a b s t r a c t

Introduction: Although preventing recurrent Violent injury is an important component of a public health ap- proach to Interpersonal violence and a common focus of violence intervention programs, the true incidence of recurrent violent injury is unknown. Prior studies have reported recurrence rates from 0.8% to 44%, and risk fac- tors for recurrence are not well established.

Methods: We used a statewide, all-payer database to perform a retrospective cohort study of emergency department visits for injury due to interpersonal violence in Florida, following up patients injured in 2010 for re- currence through 2012. We assessed risk factors for recurrence with multivariable logistic regression and esti- mated time to recurrence with the Kaplan-Meier method. We tabulated hospital charges and costs for index and recurrent visits.

Results: Of 53 908 patients presenting for violent injury in 2010, 11.1% had a recurrent violent injury during the study period. Trauma centers treated 31.8%, including 55.9% of severe injuries. Among recurrers, 58.9% went to a different hospital for their second injury. Low income, homelessness, Medicaid or uninsurance, and Black race were associated with increased odds of recurrence. Patients with visits for mental and behavioral health and un- intentional injury also had increased odds of recurrence. Index injuries accounted for $105 million in costs, and recurrent injuries accounted for another $25.3 million.

Conclusions: Recurrent violent injury is a common and costly phenomenon, and effective Violence prevention programs are needed. Prevention must include the nontrauma centers where many patients seek care.

(C) 2016

Introduction

Interpersonal violence caused 16 671 deaths, 140 343 hospitaliza- tions, and 1 615 995 emergency department (ED) visits in the United

? Sources of funding: No specific funding support was obtained for this research, but the first author was supported by a training grant from the National Heart, Lung and Blood Insti- tute (T32 HL-98054-6). Dr. Delgado was supported by the National Heart, Lung, and Blood Institute Career Development Program in Emergency Care Research (K12HL109009).

?? Presentation: Poster presentation, the 2016 Annual Scientific Assembly of the Eastern

Association for the Surgery of Trauma; January 12-16, 2016; San Antonio, TX.

? Conflicts of interest: The authors have no conflicts of interest to declare.

?? Author contributions: E.J.K. had full access to the data and takes responsibility for the

integrity and accuracy of the analysis. E.J.K. and K.L.R. are responsible for data acquisition.

E.J.K. and M.K.D. are responsible for the concept and data analysis. E.J.K. drafted the man- uscript. All authors contributed to critical review of the manuscript.

* Corresponding author at: Master of Science in Health Policy, University of Pennsylva- nia, Philadelphia, PA. Tel.: +1 215 573 2562.

E-mail address: [email protected] (E. Kaufman).

States in 2010, for a cumulative $8.5 billion in Medical costs [1]. Violence is increasingly understood as a public health issue with a broad impact on individual and community health [2]. Recurrent injury is a promising target for prevention, as prior injury is a predictor of future violent inju- ry [3] and death [4-6]. Hospital-based violence intervention programs show promising results with in reducing reinjury and costs by using an initial injury visit to initiate prevention [7,8], but appropriately scal- ing and distributing violence prevention resources requires a more ac- curate understanding of the incidence, risk factors, and impact of recurrent violent injury.

Although recurrent violent injury has been studied since the 1980s, prior studies have tracked recurrent visits limited to an individual trau- ma center [7,9-16], a particular intervention or cohort study [17-19], or a local area [3,20,21], and many have combined results for both violent and nonviolent injuries. Most report a rate of recurrence of 10% to 25% within 2 to 5 years [10-15,20,22-24]. At the population level, one study showed a 0.8% incidence of recurrence within 6 years in San

http://dx.doi.org/10.1016/j.ajem.2016.06.051

0735-6757/(C) 2016

Diego County [25], and another found a 2.0% incidence in Nevada over 5 years [26]. These 2 studies included both violent and nonviolent inju- ries, and were limited to severely injured patients. A prospective cohort study in Detroit showed a 44% rate of recurrence and a 20% rate of death within 5 years for violently injured individuals [3]. A population-based study in Philadelphia that included both violently and nonviolently in- jured patients showed a 14% rate of recurrent injury within 4 years [21]. A recent, prospective cohort study of violently injured youth in Flint, Michigan, that incorporated the full range of injury severity found a 37% recurrence rate within 2 years [27]. Past studies have re- ported costs of recurrent violent injuries in small cohorts [7,8], but not population-level costs. Although many injured patients are treated in nontrauma center hospitals [28], these institutions’ role in violence pre- vention is not established.

Understanding the health and economic impacts of recurrent violent injury at a population level is critical for clinicians to identify and inter- vene with high-risk individuals, and for the health care system to allo- cate violence prevention resources appropriately. To this end, we performed a retrospective cohort study of all ED visits for injury resulting from interpersonal violence in a large US state to provide a comprehensive assessment of the incidence of recurrent violent injury, the site of care and acute care costs of these injuries, and Demographic and clinical factors associated with risk of recurrence.

Methods

Study design and data source

We conducted a retrospective cohort study of all patients presenting to an ED or admitted to a hospital in Florida in 2010 for injuries due to interpersonal violence, and followed these patients for recurrence through the end of 2012. We used data from The Agency for Healthcare Research and quality healthcare Cost and Utilization Project (HCUP) State Emergency Department Database (SEDD) includes all ED dis- charges. The State Inpatient Database (SID) includes all hospital admis- sions. Combined, these 2 databases include all ED visits to nonfederal

hospitals. This analysis uses the HCUP revisit variables visitlink and daystoevent, which allows tracking of individual patients across years and hospitals [29,30]. We include data from Florida from January 2010 to December 2012, as Florida collects relevant risk factors; 1 one of 8 states to report person identifiers [31]; and has robust revisit variables, with greater than 95% verified [32]. Based on prior studies [20,26], we expected 3 years to capture most recurrences, while being short enough for trauma centers and violence prevention programs to feasibly repli- cate. This study was deemed exempt by the University of Pennsylvania Institutional Review Board.

