Interpersonal violence is commonly experienced, particularly among those with multiple marginalized identities, and is associated with comorbidities including mental illness and cancer.1,2 Electronic medical record (EMR) data provide a rich resource for violence-related research including nuanced descriptive data about patients’ experiences and structured data regarding health outcomes and mortality. EMR-based violence research has been limited because: (1) International Classification of Diseases (ICD) diagnoses indicative of specific types of violence have limited sensitivity and specificity.3 (2) Clinicians either do not query about violence or underuse ICD codes, thus leaving much violence unidentified.4–6 (3) Bias also exists in coding, which makes it difficult for scholars to interpret resultant data. For example, clinicians more frequently document child abuse in children of color than in White children.7 Manual chart review has greater efficacy than diagnosis codes for identifying violent experiences, but the incorporation of such reviews is limited by labor intensity.8
Novel methods to collect, record, and access data on experiences of violence have been suggested, for example, the use of an artificial intelligence tool to infer intimate partner violence (IPV) risk.9 To our knowledge, however, scholars have not further developed or implemented this approach. Some scholars suggest incorporating child abuse, IPV, and elder abuse screening tools directly into the EMR workflow.10–12 A universal child abuse screening tool was found to increase screening, identification, documentation, and reporting.13 Educational interventions to improve child abuse documentation have been attempted but have failed.14 Training physicians improves rates and quality of IPV documentation, but training is time and labor-intensive and may not be scaleable.15 Researchers have examined predictive analyses based on household social risk factors to improve early recognition of abuse. However, such analyses are biased and unreliable.16 The Joint Commission recommends screening for violence,17 but clinicians do not systematically perform such screens.
Most EMR-based approaches to assess for violence focus on IPV and child abuse. To our knowledge, no EMR-based research has examined assaults by police or strangers or violence in specific marginalized populations, including transgender populations. Filling this gap is of utmost importance to understand and intervene in the violence marginalized people experience. Transgender people experience extreme violence in myriad settings including in homes, schools, health care, jails, and prisons,18 which likely results in many health disparities.2 EMR-based data lends itself to studying connections between violence and health outcomes. However, we are unaware of studies that utilize EMR-based data to assess violence or its sequelae among transgender people. Thus, we created a multifaceted EMR-based methodology to identify experiences of violence among transgender and cisgender (nontransgender) people and tested its efficacy compared with diagnostic coding and the “Are you safe at home?” screening question. This paper presents our findings, explores the utility of keywords for identifying violence, and demonstrates the potential for this methodologic approach to improve care and inform policy change.
METHODS
Setting
We conducted our study at an academic medical center in Rochester, New York, a racially diverse city with a population of 211,328 in 2020. Black people comprise 39% of the population, White non-Latinx people 36%, and Latinx people 19%. Of the population, 30% live in poverty. Rochester was home to the first lesbian, gay, bisexual, transgender, and queer plus (LGBTQ+) community center in the United States and has a vibrant LGBTQ+ community. Transgender patients in Rochester access health resources in private practices, an LGBTQ+ health center, or one of 2 large medical centers. At the University of Rochester Medical Center (URMC), transgender people access hormone therapy through the Endocrinology Division, individual primary care physicians, or adolescent medicine. URMC hosts a medical-law and community-based agency collaborative called Healing through Education, Advocacy, and Law, which provides transdisciplinary health care as well as legal consultations and protection orders for people who have experienced violence. At URMC the screening question, “Are you safe at home?” is asked of patients in inpatient and outpatient settings.17
Keyword List Development
Based on a curated set of charts (n =15) pulled by violence-related diagnosis codes, we identified a list of keywords and word strings associated with interpersonal harm. We removed text unlikely to be specific to interpersonal violence and enriched our list with concepts from the Power and Control Wheel19 and the DANGER assessment.20 We assembled a multidisciplinary team of physicians, psychologists, and lawyers with expertise in IPV, health disparities, and ethics who suggested additions and subtractions.
Our study sample included a cohort of transgender people who were identified using structured query language (SQL) searches for keywords, diagnostic codes, sex identity, sex assigned at birth, legally documented sex, and pronoun data followed by manual reviews in a manner similar to that described.21 The cisgender cohort was identified using matching 2–1 with transgender patients based on age and follow-up time available in the EMR. For the cisgender cohort, inclusion criteria were the same as those for the transgender cohort and exclusion criteria were also the same with the additional criterion of no evidence of transgender identity. Some of the people in the cisgender cohort could have been transgender people who were not known to their clinicians as transgender.
To iteratively refine our list of keywords we utilized SQL to search for keywords in the progress notes of 500 randomly selected cisgender patients seen at our institution and pulled the 200 characters before and after the keywords to facilitate manual review. ABA and SJS reviewed these findings, categorizing matches into specific types of violence by perpetrator, setting, and whether the violence occurred in childhood or adulthood. Using this process, the investigative team identified 30 (6%) of 500 cisgender people who had experienced violence.
