Morbidity and Length of Stay After Injury Among People Experiencing Homelessness in North America

Key Points Question Do people experiencing homelessness have increased morbidity and length of hospital stay after injury compared with housed patients? Findings In this cohort study of 1 441 982 patient encounters from the American College of Surgeons Trauma Quality Programs, people experiencing homelessness demonstrated similar rates of morbidity compared with propensity score–matched housed patients, but with a longer adjusted length of stay (LOS). The association between homelessness and increased LOS was greatest among those 65 years and older and with minor injury. Meaning These findings underscore the challenges in providing safe hospital discharge for people experiencing homelessness after injury, leading to prolonged LOS.


Introduction
Approximately 580 000 people experienced homelessness on any given day in 2020 across the US, and this incidence has increased since the start of the COVID-19 pandemic. 18][9][10] Traumatic injury is the second leading reason for hospitalization among people experiencing homelessness. 11A previous study 9 demonstrated that injured people experiencing homelessness are more likely to be admitted to the hospital than housed individuals.However, little is known about their subsequent hospital course and the incidence of morbidity, surgical procedures, and intensive care unit (ICU) admissions. 3Length of stay (LOS) is a measure of resource use that closely approximates total cost of care. 12Prolonged LOS impairs bed turnover and can lead to decreased capacity and strain in the emergency department and wards.Increased morbidity and LOS among people experiencing homelessness admitted for physical and behavioral health complaints have been documented. 10,13However, injury admissions differ from other admissions because injury often causes new physical disabilities, wounds, and pain.Furthermore, many studies were limited to single centers, limiting generalizability. 14Hospital course among injured people experiencing homelessness has not been studied nationally.Defining which patients have the greatest increase in LOS can help tailor discharge planning efforts.This cohort study sought to evaluate hospital course after injury among people experiencing homelessness compared with housed patients in North America.Our objectives were to (1) assess the incidence of morbidity, surgical intervention, and ICU admission in people experiencing homelessness compared with housed patients; (2) investigate associations between homelessness and LOS; and (3) evaluate whether age and injury severity modified the association.

Methods
The Northwestern University Institutional Review Board approved this cohort study and waived the need for informed consent owing to the use of deidentified data.The study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

Data Source
We conducted a retrospective cohort study of the American College of Surgeons (ACS) Trauma Quality Programs (TQP) data.The ACS TQP includes data from ACS Trauma Quality Improvement Program-participating centers as well as a small number of non-ACS-verified centers that adhere to the National Trauma Data Standard.Together, these data represent an incident-based injury registry of over 7.5 million records.Hospitals participating in the TQP are located across the US and Canada.
They are mostly ACS-verified level I or level II trauma centers or regionally designated trauma centers.Dedicated, trained abstractors recorded patient demographic, clinical, injury, and hospital data. 15

JAMA Network Open | Equity, Diversity, and Inclusion
Morbidity and Length of Stay After Injury Among People Experiencing Homelessness
We defined a subcohort to address a priori known differences in demographic, clinical, and injury characteristics between people experiencing homelessness and housed patients.Many of these characteristics are also associated with increased morbidity and LOS. 17 This subcohort consisted of people experiencing homelessness matched by propensity score to housed patients as described below.

Exposure
Our exposure of interest was homelessness.People experiencing homelessness were identified using TQP's alternate home residence variable (including homeless individuals, undocumented citizens, migrant workers).The TQP abstractors recorded this variable when patients did not have a temporary or permanent residence zip code listed on presentation to the emergency department.
We classified those listed as homeless as people experiencing homelessness, while undocumented citizens and migrant workers were considered housed.The alternate home residence variable was not completed when a patient had a residence zip code listed; we labeled these records as housed.
This variable has been used in another study of injured people experiencing homelessness by Silver et al. 9

Outcomes of Interest
Our primary outcome was LOS (in days).We evaluated LOS in 2 ways.First, we evaluated LOS as a continuous, nonnormally distributed outcome variable.Second, we evaluated LOS as a binary variable of whether or not LOS was greater than 30 days.Hospitals' trauma center status was defined by ACS verification level or, for those not ACS verified, state designation.When hospital-level data were known for some encounters and missing for others with the same hospital identification, missing hospital data were imputed under the assumption that hospital data would be the same for all patients admitted to a given hospital (<1% of encounters).