Population

We identified all patients presenting for violent injury to a Florida hospital, excluding out-of-state residents. Visits missing revisit variables were excluded from the analysis as recurrence could not be assessed. We used HCUP Clinical Classification Software to identify all visits with an International Classification of Diseases, Ninth Revision (ICD-9) di- agnosis code associated with injury, excluding poisonings and compli- cations of medical treatment or devices, (Summarizing 2721 ICD-9 codes into Clinical Classification Software codes 225-236, 239, 240, and 244; see Supplementary Table) [33], and an external cause-of- injury code (E-code) indicating interpersonal violence as the cause, as classified by the Centers for Disease Control and Prevention [34]. Inju- ries due to intimate partner violence were included. Florida has man- dated e-coding in hospital discharge data since 1997 and in ED discharge data since 2005 [35], and more than 95% of Florida inpatient and ED discharges with an injury diagnosis include an e-code [36,37]. E-codes have been validated 95% accurate with respect to injury intent, compared with medical record review [38].

Index visits were limited to those occurring in 2010 in order to allow a minimum of 2 years of follow-up after the index injury. To ensure at least 2 years of follow-up, we excluded patients who died of their index injury, and patients who had no documented recurrence but whose death was recorded in the data set within 2 years. Patients who died of a recurrent injury were included as recurrers. The SID and

Excluded

Same CCS diagnosis: considered same injury event

Different CCS diagnosis: considered new violent

injury event

Repeat visit within 2 days: considered same injury event

Additional visits for violent injury

Visits 2-14 days after index visit

Visits > 14 days after index visit: considered new violent injury event

First visit listed in dataset: classified as

index visit

Visits coded for late effects of injury

All injury visits coded for violent injury

Fig. 1. Injury event determination.

SEDD cannot be linked to any other data set, including death certificate data, so out-of-hospital deaths could not be included, nor could injuries not leading to ED or hospital treatment. To identify recurrent injuries stemming from separate episodes of violence, rather than revisits for continuing care, we excluded visits coded as late effects or complica- tions of injury. Because a visit occurring soon after an injury might not be coded as a late effect, even if the patient had already sought care, we further classified visits as new violent injury events according to both time between visits and injury diagnoses (see Fig. 1).

Measurements

The primary outcome was recurrence. Patients were classified as recurrers if they had at least 1 visit for a new violent injury subsequent to their index violent injury visit. Demographic explanatory variables included age, sex, race/ethnicity, homelessness, urban vs rural location, median income of home zip code, and Insurance type. Injury-related co- variates included diagnosis and mechanism. Injury severity score (ISS) was calculated from ICD-9 Diagnosis codes using validated methodology [39,40]. Patients were classified as to whether their index injury result- ed in hospital admission, and whether they had a visit for an alcohol- related disorder, substance use, Mental illness, or unintentional injury at any point in the study period.

charge ratios provided by HCUP, which are based on all-payer, inpatient costs. Combining validated techniques [43], where no hospital-specific ratio was available, the HCUP group average ratio based on hospital cat- egory was substituted. When neither was available, the state average was used.

Results

There were 54 178 individuals who presented for violent injuries in the first year of the study period. One hundred fifteen died of their initial injuries, and 155 died at another hospital visit within 2 years, without being reinjured, leaving 53 908 patients for analysis. Among survivors, 11.1% (5967/53 908) recurred, with 5.6% (336) classified as severe. Just 2.2% (1192/53 908) had multiple recurrences. Overall, 31.3% of nonrecurrers and 33.3% of recurrers went to a trauma center. Of severe injuries, 55.9% of index injuries and 52.0% of recurrent injuries were treated in trauma centers.

Table 1

Patient characteristics at first visit for violent injury, 2010

2.4. Statistical analysis

Nonrecurrers (n = 47 941), n (%)

Recurrers (n = P

5967), n (%)

Bivariate analyses were performed on demographics, injury charac-	Race White	24 911 (52.3)	b.001 3410 (57.5)
teristics, and occurrence of visits for comorbid disease. Categorical vari-	Black	15 375 (32.3)	1912 (32.2)

ables were assessed using ?² tests, and continuous variables were assessed using t tests. Differences in demographic and clinical charac-

Hispanic	6333 (13.3)	524 (8.8)
Asian/Pacific Islander	260 (0.6)	16 (0.3)
Native American	63 (0.1)	13 (0.2)

teristics between recurrers and nonrecurrers were estimated using	Other	685 (1.4)	61 (1.0)
multivariable logistic regression. Multiple imputation was performed	Female	17 564 (36.6)	2280 (38.2)	.017

on missing data using the remaining covariates [41]. To identify the rel- evance of a population-based as opposed to registry-based studies, we identified severe recurrent injuries as those that would have been in- cluded in the state trauma registry: injuries resulting in death or hospi- tal admission, excluding isolated skeletal injuries [42]. We performed a secondary analysis to identify risk factors for severe recurrence. To as- sess the relevance of a statewide vs single-center study, we tracked

Age (y) b18	5824 (12.2)	529 (8.9)	b.001
18-34	24 862 (51.9)	3070 (51.5)
35-54	14 146 (29.5)	2070 (34.7)