Keywords were nonspecific when exact word matches were not sought, and 200 characters before and after often did not provide enough information to understand the words in context. The keywords identified few instances of violence in the context of police, incarceration, hate crimes, childhood, older adulthood, trafficking, or war. Thus, we engaged emergency department physicians and a social worker with expertise in elder abuse and human trafficking and added words based on their input to capture additional violence types. We piloted a keyword search with this revised list of over 400 words with the same sample, searching this time for exact matches only, pulling a greater number of characters before and after, and allowing for retrieving multiple word matches per patient. We retrieved 6821 matches. To improve the feasibility of our methods for future use, we excluded words matched in >50% of charts. We reviewed matches for words found in >25% of the charts and found no instances of actual violence. We, therefore, excluded those words (ie, abuse, cuff, denies, safe, safety, trauma, wound, aggressive, attack, break, crush, push, resolve, and shot) from our keyword list. We reviewed the findings for the 48 words with the lowest match rate and found these highly correlated with experiences of violence (17 words yielded true positives). We thus kept all low-match terms in our list. Using this revised 134-word list, we searched the final samples of transgender (923) and cisgender (1846) people. This search returned 35,565 keyword matches.
To refine our methodologic approach for feasibility, we organized keywords by violence categories and enriched not previously well-captured categories with words from validated measures,22–28 national reports,18,29,30 and qualitative research.31 The new 212-word list included the original 134 words with additional words to ensure at least 10 per violence category (military/war, childhood sexual, other childhood, trafficking, elder, hate crime, and police and prisons).
We refined our coding rubrics based on a validated measure of childhood abuse to allow for more specificity.32 (Supplemental Appendix Fig. 1, Supplemental Digital Content 1, https://links.lww.com/MLR/C623). We then performed a search on 50 randomly selected cisgender and 50 randomly selected transgender people. ABA and SJS independently reviewed results for 10% of the sample to assess for true experiences of violence and to categorize them by type. Interrater reliability was 80% for the type of violence identified. The investigators reached a consensus on words about which they disagreed. They then independently coded the remainder of the matches, jointly reviewed them, and came to a consensus on any areas of disagreement. Forty words identified experiences of violence. The words were highly successful in identifying experiences of sexual violence, IPV, and homophobic/transphobic hate crimes. ABA and SJS created a final list of 88 keywords including the 40 previously successful words along with 5–10 additional words drawn from psychometric scoring systems in each aforementioned category, for which our methods were least successful.33–35 (Supplementary Appendix, Table 1, Supplemental Digital Content 2, https://links.lww.com/MLR/C624).
Electronic Medical Record Searches
Using SQL, we searched for our final word set in the final samples of transgender (923) and cisgender (1846) people. We identified 8985-word matches, which ABA and SJS audited through independently double-coding for violence (yes/no) and type, reconciling differences in coding by consensus. After coding over 10% of the results, the interrater reliability was 81.5% for whether there was violence and, if present, what type. The investigators independently coded the remaining sample, mutually reviewing matches, about which they noted ambiguity. For matches upon which the coders disagreed, they consulted the larger investigative team. We also used SQL to search the EMR for violence-related diagnosis codes and responses of “No” to the screening question, “Are you safe at home?”
Statistical Methods
We compared the prevalence and type of violence between transgender and cisgender cohorts using the χ2 test of independence. We compared the agreement of identifying violence by structured EMR fields versus keyword searches using the McNemar test.
RESULTS