Moderation Analysis
We assessed moderation effects of age and Injury Severity Score (ISS) on the association between homelessness and LOS.These variables were selected due to increasing age of the US homeless population and prior studies demonstrating differential associations between homelessness and hospital admission by ISS. 9,26Age was considered categorically (18-35, 36-50, 51-64, and Ն65 years), given the highly skewed distribution.The ISS was also considered categorically using standard categories for minor (1-8), moderate (9-15), and severe (Ն16) injury. 23

Propensity Score Matching
We used propensity score matching to reduce confounding given known differences between people experiencing homelessness and housed patients. 27We used the Stata package psmatch2 Downloaded from jamanetwork.comby guest on 03/06/2024 differences between hospitals.Propensity matching was performed using a greedy 1:1 nearestneighbor algorithm with a specified maximum caliper width of 0.5. 28Postmatch comparisons with standardized differences demonstrated a near complete match of people experiencing homelessness with better balanced distributions for demographic, clinical, and injury characteristics, though differences in age distributions persisted (eTable 1 in Supplement 1).A standardized difference less than 0.20 was considered sufficient balance and less than 0.10 was considered ideal. 21,29

Statistical Analysis
In the unmatched cohort, we compared demographic, clinical, hospital, and injury characteristics between people experiencing homelessness and housed patients using χ 2 tests of independence.
Unadjusted rates of morbidity, hemorrhage control surgery, and ICU admission were compared using χ

Unmatched Analysis
There

Matched Analysis
The propensity score-matched cohort consisted of 8665 pairs admitted to 378 hospitals.Matched people experiencing homelessness had a trend toward increased rates of surgical site infection (25  4).This moderation of age on the direct association between homelessness and adjusted LOS was statistically significant (P < .001).The difference in unadjusted median LOS between people experiencing homelessness and housed patients was 1 day across all ISS categories (Table 4).However, on multivariable analysis, the direct association between homelessness and LOS was greatest for those with an ISS of 1 to 8 (minor injury), with a 30.0%increased adjusted LOS (IRR, 1.30 [95% CI, 1.25-1.35]),while people experiencing homelessness with an ISS of 16 or greater had an associated 14.4% increased adjusted LOS (IRR, 1.14 [95% CI, 1.09-1.20])(Table 4).This moderation of ISS on the direct association between homelessness and adjusted LOS was statistically significant (P < .001).

Discussion
Traumatic injury is the second-leading cause of hospitalization and a common cause of death for people experiencing homelessness. 11,30The hospital course among people experiencing  homelessness remains poorly understood despite increasing recognition of housing as a healthrelated social need. 2 To our knowledge, this study constitutes the first evaluation of hospital course and LOS in North America among people experiencing homelessness who are discharged alive following traumatic injury.We found that unmatched people experiencing homelessness had higher unadjusted rates of morbidity, hemorrhage control surgery, and ICU admission compared with housed patients, though these differences did not persist in the matched cohort.Nonetheless, matched people experiencing homelessness had an associated 22.1% increased adjusted LOS compared with housed patients.The relative increase in adjusted LOS was greatest among people experiencing homelessness who were 65 years or older and among those with minor injury.
Few studies have examined morbidity and incidence of surgery among injured people experiencing homelessness.A study by Decker et al 31 found that people experiencing homelessness and undergoing emergent surgery had comparable rates of in-hospital mortality and postoperative complications compared with housed patients.However, their study did not include patients with traumatic injury.In our study, the fact that differences in morbidity and surgery between people experiencing homelessness and housed patients did not persist in a balanced matched cohort suggests that differences observed in the unmatched cohort were due to high rates of underlying comorbidities and increased injury severity in people experiencing homelessness.
3][34] Reasons for increased LOS among people experiencing homelessness are multifactorial.It may be driven in part by low socioeconomic status; patients with social and/or material deprivation have been shown to have longer LOS compared with patients with higher socioeconomic status. 35Prior studies have also shown that up to 80% of avoidable hospital days are due to delays in accessing postdischarge care. 36People experiencing homelessness may not have caregivers to assist them after discharge and may lack resources to pay for postdischarge care. 37,38Length of stay may also be prolonged by concomitant physical health and behavioral health comorbidities. 39High proportions of people experiencing homelessness have public health insurance, which has been linked to increased LOS in other studies. 40Admitted people experiencing homelessness may also receive health and social service resource referrals, which may prolong LOS. 41While this study investigated LOS among traumatically injured people experiencing homelessness, results of increased LOS are likely generalizable to people experiencing homelessness hospitalized for other conditions.
Importantly, this study identifies groups of people experiencing homelessness with disproportionately increased LOS: those 65 years or older and those with minor injuries.3][44] Geriatric patients have been shown to have more complications and longer LOS than younger patients due to frailty and comorbidities. 45Older patients also frequently lack familial caregiver support. 35These challenges are likely amplified in older unhoused patients, leading to greater disparities.Our findings of increased associations between homelessness and hospital use for patients with minor injury are consistent with those of previous work.In 1 study, 9 differences in rates of hospital admission between people experiencing homelessness and housed patients were greater for patients with minor injuries than patients with severe injuries.The timing of discharge for patients with minor injury may be more clinically discretionary and therefore more susceptible to prolongation by vulnerabilities such as homelessness.Therefore, there may be greater opportunity to reduce disparities in LOS among people experiencing homelessness with minor injury.People experiencing homelessness would benefit from programs to make discharge safe and feasible, and there is increasingly a business case for society to invest in interventions that address social determinants of health. 47,48These may comprise initiatives to promote affordable housing, improved ambulatory care access, expanded respite programs to support safe and efficient discharge, and care coordination services. 49The results of our study suggest that older people experiencing homelessness and those with minor injuries in particular may benefit most from efforts to reduce LOS.