>=55 3109 (6.5) 298 (5.0)

Insurance type b.001

Private	7638 (15.9)	426 (7.1)
Medicare	2634 (5.5)	375 (6.3)
Medicaid	11 381 (23.7)	1700 (28.5)

whether recurrers sought care for a second injury at the same or at a dif-	Uninsured/other	26 288 (54.8)	3466 (58.1)
ferent hospital and whether they were treated at a designated trauma center. We calculated the proportion of recurrence for injured patients	Zip code median income below the national median	33 387 (72.0)	4307 (76.7)	b.001

survival analysis“>Struck by or against

residing in each Florida zip code and mapped these results using ArcGIS Homeless		516 (1.1)	198 (3.3)	b.001
(Esri Inc, Redlands, CA; 2015). All other analyses used Stata (Version 14; Urban location		41 131 (87.1)	5071 (88.4)	.005
injury type				b.001
StataCorp, College Station, TX; 2015). Head injury		6557 (13.7)	805 (13.5)
spinal cord injury		21 (0.04)	2 (0.03)
2.5. Survival analysis Internal injury		751 (1.6)	65 (1.1)
musculoskeletal injury		6898 (14.4)	934 (15.7)
Open wound		12 618 (26.3)	1431 (24.0)
The Kaplan-Meier method was used to estimate time to recurrence, Superficial injury		14 988 (31.3)	1994 (33.4)
including all patients presenting for violent injury between 2010 and Other		6108 (12.7)	736 (12.3
2012 and accounting for differences in follow-up time. For patients Mechanism				b.001
who did not recur or die, time under observation was given as maxi- Gunshot wound		1180 (2.5)	92 (1.5)
mum time under observation. Time to recurrence for severe vs mild re- Stab wound		3007 (6.3) 24 824 (51.8)	365 (6.1) 3075 (51.5)
currences was compared using the log-rank test.	Other or multiple	18 924 (39.5)	2435 (40.8)
	Admitted to the hospital at	3428 (7.2)	363 (6.1)	.002
2.6. cost analysis	index injury

Index presentation to trauma center

We assessed the total hospital charges and costs for violent injury for	ISS			.022
the cohort of patients with an index injury in 2010. The SID and SEDD	b9	45 218 (94.3)	5680 (95.2)
report total hospital charges for each visit. These charges were summed	9-15	2109 (4.4)	224 (3.8)
for all violent injury visits in 2010 to 2012 for this population, and sep- arated by ED treat-and-release visits, including observation stays and inpatient admissions. Charges associated with the index injury were	N15 Other visits Mental health Substance use	614 (1.3) 5240 (10.9) 1864 (3.9)	63 (1.1) 1588 (26.6) 641 (10.7)	b.001 b.001
grouped, as were charges associated with any subsequent violent inju-	Alcohol abuse	2064 (4.3)	977 (16.4)	b.001
ries, including charges for multiple visits related to a single injury. To es-	Unintentional injury	15 619 (32.6)	3416 (57.3)	b.001
timate costs, we multiplied charges by the hospital-specific cost-to-	t Test. All others ?2 test.

15 110 (31.5) 1994 (33.4) .003

Characteristics of study subjects

Demographic and clinical characteristics from the index visit for recurrers and nonrecurrers are shown in Table 1. Demographics of recurrers were as follows: mean age was 33 years, 61.8% (n = 3687) were male, 57.5% (n = 3410) were white; 76.7% (n = 4307) lived in a zip code with median income below the national median; and 88.4% (n = 5071) lived in an urban area. Most injuries were mild: ISS was less than 9 in 94.3% of nonrecurrers and 95.2% of recurrers, and 0.3% (115/53 908) of index injuries and 0.2% (11/5967) of second injuries re- sulted in death. More than 90% of injuries were blunt for both recurrers and nonrecurrers, and the most common mechanism was struck (with or without a weapon). Only 2.5% of nonrecurrers and 1.5% of recurrers had gunshot wounds. Utilization of ED was high: the mean total number of ED and Hospital visits per patient was 8 for recurrers (interquartile range 5-16) and 3 for nonrecurrers (interquartile range 1-6) over the 3-year period, although recurrent violent injury accounted for an aver- age of only 1 of these additional visits. Fig. 2 shows Recurrence rates for violently injured individuals by zip code of residence.

Risk factors for recurrence

Multivariable logistic regression results identifying risk factors for recurrence and severe recurrence are presented in Table 2. Patients liv- ing in low-income areas had 20% increased odds of recurrence, but equal odds of severe recurrence. Medicare patients had 1.7 times in- creased odds compared with privately insured patients. Patients with Medicaid or no insurance had approximately double the odds of recur- rence, and increased odds of severe recurrence. Compared with white patients, black patients had 10% higher odds of any recurrence, but 40% higher odds of severe recurrence. Hispanic patients had 30% lower odds of recurrence. Men and women had equal odds of any

recurrence, but women had 70% lower odds of severe recurrence. Rural residence was associated with 10% lower odds of recurrence, al- though nearly 90% of patients lived in urban areas. Compared with pa- tients aged 18 to 35 years, those younger than 18 years and older than 55 years had lower odds of any recurrence, whereas those aged 35 to 54 years had 60% higher odds of severe recurrence.