Transgender people experienced more violence than cisgender people overall and across most categories (Table 1). Of the transgender cohort, 47% had documented experiences of violence compared with 14% of the cisgender cohort and 20% had experienced 3 or more types of violence versus 3% of the cisgender cohort. In childhood, 30% of the transgender cohort had experienced violence compared with 6% of the cisgender cohort, 14% of the transgender cohort experienced sexual violence compared with 3% of the cisgender cohort, and 11% of the transgender cohort experienced school bullying versus 1% of the cisgender cohort. With family members, 26% of the transgender cohort had experienced violence compared with 5% of the cisgender cohort. All χ2 test P values were <0.001.
TABLE 1 -
Unique People With Each Keyword and Percentage Experiencing
Violence by Sex Identity
|
Overall (N = 2769) |
Transgender (N = 923) |
Cisgender (N = 1846) |
| Keyword |
Keyword found; n (%) |
Actual violence; n (%) |
Keyword found; n (%) |
Actual violence; n (%) |
Keyword found; n (%) |
Actual violence; n (%) |
| Violence*
|
1079 (39.0) |
22 (2.0) |
495 (53.6) |
16 (3.2) |
584 (31.6) |
6 (1.0) |
| Domestic violence*
|
726 (26.2) |
51 (7.0) |
258 (28.0) |
25 (9.7) |
468 (25.4) |
26 (5.6) |
| Hit/hitting |
670 (24.2) |
47 (7.0) |
318 (34.5) |
27 (8.5) |
352 (19.1) |
20 (5.7) |
| Abusive/abusing/abused*
|
433 (15.6) |
312 (72.1) |
269 (29.1) |
206 (76.6) |
164 (8.9) |
106 (64.6) |
| Violent |
408 (14.7) |
15 (3.7) |
282 (30.6) |
11 (3.9) |
126 (6.8) |
4 (3.2) |
| Wounds |
385 (13.9) |
7 (1.8) |
174 (18.9) |
5 (2.9) |
211 (11.4) |
2 (1.0) |
| Controlling*
|
363 (13.1) |
139 (45.1) |
180 (19.5) |
25 (13.9) |
183 (9.9) |
10 (5.5) |
| Police |
347 (12.5) |
31 (8.9) |
205 (22.2) |
20 (8.9) |
142 (7.7) |
11 (7.8) |
| Isolation |
343 (12.4) |
13 (3.8) |
221 (23.9) |
10 (4.5) |
122 (7.4) |
3 (2.5) |
| Bullying/bullied†
|
308 (11.1) |
139 (45.1) |
277 (30.0) |
120 (43.3) |
31 (1.7) |
19 (61.3) |
| Sexual abuse |
270 (9.8) |
53 (19.6) |
168 (18.2) |
38 (22.6) |
102 (5.5) |
15 (14.7) |
| Assault/assaulted/assaulting |
269 (9.7) |
151 (56.1) |
174 (18.9) |
93 (53.5) |
95 (5.1) |
58 (61.1) |
| Choking/choked |
264 (9.5) |
16 (6.1) |
127 (13.8) |
9 (7.1) |
137 (7.4) |
7 (5.1) |
| Military |
244 (8.8) |
20 (8.2) |
145 (15.7) |
14 (9.7) |
99 (5.4) |
6 (6.1) |
| Harassment |
190 (6.9) |
12 (6.3) |
184 (19.9) |
11 (6.0) |
6 (0.3) |
1 (16.7) |
| Provoked |
182 (6.6) |
2 (1.1) |
112 (12.1) |
2 (1.8) |
70 (3.8) |
0 |
| Flashbacks |
175 (6.5) |
15 (8.6) |
136 (14.7) |
12 (8.8) |
39 (2.1) |
3 (7.7) |
| Raped/rape†
|
160 (5.8) |
52 (32.5) |
99 (10.7) |
37 (37.4) |
61 (3.3) |
15 (24.6) |
| Victim |
145 (5.2) |
14 (9.7) |
99 (10.7) |
8 (8.1) |
46 (2.5) |
6 (13.0) |
| Altercation |
136 (4.9) |
45 (33.1) |
85 (9.2) |
32 (37.7) |
51 (2.8) |
13 (25.5) |
| Stolen/stole/steal/stealing |
133 (4.8) |
17 (12.8) |
80 (8.7) |
8 (10.0) |
53 (2.9) |
9 (17.0) |
| Threats |
128 (4.6) |
17 (13.3) |
92 (10.0) |
11 (12.0) |
36 (2.0) |
6 (16.7) |
| Officer |
127 (4.6) |
6 (4.7) |
75 (8.1) |
3 (4.0) |
52 (2.8) |
3 (5.8) |
| Discrimination*
|
126 (4.6) |
14 (11.1) |
80 (8.7) |
14 (17.5) |
46 (2.5) |
0 |
| Criminal |
119 (4.3) |
9 (7.6) |
79 (8.6) |
7 (8.9) |
40 (2.2) |
2 (5.0) |
| Penetration |
108 (3.9) |
2 (1.9) |
66 (7.2) |
2 (3.0) |
42 (12.3) |
0 |
| Order of protection |
70 (2.5) |
12 (17.1) |
32 (3.5) |
5 (15.6) |
38 (2.1) |
7 (18.4) |
| Accused |
69 (2.5) |
11 (15.9) |
44 (4.8) |
6 (13.6) |
25 (1.4) |
5 (20.0) |
| Verbal abuse |
67 (2.4) |
30 (44.8) |
46 (5.0) |
23 (50.0) |
21 (1.1) |
7 (33.3) |
| Alleged |
46 (1.7) |
19 (41.3) |
31 (3.4) |
14 (45.2) |
15 (0.8) |
5 (33.3) |
| Blamed |
42 (1.5) |
10 (23.8) |
30 (3.3) |
9 (30.0) |
12 (0.7) |
1 (8.3) |
| Molested |
39 (1.4) |
15 (38.5) |
29 (3.1) |
12 (41.4) |
10 (0.5) |
3 (30.0) |
| Sheriff |
21 (0.8) |
0 |
10 (1.1) |
0 |
11 (0.6) |
0 |
| War |
21 (0.8) |
1 (4.8) |
12 (1.3) |
1 (8.3) |
9 (0.5) |
0 |
| Torture/tortured/torturing |
20 (0.7) |
1 (5.0) |
13 (1.4) |
0 |
7 (0.4) |
1 (14.3) |
| Transphobic |
19 (0.7) |
8 (42.1) |
18 (2.0) |
7 (38.9) |
1 (0.1) |
1 (100.0) |
| Homophobic |
18 (0.7) |
12 (66.7) |
17 (1.8) |
12 (70.6) |
1 (0.1) |
0 |
| Bomb/bombed/bombing |
16 (0.6) |
3 (18.8) |
10 (1.1) |
2 (20.0) |
6 (0.3) |
1 (16.7) |
| Prostitute/prostitution/commercial sex |
12 (0.4) |
1 (8.3) |
8 (0.9) |
1 (12.5) |
4 (0.2) |
0 |
| Kidnaped/kidnap/captivity |
11 (0.4) |
7 (63.6) |
9 (1.0) |
7 (77.8) |
2 (0.1) |
0 |
Statistical difference between sex cohorts in the proportion of people experiencing violence among those with a specific keyword found in their notes using a χ2 test.
*P < 0.05.
†0.05 ≤ P < 0.10.
