Contributors: YT obtained funding. AY obtained data and ethics approval. All authors, contributed to the study design, analysis, interpretation of the results; and preparation, review, and approval of the manuscript. AY is the guarantor. The corresponding author attests that all, listed authors meet authorship criteria and that no others meeting the criteria have been omitted.
Corresponding author: Ayae Yamamoto, SM, PhD Candidate, Department of Health Policy and Management, UCLA Fielding School of Public Health, Address: 650 Charles Young Dr. S., 31-269 CHS Box 951772, Los Angeles, CA 90095, Phone: +1-510-565-6267, moc.liamg@5yeaya
The publisher's final edited version of this article is available at Soc Sci MedAlthough homelessness and opioid overdose are major public health issues in the U.S., evidence is limited as to whether homelessness is associated with an increased risk of opioid overdose.
To compare opioid-related outcomes between homeless versus housed individuals in low-income communities.
Cross-sectional analysis of individuals who had at least one ED visit or hospitalization in four states (Florida, Maryland, Massachusetts, and New York) in 2014.
Risk of opioid overdose and opioid-related ED visits/hospital admissions were compared between homeless versus low-income housed individuals, adjusting for patient characteristics and hospital-specific fixed effects (effectively comparing homeless versus low-income housed individuals treated at the same hospital). We also examined whether risk of opioid-related outcomes varied by patients’ sex and race/ethnicity.
Individuals with no ED visit or hospitalization in 2014 were not included.
Homeless individuals had disproportionately higher adjusted risk of opioid-related outcomes compared to low-income housed individuals treated at the same hospital. Among homeless individuals, non-Hispanic White females incurred the highest risk. These findings highlight the importance of recognizing the homeless population—especially White female homeless population—as a high-risk population for opioid overdose.
Keywords: opioid overdose, opioid abuse, homeless, homelessness, emergency departmentThe opioid overdose epidemic has become one of the most important public health emergencies in the United States. Opioid overdose was responsible for an estimated 50,000 deaths in 2017 (1), and its total economic burden is estimated to be over $500 billion annually (2). Studies have found that life expectancy in the United States declined in 2017 for the third consecutive year, in part, due to an increase in deaths from unintentional injuries, including opioid overdoses (3). Despite a number of efforts targeted at reducing the number of adverse events from opioid overdose, the effectiveness of such strategies has been limited.
Homelessness is another major public health issue in the United States, with an estimated 2.5 to 3.5 million Americans experiencing homelessness annually (4), and over 550,000 people are homeless on any given night (5, 6). Homeless individuals experience higher chronic and acute disease burdens (7, 8); higher mortality rates (9, 10); and are more likely to utilize costly emergency department and hospital inpatient services compared to housed individuals (7, 11, 12). Although these two public health problems are closely related, they are often addressed separately. Given the lack of access to healthcare and social support available to the homeless population, they may be incurring a higher burden of opioid use disorders compared to a comparable low-income housed population.
Evidence is scarce as to whether homelessness is associated with an increased risk of opioid-related adverse health outcomes. Existing studies suggest that homeless individuals are at a higher risk of opioid overdose than the general population (9, 13). However, these studies are limited as they were conducted in a single city or hospital (in Boston or New York City) (9, 13), or among Veterans (14, 15) and therefore, it remains largely unknown whether these findings are generalizable to other cities, states and non-Veterans. To our knowledge, no study to date has examined the association between homelessness and opioid-related health outcomes using multi-state data, possibly due to the lack of data that can reliably identify the homeless population. Although chronic pain guidelines recommend physicians to co-prescribe naloxone (a life-saving opioid antagonist) to patients who have a high risk for opioid overdose (16), accurate prediction models for opioid overdose have not yet been developed. Furthermore, existing prediction models do not include homeless status as a key predictor (17–36). It is, thus, critical for clinicians to accurately identify patients with a high risk of opioid overdose, employ targeted screening, and to intervene if necessary, to effectively address the current opioid overdose epidemic.
