Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2018 May 1.
Published in final edited form as: Psychol Serv. 2017 May;14(2):184–192. doi: 10.1037/ser0000111

Comprehensive Services Delivery and Emergency Department use Among Chronically Homeless Adults

David T Moore 1, Robert A Rosenheck 2
PMCID: PMC5424614  NIHMSID: NIHMS836696  PMID: 28481603

Abstract

Homeless adults use emergency department (ED) services more frequently than other adults but the relationships between homelessness, health status, outpatient service use, and ED utilization are poorly understood. Data from the Collaborative Initiative to Help End Chronic Homelessness (CICH) were used to compare ED use among chronically homeless adults who received comprehensive housing, case management, mental health/addiction and primary care services through CICH at five US sites (N=274) and ED use among control group clients receiving generally available community services (N=116) at the same sites. Multiple imputation was used to account for missing data and differential rates of attrition between the cohorts. Longitudinal models were constructed to compare ED use between the two groups during the first year after initiation of CICH services. A mediation analysis was performed to determine the relative contributions of being housed, the receipt of outpatient services, and health status to group differences in ED utilization. Participants receiving CICH services were significantly less likely to report ED use (OR = 0.78, 95% CI = 0.65–0.93) in the year after program entry. Decreased ED use by CICH participants was primarily mediated by decreased homelessness—not by increased access to case management, other outpatient services, or health status. This suggest that becoming housed is a key driver of reduced ED utilization and that efforts to provide housing for homeless adults may result in significantly decreased ED use. Further research is needed to determine the long-term effects of housing on health status and to develop services to improve health outcomes.

Keywords: homelessness, emergency department utilization, supportive housing, health outcomes, outpatient service utilization

Homeless adults use emergency department (ED) services with greater frequency than adults in the general population (D’Amore, Hung, Chiang, & Goldfrank, 2001; Doran, Raven, & Rosenheck, 2013; Hastings et al., 2011; Kushel, Perry, Bangsberg, Clark, & Moss, 2002; Mandelberg, Kuhn, & Kohn, 2000; Salit, Kuhn, Hartz, Vu, & Mosso, 1998; Tsai, Doran, & Rosenheck, 2013). High ED utilization among homeless adults reflects a complex combination of health and psychosocial factors. A number of studies have shown that high acuity of medical, psychiatric, and substance use problems account for much of the increased ED use among homeless adults (Chambers et al., 2013; Doran et al., 2013; Hansagi, Olsson, Sjoberg, Tomson, & Goransson, 2001; Hunt, Weber, Showstack, Colby, & Callaham, 2006; Mackelprang, Qiu, & Rivara, 2015; Padgett, Struening, Andrews, & Pittman, 1995; Sun, Burstin, & Brennan, 2003; Thakarar, Morgan, Gaeta, Hohl, & Drainoni, 2015; Tsai et al., 2013). However, it has also been shown that lack of stable housing and social stressors common among homeless people also directly contribute to higher rates of ED use (Brown et al., 2015; Chambers et al., 2013; Kushel, Gupta, Gee, & Haas, 2006; Rodriguez, Fortman, Chee, Ng, & Poon, 2009).

It has been suggested that high ED use reflects difficulty accessing other needed services to address the complex medical, psychiatric, and social problems in homeless populations (Chwastiak, Tsai, & Rosenheck, 2012; Ku, Scott, Kertesz, & Pitts, 2010; Kushel et al., 2006; Oates, Tadros, & Davis, 2009). However, a number of studies have found that individuals who use the ED frequently access a range of standard outpatient services at high rates (Doran et al., 2013; Hansagi et al., 2001; Hunt et al., 2006; Kushel, Vittinghoff, & Haas, 2001; LaCalle & Rabin, 2010; Sun et al., 2003; Tsai & Rosenheck, 2013), suggesting that typical community services may not provide the necessary levels of care for these homeless adults. Recently, there have been efforts to stabilize housing through supportive housing programs which provide both housing rent subsidies and clinical case management. Perhaps the largest study of supportive housing thus far, the multi-site At Home/Chez Soi (AH/CS) trial conducted in Canada produced mixed results (Stergiopoulos, Gozdzik, et al., 2015; Stergiopoulos, Hwang, et al., 2015). Participants were less likely to use ED after entry into supportive housing. However, the final analysis of this program did not find that ED use was consistently significantly lower in the group randomized to receive supportive housing than a comparison group receiving typical community services. There were significant differences at individual AH/CS sites (Currie, Moniruzzaman, Patterson, & Somers, 2014; Stergiopoulos et al., 2014) with some reporting significant reduction in ED use with supported housing. Likewise, a number of single site studies in America have found robust reductions in ED use after initiation of supportive housing programs (Basu, Kee, Buchanan, & Sadowski, 2012; Gilmer, Stefancic, Ettner, Manning, & Tsemberis, 2010; Larimer et al., 2009; Sadowski, Kee, VanderWeele, & Buchanan, 2009). Thus, further analyses of longitudinal trends in ED use are needed to better understand how supportive housing services affect ED utilization by homeless adults. In particular, since initiation of supportive housing is likely to coincide with changes in access to other health and social services, analyses should examine how different service elements may affect ED use in populations with a complex array of problems.

The Collaborative Initiative to Help End Chronic Homelessness (CICH) was a multi-site demonstration program initiated in 2004 that provided chronically homeless adults with access to permanent housing, case management, primary care, and mental health/substance use services in eleven American cities (NOFA, 2003; Rickards et al., 2010). Previous reports have shown that CICH participants were significantly more likely to have a case manager, access to primary care and mental health services, and had much reduced levels of homelessness relative to a non-randomized comparison cohort receiving usual community services (Mares & Rosenheck, 2011). Health service expenditures were found to decrease in both groups over time, however, overall costs were calculated to be higher in the CICH group, largely because of greater use of outpatient services (Mares & Rosenheck, 2011). Correlates of ED use prior to initiation of enriched CICH services have been studied and number of contributing health and psychosocial factors have been identified (Chwastiak et al., 2012; Moore & Rosenheck, 2016). A longitudinal analysis of ED use has not been conducted and it is not known how integrated housing services, case management, or clinical outcomes may have affected ED utilization over time. Furthermore, although there was significantly greater attrition among comparison participants over the course of the intervention, prior analysis of healthcare costs did not address potential biases introduced by the increased rates of dropout in the comparison cohort (Mares & Rosenheck, 2011).