Keywords were significantly more effective than structured data at identifying violence experiences among transgender and cisgender people. Among transgender people, keywords identified 126 transgender people (14%) with experiences of childhood sexual violence compared with only 9 (1.0%) identified by ICD codes (McNemar P <0.0001). All 9 transgender people identified by ICD codes were also identified by keywords. In regard to adult sexual violence, keywords identified 86 people (9%) with such experiences versus 24 identified by ICD codes (McNemar P <0.0001). Among people identified by ICD codes, 15 of 24 were also identified by keyword searches. Regarding violence in the home, 91 (10%) of transgender people had a negative response to the question, “Are you safe at home?” compared with 148 (16%) who were found to have experienced violence in the home based on keyword searches (McNemar P <0.001). Of the people identified by the screening question, 32 of 91 were also identified by keyword searches.
Among cisgender people, keywords identified 56 (3.0%) people who had experienced childhood sexual violence compared with 11 (1%) using ICD codes with 9 (1%) identified by both sources (McNemar P <0.0001). ICD codes identified 14 people (1%) who had experienced sexual violence in adulthood compared with 25 identified by keyword searches (McNemar P =0.035) and 8 (0%) identified by both sources. Regarding violence in the home, 63 (3%) of people were identified by ICD codes compared with 109 (6%) with keyword searches with 29 (2%) identified by both methods (McNemar P < 0.0001).
Table 2 includes the number of matches per keyword and the percentage of those matches that identified actual violence. The word sets that identified the most people with experiences of violence across both cohorts were “abusive/abusing/abused” (312), “assault/assaulted/assaulting” (151), “controlling” (139), and “bullying/bullied”(139). Of these, the ones with the highest percentage yield per word match were “abusive/abusing/abused” (72.1% of word matches represented true instances of violence) and “assault/assaulted/assaulting” (56.1%). Other words highly effective at identifying violence, “handcuffs” (57.1%), “kidnaped/kidnap/captivity” (63.6%), “homophobic” (66.7%), and “tazed” (100%), were identified under 20 times in the overall cohort. Words that were rarely found and rarely identified violence included “war,” “torture/tortured/torturing,” “sheriff,” “bomb/bombed/bombing,” and “prostitute/prostitution/commercial sex.” The words “violence,” “domestic violence,” “abusive/abusing/abused,” “controlling,” “raped/rape,” and “discrimination” were significantly more likely to identify actual instances of violence in the transgender versus cisgender cohorts. In contrast, the words “bullying” and “bullied” were significantly more likely to represent actual violence in the cisgender versus transgender cohorts. All χ2P values were <0.05 except for the words “raped/rape” and “bullying/bullied,” for which the χ2P values were <0.1.
TABLE 2 -
Events of
Violence Identified by Keywords Classified by Age, Perpetrator, Place, and Type
| Perpetrator/place |
Type of violence |
Total (N = 2769) |
Transgender cohort (N = 923) |
Cisgender cohort (N = 1846) |
| Childhood |
| Family in the home |
Physical |
96 (3.5) |
73 (7.9) |
23 (1.3) |
|
Sexual |
55 (2.0) |
38 (4.1) |
17 (0.9) |
|
Emotional |
197 (7.1) |
150 (16.3) |
47 (2.6) |
|
Emotional: drugs and alcohol |
55 (2.0) |
44 (4.8) |
11 (0.6) |
|
NOS |
22 (0.8) |
14 (1.5) |
8 (0.4) |
| Family outside the home |
Physical |
7 (0.3) |
6 (0.7) |
1 (0.1) |
|
Sexual |
30 (1.1) |
19 (2.1) |
11 (0.6) |
|
Emotional |
7 (0.3) |
6 (0.7) |
1 (0.1) |
|
Emotional: drugs and alcohol |
1 (0.04) |
1 (0.1) |
0 |
|
NOS |
4 (0.1) |
4 (0.4) |
0 |
| School |
Physical |
10 (0.4) |
9 (1.0) |
1 (0.1) |
|
Sexual |
5 (0.2) |
4 (0.4) |
1 (0.1) |
|
Emotional |
110 (4.0) |
94 (10.2) |
16 (0.9) |
|
NOS |
1 (0.04) |
1 (0.1) |
0 |
| NOS |
Physical |
24 (0.9) |
17 (1.8) |
7 (0.4) |
|
Sexual |
107 (3.9) |
76 (8.2) |
31 (1.7) |
|
Emotional |
37 (1.3) |
29 (3.1) |
8 (0.4) |
|
NOS |
18 (0.7) |
11 (1.2) |
7 (0.4) |
| Adulthood |
| IPV |
Physical |
74 (2.7) |
33 (3.6) |
41 (2.2) |
|
Sexual |
32 (1.2) |
29 (3.1) |
3 (0.2) |
|
Emotional |
113 (4.1) |
59 (6.4) |
54 (2.9) |
|
NOS |
51 (1.8) |
30 (3.3) |
21 (1.1) |