In this context, we used datasets that include all hospital admissions and emergency department (ED) visits from four large and diverse states to examine the association between homelessness and opioid overdose and the use of emergent care for opioid use among patients who had an ED visit(s) or hospitalization(s) in 2014. The comparison group consisted of housed individuals living in low-income neighborhoods (defined by zip code with the lowest median household income quartile), which we refer to as “low-income housed”. To identify patient populations with the highest risk, we also investigated patient’s sex and race/ethnicity associated with the highest risk of opioid-related adverse health outcomes among the homeless population and low-income housed individuals, separately.
We analyzed the 2014 State Inpatient Database (SID) and the State Emergency Department Databases (SEDD) for four states (Florida, Maryland, Massachusetts, and New York), that are made available for the Healthcare Cost and Utilization Project of the Agency for Healthcare Research and Quality (37). The SID includes all inpatient discharge records from community hospitals (including emergency visits that result in hospitalization), and the SEDD includes all emergency department (ED) visits at hospital-affiliated emergency departments that do not lead to a subsequent hospitalization. These databases capture visit information for all patients regardless of the type of insurance and insurance status. The records for each patient include a direct report of homeless status, key demographic information such as age, gender and race, insured status, and data on the primary diagnosis associated with the visit and secondary diagnoses that affect the course or cost of treatment. The non-public use data sets we used also include a unique patient linkage number that allows us to track patients across multiple visits and admissions, allowing a direct assessment of the frequency of use and variations in diagnosis across visits. We used data from four states with homeless flags to achieve the broadest range of socioeconomic and geographic diversity in the study (only 7 states [4 states included in our analysis plus Georgia, Utah and Wisconsin] reported both the homeless indicator and a unique patient linkage number for both SID and SEDD in 2014. Our internal investigation found a severe underreporting of the homeless indicator for Utah and Wisconsin’s SID/SEDD, and the hospital identifier was not available in Georgia’s SID/SEDD; therefore, these states were not included in our analyses), and homeless status is reported directly from the hospitals (38). For each inpatient and hospital-affiliated emergency department discharge, there is an indicator for each patient’s homeless status, which has been used in previous studies (39–44).
The study population was restricted to individuals aged ≥18 years old with at least one ED visit or hospital admission in 2014. We compared individuals who were identified as homeless with housed individuals living in the lowest income quartile (median household income was estimated based on residential zip code). We excluded people with a primary diagnosis related to delivery (Clinical Classification Software single-level codes: 177–192, 194–196, 218–220, 222–224) since a significant proportion of inpatient visits were delivery-related (6.1% [398,475/513,409] of all visits), and people with missing data on the homeless indicator (0.6% of individuals in our data were missing information about the homeless indicator) or any of the key adjustment variables described below.
The primary outcomes were: (1) opioid overdose and (2) opioid-related ED visit or hospital admission. Opioid overdose was defined as having any of the following ICD-9 diagnosis codes: 965.00 – 965.02, 965.09, E850.0 - E850.2 in the first 10 diagnosis codes across all ED visits or hospital admissions (45). Opioid-related ED visit or hospital admission was identified using the following ICD-9 codes: 304.00 – 304.02, 304.70 – 304.72, 305.50 – 305.52, 965.00 – 965.02, 965.09, 970.1, E850.0 - E850.2, E935.0 – E935.2, E940.1 (46).
We adjusted for patient characteristics and hospital-specific fixed effects. Patient characteristics included age (categorized as 18–34, 35–44, 45–55 and 65+ years old), sex, race, and ethnicity (Non-Hispanic White, Non-Hispanic Black, Hispanic, and other), primary insurance type (Medicare, Medicaid, private, self-pay, and no charge/other), and 26 comorbidities included in the Elixhauser Comorbidities Index (47) (excluding drug abuse). We also adjusted for hospital-specific fixed effects (indicator variables for each hospital) to account for both measured and unmeasured characteristics of hospitals that do not vary over time (48). Therefore, our models effectively compared homeless and low-income housed individuals treated at the same hospital.
