Introduction

Violence against women, including psychological, sexual, and physical violence, is a long-standing and urgent public health issue (Oram et al. 2017; Sardinha et al. 2022) in both civilian and US military populations (Kwan et al. 2020; Lofgreen et al. 2017; Wilson 2018). Historical medical record data and prior published studies are limited to administrative and assigned markers (e.g., male or female), which are not accurate indicators of sex assigned at birth or gender for many people. As such, we use terms reported by prior studies to refer to gender or assigned sex markers, and refer to assigned sex when describing medical record markers. Multiple survey and retrospective medical record reviews indicate the effects of violence, including suicidal ideation, posttraumatic stress, depression, anxiety, substance use disorders, chronic pain, back pain, fibromyalgia, chronic fatigue, insomnia, and sleep apnea (Basile et al. 2021; Dworkin et al. 2017; Kelly et al. 2008; Luterek et al. 2011; Maguen et al. 2012; Short et al. 2021; Suris and Lind 2008; Surìs et al. 2007; Ulirsch et al. 2014; Vives-Cases et al. 2011; Williams et al. 2020; Wuest et al. 2008; Yalch et al. 2018; Young-Wolff et al. 2018). Those who experience violence may require healthcare interventions, including emergency room visits, in both civilian and military settings (Brignone et al. 2017; Calhoun et al. 2018; Creech et al. 2021; Dichter et al. 2018; Kelly et al. 2011; Kimerling et al. 2016; Sadler et al. 2004; Young-Wolff et al. 2018). Although research indicates military service-connected adults (e.g., active duty service members, spouses of active duty service members) may experience abuse at elevated rates relative to civilians (Allard et al. 2011; Kwan et al. 2020; Suris and Lind 2008), it is unclear the extent to which patient disclosure of abuse is documented in the medical record, and the extent to which survivors of violence are receiving care across the US Military Health System (MHS).

Variation in health outcomes and healthcare utilization in the MHS for survivors of abuse may be related to types of violence experienced and documented in the medical record and systemic inequities. Research across civilian samples indicates that systemic and institutional factors such as racism, sexism, lack of training, lack of structural responsiveness in healthcare delivery, and lack of racial representation among healthcare providers may be significant barriers to patient disclosure of violence to their healthcare providers (Heron and Eisma 2021; Tillman et al. 2010; Ullman and Lorenz 2020). For example, structural and institutional inequities may explain why Black women are less likely than White women to disclose experiences of abuse to mental health professionals and receive mental health treatment after experiencing sexual assault (Alvidrez et al. 2011; Starzynski et al. 2007).

Additionally, inequities in clinician screening for intimate partner violence (IPV) have been documented. Across studies with civilian samples, IPV screening has been documented to occur more often in women of color than White women (Jones et al. 2020). The US Preventive Services Task Force has endorsed screening people of reproductive potential for IPV, but this recommendation did not extend to screening older women due to insufficient evidence (US Preventive Services Task Force et al. 2018). However, routine screening for IPV may improve the provision of support to survivors of abuse (Makaroun et al. 2020). In light of these inconsistencies, it is unclear what medical conditions and healthcare utilization patterns may occur in the year following medical record abuse code documentation. The goals of the present study were threefold: to identify the overall incidence of abuse code documentation in the medical record in patients assigned female, examine factors associated with abuse code documentation, and compare subsequent healthcare utilization (e.g., overall visits, emergency room visits, opioid prescription receipt), including pain and behavioral health diagnoses received, in the year following abuse code documentation in a sample of military service-connected adults.

Methods

Data sources and record selection

This observational, retrospective study was provided a non-research determination by the Brooke Army Medical Center Institutional Review Board (C.2019.156n) and conforms with the US Federal Policy for the Protection of Human Subjects. Data extracted from the MHS Data Repository were accessed and analyzed within the Army Analytics Group Person Event Data Environment (Vie et al. 2015). Adult patients (18 years of age or older) who received care in the MHS between October 2015 and December 2019 and were assigned female in the medical record were included in the study.

Cohort identification

Patients who had an initial healthcare encounter (i.e., index visit) for physical, sexual, or psychological abuse (Healthcare Documentation [HD+] cohort) were identified using ICD-10 codes (T74.11XA, T74.21XA, T74.31XA). Note that the codes and available documentation do not indicate the time frame for abuse (e.g., lifetime), but instead indicate that there was an initial healthcare encounter with a documented code corresponding to abuse. The no-documentation cohort (ND cohort) included patients who had never received documentation corresponding to abuse during the study period. For the HD+ cohort, healthcare records were extracted for the year preceding and the year following the index visit. For the ND cohort, a random, “pseudo-index” date was selected by identifying the first and last available healthcare encounters within the study period, then randomly selecting a date occurring at least 1 year after the first healthcare encounter and 1 year preceding the last healthcare encounter. Records for both HD+ and ND cohorts were excluded if the patient did not have any healthcare encounters before or after the index visit date.

