Associations of housing stress with later substance use outcomes: A systematic review

Abstract

A synthesis of existing evidence regarding the association of housing stress with later substance use outcomes can help support and inform housing interventions as a potential strategy to address problematic substance use. We conducted a comprehensive search of PubMed, Web of Science, PsycInfo, CINAHL, Social Work Abstracts, and Sociological Abstracts and systematically screened for articles examining housing stress and later substance use outcomes among U.S. adults. Across 38 relevant articles published from 1991 to 2020, results demonstrated an association of homelessness with an increased likelihood of substance use, substance use disorders (SUD), and overdose death. Results regarding the association of homelessness with receipt and completion of SUD treatment were mixed, and one study indicated no association of homelessness with motivation to change substance use behaviors. Several studies did not find an association of unstable housing with substance use or receipt of SUD treatment, while others found an association of unstable housing with intensified SUD symptoms and a decreased likelihood of completing SUD treatment. Overall, while there is evidence of an association of homelessness with later substance use, SUD, and overdose death, results for other forms of housing stress and some substance use outcomes are less consistent. There are several methodological considerations specific to selected measures of housing stress and substance use, study populations, and analytic approaches that have implications for results and directions for future research. Despite these considerations, results collectively suggest that innovative interventions to address housing stress, namely homelessness, may help mitigate some substance use outcomes.

1. Introduction

Substance use disorders (SUD) and overdose are critical public health issues in the United States. From 1999 to 2019, >800,000 people died of a drug overdose, with the rate of drug overdose death more than tripling during this period (Hedegaard, Minino, & Warner, 2020). The U.S. also has a high prevalence of substance misuse and use disorders. In 2019, 21% of U.S. adults engaged in illicit drug use and 8% had an alcohol or other substance use disorder in the past year (Substance Abuse and Mental Health Services Administration, 2020). Thus, there is a need for evidence-based treatment and prevention initiatives to reduce harm and improve quality of life among those engaging in substance use.

Housing stress, broadly defined to include housing problems such as affordability, quality, stability, and loss (U.S. Department of Housing and Urban Development, 2018), represents another critical public health issue in the U.S. On any given night in 2019, more than half a million people experienced homelessness (U.S. Department of Housing and Urban Development, 2020), and in 2019 50% of U.S. renter households spent >30% of their income on housing costs (U.S. Department of Housing and Urban Development, 2020). Several studies indicate that SUD and overdose are intricately linked to housing stress (Johnson and Chamberlain, 2008; McVicar et al., 2015; Stringfellow et al., 2016; Rhoades et al., 2018; Thompson et al., 2013). This association is complex, with substance use potentially being both a cause and a consequence of housing stress (Polcin, 2016). Theoretical support for housing stress as a cause of substance use is derived from the social causation hypothesis. The social causation hypothesis has been primarily used to understand the link between poverty and mental health conditions, holding that the stress associated with living in poverty contributes to the development psychopathology (Polcin, 2016). More recently, this hypothesis has been applied to research on the association between economic disadvantage and substance use behaviors (Wadsworth & Achenbach, 2005). Under this hypothesis, substance use may function as a mechanism for coping with the emotional and economic strain associated with housing problems, such that housing problems trigger changes in substance use behaviors (Tsai, 2020).

The social causation hypothesis lends support to the Housing First approach as a strategy to address the association between housing stress and substance use outcomes. In the Housing First approach, housing stability and affordability are prioritized first, rather than abstinence or recovery from SUD, and individuals have autonomy to choose the social services (e.g., SUD treatment) suited to their needs (Polcin, 2016). While this model is shown to improve housing stability and reduce inpatient and emergency care use (Tsai, 2020; Baxter et al., 2019; Rog et al., 2014; Hwang et al., 2005), overall evidence regarding effectiveness, particularly related to substance use outcomes, is mixed (Tsai, 2020; Baxter et al., 2019; Fitzpatrick-Lewis et al., 2011; Kertesz et al., 2009). Moreover, this intervention approach typically focus only on individuals experiencing homelessness, the most severe form of housing stress. A comprehensive understanding of the association of multiple forms of housing stress with various substance use outcomes can help refine and further inform appropriate intervention. Importantly, if existing evidence supports the social causation hypothesis with respect to multiple forms of housing stress and substance use outcomes, this suggests that housing may represent a modifiable component of the social and structural context contributing to substance use outcomes.

We aimed to systematically synthesize and critically evaluate existing research regarding the association of housing stress with later substance use outcomes among U.S. adults. We focused our review on studies in which housing stress preceded substance use in order to inform housing-focused interventions as a potential strategy to address the current SUD and overdose crises in the U.S.

2. Methods

We systematically reviewed the literature using a protocol informed by the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines (http://www.prisma-statement.org/PRISMAStatement/CitingAndUsingPRISMA). We searched multiple electronic databases including PubMed, Web of Science, PsycInfo, Cumulative Index to Nursing and Allied Health Literature (CINAHL), Social Work Abstracts, and Sociological Abstracts to identify articles published from the database start date through July 2020. In collaboration with a health sciences librarian, we conducted searches using keywords related to (1) housing stress, (2) substance use, and (3) epidemiologic research (Supplemental File 1).

2.1. Eligibility criteria

We included articles that examined the association of housing stress (i.e., housing instability, insecurity, or insufficiency, homelessness, eviction, overcrowding, poor physical housing conditions, frequent moves, high housing cost relative to income) with substance use outcomes (i.e., substance use, misuse, or use disorders, motivation or intention to change use, fatal and non-fatal overdose, treatment initiation, retention, and completion). To focus our review on articles relevant to the social causation hypothesis, we included studies in which temporality was established between the measures of housing stress and substance use examined in the study (i.e., housing stress preceded substance use) and in which housing stress was the primary exposure or a primary predictor or risk factor of interest. We excluded studies that explicitly examined the effect of a housing program or service on substance use, as we were interested in studies regarding this association in the absence of intervention. We also included only studies that were peer-reviewed, published in English, conducted in an adult (≥18 years) population in the U.S., and reported quantitative data. We limited inclusion to studies conducted among adults as the primary type of housing stress examined among youth was homelessness due to running away from home, which would likely require different intervention strategies than the types of housing stress typically examined among adults. We limited inclusion to studies conducted in the U.S. given differences across countries in housing markets and trends and social norms regarding substance use.

2.2. Screening process

Using our pre-specified eligibility criteria detailed above, two authors independently screened the title and abstract of each article identified from the electronic database search for potential eligibility. Two authors then independently reviewed the full text of potentially eligible articles to further screen for our criteria. A third author resolved any discrepancies. After the full text review, we conducted a hand search of the reference lists of included articles to identify relevant studies not captured by the electronic database search.

2.3. Data abstraction

We conducted an in-depth examination of articles meeting full eligibility criteria using a data abstraction tool specific to the aims of the review (Supplemental File 2), with abstraction domains including study aims, design, data sources, and population; measures of housing stress and substance use; data analysis methods; key results; study limitations; and potential sources of bias. For each article, one author abstracted data and a second author confirmed the abstracted information. We resolved discrepancies through discussion and consensus.

3. Results

We identified 9,445 articles through the electronic database search (Fig. 1). A total of 4,665 remained after we removed duplicates. In the initial title and abstract screen, we determined that 4,408 articles did not meet eligibility criteria, leaving 257 for full text review. Of these, 38 met full eligibility criteria (Asana et al., 2018; Ayvaci et al., 2018; Baggett et al., 2013, 2015; Baker et al., 2019; Binswanger et al., 2016; Bowser et al., 2011; Buchholz et al., 2010; Calcaterra et al., 2014; Crawford et al., 2018; Fine et al., 2020; Guerrero, 2013; Johnson et al., 1997; Johnson and Fendrich, 2007; Kerker et al., 2011; Kertesz et al., 2003; Kleinman et al., 2002; Leickly et al., 2018; LePage et al., 2014; Linton et al., 2013; Moos et al., 2002; Moss et al., 2020; Mutter et al., 2015; Mutter and Ali, 2019; Neaigus et al., 2006; Ober et al., 2018; O’Connor et al., 1991; O’Driscoll et al., 2001; O’Toole et al., 2013; Polcin and Korcha, 2017; Riley et al., 2015; Rowe et al., 2019; Shah et al., 2006; Simon et al., 2017; Stack et al., 2000; Tuten et al., 2003; Upshur et al., 2014; Watkins et al., 2018). We did not identify additional articles from the reference lists of these studies.