Homeless patients had 60% increased odds of any recurrence, but no significant increase in severe recurrences. Concomitant visits for mental illness, alcohol abuse, and unintentional injury were associated with in- creased odds of any recurrence and severe recurrence. Visits for sub- stance use were associated with recurrence, but not severe recurrence. Patients admitted to the hospital for their index injury had equal odds of any recurrence, but 50% higher odds of severe recurrence. Injury se- verity score was not an independent predictor of recurrence, nor was initial trauma center treatment. However, patients with gunshot wounds had 20% reduced odds of recurrence compared with other mechanisms of injury. Of patients who recurred within the study peri- od, median time to recurrence was 307 days, with no significant differ- ence between severe and nonsevere recurrers (P value for the log-rank test = .1204).

Charges and costs

Hospital charges and costs are summarized in Table 3. Overall, vio- lent injury accounted for $596 million in charges and $131 million in costs for patients with an index injury in 2010. Index visits accounted for $105 million and recurrent visits for $25.3 million in costs. Overall, 58.9% of recurrers presented to a different hospital for their second inju- ry. Women, black patients, rural residents, and those 35 years or older were more likely to return to the same hospital. Odds of returning to the same hospital decreased with time between injuries and in patients with mental health visits.

Fig. 2. Geographic distribution of recurrent violent injury, Florida, 2010-2012.

Missing data

We excluded 24 869 visits (12.5%) for intentional injury from 2010 to 2012 that lacked the visitlink identifier. These patients were more likely to be Hispanic, male, or homeless, and to live in low-income or urban areas. These visits more often resulted in death (1.8% vs 0.2%) or hospital admission. Injury diagnoses were similar, but excluded visits were more often due to penetrating mechanism (16.9% vs 7.4%). Re- garding other missing data, race, age, income, or rurality was missing for 4% of patients. The regression results reported here reflect imputed values, but results and standard errors were similar when performed without imputation.

Discussion

In the first statewide analysis of all ED and hospital stays for violent injury, we found that 11.1% of violently injured patients presented to the ED with another violent injury within 2 to 3 years, and 5.6% of these recurrences were severe. These patients’ recurrent violent injuries accounted for 9836 ED visits, 1244 hospitalizations, and $25.3 million in direct costs, a substantial burden on injured individuals and the health care system. Most visits were to nontrauma centers, consistent with past studies [44]. Because most hospital-based violence intervention programs are at trauma centers, this suggests new opportunities for prevention [45]. We identified key predictors of recurrence, includ- ing urban location; low-income home zip code; Public insurance or no insurance; homelessness; and visits for behavioral health or uninten- tional injury.

By using a statewide, all-payer database, we followed up patients re- gardless of where they sought care. This provides a major advantage over single-institution studies, as we saw that 58.9% of patients present- ed to a new hospital for their second injury and 68.5% initially presented to nontrauma centers, in keeping with a Los Angeles survey which found that 42% of recurrently injured patients had previously sought care elsewhere [20]. Moreover, as shown in Fig. 2, rates of recurrence varied across the state, with multiple areas of high and low rates. Our findings can serve as a baseline for interventions aimed at reducing re- currence. Our results also point to the opportunity for a comprehensive approach to preventing recurrent violent injury: by intervening in com- munity hospitals as well as trauma centers, by studying how to identify high-risk patients at their noninjury visits, and by harnessing the poten- tial of housing stability and behavioral health treatment to promote safety. Patients who present for an initial violent injury with the risk fac- tors we identify could be directed to detailed assessment or interven- tion by ED protocol or by a flag in the electronic medical record. Hospital-based violence intervention programs that provide wrap- around social and psychological support to injured patients have shown promise [8,46-48]. However, these programs have primarily targeted Youth violence, and expanded strategies may be needed to reach the older segment of the violently injured population.

Our analysis supports past findings and expands upon them. Nearly all authors have found increased risk of recurrent injury among those living in poverty [13,15,16,18,49,50] and those with limited insurance [13,16,18,50]. Although many prior studies have identified increased risk in black compared with white patients [13,16,18,23,25,50], we found that black patients had minimally higher overall recurrence (odds ratio, 1.1), but a substantial increase in severe recurrence (odds

Table 2

Multivariable logistic regression results: risk factors for recurrent violent injury and severe recurrent violent injury

	Any recurrence			Severe recurrence
	Odds ratio (95% confidence interval)	P		Odds ratio (95% confidence interval)	P
Race White	1	Reference
Black	1.1 (1.0-1.2)	.009		1.4 (1.1-1.8)	.018
Hispanic	0.7 (0.6-0.8)	b.001
Asian/Pacific Islander	0.7 (0.4-1.1)	.104
Native American	1.7 (0.9-3.2)	.109
Other	0.9 (0.7-1.1)	.287
Female Age (y) b18	1 (0.9-1.0) 0.7 (0.7-0.8)	.509 b.001		0.3 (0.2-0.5) 0.6 (0.3-1.1)	b.001 .073
18-34	1	Reference		1	Reference
35-54	1 (0.9-1.0)	.345		1.6 (1.2-2.1)	.001
>=55 Insurance type	0.7 (0.6-0.8)	b.001		1.5 (0.9-2.4)	.087
Private	1	Reference
Medicare	1.6 (1.4-1.9)	b.001
Medicaid	2.1 (1.9-2.4)	b.001		1.5 (1.0-2.4)	.061
Uninsured/other	1.9 (1.7-2.1)	b.001		1.8 (1.2-2.6)	.003
Zip code median income below national median	1.2 (1.1-1.3)	b.001		1.3 (1.0-1.9)	.085
Homeless	1.8 (1.4-2.1)	b.001
Rural residence ISS	0.9 (0.8-0.9)	.001
b9	1	Reference
9-15	1.0 (0.8-1.1)	.618
N15	0.9 (0.6-1.2)	.321
Admitted to the hospital at index injury	0.9 (0.8-1.0)	.066		1.5 (1.0-2.2)	.023
Index presentation to trauma center Mechanism of injury Gunshot wound	1 (1.0-1.1) 0.8 (0.6-1.0)	.226 .020		1.3 (1.0-1.6)	.072
Stab wound	1.0 (0.9-1.2)	.621
Struck by or against	1	Reference
Other or multiple Other visits Mental health	1.0 (1.0-1.1) 1.8 (1.7-2.0)	.688 b.001		1.8 (1.3-2.4)	b.001
Substance use	1.6 (1.4-1.7)	b.001
Alcohol abuse	2.6 (2.4-2.9)	b.001		2.7 (1.9-3.7)	b.001
Unintentional injury	2.2 (2.0-2.3)	b.001		1.7 (1.3-2.1)	b.001

