4

Cost-Effectiveness of Permanent Supportive Housing

Questions have been raised regarding the cost-effectiveness associated with permanent supportive housing (PSH). This chapter addresses this issue. In particular, this chapter responds to the question posed in the statement of task: Is permanent supportive housing cost-effective to those institutions and agencies providing funding for PSH programs? In addressing this question, the committee provides a brief overview of cost-effectiveness analysis (CEA) principles, outlines a framework for conducting CEA of PSH, reviews the available literature on the cost and effectiveness of PSH, and synthesizes the findings from the literature. It must be noted, however, that many researchers studying the potential effectiveness of PSH argue that “the cost-savings argument is problematic and that it would be better to reframe the discussion to focus primarily on the best way to meet this population’s needs” (Kertesz et al., 2016, p. 2115). The committee agrees with Kertesz and colleagues but still thought it important to investigate the cost-effectiveness question.

GENERAL COST-EFFECTIVENESS PRINCIPLES

CEA is an important tool used for evaluating various medical and health care practices and intervention polices involving resource allocation. A series of articles published in the Journal of the American Medical Association (JAMA) two decades ago provided widely accepted guidelines for conducting and reporting CEA based on the Report of the Panel on Cost-Effectiveness in Health and Medicine (RPCEHM) (Russell et al., 1996; Siegel et al., 1996; Weinstein et al., 1996). The RPCEHM considered cost-effectiveness from a societal benefit perspective and considered 13 factors, including community compassion and equity, with the overall goal to show the relative value of different health interventions.

More than 20 years later, a new set of standards for CEA was published (Sanders et al., 2016), with the overall goal of improving the quality of cost analyses. As stated by the authors, “the landscape and the set of challenges to cost-effectiveness analysis have changed since 1996” (p. 1095). The new standards include the addition of two “reference cases” that should be used in all CEAs, one

based on a health care sector perspective and the other based on a societal perspective. These reference analyses are a set of standard methodological practices that should be followed in the use of a CEA. The new recommendations also include the use of an “impact inventory,” which provides a structure for the outcomes of the two reference case analyses.

When considering the cost-effectiveness of PSH, it is reasonable to include housing, health care, and other relevant costs in the numerator of the cost-effectiveness (C/E) ratio and improvements in health, socioeconomic status, and other outcomes related to benefits in the denominator. While the former is basically monetary, the latter may include nonmonetary measures that may contain items in the quality-of-life index (e.g., the Lehman 7-scale composite index of various subjective and objective indicators) (Lehman, 1983) or the prototypical quality-adjusted life-years (QALYs) commonly used in the literature of medicine and health (Klarman et al., 1968; Sanders et al., 2016), or nonmonetary measures that can be converted to dollar values (following the concept of consumption equivalence in welfare analysis) as in the economics literature. It is informative to summarize several important principles highlighted by the new recommendations (Sanders et al., 2016):

• Reporting of CEAs must include sharing the results of both the health care reference case and the societal reference case.
• CEAs are not the only measure that should be considered when making decisions about the efficacy of a program or cost.
• QALYs are recommended for use in measuring health effects. All CEAs should include the impact inventory, which is designed to make certain that all consequences; particularly those outside the formal health care sector—are appropriately considered.¹
• Transparency and sensitivity are paramount.
• The health care reference case should include both out-of-pocket expenses paid by the patient as well as reimbursements by third-party payers.
• The social reference case should include medical costs as described in the bullet above, time costs for unpaid caregivers, transportation costs, effects on future productivity, and other costs outside the health care sector.
• Analysis of the social reference case should include efforts to quantify nonhealth consequences.
• Other principles to be considered include the cost-effectiveness framework measures described below because they have sometimes been overlooked by researchers evaluating PSH outcomes. Double counting of an aspect of a program or intervention in both the estimation of costs and effects should be avoided.

___________________

¹ The QALY concept has been criticized on technical and ethical grounds (Prieto and Sacristan, 2003). Nonetheless, it is still a primary measure of the value of a health intervention.

• Societal costs and benefits generated through externalities must be considered. Benefits generated at different times should be adjusted by a time discount.
• Costs incurred in different years should be adjusted for inflation, that is, measured in constant dollars.
• The concept of opportunity costs should be observed, which may be measured based on market prices properly adjusted by market distortions.

In developing a framework for assessing the CEA literature for PSH, the committee first considered relevant cost analysis and effectiveness measures associated with PSH. These measures are discussed below.

Identification of the Costs in PSH

Cost analysis in PSH can be categorized into two broad components: (1) incurred program cost and (2) all induced cost changes as a result of the program intervention. In the discussion below, various program costs were recategorized in a more systematic manner.

Program Costs

Program costs include, among others, rent subsidies, case management service costs, financing (or capital) cost for development of the housing project, and other operative or administrative costs. Capital costs are usually accrued over several years, and so these costs should be properly adjusted by year-specific interest rates and adjusted for inflation (i.e., imputed based on the real interest rates).