Limitations
There are several limitations to this study.Many are inherent to the use of national registry datasets.
These data are retrospectively analyzed after being collected for measurement of hospital quality.
First, there could be unobserved variable bias.However, rigorous data abstraction constitutes an important strength of such data, particularly for trauma-specific fields such as ISS. 20Second, misclassification bias may have been introduced in our use of the TQP's alternate home residence variable to define our cohort of people experiencing homelessness.This variable is completed when a patient's residential zip code is unknown.It would not capture patients whose documents list the zip code of a former residence or shelter or patients who are temporarily unhoused.The TQP likely underestimates the prevalence of people experiencing homelessness.However, demographic and injury characteristics of the cohort experiencing homelessness are similar to those in other studies, suggesting face validity of the cohort. 33,50Third, TQP does not contain income data, a known factor associated with LOS in patients with traumatic injury. 35We addressed this by including insurance as a matching criteria in our propensity score model.Insurance status is a proxy for income in US-based studies, as eligibility is largely based on modified adjusted gross income. 51,52

Conclusions
This findings of this cohort study suggest that people experiencing homelessness in North America had significantly increased adjusted LOS after hospitalization for traumatic injury compared with housed patients.Older patients and those with minor injuries had disproportionately greater increased adjusted LOS.These findings have significant implications for quality and costs of care for people experiencing homelessness and underscore potential opportunities to reduce disparities in trauma outcomes and improve hospital resource use among those with injuries.

Figure
Figure.Patient Selection Schema

SUPPLEMENT 1 . eTable 1 . 2 . 3 .
Characteristics of 8665 Matched Pairs of People Experiencing Homelessness and Housed Patients in the Trauma Quality Programs, 2017 to 2018 eTable Unadjusted Morbidity, Surgery, and Length of Stay Among 8665 Matched Pairs of People Experiencing Homelessness and Housed Patients in the Trauma Quality Programs, 2017 to 2018 eTable Multivariable Model for Length of Stay Greater Than 30 Days in 8665 Matched Pairs of People Experiencing Homelessness and Housed Patients in the Trauma Quality Programs, 2017 to 2018 SUPPLEMENT 2. Data Sharing Statement