First, we examined the association between homeless status and opioid overdose and opioid-related ED visits/hospitalizations using multivariable regression models. We used multivariable linear probability models with Huber-White robust standard errors to account for heteroscedasticity (because small cell sizes for some combinations of patient characteristics resulted in complete or quasi-complete separation in logistic regression models (49)), adjusting for patient characteristics (age, sex, race/ethnicity, primary payer, Elixhauser co-morbidities) and hospital-specific fixed effects. After fitting the regression models, adjusted outcomes were calculated using the marginal standardization form of predictive margins (also known as predictive margins or margins of responses); for each individual we calculated predicted probabilities of opioid-related outcomes with homeless indicator fixed at each category (0 or 1) and then averaged over the distribution of covariates in our sample (50).
Then, we examined patients’ sex and race/ethnicity associated with the highest risk of opioid-related adverse health outcomes among the homeless population and the low-income population with housing, respectively. In doing so, we estimated the adjusted risk of opioid-related adverse health outcomes (adjusted for patient characteristics and hospital-specific fixed effects) for each combination of patients’ sex and race/ethnicity.
We conducted several sensitivity analyses. To determine whether the lower than expected count of homeless individuals in FL and MD could influence our results, we restricted our sample to MA and NY. Next, we restricted the analysis only to male patients and conducted a separate analysis only for ED visits because low-income housed individuals may be more likely present to hospitals for pregnancy-related concerns or elective surgeries. Finally, since residual confounding may bias our results, we performed a formal test to assess the sensitivity of unmeasured confounders to regression results (51).
All analyses were conducted in SAS Enterprise Guide 4.2 (SAS Institute) and Stata, version 14 (StataCorp). This study was approved by the institutional review board of the University of California, Los Angeles Office of the Human Research Protection Program.
Our final sample consists of 96,099 homeless and 2,869,230 low-income housed individuals who had at least one ED visit/hospital admission in 2014 in these four states. Compared to low-income housed individuals, homeless individuals were slightly older, more likely to be male, more likely to be have Medicaid as primary payer, more likely to be Hispanic or other race/ethnicity, and more likely to have comorbidities such as alcohol and drug abuse, mental illness, and diabetes ( Table 1 ).
Characteristics of homeless vs. low-income housed individuals
| Homeless individuals (N=96,099) | Low-income housed individuals (N=2,869,230) | P value | |
|---|---|---|---|
| Age in years at admission, mean (SD) | 47.7 (18.0) | 47.2 (20.0) | |
| Female | 42,705 (44.5%) | 1,607,821 (56.3%) | |
| Primary expected payer | |||
| Medicaid | 51,018 (53.1%) | 748,567 (26.2%) | |
| Medicare | 21,159 (22.0%) | 742,560 (26.0%) | |
| Private insurance | 3,803 (4.0%) | 672,970 (23.6%) | |
| Self-pay | 15,850 (16.5%) | 538,431 (18.9%) | |
| No charge/Other | 4,170 (4.3%) | 152,370 (5.3%) | |
| Race/ethnicity | |||
| Non-Hispanic White | 30,021 (31.3%) | 1,160,535 (40.7%) | |
| Non-Hispanic Black | 28,990 (30.2%) | 906,190 (31.7%) | |
| Hispanic | 23,195 (24.2%) | 595,236 (20.8%) | |
| Other | 13,794 (14.4%) | 192,937 (6.8%) | |
| Selected Elixhauser co-morbidities | |||
| Alcohol abuse | 7,194 (7.5%) | 44,676 (1.6%) | |
| Drug abuse | 8,689 (9.1%) | 46,337 (1.6%) | |
| Psychoses | 4,756 (5.0%) | 37,480 (1.3%) | |
| Depression | 4,208 (4.4%) | 73,814 (2.6%) | |
| Congestive heart failure | 2,492 (2.6%) | 46,598 (1.6%) | |
| Neurological disorders | 4,882 (5.1%) | 63,391 (2.2%) | |
| Chronic pulmonary disease | 8,567 (8.9%) | 158,387 (5.5%) | |
| Diabetes | 13,896 (14.5%) | 223,028 (7.8%) | |
| Renal failure | 3,845 (4.0%) | 75,705 (2.7%) | |
| Cancer | 1,523 (1.6%) | 26,136 (0.9%) |
* Low-income housed individuals were defined as individuals who live in areas with the lowest quartile of the median household income.