Fig. 1.

PRISMA diagram.

3.1. Study characteristics

3.1.1. Study aims and design

Half of the reviewed studies (n = 19) were designed specifically to examine the association of housing stress with substance use outcomes, while half (n = 19) were designed to identify predictors of substance use outcomes, with housing stress examined as one of multiple of focal predictors. Two studies used a case-control design, 33 used a longitudinal cohort design, and three used a cross-sectional design, establishing temporality by asking participants for age at first substance use and homelessness or examining housing status prior to overdose death.

3.1.2. Study populations and data sources

Common study populations (Table 1) included individuals recently released from prison (n = 2); with current or prior housing stress (n = 5), engaging in a specific type of substance use or with SUD (n = 6), and in SUD treatment (n = 15). In most studies (n = 31), participants were predominantly male (60%–98%), though four studies were conducted among women only. Participants were predominantly white non-Hispanic (61%–81%) in 12 studies and Black non-Hispanic (64%–95%) in nine studies. Three studies had a large percentage of Hispanic participants (44%–58%). The average age of participants ranged from 27 to 51 years in three-fourths of studies (n = 29). Twenty-two studies recruited participants and collected data while 16 relied on existing administrative data.

Table 1.

Study characteristics.

Citation	Study population	Housing stress measure	Data source for housing stress measure	Substance use outcome	Data source for substance use outcome
Asana et al. (2018)	Recruited individuals experiencing homelessness from shelters and street locations in St. Louis, MO from 1999 to 2001 (N = 255; 73% male, 77% Black non-Hispanic, median age 43 years)	Stable housing in past year at 1-year study visit, defined as having been housed in their own place for most of the past year	Self-report	Alcohol use and outpatient substance use disorder treatment use in past year at 2-year study visit	Self-report, urine toxicology results, and administrative records
Ayvaci et al. (2018)	Recruited individuals experiencing homelessness from shelters and street locations in St. Louis, MO from 1999 to 2001 (N = 255; 73% male, 77% Black non-Hispanic, median age 43 years)	Stable housing in past year at 1-year study visit, defined as having been housed in their own place for most of the past year	Self-report	Cocaine use and outpatient substance use disorder treatment use in past year at 2-year study visit	Self-report and administrative records
Baggett et al. (2013)	Patients at the Boston Health Care for the Homeless Program (N = 28,033; 66% male, 43% white non-Hispanic, average age 41 years) and the Massachusetts general population from 2003 to 2008	Current homelessness at clinic enrollment, defined as an encounter at health clinic for homeless individuals	Administrative records	Unintentional or undetermined drug overdose death or death due to psychoactive substance use disorder	Death certificates
Baggett et al. (2015)	Patients at the Boston Health Care for the Homeless Program (N = 28,033; 66% male, 43% white non-Hispanic, average age 41 years) and the Massachusetts general population from 2003 to 2008	Current homelessness at clinic enrollment, defined as an encounter at health clinic for homeless individuals	Administrative records	Alcohol- or drug-attributable death	Death certificates
Baker et al. (2019)	Patients receiving long-term residential substance use disorder treatment at a non-profit in California from 2010 to 2015 (N = 2,069; 74% male, 79% white non-Hispanic, average age 39 years)	Current unstable housing at treatment admission	Administrative records from treatment facilities	Completion of long-term residential substance use disorder treatment	Administrative records from treatment facilities
Binswanger et al. (2016)	A random sample of all-cause deaths among individuals released from Washington state prisons from 1999 to 2009 (oversampling for overdoses) and a matched group of living releasees (N = 1,398; 80% male, 79% white non-Hispanic, average age 40 years)	Homelessness at prison release, defined as no residential address listed as place of release	Administrative records from the Washington State prisons	Unintentional overdose death	Death certificates
Bowser et al. (2011)	Patients at an outpatient treatment program in East Palo Alto, CA from 2001 to 2003 (N = 197, mostly Black non-Hispanic men, age not reported)	Homelessness in the past month at treatment admission	Self-rep	Substance use disorder treatment completion	Administrative records from treatment facilities
Buchholz et al. (2010)	Participants in a randomized controlled trial of integrated primary care and substance use disorder treatment at a Veteran’s Affairs facility in Seattle, WA from 2000 to 2003 (N = 622; 98% male, 64% white non-Hispanic, average age 46 years)	Homelessness in the past 90 days at study baseline and 3-, 6-, and 12-month follow-up	Self-report	Substance use disorder symptoms in past 30 days at 12-month study visit	Self-report on the Addiction Severity Index
Calcaterra et al. (2014)	Recruited individuals released from prison in Denver, CO from 2010 to 2012 (N = 155; 74% male, 37% Black non-Hispanic/33% white non-Hispanic, average age 42 years)	Current homelessness, unstable housing (hotel/motel, boarding house, or halfway house), or stable housing (own home or friend’s home) at study baseline	Self-report	Substance use in the past 30 days at 3-month study visit	Self-report on the Addiction Severity Index
Crawford et al. (2018)	Participants in the Multi-Ethnic Study of Atherosclerosis (MESA) neighborhood ancillary study from 2005 to 2012 (N = 3,807; 47% male, 37% white non-Hispanic/28% Black non-Hispanic, average age 65 years)	Neighborhood foreclosures in year prior to study baseline	Foreclosure records	Number of alcoholic drinks per week in 2005–2007 and 2010–2012	Self-report
Fine et al. (2020)	Patients at a multi-hospital health system in Boston, MA who had two encounters with buprenorphine on the active medication list from 2007 to 2018 (N = 5,948; 60% male, 81% white non-Hispanic, average age 38 years)	Current homelessness at study baseline, defined based on ICD-9/10 codes	Administrative records from health facilities	Opioid overdose death, based on cause of death codes	Death certificate
Guerrero (2013)	Patients receiving residential treatment at publicly funded substance use disorder treatment programs in Los Angeles County, CA from 2006 to 2009 (N = 52,799; 65% male, 40% Hispanic/34% white non-Hispanic, average age 36 years)	Current homelessness at treatment admission	Administrative records from treatment facilities	Substance use disorder treatment completion	Administrative records from treatment facilities
Johnson et al. (1997)	Recruited participants from emergency and transitional shelters, soup kitchens, drop-in centers, and single-room occupancy hotels in Cook County, IL in 1990 (N = 303; 62% male, 64% Black non-Hispanic, median age 35 years)	Age at first homelessness	Self-report	Age at first alcohol and drug use disorder symptoms	Self-report
Johnson and Fendrich (2007)	Recruited participants 18–40 years through multi-stage area probability sampling in Chicago, IL from 2001 to 2002 (N = 627; 50% male, 36% Black non-Hispanic/33% white non-Hispanic, average age 27 years)	Age at first homelessness	Self-report	Use of illicit drugs or non-medical use of prescription medications in the past year	Self-report
Kerker et al. (2011)	Families (adults with children, their children, and pregnant women) who stayed at an emergency shelter in New York City, NY from 2001 to 2003 (N = 102,771 family members; 69% female adults, 64% Black nonHispanic, 80% age 18–49 years)	Homelessness during study period, defined as staying in an emergency shelter	Administrative records from emergency shelters	Deaths attributable to substance use	Death certificates
Kertesz et al. (2003)	Recruited participants who used alcohol, heroin, or cocaine from a short-term inpatient detoxification unit in Massachusetts from 1997 to 1999 (N = 470; 81% male, 46% Black non-Hispanic/37% white non-Hispanic, average age 36 years)	Homelessness in the past 6 months at treatment admission, defined as having spent one or more nights in a shelter or on the streets	Self-report	Use of alcohol, heroin, or cocaine between discharge from detoxification and 6-month study visit	Self-report
Kleinman et al. (2002)	Recruited participants from detoxification units at two hospitals in New York City, NY from 1998 to 1999 (N = 279; 75% male, 58% Hispanic, average age 37 years)	Current homelessness at treatment admission	Self-report	Use of long-term substance use disorder treatment in the 30 days after discharge from detoxification	Self-report
Leickly et al. (2018)	Participants in a randomized controlled trial of contingency management who had alcohol dependence and a mental health disorder in Spokane, WA in 2017 (n = 79; 63% male, 56% white non-Hispanic, average age 46 years)	Current homelessness at study baseline	Self-report	Attrition from alcohol contingency management	Study records
LePage et al. (2014)	Patients receiving care from the Veteran’s Affairs North Texas Health Care System from 2009 to 2010 (N = 79; 93% male, 61% white non-Hispanic, average age 64 years)	Current homelessness at study baseline, defined based on ICD-9 codes or receipt of services from a homeless program	Administrative records from Veteran’s Affairs facilities	Use of inpatient substance use disorder treatment and substance use disorder diagnoses over 12 months	Administrative records from Veteran’s Affairs facilities
Linton et al. (2013)	People with current and former injection drug use recruited for the AIDS Linked to the Intravenous Experience (ALIVE) study in Baltimore, MD from 2005 to 2009 (N = 1,405; 66% male, 87% Black non-Hispanic, 77% age 35–54 years)	Homelessness in the past 6 months at annual study visits, defined as a self-report of homelessness or residence in a shelter, park, abandoned building, bus/train station, or on the street for at least 1 night in the past 6 months	Self-report	Injection drug use and injection-related risk behaviors in the past 6 months at annual study visits	Self-report
Moos et al. (2002)	Patients who engaged in substance use disorder treatment at 150 Veteran’s Affairs facilities and whose substance use problems intensified during or after treatment, with a matched control group of patients whose substance use problems remained stable or improved (N = 8,427; 97% male, 36% Black, 83% age ≥ 40 years)	Housing instability at study baseline, defined as <1 year in the same home	Administrative records from Veteran’s Affairs failities	Substance use disorder symptoms in the past 30 days at 9- and 12-month study visits	Administrative records of self-report on the Addiction Severity Index
Moss et al. (2020)	Participants in the National Longitudinal Study of Adolescent to Adult Health (Add Health), a nationally representative sample of individuals in grades 7–12 in 1994–1995 (N = 9,421; 44% male, 66% white non-Hispanic, all age 24–32 years)	Homelessness prior to 12th grade, defined as a week or more of sleeping in a place where people aren’t meant to sleep or in a homeless shelter, or did not have a regular residence	Self-report	Alcohol use disorder and cannabis use disorder in the past 12 months at 2008 study visit	Self-report
Mutter et al. (2015)	Data from publicly funded substance use disorder treatment programs collected through the 2010 Treatment Episode Data Set-Discharges (TEDS-D) (N = 104,999 treatment episodes; 61% male, 73% white non-Hispanic, 56% age 30–54 years)	Current homelessness at treatment admission	Administrative records from treatment facilities	Completion of residential substance use disorder treatment	Administrative records from treatment facilities
Mutter and Ali (2019)	Data from publicly funded substance use disorder treatment programs collected through the 2006–2014 Treatment Episode Data Set-Discharges (TEDS-D) (N not reported; 78% male, 65% white non-Hispanic, 64% age 35–54 years)	Current homelessness at treatment admission	Administrative records from treatment facilities	Completion of detoxification for alcohol use disorder	Administrative records from treatment facilities
Neaigus et al. (2006)	Recruited participants who used non-injection heroin in the past 30 days and had not injected drugs in past 6 months from targeted sampling, street outreach, and chain referral in New York City from 1996 to 2003 (N = 392; 66% male, 44% Hispanic, average age 35 years)	Current homelessness at each 6-month study visit	Self-report	Injection drug use since last study visit at each 6-month study visit	Self-report
Ober et al. (2018)	Patients with probable opioid or alcohol use disorder participating in the Substance Use Motivation and Medication Integrated Treatment (SUMMIT) trial at two clinical sites in Los Angeles, CA from 2014 to 2016 (N = 392; 79% male, 44% white/43% other or multiple races, average age 42 years)	Current homelessness at study baseline, defined as lacking a regular place to stay or spending the previous night sleeping outside, in a shelter, or in an abandoned building	Self-report	Initiation of behavioral treatment or medication-assisted treatment for opioid or alcohol use disorder within 6 months	Administrative records
O’Connor et al. (1991)	Patients 18–65 years participating in a randomized controlled trail for outpatient management of alcohol withdrawal syndrome at a Connecticut hospital (N = 179; 81% male, race/ethnicity not reported, average age 38 years)	Current homelessness at study baseline, defined as living in a shelter	Self-report	Alcohol detoxification failure	Administrative records from treatment facilities
O’Driscoll et al. (2001)	Participants from the Risk Activity Variables, Epidemiology, and Network (RAVEN) Study, a cohort of individuals who injected drugs in the past year in King County, WA from 1994 to 1997 (N = 2,849; 63% male, 65% white non-Hispanic, average age 37 years)	Current homelessness at study baseline	Self-report	Unintentional drug overdose death	Death certificates
O’Toole et al. (2013)	New primary care patients who had two visits within first 6 months of enrollment at a Veteran’ s Affairs facility in Providence, RI from 2008 to 2011 (N = 233; 95% male, 85% white non-Hispanic, average age 51 years)	Current homelessness at primary care enrollment, defined as being unsheltered, emergency sheltered, in transitional housing, or doubled up with family or friends	Administrative records from Veteran’s Affairs facilities	Substance use disorder diagnoses and outpatient substance use disorder treatment in the first 6 months of initiating primary care	Administrative records from Veteran’s Affairs facilities
Polcin and Korcha (2017)	Recruited residents from two sober living houses in northern California from 2004 to 2006 (N = 299; 80% male, 65% white non-Hispanic, average age 39 years)	Current homelessness or marginally housed at baseline and 6-, 12-, and 18-month study visits, defined as temporarily living with family or friends or in a hotel/motel	Self-report	Substance use disorder symptoms in the past 30 days at baseline and 6-, 12-, and 18-month study visits	Self-report on the Addiction Severity Index
Riley et al. (2015)	Recruited women with a history of housing instability at free meal programs, homeless shelters, and single room occupancy hotels in San Francisco, CA from 2008 to 2010 (N = 260; 100% female, 70% non-white, median age 47 years)	Current homelessness at study baseline, defined as sleeping in a shelter or public place	Self-report	Stimulant use in prior 6 months at 6-month study visit	Self-report
Rowe et al. (2019)	All methamphetamine, cocaine, and opioid overdose deaths in San Francisco, CA from 2010 to 2017 (N = 1,551; 74% male, 55% white non-Hispanic, average age 49 years)	Current residence in a single room occupancy building at the time of death	Administrative records for single room occupancy buildings	Methamphetamine, cocaine, and opioid overdose death	Death certificates
Shah et al. (2006)	People with active injection drug use from the AIDS Linked to the Intravenous Experience (ALIVE) study in Baltimore, MD from 1988 to 2000 (N = 1,327; 77% male, 95% Black non-Hispanic, average age not reported)	Homelessness in the past 6 months at baseline and each 6-month study visit	Self-report	Cessation of injection drug use and return to injection drug use at baseline and 6-month study visits	Self-report
Simon et al. (2017)	Recruited patients with an opioid use disorder diagnosis seeking office-based buprenorphine treatment at a clinic in Washington state from 2015 to 2016 (N = 100; 71% male, 81% white non-Hispanic, average age 39 years)	Homelessness or residence in a shelter in the past 3 months at first clinic contact for buprenorphine treatment	Self-report	Initiation of buprenorphine treatment within 90 days of initial clinic contact	Administrative records
Stack et al. (2000)	Patients admitted to a residential drug and alcohol treatment program at a Veteran’s Affairs facility in Hampton, VA in 1996 (N = 340; 92% male, 82% Black non-Hispanic, average age 42 years)	Current homelessness at treatment admission	Administrative records from treatment facilities	Completion of substance use disorder treatment	Administrative records from treatment facilities
Tuten et al. (2003)	Recruited pregnant women admitted to residential substance use disorder treatment at the Center for Addiction and Pregnancy in Baltimore, MD from 1994 to 1999 (N = 235; 100% female, 86% Black non-Hispanic, average age 30 years)	Homelessness at treatment admission, defined as not having an adequate, fixed, and regular nighttime dwelling address	Administrative records	Number of days in residential substance use disorder treatment and total number of substance use disorder treatment admissions during pregnancy	Administrative records
Upshur et al. (2014)	Women with diagnoses for drug or alcohol dependence and heavy use in the past 30 days recruited for the Addiction Health Evaluation and Disease Management (AHEAD) study in Boston, MA from 2006 to 2008 (N = 154; 100% female, 51% white nonHispanic, average age 37 years)	Homelessness in the past 90 days at study baseline and 3-, 6-, and 12-month study visits, defined as one or more nights in a homeless shelter or on the street	Self-report	Motivation to change alcohol or drug use behavior at baseline and 3-, 6-, and 12-month study visits	Self-report
Watkins et al. (2018)	Patients with probable opioid or alcohol use disorder participating in the Substance Use Motivation and Medication Integrated Treatment (SUMMIT) trial at two clinics in Los Angeles, CA from 2014 to 2016 (N = 290; 79% male, 41% white non-Hispanic/34% Hispanic, average age 43 years)	Current homelessness at study baseline	Self-report	Initiation of behavioral treatment or medication-assisted treatment for alcohol use disorder	Administrative records