1 432 917 Housed patients 9065 People experiencing homelessness 8665 People experiencing homelessness 8665 Housed patients 599 724 Excluded 232 249 Aged <18 y 246 467 Were
23,24We used International and Statistical Classification of Diseases and Related Health Problems, Tenth Revision diagnosis codes to classify injury body region according to the Centers for Disease Control and Prevention Injury Mortality Diagnosis Matrix.25Missingpatient-level data were not imputed.
2tests of independence.Unadjusted LOS was compared between people experiencing homelessness and housed patients using Mann-Whitney tests.Rates of LOS longer than 30 days were calculated and compared with χ 2 tests of independence.
After matching, rates of morbidity, hemorrhage control surgery, and ICU admission were compared between people experiencing homelessness and matched housed patients using McNemar tests.Unadjusted LOS was compared between matched people experiencing homelessness and housed patients using the Wilcoxon signed rank test.Rates of LOS longer than 30 days were compared with McNemar tests.All tests were 2 sided with α < .05consideredstatisticallysignificant.In the matched cohort, we estimated hierarchical multivariable negative binomial regression to evaluate associations between homelessness and adjusted LOS.We used hierarchical multivariable logistic regression to evaluate the association between homelessness and odds of adjusted LOS longer than 30 days.Models included random effects at the level of both propensity score-matched pairs and hospitals while controlling for age, ISS, morbidity, hemorrhage control surgery, and ICU admission.We created interaction terms to evaluate whether age and/or injury severity moderated the association between homelessness and LOS.These interaction terms were included in 2 separate additional hierarchical multivariable negative binomial regression models.The first interaction term was homelessness × categorical age.The second interaction term was homelessness × categorical ISS.These interaction terms tested whether the association between homelessness and LOS was strengthened or weakened by the moderator variable.We clustered observations by both propensity score-matched pairs and hospital identification while controlling for age, ISS, any morbidity, hemorrhage control surgery, and ICU admission.Data were analyzed from February 1, 2022, to May 31, 2023.Analyses were performed using Stata MP, version 17.0 (StataCorp LLC).

Table 1 .
Characteristics of Unmatched Cohort of People Experiencing Homelessness and Housed Patients in the Trauma Quality Programs, 2017 to 2018 b Derived from χ 2 tests of independence.c Includes American Indian or Alaska Native, Asian, and Native Hawaiian or Other Pacific Islander.d Patients may present with injury to more than 1 body region.e Higher scores indicate better response.JAMA Network Open | Equity, Diversity, and Inclusion Morbidity and Length of Stay After Injury Among People Experiencing Homelessness JAMA Network Open.2024;7(2):e240795.doi:10.1001/jamanetworkopen.2024.0795(Reprinted) February 28, 2024 6/14 Downloaded from jamanetwork.comby guest on 03/06/2024

Table 3 .
Multivariable Model for Length of Stay in 8665 Matched Pairs of People Experiencing Homelessness and Housed Patients in the Trauma Quality Programs, 2017 to 2018 a Estimated from hierarchical negative binomial regression models allowing for random effects at the pair level and at the hospital level.bHigher scores indicate greater severity of injury.

Table 4 .
Moderation of the Association Between Housing Status and Length of Stay by Age and ISS in People Experiencing Homelessness and Matched Housed Patients in the Trauma Quality Programs, 2017 to 2018 Morbidity and Length of Stay After Injury Among People Experiencing Homelessness b Calculated as the interaction term between the variable of interest (age and ISS) and experiencing homelessness.Interactions between age and ISS were modeled separately.Value NA indicates reference category.c Higher scores indicate greater severity of injury.JAMA Network Open | Equity, Diversity, and Inclusion JAMA Network Open.2024;7(2):e240795.doi:10.1001/jamanetworkopen.2024.0795(Reprinted) February 28, 2024 8/14 Downloaded from jamanetwork.comby guest on 03/06/2024

47 JAMA Network Open | Equity, Diversity, and Inclusion Morbidity
and Length of Stay After Injury Among People Experiencing Homelessness decreased bed availability and resources for other patients.Improving the care of high-need, high-cost patients such as those experiencing homelessness is a priority for Centers for Medicare & Medicaid Services.46Thiswork contributes to a growing body of evidence that health-related social drivers such as homelessness can exacerbate health disparities and lead to higher cost for public payer programs.JAMA Network Open.2024;7(2):e240795.doi:10.1001/jamanetworkopen.2024.0795(Reprinted) February 28, 2024 9/13 Downloaded from jamanetwork.comby guest on 03/06/2024