Variables of interest

Care- and patient-level characteristics

Demographic characteristics included age, beneficiary type (active duty versus non-active duty status), race, and ethnicity. Outpatient healthcare encounters and emergency room (ER) visits were aggregated to construct variables for healthcare (continuous) and ER utilization (yes, no) variables, respectively. Pharmacy transaction records were extracted for the following types of medications, per American Hospital Formulary Service classifications: opioid agonists, benzodiazepines, antidepressants, and non-opioid pain medications (nonsteroidal anti-inflammatory drugs [NSAID] and acetaminophen). Documented mental health conditions (e.g., adjustment, anxiety, depression, and posttraumatic stress disorders) and musculoskeletal pain conditions (e.g., spinal conditions, joint pain, osteoarthritis, bone fracture, dislocation or sprain, and muscle or tendon injury), within 1 year prior to the index visit, were consolidated into two variables, respectively.

Outcomes

The outcomes of the study were (1) level of healthcare utilization (Payne et al. 2018), (2) incidence of ER visits, (3) incidence of diagnosed mental health conditions, (4) incidence of diagnosed pain conditions, and (5) opioid prescription receipt within 1 year following the index visit.

Analytic plan

Propensity score construction and matching

Bivariate analyses, χ2 tests (categorical variables) and Wilcoxon–Mann–Whitney tests (continuous variables), assessed differences in pre-index factors between patients in the HD+ and ND cohorts. Pre-index factors that were different between cohorts, as evidenced by a standardized mean difference (SMD) < 0.2, were included in the propensity score construction. Additional factors included in the propensity score construction model were the pre-index values of all outcomes. Propensity scores were estimated using a logistic regression. Nearest-neighbor propensity score matching was conducted using R (R Foundation for Statistical Computing, Vienna, Austria) and the MatchIt R package (Ho et al. 2011). The cobalt R package (Griefer 2022) was used to evaluate the degree to which matching resulted in an adequately balanced subsample. An acceptable marginal balance was defined as an absolute standardized difference < 0.02.

Raw and matched comparisons

Linear (post-index healthcare encounters) and logistic (post-index ER visit, mental health condition, pain condition, opioid prescription receipt) regression models were used to examine outcomes. Models were initially tested using the full sample, which was unadjusted for pre-index factors, and then repeated for the primary analysis with the 1:1 matched sample. Follow-up sensitivity analyses were performed with a 1:3 matched sample, as well as with inverse probability weighting (IPW). IPW was conducted with the twang R package (Ridgeway et al. 2017). Given the exploratory nature of the present analyses, combined with the presence of multiple outcomes, statistical significance was assigned to differences of p < 0.01 in the 1:1 matched analyses.

Results

Univariate and bivariate statistics

Between October 2015 and December 2019, 5575 patients in the HD+ cohort had a documented initial encounter of physical, sexual, or psychological abuse, compared with 461,708 in the ND cohort. After removal of patient records lacking healthcare encounters before or after the index visit date, a total of 5239 patients in the HD+ cohort and 457,325 in the ND cohort met the criteria and were included for further analyses. Table 1 describes the distribution and description of sample characteristics between HD+ and ND cohorts. Overall, patients in the HD+ cohort were younger and more likely to be an active duty service member (55.9% vs. 43.2%) than those in the ND cohort. The proportion of Black patients in the HD+ cohort (22.4%) was greater than that in the ND cohort (14.1%). Healthcare utilization, defined as the number of healthcare encounters in the 12 months following an abuse code documentation, was greater in the HD+ cohort (median = 21, IQR = 11–38), relative to the ND cohort (median = 11, IQR = 6–21). Patients in the HD+ cohort had a greater incidence of mental health (46% vs. 23%) and musculoskeletal pain (58% vs. 47%) conditions, and were more likely to have an opioid prescription (39% vs. 27%) in the pre-index visit period, as compared with the ND cohort.

Table 1 Sample characteristics and covariate balancing results

Primary and sensitivity analyses

Primary and sensitivity model results are reported in Table 2. Results from the 1:1 matched sample show that mean (95% confidence interval [CI]) post-index healthcare utilization and odds (95%) of an ER visit, mental health diagnosis, and opioid prescription receipt within the year following the index date were higher in the HD+ cohort, relative to the ND cohort. In the 1:1 matched sample, there was a lack of significant difference between the HD+ and ND cohorts in the probability of having a musculoskeletal pain condition during the post-index visit period. Follow-up sensitivity analyses on the 1:3 matching and IPW models produced similar results as the 1:1 matched sample, with the only exception occurring in the IPW model, in that there was a significant difference in the probability of receiving a diagnosis of musculoskeletal pain.

Table 2 Differences in outcomes (95% CI) and significance levels between patients with and without documented abuse codes, across models

Discussion

In this observational, retrospective study of patients assigned female in the medical record receiving care in the MHS, those with psychological, physical, and sexual abuse documented in their medical record received more healthcare visits and were more likely to have a mental health diagnosis, experience an ER visit, and receive an opioid prescription in the year following abuse documentation, relative to matched patients in the ND cohort. There was a lack of significant association between cohort membership and receipt of musculoskeletal pain diagnosis. These findings are consistent with prior retrospective electronic health record studies of civilians and veterans (Brignone et al. 2017; Creech et al. 2021; Dichter et al. 2018; Kimerling et al. 2016; Young-Wolff et al. 2018) suggesting that people with documented psychological, sexual, or physical abuse receive more healthcare services than those who do not; however, the adequacy, appropriateness, and patient perception of the healthcare received is unknown. These findings indicate the need to better understand and optimize healthcare responsivity that matches patients’ needs, dignity, and autonomy.