Table 3

2010-2012 hospital charges and costs for all violent injury visits of patients with an index visit in 2010

	n	Median charges	Total charges	Median cost	Total costs
Visits for index injuries ED dischargesa	52 748	$2655	$227 million	$577	$48.1 million
Admissions	4662	$32 914	$252 million	$7239	$57.3 million
Total	58 661	$3040	$479 million	$674	$105 million
Visits for recurrent injuries ED dischargesa	9836	$3164	$49.7 million	$690	$10.5 million
Admissions	1244	$31 816	$66.9 million	$6875	$14.9 million
Total	11 110	$3835	$117 million	$860	$25.3 million
Grand total	68 490	$3161	$596 million	$699	$131 million

^a Observation stays included with ED discharges.

ratio, 1.4). Furthermore, although only 15.2% of the state population was black in 2010 [51], nearly one-third of both recurrers and nonrecurrers were black, consistent with national trends [52]. In 2010, 22.5% of the population of Florida was Hispanic [51], indicating relatively low rates of violent injury in this population, as Hispanics made up only 13.3% of nonrecurrers and 8.8% of recurrers. Moreover, we found that Hispanic patients had lower risk of recurrence than did non-Hispanic black and white patients. This trend was seen but was nonsignificant in one prior single-center study [16]. Contrary to some studies [26,50], young age was not a risk factor for recurrence here, and older age appeared to increase odds of severe recurrence. However, compared with the state population median age of 40.7 years, violently injured individuals were younger overall.

Women were the minority of violently injured patients and were underrepresented compared with a state female population of 51.1% [53], but in contrast to most literature [14,16,21,23,25,26,54], we found similar recurrence rates for women and men. This may represent the large proportion of minor injuries included here, as men did have a higher rate of severe recurrence. These findings support the suggestion of Madden et al that demographics alone do not identify those patients at greatest risk for recurrence [10]. Patients with gunshot wounds had lower odds of recurrence. This may reflect greater severity of injury not fully captured elsewhere in our model that led to longer time to re- covery, postponing risk of reinjury. Conversely, studies have shown ele- vated mortality risk after recovery from firearm injury, which may have lowered our measured recurrence rate [4,5].

Several studies have found increased risk for recurrent trauma asso- ciated with alcohol and substance abuse [3,49,50,55], along with an op- portunity for intervention [9,18]. Our findings support efforts to incorporate alcohol and substance abuse screening and treatment into trauma care, as recommended by the American College of Surgeons Committee on Trauma [9,56]. Few studies have investigated mental health comorbidity, but our results are consistent with evidence that mental illness increases risk of recurrence [13,49] and that trauma in- creases risk of mental health challenges [57], and with the efforts of vi- olence intervention initiatives to incorporate behavioral health care [8,58-60]. Homelessness was rare in all groups, but the homeless are nonetheless substantially overrepresented at 1.1% of nonrecurrers and 3.3% of recurrers, compared with less than 0.01% of the state population [61]. More than half of recurrers had visits for unintentional injury. The only study that, to our knowledge, investigated this relationship found increased risk of violence in unintentionally injured patients [23]. These injury visits may represent additional opportunities for ED and trauma clinicians to intervene in these patients’ trajectory.

Limitations

We acknowledge several limitations to this study. In this retrospec- tive analysis of administrative data, information is necessarily limited, and causal inference cannot be drawn. Injuries may have been miscoded as intentional or unintentional, and patients’ identities may have been mis-registered. We could not account for injuries treated out-of-state.

Because the SID and SEDD cannot be linked to any other data source, in- cluding death certificate data, out-of-hospital deaths, including those caused by recurrent injury, could not be captured. Because these data- bases do not include outpatient providers, we cannot account either for injuries not treated in an ED or hospital, or for injured patients’ use of outpatient care for other health needs. Important opportunities to in- tervene with patients to prevent recurrent injuries may exist in the out- patient setting and deserve further investigation. Although these deaths are significant in that they represent the most severe potential outcome of recurrent violent injury, we expect the impact of excluding them to be small. Approximately 20% of deaths in Florida occur in the hospital, implying an additional 1080 deaths in this population [62]. Accounting for these deaths, we found that all recurrers would raise the rate of re- currence to 13.1%. Accounting for all these deaths, we found that nonrecurrers censored for inadequate follow-up would increase the re- currence rate to 11.3%. Likewise, the Centers for Disease Control and Prevention reported 794 assault-related, out-of-hospital deaths occur- ring in Florida from 2010 through 2012 [63]. If all of these deaths oc- curred in recurrers not otherwise captured in our analysis, this would raise the recurrence rate to 12.5%. It is not possible to determine the im- pact of out-of-hospital deaths on our assessment of risk factors for recurrence.