| Non-IPV family in the home |
Physical |
11 (0.4) |
4 (0.4) |
7 (0.4) |
|
Sexual |
2 (0.1) |
2 (0.2) |
0 |
|
Emotional |
54 (2.0) |
41 (4.4) |
13 (0.7) |
|
NOS |
2 (0.1) |
2 (0.2) |
0 |
| Non-IPV family outside the home |
Physical |
11 (0.4) |
5 (0.5) |
6 (0.3) |
|
Sexual |
5 (0.2) |
4 (0.4) |
1 (0.1) |
|
Emotional |
61 (2.2) |
49 (5.3) |
12 (0.7) |
|
NOS |
2 (0.1) |
2 (0.2) |
0 |
| Family/NOS |
Physical |
1 (0.04) |
0 |
1 (0.1) |
|
Sexual |
0 |
0 |
0 |
|
Emotional |
1 (0.04) |
0 |
1 (0.1) |
|
NOS |
1 (0.04) |
0 |
1 (0.1) |
| Non-family/police, prison |
Physical |
10 (0.4) |
10 (1.1) |
0 |
|
Sexual |
0 |
0 |
0 |
|
Emotional |
8 (0.3) |
7 (0.8) |
1 (0.1) |
|
NOS |
0 |
0 |
0 |
| Non-family/military |
Physical |
3 (0.1) |
2 (0.2) |
1 (0.1) |
|
Sexual |
0 |
0 |
0 |
|
Emotional |
1 (0.04) |
1 (0.1) |
0 |
|
NOS |
0 |
0 |
0 |
| Non-family/health care |
Physical |
2 (0.1) |
2 (0.2) |
0 |
|
Sexual |
2 (0.1) |
2 (0.2) |
0 |
|
Emotional |
38 (1.4) |
27 (2.9) |
11 (0.6) |
|
NOS |
0 |
0 |
0 |
| Non-family/other |
Physical |
27 (1.0) |
15 (1.6) |
12 (0.7) |
|
Sexual |
32 (1.2) |
24 (2.6) |
8 (0.4) |
|
Emotional |
106 (3.8) |
83 (9.0) |
23 (1.3) |
|
NOS |
8 (0.3) |
7 (0.8) |
1 (0.1) |
| NOS |
Physical |
51 (1.8) |
24 (2.6) |
27 (1.5) |
|
Sexual |
43 (1.6) |
29 (3.1) |
14 (0.8) |
|
Emotional |
12 (0.4) |
9 (1.0) |
3 (0.2) |
|
NOS |
5 (0.2) |
3 (0.3) |
2 (0.1) |
| Summaries |
| Any violence |
697 (25.2) |
434 (47.0) |
263 (14.3) |
| Violence in childhood |
394 (14.2) |
278 (30.1) |
116 (6.3) |
| Sexual violence in childhood |
182 (6.6) |
126 (13.7) |
56 (3.3) |
| School-based violence |
115 (4.2) |
99 (10.7) |
16 (0.9) |
| Violence perpetrated by family |
345 (12.5) |
244 (26.4) |
101 (5.5) |
| Violence in adulthood |
490 (17.7) |
305 (33.0) |
185 (10.0) |
| IPV |
199 (7.2) |
107 (11.6) |
92 (5.0) |
| Sexual violence in adulthood |
111 (4.0) |
86 (9.3) |
25 (1.4) |
| Types of violence per person: |
| None |
2072 (74.8) |
489 (53.0) |
1583 (85.8) |
| 1 |
310 (11.2) |
148 (16.0) |
162 (8.8) |
| 2 |
144 (5.2) |
99 (10.7) |
45 (2.4) |
| 3 or more |
243 (8.8) |
187 (20.3) |
56 (3.0) |
Each people seems in each row only once. So, a child may experience both school-based emotional violence and school-based physical violence (appearing in both of those rows) but in the summary row for school-based violence—only seems once. Therefore, the summary rows in the bottom half of the table will not be the sum of the relevant rows above.
IPV indicates intimate partner violence; NOS, not otherwise specified.
Table 3 demonstrates that certain word groups (sets 1–6) were particularly effective at identifying specific types of violence in transgender and cisgender cohorts. For example, 86% of the transgender cohort who had experienced childhood violence and 85% of the cisgender cohort who had experienced childhood violence were identified with set 5. Set 5 identified 90% of the school violence in the transgender cohort and 100% in the cisgender cohort. Set 2 identified 65% of IPV in the transgender cohort and 76% in the cisgender cohort. Set 3 identified 77% of adult sexual violence in the transgender cohort and 80% in the cisgender cohort.
TABLE 3 -
Ability of Keyword Sets to Detect Events of
Violence by Type and Sex Identity
| Types of violence |
Childhood |
School |
IPV |
Adult sexual |
| Total events (%*) by sex identity: |
Trans; 278 (30.1) |
Cis; 116 (6.3) |
Trans; 99 (10.7) |
Cis; 16 (0.9) |
Trans; 107 (11.6) |
Cis; 92 (5.0) |
Trans; 86 (9.3) |
Cis; 25 1.4) |
| Base keyword set: abusive/abusing/abused assault/assaulted/assaulting |
156 (56.1) |
67 (57.8) |
20 (20.2) |
3 (18.8) |
64 (59.8) |
60 (65.2) |
51 (59.3) |
15 (60.0) |
| Set 2: base set domestic violence |
161 (57.9) |
75 (64.7) |
20 (20.2) |
3 (18.8) |
70 (65.4) |
70 (76.1) |
52 (60.5) |
15 (60.0) |
| Set 3: set 2 raped/rape |
167 (60.1) |
80 (69.0) |
20 (20.2) |
3 (18.8) |
76 (71.0) |
70 (76.1) |
66 (76.7) |
20 (80.0) |
| Set 4: set 3 sexual abuse |
189 (68.0) |
88 (75.9) |
20 (20.2) |
3 (18.8) |
78 (72.9) |
71 (77.2) |
70 (81.4) |
21 (84.0) |
| Set 5: set 4 bullying/bullied |
239 (86.0) |
98 (84.5) |
89 (89.9) |
16 (100.0) |
78 (72.9) |
71 (77.2) |
71 (82.6) |
21 (84.0) |
| Set 6: base set bullying/bullied |
212 (76.3)†
|
77 (66.4) |
89 (89.9) |
16 (100.0) |
64 (59.8) |
60 (65.2) |
52 (60.5) |
15 (60.0) |
*Percentage of 923 transgender or 1846 cisgender patients.
