Injury Patterns and Hospital Admission After Trauma Among People Experiencing Homelessness

Key Points Question Is homelessness associated with hospital admission following injury? Findings In this national cohort study of 12 266 people experiencing homelessness (PEH) from the American College of Surgeons Trauma Quality Improvement Program, PEH demonstrated significantly increased adjusted odds of hospital admission after injury compared with housed patients. Meaning These findings suggest that potential challenges in facilitating safe discharge from the emergency department may lead to increased hospital admission after injury for PEH.


Introduction
An estimated 580 000 people experienced homelessness in the US on any given night in 2020, with increasing volume during the COVID-19 pandemic. 1Lack of stable housing is an important healthrelated social risk factor. 2,3[11] Traumatic injury accounts for up to 28% of mortality among PEH. 12,13Despite the high incidence, to our knowledge, no national study of the epidemiology and management of traumatic injury among PEH has been conducted.5][16] However, injury mechanism can frequently vary between trauma centers and geographic regions.Further exploration of common injury mechanisms could inform injury prevention efforts for PEH.Additionally, although PEH presenting to the ED with physical and behavioral health symptoms have higher rates of hospital admission compared with housed patients, what happens to traumatically injured PEH after ED presentation has not been studied at the national level. 4,17ED disposition after physical trauma is particularly relevant among PEH because the lack of housing contributes to a continued risk of additional injury.
In this nationwide cohort study, we aimed to define the epidemiology of traumatic injury and subsequent hospital use among PEH compared with housed patients.Specifically, we sought to (1)   characterize injury patterns among PEH sustaining traumatic injury, (2) evaluate the associations of housing status with hospital admission, and (3) conduct an a priori subanalysis of PEH compared with low-income housed patients who may experience similar inequities.We hypothesized that limited options for safe ED discharge would lead to increased hospital admission among PEH compared with both all housed and low-income housed patients.

Data Source
This was a retrospective observational cohort study of patients in the American College of Surgeons (ACS) Trauma Quality Improvement Project (TQIP).TQIP is a nationwide traumatic injury registry containing more than 7.5 million incident-based encounters of trauma activations at participating hospitals.Encounters were submitted voluntarily by over 750 facilities across the US and Canada.
Most participating facilities were ACS-verified level I or II trauma centers, although nontrauma centers were also included.Although TQIP aims to include patients with at least 1 severe injury (an Abbreviated Injury Scale score of Ն3 in at least 1 body region), patients with less severe injuries are also included.Patient, injury, and hospital data were recorded by trained dedicated abstractors. 18rthwestern University's institutional review board approved the project.Informed consent was not needed because the data were anonymous, in accordance with 45 CFR §46.This study adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.

JAMA Network Open | Emergency Medicine
Injury Patterns and Hospital Admission After Trauma Among People Experiencing Homelessness

Study Population
Adult patients aged 18 years or older who presented following injury to participating TQIP hospitals from January 1, 2017, to December 31, 2018, were identified.We excluded 14 606 encounters with no signs of life upon presentation.Observations with missing ED discharge disposition data were excluded (50 531 observations), because this field determined the primary outcome.Encounters that left against medical advice (5328 observations) were also excluded, as their disposition was not determined by clinician decision (eFigure in Supplement 1).
Given the marked systematic differences in demographic, clinical, and injury characteristics by housing status, we defined 2 further subcohorts to address potential confounding on discharge disposition.The first was a subcohort of PEH propensity score-matched to all housed patients.The second was a subcohort of PEH compared with low-income housed patients.TQIP does not contain income data, and we used Medicaid insurance as a proxy for low income, because Medicaid eligibility is determined largely by income at or below the federal poverty limit. 19,20

Exposure of Interest
The exposure of interest was documented housing status.PEH were identified using TQIP's alternate home residence variable, which is completed for individuals who do not have a temporary or permanent residence ZIP code listed on identification documents.Clinical abstractors are trained to record these patients' ZIP code as not applicable and complete the alternate home residence variable.When patients did have a ZIP code listed, this variable was not completed and we classified them as housed.Encounters for which the alternate home residence variable was recorded as undocumented citizen or migrant worker rather than homeless were also considered to be housed.

Outcomes
The primary outcome was hospital admission created as a binary variable (admitted vs not admitted).
Hospital admission was derived from TQIP's ED discharge disposition variable; categories included observational unit, inpatient unit, transfer, home, other facility, and deceased.Patients discharged home from the ED account for approximately 9% of records in TQIP.Patients were considered admitted if they were admitted to an observational or inpatient unit or if they were transferred to another hospital.We considered patients not admitted if their disposition was home or other facility.TQIP's other facility category includes jail, institutional care, and mental health facilities.Although it may have been appropriate to consider individuals admitted for psychiatric care to have been admitted, TQIP lacks the granularity to distinguish these individuals from those admitted to a nonmedical facility such as a jail.Thus, we considered all individuals with this disposition of other facility to have been not admitted.Those who died in the ED were excluded from the final analysis of hospital admission.