It was challenging to determine whether each visit was related to a separate injury event, and visits may have been misclassified as new as opposed to related to the same injury event. Using a restricted period necessarily underestimated recurrence, both because later recurrences could not be captured and because the index injury we identify may not have been a patient’s true, lifetime, first injury. Studies have consis- tently found recurrences beginning within 1 month after initial injury, and Cunningham et al [27] identified the first 6 months after injury as the highest risk period in youth, in part due to high risk for retaliatory violence in the immediate postinjury period [64]. One study found that 82% of recurrences occurred within 2 years [26], but recurrences have been reported for up to 30 years [20]. We expect that our 3-year time frame captured most recurrences, while remaining short enough that a violence prevention program could feasibly follow up patients for a similar period. Early presentation may itself indicate high risk, leading to more recurrences, but we did not discern any substantive dif- ferences between patients whose index injury occurred early as op- posed to late in the study period. We defined severe injuries in keeping with Florida trauma registry criteria, but these criteria are sub- jective, and many others could have been used. We excluded visits lack- ing revisit variables, which may have biased our results. Lastly, our cost data were estimated using cost-to-charge ratios derived for inpatient stays [65], and the results for ED visits may be less reliable.

Conclusions

Recurrent violent injury is a common and costly phenomenon, with an incidence of 11.1% within 2 to 3 years in this statewide analysis. Two- thirds of violently injured patients, including half of those with severe injuries, present to nontrauma centers. Effective violence prevention

programs that collaborate with all hospitals are needed to prevent re- current injury. The homeless and those with behavioral health needs are at high risk for recurrent injury and stand to benefit from focused Prevention efforts.

Supplementary data to this article can be found online at http://dx. doi.org/10.1016/j.ajem.2016.06.051.