After adjusting for patient characteristics and hospital-specific fixed effects (effectively comparing homeless versus low-income housed individuals treated at the same hospital), homeless individuals had significantly higher risk of opioid overdose (adjusted risk, 1.8% for homeless vs. 0.3% for low-income housed individuals; adjusted risk difference [aRD], 1.5%; 95%CI, 1.0% to 2.0%; p <0.001) and opioid-related ED visit/hospital admissions (adjusted risk, 10.4% vs. 1.5%; aRD, 8.9%; 95%CI, 7.2% to 10.6%; p<0.001) compared to low-income housed individuals (Table 2 ).
Unadjusted and adjusted opioid-related adverse health outcomes between homeless compared to low-income housed individuals
| No. of individuals | Unadjusted outcomes (95% CI) | Adjusted outcomes (95% CI) | Adjusted risk difference (95% CI) | P value | ||
|---|---|---|---|---|---|---|
| Opioid Overdose | Homeless | 1,829 | 1.9% (1.8% to 2.0%) | 1.8% (1.3% to 2.2%) | 1.5% (1.0% to 2.0%) | |
| Low-income Housed | 7,170 | 0.3% (0.3% to 0.3%) | 0.3% (0.2% to 0.3%) | |||
| Opioid-related ED Visit/ Hospital Admission | Homeless | 10,792 | 11.2% (11.1% to 11.5%) | 10.4% (8.8% to 12.1%) | 8.9% (7.2% to 10.6%) | |
| Low-income Housed | 42,797 | 1.5% (1.5% to 1.5%) | 1.5% (1.5% to 1.6%) |
* Adjusted for patient characteristics and hospital-specific fixed effects. Low-income housed refers to individuals who live in areas with the lowest quartile of the median household income
We found that, among the homeless population, non-Hispanic White females experienced the highest risk of opioid overdose ( Figure 1 and Online Appendix Tables A and B). On the other hand, among low-income housed individuals, non-Hispanic White males had the highest risk. We found a similar pattern for opioid-related ED visits and hospitalizations ( Figure 2 and Online Appendix Tables A and B).
Adjusted risk of opioid overdose for homeless compared to low-income housed patients, by Sex and Race/ethnicity
1 Wald p-value for heterogeneity statistically significant at p
*Point estimate statistically significant at p
**Point estimate statistically significant at p
Adjusted risk of opioid-related emergency department (ED) visit/hospital admission for homeless compared to low-income housed patients, by Sex and Race/ethnicity
1 Wald p-value for heterogeneity statistically significant at p
*Point estimate statistically significant at p
**Point estimate statistically significant at p
Overall, our findings for both opioid outcomes were not sensitive to restricting the analysis to NY and MA (Online Appendix Table 2A), to male patients (Online Appendix Table 3A), or to patients with ED visits only (Online Appendix Table 4A). Homeless individuals, on average, had greater number of ED and inpatient visits per person compared to low-income housed individuals (Online Appendix Table 5A). The test to assess the sensitivity of our regression results to unmeasured confounders revealed that residual confounding is unlikely to explain the observed association between homeless status and the two opioid outcomes (Online Appendix Table 6A).
Using a comprehensive dataset of all ED visits and hospital admissions from four large and diverse states, we found homeless persons had significantly higher risk of opioid overdose and opioid-related ED visits/hospital admissions, even when they were compared to low-income housed individuals who were treated at the same hospital. We also found that non-Hispanic White females incurred the largest risk of opioid-related adverse health outcomes among the homeless population, whereas non-Hispanic White males incurred the highest risk among the low-income housed population. These findings suggest that homelessness is an issue that extends beyond poverty—as homeless individuals are at higher risks even compared to comparable, low-income individuals with housing—and shed light on the importance of homeless individuals, especially non-Hispanic White female homeless individuals, as the high-risk population of the opioid overdose.