In this post-hoc analysis, data from the CICH initiative are used to compare ED utilization among CICH participants receiving a comprehensive array of services and comparison clients who only had access to usual local community services. We seek to identify specific service elements and outcomes within the comprehensive CICH intervention that may mediate decreased ED use with the CICH intervention. More specifically ED utilization in the CICH and comparison cohorts were examined with longitudinal models to determine whether significant differences were observed in ED use and whether these differences were mediated by concomitant changes in housing outcomes, service use, or clinical status. As contrasted with prior analyses of the CICH dataset, multiple imputation statistical methods were used to adjust for differential loss of follow-up data between groups.

Method

Study Design

Source of data

The Collaborative Initiative to Help End Chronic Homelessness (CICH) was a multi-site demonstration program of assistance for chronically homeless adults funded jointly by the Department of Housing and Urban Development (HUD), the Department of Health and Human Services (HHS), and the Department of Veteran’s Affairs (VA). The program was implemented beginning in 2004 at 11 localities: Chattanooga, TN; Chicago, IL; Columbus, OH; Denver, CO; Fort Lauderdale, FL; Los Angeles, CA; Martinez, CA; New York City, NY; Philadelphia, PA; Portland, OR; and San Francisco, CA. Each site was responsible for development and implementation of outreach efforts to contact chronically homeless adults and provide comprehensive housing, case management, primary care and mental health services (Kresky-Wolff, Larson, O’Brien, & McGraw, 2010; Mares & Rosenheck, 2007, 2011). The primary entry criterion was chronic homelessness, defined as either having been homeless continuously for more than one year or having had four or more separate episodes of homelessness in the prior three years. There were no clinical exclusion criteria.

In Chattanooga, participants were housed in scattered-site one-bedroom apartments through housing vouchers. In Los Angeles, participants were housed in one of three Single Room Occupancy (SRO) hotels with onsite services. Participants in Martinez were placed in one-bedroom apartments with onsite services. In New York, clients were placed in SROs or apartments, most of which had onsite services. Portland participants were placed in units ranging from refurbished hotels to garden apartments. Recruitment methods were developed and implemented separately at each individual site and have been described elsewhere (Mares & Rosenheck, 2007, 2011).

Homeless adults were formally screened for program entry. Participants then consented to participation in data collection for the program evaluation. Receipt of housing and services was not contingent upon participation in the evaluation process. Participants received $15 remuneration per interview. Written informed consent was provided by each participant and approved of by the Institutional Review Boards at the 11 individual sites and the coordinating site at the VA Northeast Program Evaluation Center in West Haven, Connecticut.

To provide comparison data with which to evaluate the distinct impact of the CICH intervention five of the 11 CICH study sites also recruited comparison participants. Recruitment methods for comparison clients has been described in detail in prior publications (Mares & Rosenheck, 2007, 2011). Comparison clients were recruited from an area of the same city in which the full intervention was not available and who received standard service available to homeless people in those communities (Chattanooga, Los Angeles, Martinez, New York, and Portland). Recruitment of the comparison cohort was approved by the same local IRBs as approved primary CICH recruitment. Research staff identified local portals of entry into pre-existing non-CICH housing, case management, and supportive services at each site. Working with staff at these service portals, they research staff identified chronically homeless adults who were intended to be demographically similar to those participating in CICH. Data from a total of 274 CICH and 116 comparison participants at the 5 sites were used for the analyses presented here, excluding 24 participants (5.8%) who died during the first year of follow-up.

Data collection

CICH staff were trained in a two-day training workshop in which all procedures and measures were reviewed. Assessments were performed at baseline and every three months during follow up through face-to-face interviews. Follow up interviews were administered regardless of housing or case management status.

Measures

Emergency department use

Clients reported “yes” or “no” to three questions of whether they had received medical, psychiatric, or substance use services from an ED during the prior 90 days. For the purpose of statistical analyses, these were combined into a single dichotomous measure of “any ED use” during the prior 90 days.

Sociodemographic measures

Interviews also documented age, race, gender, employment, residential status, and legal history.

Residential status

A series of 10 questions documented where each client had slept in the previous 90 days. Homelessness included living in a shelter, the outdoors, an abandoned building, or a car. For the purpose of statistical analyses, these were combined into a single dichotomous measure of “any homelessness” during the prior 90 days.

Physical health status

The presence of 27 medical problems involving a range of body systems was evaluated by self-report. The Medical Outcomes Study Short Form (SF-12) physical component score was used to assess physical functioning and related quality of life (Larson, 2002). Scores ranged from 0 to 100, with higher scores reflecting increased functioning.

Mental health status

Participants reported whether they had ever been diagnosed with each the following eight psychiatric conditions: schizophrenia, another psychotic disorder, major depression, bipolar disorder, a personality disorder, PTSD, an adjustment reaction, or an anxiety disorder. The Medical Outcomes Study Short Form (SF-12) SF-12 mental health component score (Larson, 2002) was used to assess mental health related quality of life. Scores ranged from 0 to 100, with higher scores indicating better functioning. In a study of homeless individuals, the SF-12 physical health and mental health were found to have reliabilities of α = .82 and .79, respectively, and convergent validities ranging from .62 to .88 (Larson, 2002).

Substance use

Items from the Addiction Severity Index (ASI) (Mclellan, Luborsky, Woody, & Obrien, 1980) were used to assess alcohol and drug use. Scores ranged from 0 to 1, with higher scores indicating more severe substance use. In a homeless cohort, the ASI alcohol and drug scales were found to have reliabilities of α = .87 and .70, respectively. The validity of the ASI alcohol and drug scales in a group of homeless adults was found to be acceptable with correlations with other metrics ranging from .31 to .54 (Zanis, McLellan, Cnaan, & Randall, 1994).