3.1.3. Housing stress measures

Measures of housing stress largely focused on homelessness (n = 32) and unstable housing (n = 5), though one study examined neighborhood level housing foreclosures and one examined residence in a single room occupancy building (Table 1). Of the 32 studies that examined homelessness, 17 did not present a clear definition. Nineteen measured homelessness at study baseline or admission to SUD treatment. Nine measured homelessness within a specified timeframe (e.g., past 6 months), with five studies assessing homelessness at multiple time points throughout the study period. Of the five studies that examined unstable housing, four presented a clear definition, with unstable housing defined as less than one year in the same home at study baseline.

3.1.4. Substance use outcomes

Substance use outcomes included measures of various types of substance use (n = 11), SUD (n = 7), motivation to change substance use behaviors (n = 1), receipt of SUD treatment (n = 9), completion of SUD treatment (n = 8), and overdose death (n = 7), with some studies examining multiple measures (Table 1). A total of 12 studies focused on specific substance types, with the most common being alcohol, opioids, and other drugs typically injected. Of the studies that examined substance use, SUD, or treatment receipt as the outcome, outcomes were assessed over 30-day (n = 6), 90-day (n = 1), 6-month (n = 8), and 12-month periods (n = 8). Timeframes for SUD treatment completion were largely unspecified, and nearly all studies examining SUD treatment used a binary indicator for treatment receipt and completion (n = 12).

3.2. Key results

To present key results from each study, we organized results by category of substance use outcome. Given the breadth of outcomes examined, quantitative measures are included in Table 2. Additionally, a detailed overview of comparison groups and covariate sets is available in Table 2.

Table 2.

Key results.