Although chronic pain guidelines currently recommend that physicians co-prescribe naloxone, a life-saving opioid antagonist, to patient at high risk of opioid overdose (16), clinical tools to effectively identifying patients who could benefit from naloxone are lacking. A recent study using machine-learning algorithms to predict patients with high risk of opioid overdose identified 268 potential predictors of opioid overdose, but the homelessness was not included as one of the potential predictors (27). Indeed, homelessness has not been identified as an important predictor of opioid overdose in the currently-available clinical prediction models. Our findings, indicating a higher risk of opioid-related outcomes among the homeless population, underscore the importance of including homelessness as the key predictor in the clinical tools for predicting patients at an increased risk of opioid overdose.
There are several potential mechanisms that could explain our findings. First, it is possible that homeless people may use opioids as a way to cope with their emotional suffering and distress from living on the streets (52, 53). Alternatively, it is also possible that addiction to opioids makes people more likely to become homeless due to their limited ability to work, strained relationships with family and friends, and challenges in accessing and motivation for receiving treatment for their addiction (53–55). We tried to isolate the impact of homelessness by using low-income individuals with housing as the control, and our findings suggest that homelessness per se—after controlling for the impact of poverty—is an independent risk factor for opioid-related adverse health outcomes. Second, in addition to their higher risks of opioid overdose, homeless individuals may also face a major barrier to accessing any drug treatment (56), and once treatment is completed, they confront additional barriers transitioning to life without opioids. Even for homeless persons who complete treatment programs, overdose risks may remain high when they leave that treatment as they lapse to opioid use on the streets (57). These increased risks for overdose observed in our study may be linked with a lower tolerance and the lack of access to any treatments for opioid addiction, but especially medication-assisted treatments (58). Moreover, not all community health centers provide medication-assisted treatments, and among those that do, the majority of the clinics face provider shortages (59).
We also found that the highest risk population subgroup for opioid overdose was different for homeless compared to low-income housed individuals, and to our knowledge, this is the first to study race-sex combinations for opioid-related emergency and hospitalization risk. Non-Hispanic White females incurred the highest risk among the homeless population, whereas non-Hispanic White males exhibited the highest risk among the low-income housed population. These findings are consistent with a recent report using a nationally-representative sample of Medical Expenditure Panel Study (MEPS) that found that women were more likely than men to have had any opioid use as well as frequent opioid use during the year (60). Another study using national survey data found that non-Hispanic White females may be slightly more likely to receive prescription opioids compared to non-Hispanic White males (61). The differences in risks based on sex and race/ethnicity may be explained, in part, by differences in cultural perspectives on pain, access to pain treatment, and/or provider bias between non-Hispanic White persons and racial and ethnic minorities (62–64).
To our knowledge, this is the first study using data from multiple states to show that homeless individuals experience higher risk of opioid-related adverse health outcomes in emergent care settings. Existing studies that assessed the relationship between homelessness and opioid overdose are limited as they are conducted using a convenient sample collected in a single city or hospital (mostly in Boston and New York City) (9, 13) or among Veterans (14, 15). The SID/SEDD database from 4 states used in this study contains all ED visits and hospital admissions, and covers more than 10% of the US population and 17% of the homeless population, according to estimates from the U.S. Department of Housing and Urban Development (5). We are unaware of any other databases that allow detailed analysis of the healthcare needs and use of a broad and representative sample of the homeless population.
Our findings are consistent with previous smaller studies suggesting that homeless individuals may have higher opioid misuse compared to housed individuals (13, 65). A study conducted in Boston found that a third of the deaths for homeless individuals younger than 45 years were associated with a drug overdose, and opioids were implicated in 81% of all overdose deaths (9). A recent interview conducted by a public New York City emergency department showed that homeless patients had higher risk of heroin and prescription opioid use compared with housed patients (13).