†The proportion of events detected with the keyword set is significantly different between sex cohorts using the χ2 test (P < 0.05).
IPV indicates intimate partner violence.
Table 4 demonstrates that demographic characteristics were not significantly different between cohorts. Within the cisgender cohort, many types of violence experiences differed significantly by race, ethnicity, and age. Within the transgender cohort, violence of any type and childhood violence differed significantly by ethnicity and age, and violence by family also differed significantly by age, but no violence categories differed significantly by race.
TABLE 4 -
Demographics and
Violence
|
Transgender cohort (N = 923) |
Cisgender cohort (N = 1846) |
|
N (%) |
Any violence (N=434) |
Childhood violence (N=278) |
Any adult violence (N=305) |
Violence by family (N=244) |
N (%) |
Any violence (N=263) |
Childhood violence (N=116) |
Any adult violence (N=185) |
Violence by family (N=101) |
| Index age |
|
(<0.001)*
|
(<0.001) |
(0.211) |
(<0.001) |
— |
(0.004) |
(0.003) |
(0.119) |
(0.009) |
| <25 |
524 (57) |
278 (53) |
196 (37) |
182 (35) |
174 (33) |
1048 (57) |
171 (16) |
81 (8) |
115 (11) |
70 (7) |
| ≥25 |
399 (43) |
156 (39) |
82 (21) |
123 (31) |
70 (18) |
798 (43) |
92 (12) |
35 (4) |
70 (9) |
31 (4) |
| Race |
— |
(0.080) |
(0.882) |
(0.062) |
(0.673) |
— |
(<0.001) |
(0.042) |
(<0.001) |
(0.072) |
| Black |
107 (12) |
57 (53) |
32 (30) |
46 (43) |
25 (23) |
267 (14) |
73 (27) |
26 (10) |
58 (22) |
22 (8) |
| White |
733 (79) |
331 (45) |
219 (30) |
231 (32) |
195 (27) |
1415 (77) |
171 (12) |
81 (6) |
113 (8) |
73 (5) |
| Another race†
|
83 (9) |
46 (55) |
27 (33) |
28 (34) |
24 (29) |
164 (9) |
19 (12) |
9 (5) |
14 (9) |
6 (4) |
| Ethnicity |
— |
(0.039) |
(0.001) |
(0.886) |
(0.094) |
— |
(0.010) |
(0.104) |
(0.004) |
(0.407) |
| Latinx |
62 (7) |
37 (60) |
30 (48) |
21 (34) |
22 (35) |
111 (6) |
25 (22) |
11 (10) |
20 (18) |
8 (7) |
| Non-Latinx‡
|
861 (93) |
397 (46) |
248 (29) |
284 (33) |
222 (26) |
1735 (94) |
238 (14) |
105 (6) |
165 (10) |
93 (5) |
*The proportion with violence was compared by demographic characteristics using the χ2 test within sex identity cohorts (P value).
†Other race category includes for transgender cohort: American Indian (9), Asian (15), Asian Indian (1), Filipino (1), Japanese (2), Pacific Islander (1), other (38), not provided (16); and for cisgender cohort: American Indian (4), Asian (38), Asian Indian (2), Chinese (2), Filipino (1), Laotian (2), Malaysian (1), Pacific Islander (2), Vietnamese (1), other (89), not provided (22).
‡Includes patients who did not report this demographic: transgender = 24 (3%); cisgender = 61 (3%).
DISCUSSION
Among cohorts of cisgender and transgender patients in this single-institution study, keyword searches were highly effective at identifying experiences of violence and more effective than diagnostic codes and other structured data at identifying experiences of violence in both cohorts. Transgender patients had an extreme burden of violence, with much higher rates of violence overall and nearly every type of violence than cisgender people. These rates are similar to those demonstrated in prior studies.18 Violence against transgender people is an urgent public health issue in need of immediate attention including policy change to decrease structural stigma and improve access to housing, employment, food, and shelter. These changes could render transgender people less vulnerable to violence. Future studies should assess the ways, in which experiences of violence influence health and develop and test policy and other interventions to decrease violence against transgender people.
Among cisgender people, Latinx, Black, and other people of color experienced significantly more violence than White, non-Latinx people. Among transgender people, Latinx people also experienced more violence while no significant differences were demonstrated related to race. Certain types of violence were also more common in younger compared with older people in both cohorts. We identified sets of words, which were particularly effective for identifying IPV, child abuse, adult sexual violence, and bullying. Using these word sets, future researchers can more effectively identify specific violence types in their institution’s EMR instead of relying on diagnostic codes and other structured data. These shorter lists may require less time reviewing the EMR than spent by the investigators in this study.
The only word that was less effective at identifying violence experiences in the transgender cohort than in the cisgender cohort was “bullying.” This may be because the note templates for the transgender adolescent medicine clinic at our institution incorporate a bullying screen, and the word “bullying,” thus appears in these notes whether or not patients have experienced it.
Many patients with a positive IPV screen were not identified through keyword searches, and this may be attributable to positive screens entered in error or typographical errors. Alternatively, the keywords may not have been effective at picking up specific instances of violence. Another, more concerning, possibility is that these positive screens were not followed up with more detailed conversations.