Citation	Comparison groups	Covariates	Key results
Asana et al. (2018)	Unstable vs. stable housing in past year at 1-year study visit	Age, sex, race/ethnicity, lifetime alcohol and cocaine use disorder, lifetime serious mental illness, current alcohol use disorder, stable housing, and medical, psychiatric, and substance use disorder treatment	Unstable housing in past year at 1-year study visit not associated with alcohol use or outpatient substance use disorder treatment in past year at 2-year study visit (p > 0.05)
Ayvaci et al. (2018)	Unstable vs. stable housing in past year at 1-year study visit	Age, sex, race/ethnicity, lifetime alcohol and cocaine use disorder, lifetime serious mental illness, current cocaine use disorder, stable housing, and medical, psychiatric and substance use disorder treatment	Unstable housing in past year at 1-year study visit not associated with cocaine use or outpatient substance use disorder treatment in past year at 2-year study visit (p > 0.05)
Baggett et al. (2013)	Homeless individuals who received services at the Boston Health Care for the Homeless Program vs. the Massachusetts general population	Race/ethnicity	Rate of drug overdose deaths higher among homeless individuals who received services at the Boston Health Care for the Homeless Program compared to the Massachusetts general population (RR = 16.0, 95% CI 12.6, 20.3 for men 25–44 years; RR = 17.5, 95% CI 12.6, 22.5 for men 45–54 years; RR = 23.6, 95% CI 15.2, 36.6 for women 25–44 years; RR = 21.2, 95% CI 11.4, 39.5 for women 45–64 years)
Baggett et al. (2015)	Homeless individuals who received services at the Boston Health Care for the Homeless Program vs. the Massachusetts general population	None	Rate of alcohol-attributable deaths higher among homeless individuals who received services at the Boston Health Care for the Homeless Program than among the Massachusetts general population (RR = 6.4 for men 20–34 years; RR = 6.5 for women 20–34 years; RR = 7.8 for men 35–49 years; RR = 10.2 for women 35–39 years; RR = 6.8 for men 50–64 years; RR = 6.6 for women 50–64 years) Rate of substance-attributable deaths higher among homeless individuals who received services at the Boston Health Care for the Homeless Program than among the Massachusetts general population (RR = 10.9 for men 20–34 years; RR = 7.6 for women 20–34 years; RR = 10.0 for men 35–49 years; RR = 16.9 for women 35–39 years; RR for men 50–64 years; RR = 17.4 for women 50–64 years)
Baker et al. (2019)	Unstable vs. stable housing prior to admission to long-term residential substance use disorder treatment	Age, sex, race/ethnicity, education, employment status, health insurance type, diagnosed mental illness, current cigarette smoker, past month use of primary and secondary substances and alcohol, past year use of needles, number of prior treatment episodes, recommended length of stay	Unstable housing prior to treatment associated with a decreased likelihood of completing longterm residential substance use disorder treatment compared to stable housing (OR = 0.84, 95% CI = 0.57, 0.96)
Binswanger et al. (2016)	Homeless vs. not homeless at prison release	None	Homelessness at prison release associated with an increased likelihood of unintentional drug overdose death post-release compared to no homelessness (OR = 1.86, 95% CI 1.19, 2.29)
Bowser et al. (2011)	Homeless vs. not homeless in past month at treatment admission	Days used heroin, nights in jail	Homelessness in past month at treatment admission associated with a decreased likelihood of completing outpatient substance use disorder treatment compared to no homelessness (b = −0.60, 95% CI −0.94, −0.18)
Buchholz et al. (2010)	Consistent homelessness vs. consistent housing over 12 months	Age, primary substance used, psychiatric comorbidities, baseline alcohol or drug abstinence, final assessment time point, treatment assignment	Consistent homelessness over the 12-month study period associated with more severe substance use disorder symptoms at 12-month study visit compared to consistent housing (p < 0.0001)
Calcaterra et al. (2014)	Homeless vs. unstable housing vs. stable housing at baseline	None	Homelessness at baseline associated with an increased likelihood substance use in the past 30 days at 3-month study visit compared to stable housing (OR = 2.68, 95% CI = 1.02, 7.04) Unstable housing at baseline not associated with an increased likelihood substance use in the past 30 days at 3-month study visit compared to stable housing (OR = 0.71, 95% CI = 0.19, 2.65)
Crawford et al. (2018)	Number of neighborhood foreclosures in year prior to baseline	Age, sex, race/ethnicity, education income, employment, moving status duration between study visits, neighborhood percent poverty and unemployment, interactions between time and age, sex, race/ethnicity, and education	A greater number of neighborhood foreclosures within 0.4 km of participant’s residence in the prior year associated with a greater number of alcoholic drinks per week at later study visits (b = 0.11, 95% CI 0.05, 0.17)
Fine et al. (2020)	Homeless vs. not homeless at baseline	Age, sex, race, language, chronic conditions, substance use disorders, prescription medication use, buprenorphine encounter site, study entry year	Homelessness at baseline associated with an increased likelihood of opioid overdose mortality compared to no homelessness (HR = 1.77, 95% CI 1.25, 2.30)
Guerrero (2013)	Homeless vs. not homeless at admission to residential substance use disorder treatment	Age, sex, ethnicity, education, mental health disorder diagnoses, age at first drug use, primary substance used, type of substance use disorder treatment, number of treatment episodes	Homelessness at treatment admission associated with a decreased likelihood of completing residential substance use disorder treatment compared to no homelessness (HR = 0.76, 95% CI 0.73, 0.79)
Johnson et al. (1997)	Prior homelessness vs. never homeless	Sex, race, marital status, parental substance abuse, ever unemployed, ever homeless	Prior homelessness associated with an increased likelihood of later alcohol use disorder symptoms (RR = 1.65, 95% CI 1.05, 2.60) and substance use disorder symptoms (RR = 3.36, 95% CI 2.13, 5.30) compared to no homelessness
Johnson and Fendrich (2007)	Homeless vs. no homelessness prior to age 18 years	Age, sex, race/ethnicity, education, age at first drug use	Homelessness prior to age 18 years associated with an increased likelihood of past year substance use (b = 0.23, p-0.01)
Kerker et al. (2011)	Homeless vs. living in lowest income neighborhoods vs. general population of New York City, NY	None	Rate of substance use deaths among homeless adults was 2.1 times the rate for adults in lowest-income neighborhoods and 4.7 times the rate for those in general population
Kertesz et al. (2003)	Homeless vs. no homelessness in the past 6 months at treatment admission	Age, sex, race/ethnicity, employment, substance of choice, Stages of Change Readiness and Treatment Eagerness Scale instrument, treatment assignment	Homelessness not associated with time to first recurrent substance use after discharge from detoxification compared to no homelessness (HR = 1.26, 95% CI 0.88, 1.80)
Kleinman et al. (2002)	Homeless vs. not homeless at treatment admission	Age, sex, race/ethnicity, education, currently on parole, social integration, medical history, substance use history, substance use disorder treatment, intention to enter treatment, behavioral beliefs favoring treatment, self-efficacy, perceived benefits of treatment, perceived barriers to treatment, perceived susceptibility to relapse, social support processes	Homelessness associated with an increased likelihood of long-term substance use disorder treatment in the 30 days after discharge from detoxification compared to no homelessness (b = 0.166, p < 0.05)
Leickly et al. (2018)	Homeless vs. not homeless at study baseline	Baseline urine screen results, treatment assignment, pre-intervention drinking	Homelessness at baseline associated with an increased likelihood of attrition from outpatient alcohol contingency management compared to no homelessness (p < 0.05) Homelessness at baseline not associated with the proportion of urine screens positive for alcohol over a 12-week period compared to no homelessness (p > 0.05)
LePage et al. (2014)	Homeless vs. not homeless at study baseline	Age, sex, race/ethnicity	Homelessness at baseline associated with use of inpatient substance use disorder treatment (OR = 14.13, 95% CI 11.11, 17.97) and new substance use disorder diagnoses (OR = 2.50, 95% CI 1.97, 3.17) during follow-up compared to no homelessness
Linton et al. (2013)	Homeless vs. no homeless in past 6 months at annual study visits	Age, sex, race/ethnicity, employment, income, incarceration, depressive symptoms, frequency of injection, crack use, needle and cotton sharing, shooting gallery attendance, methadone maintenance, detoxification, training sex for money, drugs, shelter, or food, sex with an injection drug user partner, past homelessness	Among those who stopped injection drug use, homelessness associated with an increased likelihood of return to use compared to no homelessness (OR = 1.67, 95% CI 1.01, 2.74) Among those with active injection drug use, homelessness not associated with sustained injection drug use compared to no homelessness (OR = 1.03, 95% CI 0.71, 1.49) Among those with injection drug use at two consecutive study visits, homelessness associated with subsequent injection-related risk behavior compared to no homelessness (OR = 1.61, 95% CI 1.06, 2.45)