References

1. US Centers for Disease Control and Prevention. Web-based injury statistics query and reporting system. http://www.cdc.gov/injury/wisqars/index.html. [accessed 12.10.14].
2. Dahlberg LL, Mercy JA. The history of violence as a public health issue. AMA Vitual Mentor 2009;11(2):167-72.
3. Sims DW, Bivians BA. Urban trauma: a chronic recurrent disease. J Trauma Injury In- fect Crit Care 1988;28(7):940-6. http://dx.doi.org/10.1097/00005373-198807000- 00092.
4. Fahimi J, Larimer E, Hamud-Ahmed W, Anderson E, Schnorr D, Yen I, et al. Long-term mortality of patients surviving firearm violence. Inj Prev 2015. http://dx.doi.org/10. 1136/injuryprev-2015-041710.
5. Rowhani-Rahbar A, Zatzick D, Wang J, Mills BM, Simonetti JA, Fan MD, et al. Firearm- related hospitalization and risk for subsequent violent injury, death, or crime perpetra- tion. Ann Intern Med 2015;162(7):492-500. http://dx.doi.org/10.7326/m14-2362.
6. Caputo ND, Shields CP, Ochoa C, Matarlo J, Leber M, Madlinger R, et al. Violent and fatal youth trauma: is there a missed opportunity? West J Emerg Med 2012; 13(2):146-50. http://dx.doi.org/10.5811/westjem.2011.6.6765.
7. Chong VE, Smith R, Garcia A, Lee WS, Ashley L, Marks A, et al. Hospital-centered vi- olence intervention programs: a cost-effectiveness analysis. Am J Surg 2014;209(4): 597-603. http://dx.doi.org/10.1016/j.amjsurg.2014.11.003.
8. Purtle J, Dicker R, Cooper C, Corbin T, Greene MB, Marks A, et al. Hospital-based vi- olence intervention programs save lives and money. J Trauma Acute Care Surg 2013; 75(2):331-3. http://dx.doi.org/10.1097/TA.0b013e318294f518.
9. Gentilello LM, Rivara FP, Donovan DM, Jurkovich GJ, Daranciang E, Dunn CW, et al. Alcohol interventions in a trauma center as a means of reducing the risk of injury re- currence. Ann Surg 1999;230(4):473-83.
10. Madden C, Garrett JM, Cole TB, Runge JW, Porter CQ. The urban epidemiology of re- current injury: beyond age, race, and gender stereotypes. Acad Emerg Med 1997; 4(8):772-5.
11. Dixon SD, Como JJ, Banerjee A, Claridge JA. Trauma recidivists: surprisingly better outcomes than initially injured trauma patients. Am J Surg 2013;207:423-31. http://dx.doi.org/10.1016/j.amjsurg.2013.09.019.
12. Morrissey TB, Byrds CR, Deitch EA. The incidence of recurrent penetrating trauma in an urban trauma center. J Trauma Injury Infect Crit Care 1991;31(11):1536-8. http://dx.doi.org/10.1097/00005373-199111000-00013.
13. Buss TF, Abdu R. Repeat victims of violence in an urban trauma center. Violence Vict 1995;10(3):183-94.
14. McCoy AM, Como JJ, Greene G, Laskey SL, Claridge JA. A novel prospective approach to evaluate trauma recidivism: the concept of the past trauma history. J Trauma Acute Care Surg 2013;75(1):116-21. http://dx.doi.org/10.1097/TA. 0b013e31829231b7.
15. Goins WA, Thompson J, Simpkins C. Recurrent intentional injury. J Natl Med Assoc 1992;84(5):431-5.
16. Chong VE, Lee WS, Victorino GP. Neighborhood socioeconomic status is associated with violent reinjury. J Surg Res 2015;199(1):177-82. http://dx.doi.org/10.1016/j. jss.2015.03.086.
17. Hankin A, Wei S, Foreman J, Houry D. Screening for violence risk factors identifies young adults at risk for return emergency department visit for injury. West J Emerg Med 2014;15(5):609-14. http://dx.doi.org/10.5811/westjem.2014.4.21275.
18. Rivara FP, Koepsell TD, Jurkovich GJ, Gurney JG, Soderberg R. The effects of alcohol abuse on readmission for trauma. JAMA 1993;270(16):1962-4.
19. Juillard C, Smith R, Anaya N, Garcia A, Kahn JG, Dicker R. Saving lives and saving money: Hospital-based violence intervention is cost-effective. J Trauma Acute Care Surg 2015;78(2):252-8. http://dx.doi.org/10.1097/TA.0000000000000527.
20. Kennedy F, Brown JR, Brown KA. Geographic and temporal patterns of recurrent in- tentional injury in south-central Los Angeles. J Natl Med Assoc 1996;88(9):970-2.
21. Schwarz DF, Grisso JA, Miles CG, Holmes JH, Wishner AR, Sutton RL. A longitudinal study of injury morbidity in an African-American population. JAMA 1994;271(10):755-60.
22. Griffin RL, Davis GC, Levitan EB, MacLennan PA, Redden DT, McGwin G. The effect of previous traumatic injury on homicide risk. J Forensic Sci 2014;59(4):986-90. http://dx.doi.org/10.1111/1556-4029.12416.
23. Litacker D. Preventing recurring injuries from violence: the risk of assault among Cleveland youth after hospitalization. Am J Public Health 1996;86(11):1633-6.
24. Brooke BS, Efron DT, Chang DC, Haut ER, Cornwell EE. Patterns and outcomes among penetrating trauma recidivists: it only gets worse. J Trauma 2006;61(1). http://dx. doi.org/10.1097/01.ta.0000224143.15498.bb.
25. Hedges BE, Dimsdale JE, Hoyt DB. Characteristics of repeat trauma patients, San Diego County. Am J Public Health 1995;85(7):1008-10.
26. Kaufmann CR, Branas CC, Brawley ML. A population-based study of trauma recidi- vism. J Trauma 1998;45(2):325-32.
27. Cunningham RM, Carter PM, Ranney M, Newton M, Woodman W, Zimmerman M, et al. Violent reinjury and mortality among youth seeking emergency department care for assault-related injury. JAMA Pediatr 2015;169:e96-104. http://dx.doi.org/ 10.1001/jamapediatrics.2014.1900.
28. Nathens AB, Jurkovich GJ, MacKenzie EJ, Rivara FP. A resource-based assessment of trauma care in the United States. J Trauma 2004;56:173-8. http://dx.doi.org/10. 1097/01.TA.0000056159.65396.7C.
29. Barrett M, Steiner C, Andrews R, Kassed C, Nagamine M. Methodological issues when studying readmissions and revisits using hospital adminstrative data. HCUP Methods Series Report # 2011-01; 2011.
30. Agency for Healthcare Research and Quality. Healthcare Cost and Utilization Project (HCUP). Available at: http://www.hcup-us.ahrq.gov. [Accessed January 1, 2015].
31. Agency for Healthcare Research and Quality. Data Element Availability Map: HCUP Revisit Variables. Available at: https://www.hcup-us.ahrq.gov/maps/revisit.jsp. [Accessed August 15, 2015].
32. Rising KL, Victor TW, Hollander JE, Carr BG. Patient returns to the emergency depart- ment: the time-to-return curve. Acad Emerg Med 2014;21(8):864-71. http://dx.doi. org/10.1111/acem.12442.
33. Agency for Healthcare Research and Quality. Clinical Classifications Software (CCS) for ICD-9-CM. Available at: https://www.hcup-us.ahrq.gov/toolssoftware/ccs/ccs. jsp#pubs. [Accessed May 14, 2015].
34. US Centers for Disease Control and Prevention. Recommended framework for pre- senting injury mortality data. MMWR 1997;46(RR-14):1-33.
35. US Centers for Disease Control and Prevention. Strategies to improve external cause- of-injury coding in state-based hospital discharge and emergency department data systems. MMWR Recom Rep 2008;57(RR01):1-15 [http://www.cdc.gov/mmwr/ preview/mmwrhtml/rr5701a1.htm Accessed February 25, 2016].
36. The Trust for America’s Health. The facts hurt: a state-by-state injury prevention pol- icy report; 2013[Washington, DC].
37. Barrett M, Steiner C. Addendum to Healthcare Cost and Utilization Project (HCUP) E- code evaluation report (updated for 2009). HCUP Methods Series Report #2004-06. U.S.: Agency for Healthcare Research and Quality; 2011 [http://www.hcup-us.ahrq. gov/reports/methods/methods.jsp Accessed February 25, 2016].
38. LeMier M, Cummings P, West TA. Accuracy of external cause of injury codes report- ed in Washington state hospital discharge records. Inj Prev 2001;7:334-8. http://dx. doi.org/10.1136/ip.7.4.334.
39. Clark DE, Osler T, Hahn DR. ICDPIC: Stata module to provide methods for translating International Classification of Diseases (Ninth Revision) diagnosis codes into standard injury categories and/or scores. Available at: https://ideas.repec.org/c/boc/bocode/ s457028.html. [Accessed July 20, 2015].
40. Greene NH, Kernic MA, Vavilala MS, Rivara FP. Validation of ICDPIC software injury severity scores using a large regional trauma registry. Inj Prev 2015;21(5):325-30. http://dx.doi.org/10.1136/injuryprev-2014-041524.
41. Schafer JL, Graham JW. Missing data: Our view of the state of the art. Psychol Methods 2002;7(2):147-77. http://dx.doi.org/10.1037//1082-989X.7.2.147.
42. National Trauma Data Bank. Florida State Information. Available at: http://www. ntdsdictionary.org/ntdbParticipants/stateInformation/florida.html;. [Accessed Noverber 8, 2015].
43. Yu T-C, Zhou H. Evaluation of techniques for handling missing cost-to-charge ratios in the USA Nationwide Inpatient Sample: a simulation study. J Comp Effect Res 2015;4:473-83. http://dx.doi.org/10.2217/cer.15.28.
44. MacKenzie EJ, Steinwachs DM, Ramzy AI. Evaluating performance of statewide re- gionalized systems of trauma care. J Trauma 1990;30(6):681-8.
45. National Network of Hospital-Based Violence Intervention Programs. Available at: http://nnhvip.org/network-membership/. [Accessed September 16, 2015].
46. Karraker N, Cunningham R, Becker M, Fein JA, Knox LM. Violence is preventable: a best practices guide for launching & sustaining a hospital-based program to break the cycle of violence. Chicago, IL: Youth ALIVE!; 2011.
47. Cunningham R, Knox L, Fein JA, Harrison S, Frisch K, Walton M, et al. Before and after the trauma bay: the prevention of violent injury among youth. Ann Emerg Med 2009;53:490-500. http://dx.doi.org/10.1016/j.annemergmed.2008.11.014.
48. Shibru D, Zahnd E, Becker M, Bekaert N, Calhoun D, Victorino GP. Benefits of a hospital-based peer intervention program for violently injured youth. J Am Coll Surg 2007;205(5):684-9. http://dx.doi.org/10.1016/j.jamcollsurg.2007.05.029.
49. Smeltzer SC, Redeker NS. A framework of trauma and trauma recidivism in adoles- cents and young adults. J Trauma Nurs 1995;2:93-9.
50. Cooper C, Eslinger D, Nash D, al-Zawahri J, Stolley P. Repeat victims of violence: re- port of a large concurrent case-control study. Arch Surg 2000;135(7):837-43.
51. United States Census Bureau. http://www.census.gov. [Accessed February 9, 2015].
52. Kalesan B, Vasan S, Mobily ME, Keiser O, Fagan JA, Galea S. State-specific, racial and ethnic heterogeneity in trends of firearm-related fatality rates in the USA from 2000 to 2010. BMJ Open 2014;4:1-9. http://dx.doi.org/10.1136/bmjopen-2014-005628.
53. U.S. Census Bureau. 2010 Census of population and housing, population and housing unit counts, CPH-2-11, Florida, US. Washington, DC: Government Printing Office; 2012.
54. Greenberg M, Schneider D. Violence in American cities: young black males in the an- swer, but what was the question? Soc Sci Med 1994;39(2):179-87. http://dx.doi. org/10.1016/0277-9536(94)90326-3.
55. Claassen CA, Larkin GL, Hodges G, Field C. Criminal correlates of injury-related emer- gency department recidivism. J Emerg Med 2007;32(2):141-7. http://dx.doi.org/10. 1016/j.jemermed.2006.05.041.
56. American College of Surgeons Committee on Trauma. Resources for Optimal care of the injured patient. Chicago, IL: American College of Surgeons; 2014.
57. Richmond TS, Kauder D, Strumpf N, Meredith T. Characteristics and outcomes of se- rious traumatic injury in older adults. J Am Geriatr Soc 2002;50:215-22. http://dx. doi.org/10.1046/j.1532-5415.2002.50051.x.
58. Cheng TL, Haynie D, Brenner RA, Wright JL. Effectiveness of a mentor-implemented, violence prevention intervention for assault-injured youths presenting to the emer- gency department: results of a randomized trial. Pediatrics 2008;122:938-46. http://dx.doi.org/10.1542/peds.2007-2096.