Induced Cost Changes

An intervention program may induce changes in many personal, governmental, and societal costs. The most important induced cost changes that the committee considered are changes in health care costs. Three primary cost changes that are typically observed are due to differences in the utilization of emergency services (including hospital emergency department and prehospital emergency medical services/ambulance costs), hospitalization and other inpatient costs, and ambulatory care and other outpatient service costs. Generally, actual costs accrued are not available. Thus, these costs are imputed from changes in ambulance calls, emergency department visits, hospitalization days, and outpatient visits, in conjunction with their respective unit costs. These unit costs vary across different hospitals and clinics, and hospitalization per-night costs sometimes are lower for longer stays.

There are two additional induced cost changes under the health care category. One concerns changes in residential treatment costs associated with substance abuse, which are particularly important to the subpopulation of individuals experiencing homelessness having high substance use rates. Another concerns

changes in prescribed drug costs, which are important to those individuals experiencing homelessness who have severe mental illness and other chronic health conditions. Whereas the former can be easily imputed based on number of treatments and the unit costs, the latter is more difficult to impute without detailed medical data.

The next set of induced cost changes, specific to PSH studies, concerns changes in shelter stay cost. They include cost changes resulting from using nursing homes (costs that are covered by Medicaid), shelters, and other housing. These costs may be imputed using data based on adding the number of changes in days spent at various places together with their respective unit costs.

Similar to many socioeconomic and health care programs, PSH interventions also lead to changes in legal and community support costs. With regard to legal-cost changes, analyses should include changes in law enforcement contacts and incarceration (including legal proceedings and court costs and jail and prison costs). Again, these costs may be imputed based on changes in police contacts, number of convictions and days in jail and prison, and the associated unit costs.

Identification of Effectiveness Measures

Effectiveness measures include those to be aggregated into the individual quality of life index or the QALYs measures and those at the societal level through externalities.²

Individual Quality-of-Life or QALY Measures

Individual benefits included in the overall quality-of-life or QALY measures are from increases in stable housing days and/or reductions in homeless days and objective or subjective measures based on improvements in health. Useful measures such as healthcare utilization and various physical and mental health outcomes, including reductions in sick days and substance dependence, may help measure quality of life. Subjective measures may also be used to produce quality-of-life scores. Notably, reductions in sick days and substance dependence may be used directly, but health care utilization should not be used. Health care utilization figures should be regarded as “inputs” (or explanatory variables) that can be used to estimate the “outputs” (or dependent variables, such as number of days housed that serve as a proxy for health quality.

Another benefit is accrued from gains in productivity, which may be measured by increases in work days or earned income. Improvements in social skills and community functioning (including network size and social contacts) should also be considered as a part of the social case reference.

___________________

² Externalities are defined as the indirect effects of economic production and consumption. There are both positive externalities and negative externalities. In this case, externalities refer to program costs that can have an indirect positive effect on society and the community.

To avoid double counting, monetary outcome change measures included in cost analysis should not be included in the effectiveness measure. By contrast, relevant time cost reduction or leisure benefit should be incorporated. As noted above, benefits generated at different times should be adjusted by a time discount, whereas QALYs will differ based on different subgroups of people or people at different stages in life.

Net Societal Benefits from Externalities

Analogous to many socioeconomic and health care programs, PSH interventions may generate societal benefits through positive externality spillovers such as reduced arrests and less time spent in jail or prison. Such positive externalities (indirect effects from a program or intervention) may arise in public health outcomes or community safety issues. They may also be due to improvements in neighborhood environments. The values of these externalities are usually obtained based on regression estimation. As described earlier, these positive externalities must be included in the social reference case. Additionally, this argument is made by Kertesz et al. (2016) regarding the framing of the cost-effectiveness argument. The committee believes that the net societal benefits, or positive externalities, of housing even some individuals experiencing homelessness are in fact benefits that should not be overlooked. Although these benefits may not be completely tangible, they are real and in fact likely contribute to overall community harmony.

A COST-EFFECTIVENESS FRAMEWORK FOR PSH

Combining the cost analysis and effectiveness measures discussed above, the committee developed a general framework for what should be included in a cost-effectiveness analysis of PSH for persons experiencing chronic homelessness who are exiting homelessness. The framework and measures are outlined below.