Moos et al. (2002)	Unstable vs. stable housing at study baseline	Age, marital status, substance use history, alcohol use disorder treatment, arrest history, alcohol or drug dependence, cocaine dependence, severity of drug problem, psychiatric symptoms, hallucinations, inpatient psychiatric treatment	Unstable housing at baseline associated with an increased likelihood of intensified substance use disorder symptoms in past 30 days at 9- and 12-month study visits compared to stable housing (OR = 1.36, p < 0.001)
Moss et al. (2020)	Homeless vs. not homeless prior to 12th grade	Sex, race/ethnicity, education, family economic conditions, adverse childhood experiences score	Homelessness prior to 12th grade not associated with mild (OR = 1.27, 95% CI 0.72, 2.22) or moderate (OR = 1.71, 95% CI 0.96, 3.04) alcohol use disorder symptoms compared to no homelessness Homelessness prior to 12th grade associated with an increased likelihood of severe alcohol use disorder symptoms (OR = 2.74, 95% CI 1.48, 5.08) and mild (OR = 2.56, 95% CI 1.54, 4.25), moderate (OR = 3.59, 95% CI 1.99, 6.48), and severe (OR = 2.37, 95% CI 1.17, 4.77) cannabis use disorder symptoms compared to no homelessness
Mutter et al. (2015)	Homeless vs. independent or dependent living at treatment admission	Age, sex, race/ethnicity, marital status, education, employment, insurance type, primary substance problem, 2 or more substances on record, medication assisted treatment, long-term treatment	Homelessness at admission associated with an increased likelihood of completing residential substance use disorder treatment compared to independent living (OR = 1.19, 95% CI 1.14, 1.24) and dependent living (OR = 1.14, 95% CI 1.09, 1.19)
Mutter and Ali (2019)	Homeless vs. independent or dependent living at treatment admission	Age, sex, race/ethnicity, education, employment, marital status, number of days in detoxification, source of referral to detoxification, service setting, age of alcohol initiation, other substance use, substance use disorder treatment, psychiatric conditions, year of detoxification	Homelessness at admission associated with an increased likelihood of completing residential detoxification for alcohol use disorder compared to independent living (OR = 1.05, 95% CI 1.03, 1.08) and dependent living (OR = 1.10, 95% CI 1.05, 1.14)
Neaigus et al. (2006)	Homeless vs. not homeless at each 6-month study visit	None	Homelessness at the prior study visit associated with an increased likelihood of transition to injection drug use among those who had never injected (HR = 1.2, 95% CI 0.5, 3.1) and former injectors (HR = 1.4, 95% CI 0.8, 2.5) compared to no homelessness
Ober et al. (2018)	Homeless vs. not homeless at study baseline	Age, sex, race/ethnicity, education income, employment, marital status, stigma, receipt of medication assisted treatment or behavioral therapy, negative consequences of substance use, substance use disorder treatment, arrest history	Homelessness at baseline associated with a decreased likelihood of initiating behavioral therapy for opioid or alcohol use disorder during the study period compared to no homelessness (OR = 0.45, 95% CI 0.21, 0.97) Homelessness at baseline not associated with initiation of medication assisted treatment for opioid or alcohol use disorder during the study period compared to no homelessness (OR = 0.51, 95% CI 0.20, 1.20)
O’Connor et al. (1991)	Homeless vs. not homeless at study baseline	None	Homelessness at baseline not associated with outpatient alcohol detoxification failure compared to no homelessness (39% vs. 47%, p > 0.05)
O’Driscoll et al. (2001)	Homeless vs. not homeless at study baseline	Age, sex, race/ethnicity, sexual orientation, injection speedball use, injection cocaine use, noninjection cocaine use, simulant use, poppers use	Homelessness at baseline associated with an increased likelihood of unintentional drug overdose death compared to no homelessness (RR = 2.30, 95% CI 1.06, 5.01)
O’Toole et al. (2013)	Homeless vs. not homeless at primary care enrollment	None	Homelessness at enrollment associated with an increased likelihood of outpatient substance use disorder treatment (37.8% vs. 7.5%, p < 0.001) and new substance use disorder diagnosis (28.3% vs. 4.2%, p-value not given) over 6 months compared to no homelessness
Polcin and Korcha (2017)	Homeless/marginally housed vs. stable housing at baseline and 6-, 12-, and 18-month study visits	Age, length of stay, sober living housing program type, study visit, psychiatric distress	Homelessness/marginal housing at prior study visit associated with higher peak density of substance use compared to stable housing (b = 1.14, 95% CI 1.05, 1.13) Homelessness/marginal housing at prior study visit not associated with an increased likelihood of more severe alcohol (b = −0.02, 95% CI −0.06, 0.01) or other substance (b = 0.00, 95% CI −0.01, 0.01) use disorder symptoms compared to stable housing
Riley et al. (2015)	Homeless vs. not homeless at study baseline	History of sexual violence, unprescribed opioid analgesic use, inpatient or outpatient substance use disorder treatment	Among women who did not use stimulants at baseline, homelessness at baseline associated with an increased likelihood of stimulant use over 6 months compared to no homelessness (RR = 2.75, 95% CI 1.15, 6.57)
Rowe et al. (2019)	Residence in a single room occupancy building vs. other residence at time of death	None	Residence in a single room occupancy building at time of death associated with a higher rate of opioid, cocaine, or methamphetamine overdose death (RR = 19.3, 95% 17.1, 21.7), opioid overdose death (RR = 17.2, 95% CI 14.7, 20.1), cocaine overdose death (RR = 24.5, 95% CI 20.5, 29.3), and methamphetamine overdose death (RR = 22.7, 95% CI 18.4, 27.9) compared to other residence
Shah et al. (2006)	Homeless vs. not homeless in the past 6 months at baseline and each 6-month study visit	Age, sex, HIV status, methadone maintenance treatment, detoxification program, daily injection, injected speedballs, cigarette use, alcohol use, non-injection cocaine use, sex with injection drug using partner, time to first cessation of injection drug use	Homelessness at prior visits associated with a longer time to injection drug use cessation (HR = 1.36, 95% CI 1.12, 1.65) and shorter time to return to use following cessation (HR = 0.56, 95% CI 0.47, 0.73) compared to no homelessness
Simon et al. (2017)	Homelessness or shelter residence vs. no homelessness in past 3 months at first clinic contact for buprenorphine treatment	Age, sex, race/ethnicity, employment, relationship status, incarceration, polysubstance use, substance use disorder treatment, mental health diagnoses	Homelessness or shelter residence in past 3 months at first clinic contact for buprenorphine treatment associated with a decreased likelihood of initiation of buprenorphine treatment (OR = 0.32, 95% CI 0.10, 1.02)
Stack et al. (2000)	Homeless vs. not homeless at treatment admission	None	Homelessness at admission not associated with completing outpatient substance use disorder treatment (69% homeless among those completing treatment vs. 63% among those not completing treatment, p > 0.05)
Tuten et al. (2003)	Homeless vs. not homeless at treatment admission	None	Homelessness at treatment admission not associated with number of days in residential substance use disorder treatment (90 vs. 127 days) or number of repeat admissions during the same pregnancy (data not shown) compared to no homelessness
Upshur et al. (2014)	Homeless vs. not homeless in past 90 days at study baseline and 3-, 6, and 12-month study visits	Age, race/ethnicity, income, education, insurance status, partner status, lifetime alcohol and drug use, outpatient treatment and 12-step meetings attended in the last 90 days, time	Homelessness at prior study visit not associated with importance of changing alcohol use behavior (OR = 0.67, 95% CI 0.34, 1.31), readiness to change alcohol use behavior (OR = 0.84, 95% CI 0.44, 1.60), confidence in success in changing alcohol use behavior (OR = 0.76, 95% CI 0.46, 1.26), importance of changing drug use behavior (OR = 0.75, 95% CI 0.33, 1.73), readiness to change drug use behavior (OR = 0.79, 95% CI 0.43, 1.42), or confidence in success in changing drug use behavior (OR = 0.78, 95% CI 0.48, 1.27) compared to no homelessness
Watkins et al. (2018)	Homeless vs. not homeless at study baseline	Enrollment site, treatment assignment, receipt of other evidence-based treatment	Homelessness at study baseline associated with a decreased likelihood of receiving behavioral therapy or medication-assisted treatment (OR = 0.18, 95% CI 0.09, 0.39), behavioral therapy (OR = 0.29, 95% CI 0.14, 0.62), or medication-assisted treatment (OR = 0.39, 95% CI 0.14, 1.06) for alcohol use disorder during the study period compared to no homelessness