Our present results indicate racialized inequities experienced by Black patients, as Black patients had disproportionate incidence of positive abuse documentation relative to White patients. Additionally, patients with abuse codes in the present sample were significantly younger on average than those without documented abuse codes. These data do not necessarily indicate that younger or Black patients are more likely to experience abuse in the MHS, but rather that they are more likely to have abuse documented in their medical records. It is also unclear whether the documented abuse is current or more remote. In part, whom a patient discloses abuse to (e.g., healthcare provider, law enforcement, commanding officer) and when the abuse is disclosed to a medical provider (e.g., when compounding stressors are high) may be impacted by systemic racism, sexism, anti-queerness, and other complex contextual factors such as patients’ desire to disclose, practice organizations’ screening guidance, state laws, safety, economics, children, and social support. Conclusions regarding inequities in abuse documentation remain limited due to the dearth of available and accessible data on the experiences of abuse and violence in service members and their family members that can be evaluated within a health equity measurement framework (Dover and Belon 2019), to include multiple aspects of identities (e.g., race, ethnicity, sex assigned at birth, gender, sexual orientation, disability, age) and system, institutional, and personally mediated factors.

Research in civilian populations indicates that early intervention after experiencing assault or abuse can mitigate negative mental outcomes (Oosterbaan et al. 2019). However, studies indicate that women who experience sexual assault during military service often do not seek care after the assault (Calhoun et al. 2018; Mengeling et al. 2015). Reasons cited for not seeking healthcare after experiencing sexual assault include, but are not limited to, believing that healthcare following a sexual assault was not needed or warranted, feeling embarrassed and concerned about privacy and confidentiality, beliefs that they would be blamed by members of their unit for the sexual assault, and fearing that seeking healthcare services would negatively impact their military career (Mengeling et al. 2015). The fear of social and professional retaliation in active duty servicewomen who have experienced military sexual assault is common (28%), with more than half (52%) of those who filed an official report following a sexual assault experiencing retaliation (Farris et al. 2021; Jaycox et al. 2023). The development of interventions and targeted programming in the MHS for individuals following sexual assault must be cognizant of and account for systemic barriers that may impact the likelihood of disclosing physical, sexual, or psychological abuse to their healthcare clinicians.

This study has several limitations, including the retrospective design that does not allow for causal conclusions. Furthermore, the sample was limited to patients with documented abuse codes in the medical record. Considering the likelihood of underreporting of abuse, it is possible that patients in the ND cohort may have previously experienced abuse and did not disclose or have disclosures documented and coded in the medical record. We are unable to determine the timing of any documented abuse, or whether the observed difference in healthcare utilization between HD+ and ND cohorts reflects a true increased need for healthcare services or if some services are utilized based on standard referrals following identification of abuse. Patients in the ND cohort in abusive relationships that have not been documented in the medical record may have similar mental healthcare needs but decline care or are prevented from accessing care by their abuser, or they decline or are not referred for additional care. Furthermore, service members are much more likely to voluntarily separate from the military early after experiencing assault (Morral et al. 2014). Healthcare utilization for service members who separate from the military soon after experiencing or reporting abuse may differ from those who remain military-connected, but MHS data do not include post-separation care.

Another notable limitation of the present study is that the sample was limited to those whose gender markers in the medical record were recorded as female. Analyses were unable to identify people with one or more socially minoritized gender identities or sexual orientations, as the MHS electronic health records and Defense Enrollment Eligibility and Reporting System (DEERS) do not capture these data. Currently, the MHS uses an administrative gender or sex marker as a synonymous single binary field (male vs. female), known as the gender marker, without a clear definition (e.g., sex assigned at birth, gender). This documentation injustice does not allow for self-reported and inclusive health systems action. Additionally, the present study did not examine patients who were assigned male in the medical record. Evidence suggests that cisgender men, including active duty military service members, also experience emotional, physical, and sexual abuse, and may face distinct gendered barriers to disclosing these experiences more than cisgender women and genderqueer people (Castro et al. 2015; Hoyt et al. 2011; Morris et al. 2014). Therefore, any screening and care pathway implementation must include clinician education and evaluation to ensure equity in care provision across people of all genders and sexual orientations.

Overall, the present findings indicate a resounding need to improve screening procedures, documentation practices, and responsive care pathways. Further research is needed to support healthcare clinicians in the MHS in recognizing and mitigating barriers to patient disclosure of abuse. High-quality healthcare addressing abuse sequelae may support military retention and long-term well-being of military-connected adults. Determination of the timing of abuse may also allow for tailored and individualized services. Our findings also underscore the need for future qualitative and quantitative research to address whole-person health after disclosure of experiencing violence (to include both mental health and physical health symptoms such as pain), and how to document such information in a manner that does not increase stigma or bias in treating clinicians and the healthcare system.