Our searches found almost no violence in the context of racism, police and prisons, war, trafficking, or older age, which may be because those types of experiences are not documented or may represent a limitation of our keywords. In the United States, carceral and medical systems are interlinked through multiple mechanisms potentially limiting the safe discussion of police and carceral violence. Many emergency departments, including ours, staff public safety officers, some of whom are employed by the police department. In addition, when patients present in the custody of police or while incarcerated, they are accompanied by police or corrections officers, which limits the ability of clinicians to query about violence with police or in carceral systems. Patients may also experience stigma if their charts contain documentation regarding incarceration or interactions with police, so clinicians may avoid charting about these experiences. Clinicians are also not routinely trained to ask about police or carceral violence, elder abuse, or war and trafficking. Of note, these types of violence are rarely documented through diagnostic coding in our EMR. For example, in our EMR of over a million people, we identified 7 instances of diagnosis codes related to police violence in patient charts and most had no narrative evidence to corroborate them. These data suggest that diagnostic coding related to police violence is grossly underutilized. Clinicians may not know that these codes exist or may not use them given concerns about stigma for patients. Alternatively, clinicians may only code what is necessary for billing or insurance.
Our findings are consistent with work demonstrating high rates of violence experienced by transgender people and prior EMR-based literature demonstrating a lack of assessment for and documentation of violence experiences, particularly with diagnostic codes. The marked difference between violent experiences recorded through diagnostic codes and structured variables versus narrative-free text is a novel finding in our data. Our work also provides a path forward to use EMR data to investigate specific experiences of violence and their sequelae. Past literature with findings from multiple data sources indicate that transgender people experience violence in every aspect of their lives, and this is corroborated with our findings, which also provide evidence that the violence experienced by transgender people is documented and can be found in EMRs.17,29
Although our data are novel, they have limitations. Our work was performed at a single institution and violence documentation in the EMR may vary by institution, geography, rurality, clinician, and other factors. Also, although our algorithm was highly effective for identifying violence and elucidated the limitations of EMR-based research that relies on diagnostic coding or structured variables, it was also labor-intensive and required programming in SQL. Also, given small numbers, we were unable to assess experiences of violence by more specific racial categories.
Future studies will need to repeat this work in other settings, perhaps with smaller keyword lists, and create machine learning algorithms based on our keyword searches. In addition, transdisciplinary efforts are needed to investigate barriers to structured documentation of violence by clinicians, reassess screening tools for effectiveness, and develop, test, and implement interventions to improve violence screening, documentation, and interventions across demographic groups. Future work must also investigate whether and how clinicians assess and document violence in the context of racism, police and prisons, war, trafficking, and older age as well as the barriers to doing so. Many barriers may exist to the transmission of this information between patients and clinicians including stigma and issues of data privacy and safety. Given the health consequences of violent experiences, future studies are needed to understand these barriers, increase the safety of health communication, and improve the ability of patients with experiences of violence to access related health resources. Also, given the extreme burden of violence experienced by transgender people, interventions are urgently needed to stop this violence and provide safe and effective health interventions for the people who have experienced it.
In the interim, we recommend that clinicians take the following steps to ensure the safe communication of violence-related health data: (1) Before asking about experiences of violence, ensure that patients are alone. (2) Use language that patients will understand. (3) Assess who has access to the patient’s EMRs and make decisions about documentation based on that knowledge. (4) Prioritize the safety of patients over documentation. (5) Make relevant referrals to support patients who have experienced violence.
In summary, current EMR-based violence research is limited by the lack of systematic use of structured variables and diagnostic codes to record violent experiences. Keyword searches are more useful for understanding violent experiences and specific word groups can be used to identify various types of violence. Transgender people experience violence throughout their life course in various settings. Transgender and cisgender people urgently need intervention studies to curb these occurrences and provide medical support to the people who have experienced them. Interventions may also be needed to disentangle medical and carceral systems to promote the safety of querying and treating violence experienced by police and in carceral systems.
REFERENCES
1. Xia D, Chen Y, Chang R, et al. Psychosocial problems and condomless anal sex among transgender women in two cities of China: study based on the syndemic framework. Int J Environ Res Public Health. 2022;19:16161.
2. Sonu S, Post S, Feinglass J. Adverse childhood experiences and the onset of chronic disease in young adulthood. Prev Med. 2019;123:163–170.
3. Hooft AM, Asnes AG, Livingston N, et al. The accuracy of ICD codes: identifying physical abuse in 4 children’s hospitals. Acad Pediatr. 2015;15:444–450.
4. Karatekin C, Almy B, Mason SM, et al. Documentation of child maltreatment in
electronic health records. Clin Pediatr (Phila). 2018;57:1041–1052.
5. Vonkeman J, Atkinson P, Fraser J, et al.
Intimate partner violence documentation and awareness in an urban emergency department. Cureus. 2019;11:e6493.
6. Karakurt G, Patel V, Whiting K, et al. Mining
electronic health records data: domestic
violence and adverse health effects. J Fam
Violence. 2017;32:79–87.
7. Hymel KP, Laskey AL, Crowell KR, et al. Racial and ethnic disparities and bias in the evaluation and reporting of abusive head trauma. J Pediatr. 2018;198:137–143.e131.