Healthcare and social services access and utilization

Participants reported whether they received outpatient mental health, or substance treatment in the previous 90 days. Participants were asked whether they had a case manager, a primary medical provider, and a regular source of mental health and/or substance use treatment.

Statistical Analysis

All statistical analyses were performed in SAS 9.4. Baseline characteristics of CICH and comparison groups were compared using chi-squared and T-tests. There was a substantial difference in attrition over the first year with 17% of CICH participants lost to follow-up and 41% of comparison participants. To adjust for potential selection biases introduced by these differential rates of attrition multiple imputation was performed on each cohort. Multiple imputation is a statistical technique that uses the characteristics of participants lost to follow-up to impute the likely outcomes that would have been observed in the absence of missing data and incorporates the uncertainty generated by this procedure. The imputations are repeated numerous times and then synthesized. Forty imputations were performed using PROC MI (SAS). Because the missing data pattern was not monotone and the dataset contained a mixture of continuous and dichotomous variables, multiple imputation by chained equations was used to impute missing values (van Buuren, 2007). Continuous and binary variables were modeled using linear and logistic regression respectively.

Data were assumed to be missing at random (MAR). Techniques for testing the assumption of MAR are areas of active research (Sterne et al., 2009). In order for MAR to be satisfied, the probability that data are missing must depend on the observed data and not the missing data being imputed. A broad range of variables were included in the imputation model (legal problems; physical health; psychiatric problems; substance use; housing instability; demographic characteristics; site location) each of which was hypothesized to be likely to affect loss to follow-up and therefore likely to account for observed missingness (Sterne et al., 2009).

To further improve the accuracy of the modeling of missing outcome data and increase the plausibility of the MAR assumption, auxiliary sociodemographic factors (childhood homelessness; current homelessness at intake interview; recent legal history; veteran status), service use characteristics (social services; health insurance status; use of inpatient services), and clinical characteristics (self-reported medical, psychiatric, and substance related problems; psychiatric symptomatology) were included. Baseline values as well as the most recent values prior to each missing observation were included as independent variables in the imputation analysis.

In the primary analysis of ED use, a longitudinal logistic model of quarterly ED use (prior 90 days) during the first year was constructed. Generalized estimating equations (GEEs) with exchangeable correlation structures were used to account for within-individual correlation of observations. To determine the difference in ED use between the CICH participants and the comparison group, program participation (CICH vs. comparison) was evaluated as the primary independent variable of interest. The model was adjusted for baseline ED use and baseline characteristics that significantly differed between the two cohorts (P<.15). A cutoff of P<.15 was chosen in order to error on the side of including potential confounders. To determine whether results were sensitive to baseline characteristics, sensitivity analysis was conducted using sets of baseline covariates with more stringent (P<.05) and more relaxed (P<.25) cutoffs, as well as with the entire set of possible covariates described in Table 1. These different combinations of covariates did not significantly affect the relationship between ED use and CICH participation. GEE analyses of the 40 imputed datasets were combined using PROC MIANALYZE (SAS) to generate estimates of odds ratios and confidence intervals.

Table 1.

Baseline Characteristics of CICH and Comparison Participants

CICH (N=274) Comparison (N=116)
Characteristic N (%)/Mean (SD) N (%)/Mean (SD) χ2/F p
Male (%) 205 (74.8) 95 (81.9) 2.3 0.129
Mean age (SD) 44.9 (9.3) 46 (9.9) −1.1 0.295
Site
 Chattanooga (%) 49 (17.9) 19 (16.4) 5.33 0.255
 Los Angeles (%) 60 (21.9) 18 (15.5)
 Martinez (%) 50 (18.3) 31 (26.7)
 New York (%) 49 (17.9) 24 (20.7)
 Portland (%) 66 (24.1) 24 (20.7)
Race
 White (%) 112 (41.3) 39 (33.9) 5.52 0.138
 Black (%) 111 (41) 54 (47)
 Hispanic (%) 31 (11.4) 19 (16.5)
 Other (%) 17 (6.3) 3 (2.6)
Veteran 91 (33.2) 41 (35.3) 0.17 0.674
Employed (%) a 56 (20.4) 30 (25.9) 1.39 0.238
Convicted of felony (%) b 129 (47.3) 64 (55.2) 2.04 0.153
Recently homeless (%) c 237 (86.5) 94 (81) 1.89 0.169
Homeless 4 episodes (%) d 189 (69) 64 (55.2) 6.82 0.009
Visited ED c 106 (38.7) 43 (37.4) 0.06 0.810
Mean medical problems (SD) e 4.8 (3.4) 4.3 (3) 1.25 0.210
Mean SF12 physical health (SD) f 43.9 (9.9) 43.2 (11.1) 0.62 0.539
Mean mental health problems (SD) g 2.4 (1.9) 1.4 (1.6) 5.05 <.001
Mean SF12 mental health (SD)f 37.5 (8.3) 40.3 (9.4) −2.9 0.004
Mean ASI alcohol use (SD) h 0.1 (0.2) 0.1 (0.2) −0 0.976
Mean ASI drug use (SD) h 0.1 (0.1) 0.1 (0.1) 0.9 0.368
Open in a new tab
a

Past 30 days

b

Lifetime

c

Past 90 days

d

Past two years

e

Possible scores ranging from 0 to 27, with higher scores indicating more medical problems.

f

SF12 = Medical Outcomes Study Short Form-12. Possible scores ranging from 0 to 100, with higher scores indicating better health

g

Possible scores ranging from 0 to 7, with higher scores indicating more mental health problems.

h

ASI = Addiction Severity Index. Possible scores ranging from 0 to 1, with higher scores indicating greater addiction severity