b = beta estimate; CI = confidence interval; HR = hazard ratio; OR = odds ratio; RR = rate ratio.

3.2.1. Substance use

Seven studies demonstrated an association between homelessness and an increased likelihood of subsequent substance use (Calcaterra et al., 2014; Johnson and Fendrich, 2007; Linton et al., 2013; Neaigus et al., 2006; Polcin and Korcha, 2017; Riley et al., 2015; Shah et al., 2006). In particular, three studies found associations of homelessness with an increased likelihood of return to injection drug use among people who previously injected drugs (Linton et al., 2013; Shah et al., 2006), injection-related risk behaviors among people currently injecting drugs (Linton, Celentano, Kirk, & Mehta, 2013), transition to injection drug use among people injecting drugs (Neaigus et al., 2006), and a longer time to cessation among people injecting drugs (Shah, Galai, Celentano, Vlahov, & Strathdee, 2006). In addition, three studies demonstrated an association of homelessness with an increased likelihood of substance use, broadly defined (Calcaterra et al., 2014; Johnson and Fendrich, 2007), and stimulant use (Riley et al., 2015), and one study demonstrated an association of homelessness or marginal housing with higher peak density substance use (Polcin & Korcha, 2017). In contrast, one study did not find an association between homelessness and return to alcohol, heroin, or cocaine use following discharge from inpatient detoxification (Kertesz, Horton, Friedmann, Saitz, & Samet, 2003).

One study found that an increasing number of neighborhood foreclosures in the prior year was associated with a greater number of alcoholic drinks per week (Crawford et al., 2018). Three studies did not observe an association between unstable housing and later alcohol use (Asana, Ayvaci, Pollio, Hong, & North, 2018), cocaine use (Ayvaci, Obiri, Pollio, & North, 2018), or overall substance use (Calcaterra, Beaty, Mueller, Min, & Binswanger, 2014).

3.2.2. Substance use disorders

Five studies found an association of homelessness with an increased likelihood of SUD diagnoses or symptoms (Johnson et al., 1997; LePage et al., 2014; Moss et al., 2020; O’Toole et al., 2013). Specifically, prior homelessness was associated with an increased likelihood of later alcohol and other SUD symptoms (Johnson, Freels, Parsons, & Vangeest, 1997), severe alcohol (Moss et al., 2020) and SUD disorder symptoms (Buchholz et al., 2010), mild, moderate, and severe cannabis use disorder symptoms (Moss et al., 2020), and new SUD diagnoses (LePage et al., 2014; O’Toole et al., 2013). However, in one study, homelessness was not associated with mild or moderate alcohol use disorder symptoms (Moss et al., 2020), and in another, homelessness or marginal housing was not associated with more severe alcohol or other SUD symptoms (Polcin & Korcha, 2017).

In one study, unstable housing was associated with an increased likelihood of intensified SUD symptoms over time (Moos, Nichol, & Moos, 2002).

3.2.3. Motivation to change substance use

In one study, homelessness was not associated with subsequent motivation to change substance use behaviors, including feelings regarding the importance of changing alcohol or drug use behavior, readiness to change alcohol or drug use behavior, and confidence in potential success in changing alcohol or drug use behavior (Upshur et al., 2014).