8. Durand MB, McLaughlin CM, Imagawa KK, et al. Identifying targets to improve coding of child physical abuse at a pediatric trauma center. J Trauma Nurs. 2019;26:239–242.
9. Khurana B, Seltzer SE, Kohane IS, et al. Making the ‘invisible’ visible: transforming the detection of
intimate partner violence. BMJ Qual Safe. 2020;29:241–244.
10. Gonzalez DO, Deans KJ. Hospital-based screening tools in the identification of non-accidental trauma. Semin Pediatr Surg. 2017;26:43–46.
11. Lee ASD, McDonald LR, Will S, et al. Improving provider readiness for
intimate partner violence screening. Worldviews Evid Based Nurs. 2019;16:204–210.
12. Rosen T, Platts-Mills TF, Fulmer T. Screening for elder mistreatment in emergency departments: current progress and recommendations for next steps. J Elder Abuse Negl. 2020;32:295–315.
13. Rumball-Smith J, Fromkin J, Rosenthal B, et al. Implementation of routine electronic health record-based
child abuse screening in general emergency departments.
Child Abuse Negl. 2018;85:58–67.
14. Guenther E, Olsen C, Keenan H, et al. Randomized prospective study to evaluate
child abuse documentation in the emergency department. Acad Emerg Med. 2009;16:249–257.
15. Edwardsen EA, Horwitz SH, Pless NA, et al. Improving identification and management of partner
violence: examining the process of academic detailing: a qualitative study. BMC Med Educ. 2011;11:36.
16. Berger RP, Lindberg DM. Early recognition of physical abuse: bridging the gap between knowledge and practice. J Pediatr. 2019;204:16–23.
17. The Joint Commission. Quick Safety Issue 63: addressing
intimate partner violence and helping to protect patients. 2022. Retrieved from:
www.jointcommission.org/resources/news-and-multimedia/newsletters/quick-safety/quick-safety-issue-63/#.ZCirLMrMLrc
18. James SE, Herman JL, Rankin S, et al. Executive summary of the report of the 2015 U.S. transgender survey. Washington, D.C: National Center for Transgender Equality; 2016.
www.USTransSurvey.org
19. Pence EPM. Education Groups for Men Who Batter: The Duluth Model. New York, NY: Springer Publishing Company, Inc; 1993.
https://doi.org/10.1891/9780826179913
20. Campbell JC, Webster DW, Glass N. The danger assessment: validation of a lethality risk assessment instrument for intimate partner femicide. J Interpers
Violence. 2009;24:653–674.
21. Getahun D, Nash R, Flanders WD, et al. Cross-sex hormones and acute cardiovascular events in
transgender persons: a cohort study. Ann Intern Med. 2018;169:205–213.
22. Straus MA, Hamby SL, Boney-McCoy SUE, et al. The revised conflict tactics scales (CTS2): development and preliminary psychometric data. J Fam Issues. 1996;17:283–316.
23. World Health Organization. Adverse Childhood Experiences International Questionnaire (ACE-IQ). Geneva, Switzerland: World Health Organization; 2018.
https://www.who.int/publications/m/item/adverse-childhood-experiences-international-questionnaire-(ace-iq)
24. Finkelhor D, Hamby SL, Ormrod R, et al. The juvenile victimization questionnaire: reliability, validity, and national norms.
Child Abuse Negl. 2005;29:383–412.
25. Pennebaker JW, Susman JR. Disclosure of traumas and psychosomatic processes. Soc Sci Med. 1982;988:327–332.
26. Yaffe MJ, Wolfson C, Lithwick M, et al. Development and validation of a tool to improve physician identification of elder abuse: the Elder Abuse Suspicion Index (EASI). J Elder Abuse Negl. 2008;20:276–300.
27. Schofield MJ, Mishra GD. Validity of self-report screening scale for elder abuse: WOMEN’S Health Australia Study. Gerontologist. 2003;43:110–120.
28. Neale AV, Hwalek MA, Scott RO, et al. Validation of the Hwalek-Sengstock elder abuse screening test. J Appl Gerontol. 1991;10:406–418.
29. Lambda Legal. Protected and Served? New York, NY: Lambda Legal;1 ed. 2015.
www.protectedandserved.org/previous-survey
30. Sylvia rivera law project. It’s war in here: a report on the treatment of transgender and intersex people in New York state men’s prisons. New York, NY: Sylvia Rivera Law Project; 2007. Retrieved from:
https://srlp.org/its-war-in-here/
31. Harris OO, Dunn LL. “I kept it to myself”: Young Jamaican men who have sex with men’s experiences with childhood sexual abuse and sexual assault. Arch Sex Behav. 2019;48:1227–1238.
32. English, DJ, & the LONGSCAN investigators. modified maltreatment classification system (MMCS). 1997. For more information visit the LONGSCAN website at
http://www.iprc.unc.edu/longscan/
33. Torres-Harding SR, Andrade AL, Romero Diaz CE. The racial microaggressions scale (RMAS): a new scale to measure experiences of racial microaggressions in people of color. Cultur Divers Ethnic Minor Psychol. 2012;18:153–164.
34. Landrine H, Klonoff EA, Corral I, et al. Conceptualizing and measuring ethnic discrimination in health research. J Behav Med. 2006;29:79–94.
35. Williams MT, Printz DMB, DeLapp RCT. Assessing racial trauma with the trauma symptoms of discrimination scale. Psychol
Violence. 2018;8:735–747.