Next a mediation analysis was performed to determine which of the following might account for (or mediate) differences in ED use observed between CICH and comparison participants: clinical outcomes; housing outcomes; access to specialized case management, primary care, or mental health services; or use of general outpatient medical or mental health care. For each potential mediator, a determination was made whether it satisfied the three criteria for mediation as described by Baron and Kenny (Baron & Kenny, 1986). These three criteria for significant mediation effects, as applied here, require that: 1) CICH participation is significantly associated with the mediator (e.g. access to case management services), 2) the mediator is itself significantly associated with the outcome of interest (ED use), and 3) adjustment for the mediator in multivariate analysis results in disappearance of the statistical significance of the association between CICH participation and ED use. To test criterion 1, each potential mediator was modeled using longitudinal GEEs to determine the association between CICH participation and the potential mediator. Logistic GEEs were used to model homelessness, service access, and service utilization as dichotomous variables indicating whether the participant had been homeless, had access to a service, or used a service during the prior 90 days. Clinical measures of health status were measured using longitudinal linear GEE models. To test criteria 2 and 3, ED use was modeled with the mediator included as a covariate. These models allowed evaluation of the relationship of each potential mediator with ED (criterion 2) use as well its effect on the statistical significance of the previously observed relationship between CICH participation and ED use (criterion 3).

To summarize, these analyses sought to determine whether receipt of the CICH intervention package was associated with reduced ED use and which treatment elements and outcomes of the intervention were likely to be most responsible for the observed effects of the overall multi-component intervention.

Results

As reported in prior studies (Mares & Rosenheck, 2011), the CICH and comparison cohorts did not differ significantly on sociodemographic measures at the time of program entry with the exception that CICH participants reported a greater number of episodes of homelessness in the previous two years (Table 1). There were no significant differences in measures of physical health status and substance use, however there were increased rates of mental illness and worse psychiatric functional status as measured by the SF-12 at baseline among those receiving the full CICH intervention (Table 1). The two groups also differed in attrition rate, with nearly 17% of CICH and 41% of comparison participants not reporting data by the end of the first year (p < .001). In the face of such a significant difference in attrition, multiple imputation was used to estimate outcomes for missing data.

At baseline, prior to initiation of services, 106 CICH participants (39%) and 43 members of the comparison group (38%) reported presenting to an emergency department during the prior 90 days. Participants were asked quarterly, whether they had visited an ED during the prior 90 days. When assessed at the end of the first year, only 25% of CICH participants reported any ED during the prior 90 days, as opposed to 41% of the comparison group. Longitudinal logistic Generalized Estimating Equations were used to model quarterly reports of any ED during the first year and compare whether CICH participants were more or less likely to report ED use than individuals in the comparison group receiving typical community services. The CICH cohort was significantly less likely to report ED use (OR = .78, 95% CI .65 – .93, p = .004).

A mediation analysis was performed (Baron & Kenny, 1986) to determine whether this relative decrease in reported ED use was attributable to changes in outpatient services, stabilization of housing, or improvements in health status. Longitudinal GEE models of potential mediators were constructed to determine whether participation in CICH was associated with other service or clinical outcomes (Table 2, Mediation Criterion 1). The original GEE model of longitudinal ED use was then adjusted for these potential mediators to determine whether there were significant relationships between these outcomes and report of an ED visit (Table 3, Mediation Criterion 2) and whether adjustment for a potential mediator significantly diminished the relationship between CICH participation and ED use (Table 4, Mediation Criterion 3). Of identified service and clinical outcomes, only decreased homelessness in the CICH cohort was found to satisfy these three criteria for mediation: CICH participants were less likely to report being homeless during the prior 90 days (Table 2); homelessness was strongly associated with ED use (Table 3); and adjustment for homelessness significantly attenuated the relationship between CICH participation and ED use (Table 4).

Table 2.

Mediation Criterion 1: Relationships Between CICH Participation and Potential Mediators of ED Use

Dependent variable: potential mediators of ED use OR (95% CI) a βb t c p
Recently homeless d 0.29 (0.23 – 0.37) −1.22 −10.16 <.001
Access to a case manager d 3.15 (2.37 – 4.19) 1.15 7.92 <.001
Access to primary medical provider d 1.28 (0.97 – 1.69) 0.25 1.77 0.077
Access to MH/SA provider d 1.7 (1.38 – 2.09) 0.53 4.97 <.001
Used outpatient medical services d 1.1 (0.91 – 1.33) 0.09 0.97 0.335
Used outpatient MH/SA services d 1.34 (1.11 – 1.63) 0.29 2.98 0.003
SF12 physical health e 0.40 0.90 0.371
SF12 mental health e 0.03 0.08 0.935
ASI alcohol use f −0.003 −0.39 0.696
ASI drug use f −0.004 −1.16 0.248
Open in a new tab

Note: Longitudinal GEEs for service outcomes in column 1 were adjusted for baseline values of the dependent variable as well as characteristics that significantly differed (P<.15) in CICH and comparison cohorts (gender, race, four or more episodes of homelessness in the prior two years, number of mental health problems, SF12 mental health measure). In each case, the dependent variable is the outcome in column 1. Both odds ratios and regression coefficients reflect the effect of participation in CICH on the outcome.

a

Odds Ratio for CICH vs Comparison cohort in predicting potential mediators of ED use.

b

Regression coefficient for CICH vs Comparison cohort.

c

t-statistic

d

Past 90 days

e

SF12 = Medical Outcomes Study Short Form-12. Possible scores ranging from 0 to 100, with higher scores indicating better health

f

ASI = Addiction Severity Index. Possible scores ranging from 0 to 1, with higher scores indicating greater addiction severity

Table 3.

Mediation Criterion 2: Relationships Between Potential Mediators and Longitudinal ED Utilization Following Study Entry.