Our findings suggest that the homeless population, and in particular White female homeless population, is at a higher risk of opioid overdose. While it has been recognized that opioid use may be higher among homeless individuals, evidence has been limited. For clinicians, identifying homelessness as an important predictor of opioid overdose would allow them to refer patients to appropriate care and precautions, and to co-prescribe naloxone if necessary. Our findings demonstrating high levels of opioid use among homeless patients, its association with an increased risk of opioid overdose, and non-Hispanic White female homeless individuals having the highest risk of opioid overdose, should be informative for policymakers and frontline clinicians to recognize the high-risk population of opioid overdose, employ more targeted screening, and use interventions (such as co-prescription of naloxone) that can more effectively reduce opioid overdose at the population level.
Our study has limitations. First, as is the case with any cross-sectional studies, the temporal relationship between exposure and outcome could not be assessed. Therefore, we are not able to rule out the possibility of reverse causation. It is possible that opioid addiction may lead individuals to lose employment and become homeless or hinder their efforts to get off the streets rather than homeless status causing higher risk of opioid overdose.
Second, exposure and outcome misclassification is another limitation. For instance, homelessness is a dynamic status and we are not able to capture the severity of homelessness (i.e., chronic versus temporary homeless) in our data. If temporary homeless individuals are coded as homeless, it will only bias the effect towards the null. Furthermore, when homeless counts were compared with the Housing and Urban Development’s 2014 point-in-time estimates, they appeared undercounted for MD and FL. However, sensitivity analysis restricted to NY and MA confirmed that our findings did not qualitatively change based on this restriction (Online Appendix Table 2A). This may be due to individuals being under-coded for these states or that homeless individuals are not being admitted to the ED or hospital. If homeless status was collected based on a self-report, the homeless count may be underreported due to social desirability bias. Homeless status in our data, however, is collected by hospitals, and hospitals have strong financial incentives associated with billing and collection to accurately determine where the patient lives. Nevertheless, some patients may underreport to the hospitals that they are homeless and/or are living in homelessness because of fear of stigma and discrimination, or because they do not consider themselves to be homeless.
Third, although we used low-income housed individuals as the control in order to isolate the impact of homelessness from poverty, our control group may not be perfect. In our data, the lowest quartile of household income was $1-$39,999 in 2014, which includes individuals with a substantially higher income than homeless individuals. However, the use of a control group in our study is more robust than existing studies evaluating the impact of homelessness by comparing their health outcomes with the general population (without restricting to a low-income population). Fourth, the SID/SEDD database captures only individuals who had at least one ED visit or hospital admission in 2014, and therefore, our findings may not be generalizable to healthier or sicker homeless individuals who had no encounter with ED or hospitals in a given year. Further, it is possible that housed individuals with opioid use disorders are getting treated in non-emergent care settings and thus, we are not able to compare the risk of opioid-use disorders between the two groups. Lastly, although we used all ED visits and hospital discharge data from four large and diverse states, our findings may not be generalizable to homeless patients in states not included in our analysis.
In summary, among homeless and low-income housed individuals who sought care in inpatient and emergency departments in 2014, homeless individuals experienced significantly higher risk of opioid overdose and opioid-related ED visits/hospitalizations, even compared to low-income housed individuals treated at the same hospital. Among the homeless population, non-Hispanic White females exhibited the highest risk of opioid-related adverse health outcomes, whereas non-Hispanic White males experienced the highest risk among the low-income housed population. Our findings highlight the importance of recognizing the homeless population—especially the non-Hispanic White female homeless population—as a high-risk population for opioid overdose. EDs and hospitals may be able to help address this epidemic by screening homeless individuals for opioid use disorders and have a system in place to refer these patients to community clinics for medication-assisted treatment. Additional research is warranted to understand the specific characteristics of individuals, geography, and health and social policy—such as policies on providing “housing for health,” access to substance use treatment and health care, case management for individuals discharging from treatments to address risks for relapse during transitions, and care when opioid prescribing—that may contribute to excessive opioid dependence and overdose deaths among homeless persons.