3.2.4. Receipt of substance use disorder treatment.

Three studies demonstrated an association of homelessness with a decreased likelihood of initiating behavioral therapy for alcohol or opioid use disorder (Ober et al., 2018), buprenorphine (Simon et al., 2017), and behavioral therapy or medication-assisted treatment for alcohol use disorder (Watkins et al., 2018), though one did not observe an association between homelessness and initiation of medication-assisted treatment for alcohol or opioid use disorder (Ober et al., 2018). However, three studies found that homelessness was associated with an increased likelihood of SUD treatment, including inpatient (LePage, Bradshaw, Cipher, Crawford, & Hoosyhar, 2014), outpatient (O’Toole et al., 2013), and long-term treatment (Kleinman, Millery, Scimeca, & Polissar, 2002). One study did not find an association of homelessness with the number of days in residential SUD treatment or the number of repeat treatment admissions among pregnant women (Tuten, Jones, & Svikis, 2003).

Two studies did not observe an association of unstable housing with outpatient SUD treatment (Asana et al., 2018; Ayvaci et al., 2018).

3.2.5. Completion of substance use disorder treatment

Three studies found an association of homelessness with a decreased likelihood of SUD treatment completion, with one finding a decreased likelihood of completing outpatient SUD treatment (Bowser, Lewis, & Dogan, 2011), one finding a decreased likelihood of completing inpatient SUD treatment (Guerrero, 2013), and one finding an increased likelihood of attrition from outpatient alcohol contingency management (Leickly et al., 2018). Two studies found an association of homelessness with an increased likelihood of completing residential SUD treatment (Mutter, Ali, Smith, & Strashny, 2015) and residential alcohol detoxification (Mutter & Ali, 2019). Two studies did not observe an association of homelessness with completion of residential SUD treatment (Stack, Cortina, Samples, Zapata, & Arcand, 2000) or outpatient alcohol detoxification (O’Connor, Gottlieb, Kraus, Segal, & Horwitz, 1991).

One study found an association of unstable housing prior to treatment admission with a decreased likelihood of completing residential SUD treatment (Baker et al., 2019).

3.2.6. Overdose death

Six studies observed an association of homelessness with an increased likelihood of overdose death (Baggett et al., 2013, 2015; Binswanger et al., 2016; Fine et al., 2020; Kerker et al., 2011; O’Driscoll et al., 2001). Specifically, three studies found a higher rate of unintentional or undetermined overdose death (Baggett et al., 2013) and alcohol- or drug-attributable death (Baggett et al., 2015; Baker et al., 2019) among homeless individuals, two studies found an association of homelessness with an increased likelihood of unintentional drug overdose death (Binswanger et al., 2016; O’Driscoll et al., 2001), and one study found an association of homelessness with an increased likelihood of opioid overdose death (Fine, Yu, Triant, Baggett, & Metlay, 2020).

One study demonstrated a higher rate of methamphetamine, cocaine, and opioid overdose deaths among those residing in a single room occupancy building at the time of death (Rowe, Riley, Eagen, Zevin, & Coffin, 2019).

4. Discussion

We identified 38 peer-reviewed studies that examined the association of housing stress with later substance use outcomes. Overall, results demonstrated an association of homelessness with an increased likelihood of substance use (Calcaterra et al., 2014; Johnson and Fendrich, 2007; Linton et al., 2013; Neaigus et al., 2006; Polcin and Korcha, 2017; Riley et al., 2015; Shah et al., 2006), SUD diagnoses and symptoms (Johnson et al., 1997; LePage et al., 2014; Moss et al., 2020; O’Toole et al., 2013), and overdose death (Baggett et al., 2013, 2015; Binswanger et al., 2016; Fine et al., 2020; Kerker et al., 2011; O’Driscoll et al., 2001). However, results regarding the association of homelessness with receipt (Kleinman et al., 2002; LePage et al., 2014; Ober et al., 2018; O’Toole et al., 2013; Simon et al., 2017; Tuten et al., 2003; Watkins et al., 2018) and completion (Bowser et al., 2011; Guerrero, 2013; Leickly et al., 2018; Mutter et al., 2015; Mutter and Ali, 2019; O’Connor et al., 1991; Stack et al., 2000) of SUD treatment were varied, and one study did not observe an association of homelessness with motivation to change substance use behaviors (Upshur et al., 2014). Several studies did not show an association of unstable housing with substance use (Asana et al., 2018; Ayvaci et al., 2018; Calcaterra et al., 2014) or receipt of outpatient SUD treatment (Asana et al., 2018; Ayvaci et al., 2018) while others found an association of unstable housing with intensified SUD symptoms (Moos et al., 2002) and a decreased likelihood of completing SUD treatment (Baker et al., 2019). Broadly, these results lend support to the social causation hypothesis with respect to homelessness and later substance use, SUD, and overdose death, though results for other forms of housing stress and for some substance use outcomes are less consistent.

4.1. Methodological considerations and implications for research

Within this body of literature, there are several methodological considerations that have implications for interpretation of results and directions for future studies. Notable considerations specific to measures of housing stress and substance use outcomes, study populations, and analytic approaches are detailed below.

4.1.1. Housing stress measures

Most studies focused on the association of homelessness with substance use outcomes. More than half of these studies did not present a clear definition of homelessness (Baggett et al., 2013, 2015; Kleinman et al., 2002; Leickly et al., 2018; O’Driscoll et al., 2001; Shah et al., 2006; Stack et al., 2000; Watkins et al., 2018; Bowser et al., 2011; Buchholz et al., 2010; Calcaterra et al., 2014; Guerrero, 2013; Johnson et al., 1997; Johnson and Fendrich, 2007; Mutter et al., 2015; Mutter and Ali, 2019; Neaigus et al., 2006). In studies that did present a definition, definitions varied and represented heterogeneous housing experiences (Binswanger et al., 2016; Fine et al., 2020; Kerker et al., 2011; Kertesz et al., 2003; LePage et al., 2014; Linton et al., 2013; Moss et al., 2020; Ober et al., 2018; O’Connor et al., 1991; Simon et al., 2017; Tuten et al., 2003; Upshur et al., 2014; O’Toole et al., 2013; Polcin and Korcha, 2017; Riley et al., 2015). Moreover, most studies assessed homelessness at a single point in time and did not examine various dimensions of the experience of homelessness, such as frequency, duration, and setting (e.g., emergency shelter vs. street vs. motel), and whether housing status changed over time. Each of these dimensions may have differing implications for subsequent substance use outcomes (Fazel et al., 2014; Creech et al., 2015; Tong et al., 2019). An important direction for future research is to explicitly present definitions of homelessness to facilitate clear interpretation of the ways in which diverse homeless experiences are related to different substance use outcomes. Use of time-varying measures that account for the various dimensions of homelessness can inform our understanding of patterns and changes in these associations over time.

As noted above, most of the reviewed studies focused on homelessness, with a few additional studies examining housing instability (Asana et al., 2018; Ayvaci et al., 2018; Baker et al., 2019; Calcaterra et al., 2014; Moos et al., 2002). Multiple forms of housing stress, such as difficulty paying rent or mortgage, eviction or foreclosure, poor physical housing conditions (e.g., lack of hot water or electricity), and overcrowding, are currently unaddressed in the existing literature. The U.S. Department of Housing and Urban Development is developing a comprehensive index to capture multiple facets of housing stress including affordability, safety, quality, and loss of housing (U.S. Department of Housing and Urban Development, 2018). Incorporation of this index in future research could help standardize the measures used and enhance our understanding of the association of specific facets of housing stress with substance use outcomes.

4.1.2. Substance use outcomes

Overall, broad substance use measures were used in many of the reviewed studies, with a few studies examining specific substance types. Given the current opioid crisis (Centers for Disease Control and Prevention., 2018) and the burden of alcohol-attributable deaths in the U.S. (Esser et al., 2020), a continued focus on these substances is warranted. In addition, given recent increases in overdose deaths involving cocaine and stimulants (Kariisa, Scholl, Wilson, Seth, & Hoots, 2019), a closer examination of the ways in which housing stress may impact cocaine and stimulant use can inform targeted intervention.