Covariate OR (95% CI) a βb t c p
Recently homeless d 1.78 (2.48 – 1.28) −0.58 −3.46 0.001
Access to a case manager d 0.81 (1.21 – 0.54) 0.21 1.03 0.304
Access to primary medical provider d 1.19 (1.82 – 0.77) −0.17 −0.79 0.430
Access to MH/SA provider d 1.16 (1.55 – 0.87) −0.15 −0.98 0.326
Used outpatient medical services d 1.10 (1.45 – 0.84) −0.10 −0.68 0.494
Used outpatient MH/SA services d 1.16 (1.53 – 0.87) −0.14 −1.00 0.316
SF12 physical health e 0.97 (0.98 – 0.95) 0.03 3.99 <.0001
SF12 mental health e 1.00 (1.02 – 0.98) 0.00 0.03 0.973
ASI alcohol use f 2.3 (5.35 – 0.99) −0.83 −1.94 0.053
ASI drug use f 1.92 (10.71 – 0.34) −0.65 −0.74 0.458
Open in a new tab

Note: Longitudinal GEEs of whether participants visited an ED in the prior 90 days were adjusted for baseline values of the dependent variable as well as characteristics that significantly differed (P<.15) in CICH and comparison cohorts (gender, race, four or more episodes of homelessness in the prior two years, number of mental health problems, SF12 mental health measure). In each case, the model was adjusted for the possible mediator in column 1. Both odds ratios and regression coefficients reflect the effect the mediator on ED use.

a

Odds Ratio for any ED use in the prior 90 days. For continuous measures, the odds ratio is for each unit change.

b

Regression coefficient for any ED use in the prior 90 days.

c

t-statistic

d

Past 90 days

e

SF12 = Medical Outcomes Study Short Form-12. Possible scores ranging from 0 to 100, with higher scores indicating better health

f

ASI = Addiction Severity Index. Possible scores ranging from 0 to 1, with higher scores indicating greater addiction severity

Table 4.

Mediation Criterion 3: Statistical Significance of Relationship between CICH Participation and ED Use After Adjustment for Possible Mediators

Covariate OR (95% CI) a βb t c p
Recently homeless d 0.87 (0.73 – 1.04) −0.14 −1.54 0.124
Access to a case manager d 0.81 (0.67 – 0.98) −0.22 −2.21 0.028
Access to primary medical provider d 0.77 (0.65 – 0.92) −0.26 −2.90 0.004
Access to MH/SA provider d 0.76 (0.63 – 0.91) −0.28 −2.98 0.003
Used outpatient medical services d 0.77 (0.65 – 0.92) −0.26 −2.90 0.004
Used outpatient MH/SA services d 0.76 (0.64 – 0.91) −0.27 −2.94 0.003
SF12 physical health e 0.77 (0.64 – 0.92) −0.26 −2.92 0.004
SF12 mental health e 0.78 (0.65 – 0.92) −0.25 −2.86 0.004
ASI alcohol use f 0.77 (0.65 – 0.92) −0.26 −2.92 0.004
ASI drug use f 0.77 (0.65 – 0.92) −0.26 −2.91 0.004
Open in a new tab

Note: Longitudinal GEEs of whether participants visited an ED in the prior 90 days were adjusted for baseline values of the dependent variable as well as characteristics that significantly differed (P<.15) in CICH and comparison cohorts (gender, race, four or more episodes of homelessness in the prior two years, number of mental health problems, SF12 mental health measure). In each case, the model was adjusted for the possible mediator in column 1. Both odds ratios and regression coefficients reflect the effect of participation in CICH on ED use after adjustment for the possible mediator.

a

Odds Ratio for any ED use in the prior 90 days. For continuous measures, the odds ratio is for each unit change.

b

Regression coefficient for any ED use in the prior 90 days.

c

t-statistic

d

Past 90 days

e

SF12 = Medical Outcomes Study Short Form-12. Possible scores ranging from 0 to 100, with higher scores indicating better health

f

ASI = Addiction Severity Index. Possible scores ranging from 0 to 1, with higher scores indicating greater addiction severity

While CICH participants were more likely to report access to a case manager during the follow up year (Table 2), this access was not significantly associated with ED use (Table 3), and adjustment for case management did not significantly diminish the relationship between CICH and ED use (Table 4). Likewise, though associated with CICH participation, neither access to nor utilization of mental health services was strongly associated with ED use (Table 3) and adjustment for these factors did not significantly affect the relationship between ED use and participation in CICH (Table 4). Relative to the comparison group, at quarterly assessments, CICH participants did not report significantly improved subjective physical or mental health as measured by the SF12 or decreased alcohol or drug use as measured by the ASI (Table 2). Though, better subjective physical health as measured by the SF12 was strongly associated with lower levels of ED utilization (Table 3), adjustment for SF12 physical health status did not affect the relationship between CICH and ED use (Table 4), suggesting that decreased ED utilization was not mediated by differential improvement in physical health status. Quarterly mental health and substance use measures were not significantly associated with ED use during follow up and did not affect the relationship between CICH participation and ED use (Table 4).

Discussion

This longitudinal analysis of a multi-site supported housing and intensive heath service intervention suggests that coordinated efforts providing both housing and clinical resources to chronically homeless adults can lead to improved housing outcomes, increased access to case management and ambulatory mental health services, and decreased emergency department utilization. It appears that CICH participants were less likely to use emergency department services and this difference was primarily mediated by the fact that they were less likely to be homeless after initiation of enriched CICH services. Though health status is clearly a factor influencing ED utilization by homeless adults, there was no evidence that decreased ED use by participants receiving CICH services could be attributed to improved access to health services or measured changes in substance use or in physical or mental health wellbeing.