    Carr BG, Wiebe DJ, Richmond TS, Cheney R, Branas CC. A randomised controlled fea- sibility trial of alcohol consumption and the ability to appropriately use a firearm. Inj Prev 2009;15(6):409-12. http://dx.doi.org/10.1136/ip.2008.020768.

59. Rich JA, Grey CM. Pathways to recurrent trauma among young black men: traumatic stress, substance use, and the “code of the street”. Am J Public Health 2005;95(5): 816-24. http://dx.doi.org/10.2105/AJPH.2004.044560.
60. U.S. Census Bureau. 2010 Census special reports, the emergency and transitional shelter population: 2010, C2010SR-02. Washington, DC: U.S. Government Printing Office; 2012.
61. Dartmouth Atlas of Healthcare: Percent of Deaths Occurring in Hospital. Available at: http://www.dartmouthatlas.org/data/table.aspx?ind=15. [Accessed August 25, 2015].
62. U.S. Centers for Disease Control and Prevention. Compressed Mortality File 1999- 2013 on CDC WONDER Online Database. Compressed Mortality File 1999-2013 20;2S. Available at: http://wonder.cdc.gov/cmf-icd10.html; 2014. [Accessed April 2, 2015].
63. Wiebe DJ, Blackstone MM, Mollen CJ, Culyba AJ, Fein JA. Self-reported violence- related outcomes for adolescents within eight weeks of emergency department treatment for assault injury. J Adolesc Health 2011;49:440-2. http://dx.doi.org/10. 1016/j.jadohealth.2011.01.009.
64. Agency for Healthcare Research and Quality. Cost-to-Charge Ratio Files. Healthcare Cost and Utilization Project (HCUP). Available at: www.hcup-us.ahrq.gov/db/state/ costtocharge.jsp; 2014. [Accessed February 25, 2015].