Results regarding the association of housing stress with completion of residential and outpatient SUD treatment were mixed. Notably, measures of treatment completion lacked specificity regarding the type (e.g., medication-assisted treatment, behavioral therapy), length, and continuity of treatment. Most studies did not describe the requirements for “successful” treatment completion or the additional services offered as part of treatment (e.g., referral to social services). Thus, it is difficult to compare results across studies, determine the factors that may have contributed to the inconsistent findings, and understand which treatment models may be best suited for those with housing stress. In future research, greater specificity regarding treatment outcomes, including treatment type, length, and requirements, can help to clarify the association of housing stress with SUD treatment receipt and completion, furthering our understanding of the treatment modalities that are most accessible and effective for populations with housing stress.

4.1.3. Study populations

In most studies, the study population was primarily male. While SUD (McHugh, Votaw, Sugarman, & Greenfield, 2018), overdose death (Hedegaard, Minino, & Warner, 2019), and homelessness (U.S. Department of Housing and Urban Development, 2020) are more common among men than women, there have been significant increases in each among women in recent years (U.S. Department of Housing and Urban Development., 2020; McHugh et al., 2018; VanHouten et al., 2019). As rates of SUD, overdose, and homelessness continue to increase among women, an assessment of the association of housing stress with substance use outcomes specifically among women, and whether results differ from those among men, can guide interventions targeted to these growing public health issues.

Of the 12 studies that had predominantly white non-Hispanic participants, nine were among patients at SUD treatment or other healthcare facilities (Baker et al., 2019; Buchholz et al., 2010; Fine et al., 2020; LePage et al., 2014; Mutter et al., 2015; Mutter and Ali, 2019; O’Toole et al., 2013; Polcin and Korcha, 2017; Simon et al., 2017). In contrast, of the nine studies that had predominantly Black non-Hispanic participants, five were among individuals recruited from shelters or street locations (Asana et al., 2018; Ayvaci et al., 2018; Johnson et al., 1997; Linton et al., 2013; Shah et al., 2006). Structural racism plays an important role in determining which populations experience housing stress (Riley, 2018) and which have access to SUD treatment and other healthcare resources (Matsuzaka & Knapp, 2020; Feagin and Bennefield, 2014). In this body of literature, there is potential for structural racism to contribute to selection bias, depending on the methods used to identify participants, by impacting the racial/ethnic composition of the study population. In addition, depending on the substance use outcome examined, structural racism may confound results by contributing both to homelessness and the development of SUD, for example. Future studies would benefit from incorporation of measures of structural racism (Groos, Wallace, Hardeman, & Theall, 2018) and a careful consideration of how structural racism may impact the composition of study populations and, subsequently, study results.

4.1.4. Analytic approaches

Half of the reviewed studies examined predictors of substance use outcomes, with housing stress one of multiple predictors examined. In these studies, the analytic models were not constructed to assess a specific exposure-outcome association, with model covariates selected to adjust for potential confounding. Thus, in these studies there is likely residual or unmeasured confounding specific to the association of housing stress with the substance use outcome. Moreover, across all of the reviewed studies, including those that were designed specifically to assess the association of housing stress with substance use outcomes, key potential confounders were consistently missing, including measures of prior trauma (e.g., sexual violence victimization) and socioeconomic status (e.g., employment, education). As prior trauma and low socioeconomic status have been found to contribute to housing stress and substance use (Dasgupta et al., 2018; Hamilton et al., 2011), failure to adjusted for these variables may bias results away from the null, overestimating the association of housing stress with substance use outcomes. The literature base would benefit from additional studies focused specifically on quantifying the association of housing stress with substance use outcomes and use of causal inference tools, such as directed acyclic graphs (DAGS) (Austin, Desrosiers, & Shanahan, 2019), to identify potential confounders and construct analytic models.

4.2. Implications for practice

Despite these methodological considerations, this research offers potential directions for practice. The current literature suggests that housing stress, specifically homelessness, may increase substance use (Calcaterra et al., 2014; Johnson and Fendrich, 2007; Linton et al., 2013; Neaigus et al., 2006; Polcin and Korcha, 2017; Riley et al., 2015; Shah et al., 2006) and SUD (Johnson et al., 1997; LePage et al., 2014; Moss et al., 2020; O’Toole et al., 2013), and may also be an important risk factor for overdose death (Baggett et al., 2013, 2015; Binswanger et al., 2016; Fine et al., 2020; Kerker et al., 2011; O’Driscoll et al., 2001). Existing evidence regarding the impact of Housing First approaches on substance use outcomes is mixed or lacking, and fidelity to the intervention model and high intervention costs can be challenging (Tsai, 2020; Baxter et al., 2019; Fitzpatrick-Lewis et al., 2011; Kertesz et al., 2009; Kirst et al., 2015; Davidson et al., 2014; Padgett et al., 2006). There are, however, other innovative interventions to prevent or address homelessness that may function to mitigate acute substance use outcomes. The Vancouver New Leaf pilot project selected recently homeless adults to receive a one-time cash transfer of $7,500 (Zhao, Dwyer, & Palepu, 2020). A recent evaluation found that these individuals increased their spending on basic essentials, such as food and clothing, reduced their spending on alcohol, cigarettes, and drugs, and reduced their use of the emergency shelter system over the next 12 months (Zhao et al., 2020). The Homelessness Prevention Call Center in Chicago provides one-time financial assistance (average of $1,000) for individuals experiencing a crisis (e.g., job loss, medical emergency) and at imminent risk of homelessness (Evans, Sullivan, & Wallskog, 2016). A recent study indicated that those who received financial assistance, compared to those who did not receive assistance due to lack of available funds, were less likely to experience homelessness over the next 6 months (Evans et al., 2016). While additional research is needed, it is possible that cash transfer interventions may prevent “drift” into homelessness and subsequent problematic substance use. For those with SUD, a combination of financial assistance and treatment may be needed to achieve sustained improvement.

One reviewed study examined the association of homelessness with motivation to change substance use behaviors, finding high motivation among both women with recent homelessness and those consistently housed (Upshur et al., 2014). Prior studies have identified motivation as a key factor predicting SUD treatment initiation and completion (Brorson et al., 2013; DiClemente et al., 2004). Though additional studies are needed, this initial study suggests that motivation may not be a barrier to engaging in and completing SUD treatment among those experiencing housing stress. Individuals with housing stress and SUD may desire treatment, but face other barriers, including lack of financial resources and practitioner assumptions about their interest in or ability to benefit from treatment, that prevent successful initiation and completion (Wen et al., 2007; Nickasch and Marnocha, 2009). Identifying and addressing such barriers unique to individual practice settings may facilitate improved substance use outcomes among populations with housing stress.

4.3. Limitations

Several limitations are worth note. Our review was limited to peer-reviewed articles and thus excludes studies published in non-peer reviewed sources and unpublished articles that may have had null results. Though we conducted a systematic search of the peer-reviewed literature, it is also possible that relevant studies were not captured by the electronic database search. It is also possible that we missed relevant articles in our review of search results or that we did not correctly capture specific details during data abstraction. Last, given that consensus is currently lacking as to the “gold standard” tool to use to assess results from observational studies (Farrah et al., 2019; Page et al., 2018), we did not use a risk of bias tool. Instead, we created a thorough data abstraction form with abstraction domains covering common forms of bias in observational studies (Supplemental File 2).

4.4. Conclusions

Results from this systematic review provide some evidence in support of the social causation hypothesis, indicating that initiatives to address housing stress, a factor of the social and environmental context that is amenable to programmatic and policy intervention, may help to mitigate some substance use outcomes. Future research can enhance this body of literature by focusing on specific methodological issues, including housing stress measures, substance use outcomes, and analytic approaches, which will advance our understanding of these complex associations. Additional research is needed to evaluate the impact of innovative housing interventions, and subsequent substance use, that show early promise.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.addbeh.2021.107076.