This is the first multi-site evaluation of the effects of a housing program on ED utilization in American cities and is consistent with multiple prior single-site efforts that have found significantly decreased use of a variety of acute health services in the context of supportive housing (Basu et al., 2012; Gilmer et al., 2010; Larimer et al., 2009; Sadowski et al., 2009). The one other large multi-site study, the At Home/Chez Soi (AH/CS) trial, while extremely successful at stabilizing housing, produced unclear results regarding ED use (Goering et al., 2014; Stergiopoulos, Gozdzik, et al., 2015; Stergiopoulos, Hwang, et al., 2015). In AH/CS, ED utilization decreased in the group receiving the housing intervention, however it also decreased in the comparison cohort, resulting in a non-significant trend towards less use in those receiving the housing intervention at one year, though at other time-points there were significant differences between the groups. It is therefore difficult to make a direct comparison between the CICH and AH/CS findings, however, the fact that the treatment as usual cohort in AH/CS saw substantial decreases in ED use while the comparison group in CICH did not suggests that there may have been significant differences in the non-intervention services available in the two studies. It is possible that differences between the Canadian and American healthcare systems—in particular the availability of universal health insurance in Canada—could have affected the extent to which homeless individuals accessed emergency departments and other services. Since, community service providers may have been more readily able to link homeless adults not receiving the AH/CS with specific health services in the Canadian context, alternatives to ED services may have been more accessible. There was also substantially greater attrition by the comparison cohort in CICH than in AH/CS, suggesting that typical community services in CICH were less able to retain participants, possibly predisposing them to loss of housing or worse clinical outcomes that may precipitate ED use.

There was no evidence in this study that CICH resulted in health status improvements during the first year of CICH program participation, despite increased access to outpatient care. Therefore, not surprisingly, decreased ED use within the CICH cohort was not mediated by changes in medical or mental health wellbeing. A number of other analyses of supportive housing programs have also found decreased use of acute inpatient and ED services without appreciable changes in health measures (Gilmer et al., 2010; Sadowski et al., 2009; Stergiopoulos, Hwang, et al., 2015), although some reports have found evidence of greater reductions in in substance use (Cheng, Lin, Kasprow, & Rosenheck, 2007; Larimer et al., 2009) and overall quality of life (Larimer et al., 2009). It is possible that the relatively short time frame of this current study and other housing interventions may not be sensitive to long-term improvements in chronic conditions. Further research is needed to better understand how housing affects chronic conditions—both in the short and long term.

Though the interplay between housing, clinical symptoms, and health requires further study, this current study speaks to the likelihood that lack of stable housing is a key modifiable factor in why homeless adults use ED services at relatively high rates. The CICH intervention was not specifically intended to decrease the use of emergency departments, yet appears to have mediated a significant drop in ED utilization relative to adults receiving typical community services primarily through housing stabilization. Beyond fostering the goal of providing stable housing and an exit from homelessness, housing services appear to significantly reduce the need and demand for ED services. Exactly why being homeless or having unstable housing is tied to use of emergency services is not directly addressed by our data. It is possible that measures such as the SF12 or the ASI, used in this current study, were not sensitive to important changes in health status that might have been associated with reduced homelessness and reduced need for ED services. However, it is also possible that emergency departments provide a refuge, a place of safety, humane contact, and shelter for people with no place to call home.

There are a number of limitations to this study that deserve comment. While this analysis included multiple sites, there were only a modest number of participants and both the CICH and comparison cohorts may not be representative of chronically homeless adults in the US as participation may reflect differences in clinical status, access to services, or functioning across sites and in different regions of the country. Furthermore, measures of housing, service use, and health status were all based on client self-report and are therefore of unknown reliability and validity. Though some clinical measures used in this study have been validated in homeless adults, it is possible that the measures used did not capture unknown clinical factors that could have changed after initiation of CICH and could have therefore mediated the observed decrease in ED utilization. ED use was modeled as a dichotomous variable and therefore does not provide information regarding individuals who frequently use the ED. While efforts were made to adjust for differential dropout rates analytically, given the significantly different dropout rates between the two groups, unknown biases that could potentially have skewed our results may have been introduced by multiple imputation—especially if MAR was not an accurate assumption. Mediation was defined as loss of statistical significance between CICH and ED use and did not attempt to differentiate subtler partial mediation affects. As this study was carried out between 2004 and 2008, it is possible that changes in available homeless services and the general shift towards housing first programs in many communities may limit the applicability of these results. Likewise, the Affordable Care Act and Medicaid expansion have been implemented in the meantime and may affect how homeless adults access health services in some states. Finally, we acknowledge that the study was not based on random assignment and while analyses were adjusted for differences between CICH participants and the comparison cohort, it is conceivable that differences between the groups could have affected outcomes in unmeasured ways.

Conclusion

These results add to a growing body of evidence that supportive housing programs can substantially decrease the use of ED services by chronically homeless adults and conversely that lack of housing is a major reason for elevated used of ED services among homeless, and particularly chronically homeless, adults. The mediation analysis presented here suggests that among the outcomes resulting from CICH, reduced homelessness was the primary mediator of decreased ED use and that, by implication, lack of stable housing is a key driver of increased ED use among homeless adults. It also implies that providing housing for homeless adults is likely to result in significantly decreased ED use, even in the absence of significant changes in short-term health status. Further research is needed to better understand how health outcomes can be improved in this population, which can be expected to require a complex array of psychosocial and health related interventions.

Acknowledgments

This research was supported by the National Institutes of Mental Health and the Department of Veteran’s Affairs. David Moore conceptualized the study, analyzed the data, and wrote the article. Robert Rosenheck conceptualized the study and assisted with writing the article.

Contributor Information

David T. Moore, Yale University

Robert A. Rosenheck, Yale University and Veterans Affairs New England Mental Illness Research, Education, and Clinical Center