Covariates
TQIP contained demographic, clinical, and hospital data.Demographic variables included in this study were sex, age, race, ethnicity, and insurance status.Race and ethnicity were characterized using separate race and ethnicity variables as Hispanic, non-Hispanic Black, non-Hispanic White, and other (ie, American Indian, Asian, and Pacific Islander per the National Trauma Databank definitions).
Race and ethnicity data in TQIP are based on self-report or report of a family member. 21Racial and ethnic disparities in admission have been widely documented elsewhere 22 ; therefore, we controlled for race and ethnicity in the current study.Clinical comorbidities in TQIP include standard Elixhauser physical and behavioral health comorbidities. 23Physical health comorbidities in this study were heart disease, hypertension, chronic obstructive pulmonary disease, chronic kidney disease, diabetes, malignant neoplasms, and liver disease.Behavioral health comorbidities included schizophrenia, bipolar disorder, major depressive disorder, social anxiety disorder, posttraumatic stress disorder, and antisocial personality disorder. 21Injury characteristics included alcohol and drug positivity, trauma type (blunt, penetrating, or other), mechanism, intent (unintentional, self-inflicted, assault,

JAMA Network Open | Emergency Medicine
Injury Patterns and Hospital Admission After Trauma Among People Experiencing Homelessness or other), body region, Injury Severity Score (ISS), and Glasgow Coma Scale (GCS) score.Injury body region was identified using International Statistical Classification of Diseases and Related Health Problems, Tenth Revision diagnosis codes in accordance with the Injury Mortality Diagnosis Matrix. 24S and GCS scores were grouped into clinically meaningful categories (ISS: minor injury, 1-8; moderate, 9-15; and severe, Ն16; GCS score: severe head injury, 3-8; moderate, 9-12; and minor, 13-15).25,26 Trauma center designation was determined using ACS verification level, and state designation was used when verification was unavailable.When hospital-level data were known for some encounters and missing for others in the same hospital, missing hospital variable data were imputed under the assumption that hospital-level data would be the same for all encounters at a given hospital.Missing patient-level data were not imputed.

Statistical Analysis
Data were analyzed from December 2021 to November 2022.ED discharge disposition was compared between PEH and housed patients using χ 2 tests.We then fitted hierarchical multivariable logistic regression models with hospital-level random intercepts to assess the association between being unhoused and odds of admission.These models controlled for age, race, insurance, trauma type, GCS score, intent, and ISS.All tests were 2-sided with α < .05indicating statistical significance.
Analyses were performed using Stata MP statistical software version 17.0 (StataCorp).
Two separate subgroup analyses were performed to reduce the effect of confounding.The first was an analysis of PEH propensity score-matched to all housed patients. 27Details of the matching process are described in the eMethods in Supplement 1. 28 PEH and housed patients were matched on sex, age, insurance, injury type, body region, and physical and/or behavioral health comorbidity.
Hospital identity was used as an exact matching criterion to account for hospital-level differences in admission practices.Postmatch characteristics were compared using standardized differences and showed nearly complete matching of PEH with better balanced distributions of demographic and clinical (eTable 1 in Supplement 1) as well as injury characteristics (eTable 2 in Supplement 1).

Stratified analysis compared differences in admission rates among matched pairs across ISS with
McNemar tests.Hierarchical logistic regression clustering by both matched pairs and hospitals evaluated the association between homelessness and odds of admission.
The second subgroup analysis evaluated differences between PEH and low-income housed patients to account for socioeconomic status as a potential confounder.Demographic and injury characteristics were compared with χ 2 tests.Unadjusted admission rates were compared between groups.Hierarchical multivariable logistic regression with hospital-level random effects was used to assess associations between homelessness and admission.These models controlled for age, race, trauma type, GCS score, intent, and ISS.Subgroup analyses were hypothesis generating; results were not adjusted for multiple testing.

Results
There

Discussion
More than 500 000 people experience homelessness each night in the US. 1 To our knowledge, this cohort study is the first nationwide study of traumatic injury among PEH.We found that PEH more frequently sustained pedestrian-strike injury and assault than housed patients.PEH were significantly more likely to be admitted to the hospital compared with both all housed patients and low-income housed patients.Differences in admission rates were greatest in cases of minor injury, potentially when disposition is influenced by clinical discretion.
Our findings of injury characteristics among PEH build on other single-center studies. 29[36] However, these studies were single center or city focused.Exposure to different mechanisms of Our study is the first to elucidate national injury patterns among PEH.Similar to the high rates of assault in our study, Kushel et al 37 found that up to 30% of PEH in San Francisco experience intentional physical assault.9][40] Being a target of violence has been associated with increased ED use and poor health. 14,41In addition, homicides are a common cause of death among unhoused young adults. 9,13,42Our findings can be used to inform injury prevention initiatives for PEH.
Although the association between homelessness and hospitalization has been shown for other diagnoses, it has not been demonstrated in trauma patients. 4,29Reasons for increased adjusted odds of admission of injured PEH were likely multifactorial.A major factor may be clinicians' perceived risks of discharging an unhoused person back to the street.Qualitative studies have shown that ED clinicians consider safety concerns when deciding to admit PEH with low medical acuity. 43The persistence of increased odds of admission among PEH compared with low-income housed patients suggests that the observed associations are due to unstable housing independently of other health-related social needs.For many PEH, discharge home would mean returning to the place where they were injured, potentially putting them at risk for reinjury.Furthermore, wound care and follow-up can be challenging in unstable housing conditions. 6,44

