p < 0.0001; median [2.0] vs. 1.0 [0.0], p < 0.0001) (Table S7).Among clinical encounters with at least one injury detected, synchronous injuries were observed more frequently for cases than controls (17.2% vs. 3.6%; aOR: 4.7 [1.5–14.1], p = 0.006) (Table S7). After a first encounter with injury detected, the probability of a subsequent encounter with a unique injury detected (an “asynchronous injury”) was also higher for cases (HR: 4.9 [2.6–9.3], p < 0.0001) (Table S7, Figure 2A). The time distributions between asynchronous injuries are shown in Figure 2B. There were 7 cases with at least five asynchronous injury encounters, while no controls had more than two such encounters (Table S7).Injuries occurred at significantly higher rates among cases than controls for every anatomic site analyzed except for breast and cervical spine (Table 3, Figure 3A–3B). Particularly large discrepancies were observed for cranial (aIRR: 104.5 [14.2–770.1], p < 0.0001), facial (aIRR: 47.0 [18.9–117.0], p < 0.0001), thoracic (aIRR: 32.6 [11.5–92.2], p < 0.0001), and abdominal (aIRR: 24.6 [5.4–113.0], p < 0.0001) injuries. Injuries of the upper extremities (aIRR: 6.6 [4.3–10.2], p < 0.0001) and lower extremities (aIRR: 3.4 [2.4–5.0], p < 0.0001) were among the most common in both groups (Table 3, Figure 3A). Spinal injuries were rare in both groups, and only a single breast injury was observed, which occurred in the IPV group (Table 3). Specific injury patterns with the highest aIRR included rib fractures (aIRR: 26.7 [6.4–236.5], p < 0.0001) and mid-facial fractures (aIRR: 20.8 [6.2–69.8], p < 0.0001) (Figure 3C–3D, Table S8).
Discussion
IPV is highly prevalent and poses risks of harm and fatality.1 Early detection is critical but challenging, and imaging can play an important role in this task.15 Prior studies have described radiologic injury patterns in patients experiencing IPV for individual organ systems or anatomic sites.19–21,24–29 This analysis expands our understanding by considering all available imaging data rather than only imaging studies performed in a particular time window around the initial reporting of IPV and by assessing new features such as temporal patterns, ED imaging utilization, and cancellation and completion rates of screening exams. In the current analysis, cases were more likely to undergo imaging studies overall, had higher cancellation rates for screening imaging exams, and underwent imaging studies at disproportionately high rates in the ED, overnight, and on weekends. They experienced injuries at a higher rate and were more likely to have multiple injuries identified during a single encounter or separate encounters.Patients experiencing IPV had higher incidence rates of radiologically evident injuries for every organ system analyzed except for breast and cervical spine. Cranial, facial, and thoracic injuries were highly prevalent among cases and had aIRRs exceeding 30 when compared to controls. These patterns corroborate previous studies that demonstrated increased rates of craniofacial injuries in patients experiencing IPV.27,29,30 Although both upper and lower extremity injuries had higher incidence rates in the IPV cohort, their aIRRs were modest because they were by far the most common injury categories among controls. The higher rates of both synchronous and asynchronous injuries in cases are consistent with the previously appreciated increase in “new-plus-old” fractures associated with IPV as compared to stranger abuse.29 These patterns, which suggest repeated cycles of physical abuse, underscore the importance of historical imaging data and emphasize that recurrent presentations with new physical injuries should trigger consideration of whether the injury pattern is consistent with the provided history.The finding that cases underwent more imaging studies with disproportionate ED imaging utilization has important implications. Patients experiencing IPV have more frequent ED encounters31, and various studies have shown that the rates of ED imaging are steadily increasing in recent years. It was estimated that about one imaging study was performed per ED encounter in 2014, and about one CT scan was performed per ten ED encounters as of 2017.32,33 Therefore, radiologists will likely have increasing opportunities to detect IPV in the coming years.Some previous studies have found reduced usage rates of preventive healthcare services in patients experiencing IPV.34,35 For example, among African American women, domestic abuse negatively impacted the rates of repeat mammography.36 In our population, rates of radiologic screening exam completion were similar between cases and controls, but cases were more likely to cancel screening mammograms. This elevated cancellation rate is concerning because it may indicate missed opportunities for early cancer detection, injury identification, and interactions with clinical providers. Reasons contributing to increased cancellation could include factors related to IPV (e.g., safety concerns, coercive behavior) and merit further study.This study has several limitations. First, as a retrospective analysis of patients in a single healthcare system, it is subject to confounding factors and selection bias. The imaging utilization and injury patterns reflect only the data available in the EHR of our integrated healthcare system. Although the case and control cohorts were matched based on age, sex, and race, there was a noted imbalance in ethnicity. Further, dedicated investigation of differences in social determinants of health (e.g., income, education, insurance) are warranted. Second, the radiologists reviewing the reports were not blinded to case/control status. Third, our study analyzed female patients, but IPV also affects males with a lifetime prevalence estimate of 44%.1 Fourth, many patients experiencing IPV never report it. The case population is likely enriched for patients who have more severe injuries and are socially able to report it, and some controls may have experienced IPV without reporting it. Finally, we included all available imaging studies, irrespective of when the patient reported experiencing IPV. We did this because IPV is often undetected for extended periods, making it challenging to establish the exact time of onset of abuse.In conclusion, this study demonstrates that patients experiencing IPV have higher rates of imaging utilization and radiologically evident injuries across various anatomic sites, particularly cranial, facial, thoracic, and abdominal regions. Neurologic, thoracic, and musculoskeletal imaging are notably more common in IPV victims. Additionally, these patients are more likely to undergo imaging studies in emergency setting, often overnight and on weekends. Specific injury patterns such as rib and mid-facial fractures are particularly indicative of IPV, and the presence of synchronous (multiple injuries during a single encounter) and asynchronous injuries (recurring injuries over time) further heightens the suspicion of IPV. These findings highlight the significant impact of IPV on both patients and healthcare systems. By carefully considering current and historic imaging studies, radiologists can detect IPV more efficiently and facilitate efforts towards early detection and intervention.
IW
Dr. Ilana Warsofsky
Radiology · Abdominal ImagingBrigham and Women's Hospitalilana