References

  1. Baron RM, Kenny DA. The moderator-mediator variable distinction in social psychological research: conceptual, strategic, and statistical considerations. J Pers Soc Psychol. 1986;51(6):1173–1182. doi: 10.1037//0022-3514.51.6.1173. [DOI] [PubMed] [Google Scholar]
  2. Basu A, Kee R, Buchanan D, Sadowski LS. Comparative cost analysis of housing and case management program for chronically ill homeless adults compared to usual care. Health Services Research. 2012;47(1 Pt 2):523–543. doi: 10.1111/j.1475-6773.2011.01350.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Brown RT, Miao Y, Mitchell SL, Bharel M, Patel M, Ard KL, … Steinman MA. Health Outcomes of Obtaining Housing Among Older Homeless Adults. Am J Public Health. 2015;105(7):1482–1488. doi: 10.2105/AJPH.2014.302539. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Chambers C, Chiu S, Katic M, Kiss A, Redelmeier DA, Levinson W, Hwang SW. High utilizers of emergency health services in a population-based cohort of homeless adults. Am J Public Health. 2013;103(Suppl 2):S302–310. doi: 10.2105/AJPH.2013.301397. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Cheng AL, Lin H, Kasprow W, Rosenheck RA. Impact of supported housing on clinical outcomes: analysis of a randomized trial using multiple imputation technique. Journal of Nervous and Mental Disease. 2007;195(1):83–88. doi: 10.1097/01.nmd.0000252313.49043.f2. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Chwastiak L, Tsai J, Rosenheck R. Impact of health insurance status and a diagnosis of serious mental illness on whether chronically homeless individuals engage in primary care. Am J Public Health. 2012;102(12):e83–89. doi: 10.2105/AJPH.2012.301025. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Currie LB, Moniruzzaman A, Patterson ML, Somers JM. At Home/Chez Soi Project: Vancouver Site Final Report. Mental Health Commission of Canada; 2014. [Google Scholar]
  8. D’Amore J, Hung O, Chiang W, Goldfrank L. The epidemiology of the homeless population and its impact on an urban emergency department. Acad Emerg Med. 2001;8(11):1051–1055. doi: 10.1111/j.1553-2712.2001.tb01114.x. [DOI] [PubMed] [Google Scholar]
  9. Doran KM, Raven MC, Rosenheck RA. What drives frequent emergency department use in an integrated health system? National data from the Veterans Health Administration. Ann Emerg Med. 2013;62(2):151–159. doi: 10.1016/j.annemergmed.2013.02.016. [DOI] [PubMed] [Google Scholar]
  10. Gilmer TP, Stefancic A, Ettner SL, Manning WG, Tsemberis S. Effect of full-service partnerships on homelessness, use and costs of mental health services, and quality of life among adults with serious mental illness. Arch Gen Psychiatry. 2010;67(6):645–652. doi: 10.1001/archgenpsychiatry.2010.56. [DOI] [PubMed] [Google Scholar]
  11. Goering P, Veldhuizen S, Watson A, Adair C, Kopp B, Latimer E, … Aubry T. National Final Report Cross-Site At Home/Chez Soi Project 2014 [Google Scholar]
  12. Hansagi H, Olsson M, Sjoberg S, Tomson Y, Goransson S. Frequent use of the hospital emergency department is indicative of high use of other health care services. Ann Emerg Med. 2001;37(6):561–567. doi: 10.1067/mem.2001.111762. [DOI] [PubMed] [Google Scholar]
  13. Hastings SN, Smith VA, Weinberger M, Schmader KE, Olsen MK, Oddone EZ. Emergency department visits in Veterans Affairs medical facilities. Am J Manag Care. 2011;17(6 Spec No):e215–223. [PMC free article] [PubMed] [Google Scholar]
  14. Hunt KA, Weber EJ, Showstack JA, Colby DC, Callaham ML. Characteristics of frequent users of emergency departments. Ann Emerg Med. 2006;48(1):1–8. doi: 10.1016/j.annemergmed.2005.12.030. [DOI] [PubMed] [Google Scholar]
  15. Kresky-Wolff M, Larson MJ, O’Brien RW, McGraw SA. Supportive housing approaches in the Collaborative Initiative to Help End Chronic Homelessness (CICH) J Behav Health Serv Res. 2010;37(2):213–225. doi: 10.1007/s11414-009-9206-y. [DOI] [PubMed] [Google Scholar]
  16. Ku BS, Scott KC, Kertesz SG, Pitts SR. Factors associated with use of urban emergency departments by the U.S. homeless population. Public Health Rep. 2010;125(3):398–405. doi: 10.1177/003335491012500308. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Kushel MB, Gupta R, Gee L, Haas JS. Housing instability and food insecurity as barriers to health care among low-income Americans. J Gen Intern Med. 2006;21(1):71–77. doi: 10.1111/j.1525-1497.2005.00278.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Kushel MB, Perry S, Bangsberg D, Clark R, Moss AR. Emergency department use among the homeless and marginally housed: results from a community-based study. Am J Public Health. 2002;92(5):778–784. doi: 10.2105/ajph.92.5.778. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Kushel MB, Vittinghoff E, Haas JS. Factors associated with the health care utilization of homeless persons. JAMA. 2001;285(2):200–206. doi: 10.1001/jama.285.2.200. [DOI] [PubMed] [Google Scholar]
  20. LaCalle E, Rabin E. Frequent users of emergency departments: the myths, the data, and the policy implications. Ann Emerg Med. 2010;56(1):42–48. doi: 10.1016/j.annemergmed.2010.01.032. [DOI] [PubMed] [Google Scholar]
  21. Larimer ME, Malone DK, Garner MD, Atkins DC, Burlingham B, Lonczak HS, … Marlatt GA. Health care and public service use and costs before and after provision of housing for chronically homeless persons with severe alcohol problems. JAMA. 2009;301(13):1349–1357. doi: 10.1001/jama.2009.414. [DOI] [PubMed] [Google Scholar]