Implications
Our findings have several implications.Injury prevention efforts among PEH must be tailored to the unique injury patterns demonstrated, particularly the striking vulnerability to assault.High rates of admission suggest that hospitals are acting as social safety nets for injured PEH.Hospitals disproportionately admit PEH, likely recognizing the challenges of recovering and preventing reinjury in unstable housing.Admitted PEH may receive needed health and social service resource referrals. 45r results underscore hospitals' current role in providing wraparound social care services.This has important implications given the rising cost of health care as well as the inability of most acute care hospitals to provide such services to patients once they return to the community.

Limitations
This study has several limitations.First, TQIP contains records from mostly level I and II trauma centers, and results may not be generalizable to PEH presenting to nonspecialized centers.However, the inclusion of nontrauma centers improves the generalizability of our findings.Similarly, records in TQIP demonstrate high rates of admission, and results may not be generalizable to hospitals that see more minor injuries and admit a lower proportion of injured patients.However, it is likely that differences between PEH and housed patients would be more pronounced in hospitals with greater variation in admission practices.Selection bias may have been introduced in the severity of injury captured, as well as clinician bias to admit patients.Nonetheless, we conducted multivariable regression and propensity score matching to control for variables that could affect bias in admission.
Second, hospital and state policy may affect whether discharge home from the ED is permissible for PEH.However, we accounted for this with the use of hospital ID as an exact matching criterion in the propensity-matched analysis.Third, the alternate home residence variable used to identify PEH was completed when patients' ZIP code of primary residence was unknown.This variable would not capture patients who were temporarily unhoused or those whose documents listed a shelter, former residence, or home of a family member.This could have led to misclassification bias and likely underestimated the prevalence of PEH.However, demographic characteristics of patients identified as PEH were similar to those described in other studies, suggesting specificity of our PEH cohort. 33,36e alternate home residence variable likely identifies those who chronically experience homelessness.Fourth, TQIP does not contain socioeconomic status or income data, which are known factors associated with injury mechanism and outcomes. 46,47We addressed this by conducting a subanalysis of a low-income cohort as defined by Medicaid insurance, because eligibility is largely determined according to an individual's Modified Adjusted Gross Income. 19,20Fifth, TQIP's ED discharge disposition category of other facility included a wide range of facilities, such as jail and mental health facilities.We considered individuals with this disposition to not have been admitted.
This disposition was more common among PEH, likely because a larger proportion of PEH were admitted to psychiatric care. 48Thus, we may be underestimating the percentage of PEH who received medical care, including psychiatric care, after injury.

Conclusions
This national cohort study demonstrated that PEH are more likely than housed individuals to experience assault and pedestrian-strike injuries.Injury prevention efforts among PEH must be tailored to these unique injury patterns.PEH had increased adjusted odds of hospital admission after injury.These findings underscore potential opportunities for policy and social programming initiatives to improve the care and hospital use of injured PEH.

Figure . a
Figure.Differences in Rates of Admission Between Injured People Experiencing Homelessness (PEH) and Propensity-Matched All Housed Patients by Injury Severity Score4

Table 1 .
Demographic and Clinical Characteristics of Injured PEH and All Housed Patients a P values were derived from χ 2 tests of independence.bReferstoAmerican Indian, Asian, and Pacific Islander.cPhysicalcomorbidities include heart disease, hypertension, chronic obstructive pulmonary disease, chronic kidney disease, diabetes, malignant neoplasm, or liver disease.dBehavioral comorbidities include schizophrenia, bipolar disorder, major depressive disorder, social anxiety disorder, posttraumatic stress disorder, and antisocial personality disorder.

Table 4 )
compared with housed patients when controlling for age, race, insurance, trauma type, GCS score, intent, and ISS.

Table 3 .
Unadjusted Emergency Department Discharge Disposition Rates of Injured PEH and All Housed Patients a P < .001forall comparisons (χ 2 test of independence).bOther facility includes jail, institutional care, and mental health facilities.

Table 4 .
Multivariable Model for Hospital Admission in Injured People Experiencing Homelessness and All Housed Patients Refers to American Indian, Asian, and Pacific Islander.

on 09/28/2023 injury
is heavily influenced by local factors, such as weather, traffic, recreational activities, and local laws.Thus, results from a single center or city cannot be generalizable to other areas.
Demographic and Clinical Characteristics of the Matched Cohort of People Experiencing Homelessness and All Housed Patients eTable 2. Injury Characteristics of the Matched Cohort of People Experiencing Homelessness and All Housed Patients eTable 3. Injury Characteristics of Injured People Experiencing Homelessness and Low-Income Housed Patients eTable 4. Multivariable Models for Hospital Admission in Injured People Experiencing Homelessness and Low-Income Housed Patients