  22. Larson CO. Use of the SF-12 instrument for measuring the health of homeless persons. Health Services Research. 2002;37(3):733–750. doi: 10.1111/1475-6773.00046. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Mackelprang JL, Qiu Q, Rivara FP. Predictors of Emergency Department Visits and Inpatient Admissions Among Homeless and Unstably Housed Adolescents and Young Adults. Med Care. 2015;53(12):1010–1017. doi: 10.1097/MLR.0000000000000436. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Mandelberg JH, Kuhn RE, Kohn MA. Epidemiologic analysis of an urban, public emergency department’s frequent users. Acad Emerg Med. 2000;7(6):637–646. doi: 10.1111/j.1553-2712.2000.tb02037.x. [DOI] [PubMed] [Google Scholar]
  25. Mares AS, Rosenheck RA. HUD/HHS/VA Collaborative Initiative to Help End Chronic Homelessness National Performance Outcomes Assessment Preliminary Client Outcomes Report. West Haven, CT: North East Program Evaluation Center; 2007. [Google Scholar]
  26. Mares AS, Rosenheck RA. A comparison of treatment outcomes among chronically homelessness adults receiving comprehensive housing and health care services versus usual local care. Adm Policy Ment Health. 2011;38(6):459–475. doi: 10.1007/s10488-011-0333-4. [DOI] [PubMed] [Google Scholar]
  27. Mclellan AT, Luborsky L, Woody GE, Obrien CP. Improved Diagnostic Evaluation Instrument for Substance Abuse Patients - Addiction Severity Index. Journal of Nervous and Mental Disease. 1980;168(1):26–33. doi: 10.1097/00005053-198001000-00006. [DOI] [PubMed] [Google Scholar]
  28. Moore DT, Rosenheck RA. Factors Affecting Emergency Department Use by a Chronically Homeless Population. Psychiatr Serv. 2016:appi.ps.201500526. doi: 10.1176/appi.ps.201500526. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. NOFA. Notice of funding availability for the Collaborative Initiative to Help End Chronic Homelessness. Federal Register. 2003;68(17) [Google Scholar]
  30. Oates G, Tadros A, Davis SM. A comparison of National Emergency Department use by homeless versus non-homeless people in the United States. J Health Care Poor Underserved. 2009;20(3):840–845. doi: 10.1353/hpu.0.0192. [DOI] [PubMed] [Google Scholar]
  31. Padgett DK, Struening EL, Andrews H, Pittman J. Predictors of emergency room use by homeless adults in New York City: the influence of predisposing, enabling and need factors. Soc Sci Med. 1995;41(4):547–556. doi: 10.1016/0277-9536(94)00364-y. [DOI] [PubMed] [Google Scholar]
  32. Rickards LD, McGraw SA, Araki L, Casey RJ, High CW, Hombs ME, Raysor RS. Collaborative initiative to help end chronic homelessness: introduction. J Behav Health Serv Res. 2010;37(2):149–166. doi: 10.1007/s11414-009-9175-1. [DOI] [PubMed] [Google Scholar]
  33. Rodriguez RM, Fortman J, Chee C, Ng V, Poon D. Food, shelter and safety needs motivating homeless persons’ visits to an urban emergency department. Ann Emerg Med. 2009;53(5):598–602. doi: 10.1016/j.annemergmed.2008.07.046. [DOI] [PubMed] [Google Scholar]
  34. Sadowski LS, Kee RA, VanderWeele TJ, Buchanan D. Effect of a housing and case management program on emergency department visits and hospitalizations among chronically ill homeless adults: a randomized trial. JAMA. 2009;301(17):1771–1778. doi: 10.1001/jama.2009.561. [DOI] [PubMed] [Google Scholar]
  35. Salit SA, Kuhn EM, Hartz AJ, Vu JM, Mosso AL. Hospitalization costs associated with homelessness in New York City. N Engl J Med. 1998;338(24):1734–1740. doi: 10.1056/NEJM199806113382406. [DOI] [PubMed] [Google Scholar]
  36. Stergiopoulos V, Gozdzik A, Misir V, Skosireva A, Connelly J, Sarang A, … McKenzie K. Effectiveness of Housing First with Intensive Case Management in an Ethnically Diverse Sample of Homeless Adults with Mental Illness: A Randomized Controlled Trial. PLoS One. 2015;10(7):e0130281. doi: 10.1371/journal.pone.0130281. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Stergiopoulos V, Hwang SW, Gozdzik A, Nisenbaum R, Latimer E, Rabouin D … At Home/Chez Soi, I. Effect of scattered-site housing using rent supplements and intensive case management on housing stability among homeless adults with mental illness: a randomized trial. JAMA. 2015;313(9):905–915. doi: 10.1001/jama.2015.1163. [DOI] [PubMed] [Google Scholar]
  38. Stergiopoulos V, O’Campo P, Hwang SW, Gozdzik A, Jeyaratnam J, Misir V, … Kirst M. At Home/Chez Soi Project: Toronto Site Final Report. Mental Health Commission of Canada; 2014. [Google Scholar]
  39. Sterne JA, White IR, Carlin JB, Spratt M, Royston P, Kenward MG, … Carpenter JR. Multiple imputation for missing data in epidemiological and clinical research: potential and pitfalls. BMJ. 2009;338:b2393. doi: 10.1136/bmj.b2393. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Sun BC, Burstin HR, Brennan TA. Predictors and outcomes of frequent emergency department users. Acad Emerg Med. 2003;10(4):320–328. doi: 10.1111/j.1553-2712.2003.tb01344.x. [DOI] [PubMed] [Google Scholar]
  41. Thakarar K, Morgan JR, Gaeta JM, Hohl C, Drainoni ML. Predictors of Frequent Emergency Room Visits among a Homeless Population. PLoS One. 2015;10(4):e0124552. doi: 10.1371/journal.pone.0124552. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Tsai J, Doran KM, Rosenheck RA. When health insurance is not a factor: national comparison of homeless and nonhomeless US veterans who use Veterans Affairs Emergency Departments. Am J Public Health. 2013;103(Suppl 2):S225–231. doi: 10.2105/AJPH.2013.301307. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Tsai J, Rosenheck RA. Risk factors for ED use among homeless veterans. Am J Emerg Med. 2013;31(5):855–858. doi: 10.1016/j.ajem.2013.02.046. [DOI] [PubMed] [Google Scholar]
  44. van Buuren S. Multiple imputation of discrete and continuous data by fully conditional specification. Stat Methods Med Res. 2007;16(3):219–242. doi: 10.1177/0962280206074463. [DOI] [PubMed] [Google Scholar]
  45. Zanis DA, McLellan AT, Cnaan RA, Randall M. Reliability and validity of the Addiction Severity Index with a homeless sample. J Subst Abuse Treat. 1994;11(6):541–548. doi: 10.1016/0740-5472(94)90005-1. [DOI] [PubMed] [Google Scholar]

RESOURCES