1. Cost analysis: Based on comparison of average cost per person per year
   1. Program costs
      1. Financing/capital cost for development of the housing project
      2. Rent subsidies
      3. Case management service costs
      4. Other operative and administrative costs
   2. Induced cost changes
      1. Changes in health care costs:
         1. Cost changes in emergency services, including emergency department and ambulance costs
         2. Cost changes in inpatient services
         3. Cost changes in outpatient services
         4. Changes in residential treatment costs associated with substance abuse
         5. Changes in prescription costs

2. 2. Changes in shelter stay cost, including nursing home, shelter, and other housing options (doubling up; transitional housing)
   3. Changes in law enforcement and incarceration costs
      1. Changes in police contact costs
      2. Changes in correction/incarceration costs, including legal proceedings, jail, or prison time
   4. Changes in community support cost for avoiding law enforcement and incarceration costs
3. Effectiveness measures
   1. Aggregate measure of individual benefits based on improvements in the overall quality-of-life index or QALYs inclusive of measurable benefits such as
      1. Increases in stable housing days
      2. Improvements in health (health care utilization and broader measures of health outcomes including reductions in sick days and substance dependence and subjective health quality measures may all be used to generate health benefit)
      3. Gains in productivity (increases in workdays and earned income)
      4. Improvements in social skills and community functioning
   2. Other societal benefits generated by positive externalities inclusive of those in:
      1. Public health
      2. Criminal justice system
      3. Neighborhood environments

To facilitate better understanding of the CEA results, the following supporting information is also included:

• Location(s) in which the intervention program is conducted;
• PSH program details, indicating whether there is on-site case management support and on-site clinics, whether housing is scattered site or single site, and whether there is an outreach team;
• Sample size (experimental versus control and attrition at various stages) and types of participants (general, people with severe mental illness, or substance abuse, or dually diagnosed, Medicaid users, VA clients, etc.); this information is particularly important because the costs, outcomes, and possible exit from homelessness all depend on the characteristics of the participants;
• Time of study length and months followed up after intervention;
• Number of days spent homeless;
• Other supporting data, including unit cost data for health care and detailed demographics.

The committee used this framework as it identified and assessed studies that might be informative to responding to the question at hand.

ASSESSMENT OF EXISTING STUDIES

The literature is sparse on the cost effectiveness of PSH (Ly and Latimer, 2015). In locating the appropriate studies to consider for this review, the committee conducted an extensive search of relevant databases, including Embase, the Cochrane Database of Systematic Reviews, Lexis, Lexis Law Reviews, Med-line/PubMed, Web of Science, WorldCat, and PsycINFO. The committee also identified 13 published or advance online publications that address, to some extent, CEA of PSH.

Among these studies, five used a randomized design (Gulcur et al., 2003; Rosenheck et al., 2003; Basu et al., 2012; Aubry et al., 2015; Stergiopoulos et al., 2015). However, three of these studies did not provide sufficient information to be used for the committee’s assessment: Aubry et al. (2015) has no monetary cost measures, Stergiopoulos et al. (2015) reported program cost but not induced cost changes, and Gulcur et al. (2003) provided health care cost reductions but no program cost. As a result, only Rosenheck et al. (2003) and Basu et al. (2012) are informative for the purposes of this CEA.

Although less robust than randomized studies, a total of eight quasi-experimental, pre/post studies were identified.³ Among these, only five included an intervention and comparison group (Culhane et al., 2002; Martinez and Burt, 2006; Gilmer et al., 2009; Larimer et al., 2009; Srebnik et al., 2013). These studies compare pre/post placement changes in the experimental group with the comparable changes in the control group, providing an array of “difference-in-difference” measures in cost and effectiveness outcomes. Such measures need to be adjusted if the characteristics between the experimental and the control group are different before the treatment. Despite potential statistical or econometric biases, these figures can be directly compared to those obtained under randomized design. The study by Martinez and Burt (2006) was excluded because the program cost was not precisely measured and the cost offset was limited to emergency and inpatient care only.

Three additional quasi-experimental, pre/post studies that did not include a comparison group were identified from the literature (Mares and Rosenheck, 2010; McLaughlin, 2011; Greenberg et al., 2013). Two of these studies (Mares and Rosenheck, 2010; Greenberg et al., 2013) were excluded because they did not include program cost measures. Only the McLaughlin (2011) study contained a comprehensive set of induced cost changes and could be used for the committee’s

___________________

³ While the committee recognizes that there have been many pre/post studies of PSH, these eight quasi-experimental, pre/post studies were identified. There are numerous other pre/post studies not included in this report, including the following: Henwood et al. (2015b), Massachusetts Housing and Shelter Alliance (2015), and New York State Department of Health (2017).

assessment. Nonetheless, the results of this study should not be compared to those studies with comparison groups (either randomized or quasi-experimental) for which a “difference-in-difference” estimate can be calculated, because of the lack of a comparison group.

In summary, seven studies contributed to the committee’s analysis of the CEA of PSH: two randomized design studies (Rosenheck et al., 2003; Basu et al., 2012); four quasi-experimental, pre/post analysis studies using a comparison group (Culhane et al., 2002; Gilmer et al., 2009; Larimer et al., 2009; Srebnik et al., 2013); and one quasi-experimental, pre/post analysis study without a comparison group (McLaughlin, 2011). Table 4-1 summarizes the basic program information for each study.

Findings

To facilitate comparisons among the studies, the committee took a number of steps to harmonize the data (see Annex 4-1 for additional information about this process). The measures from the different studies were converted to and reported on a per-person per-year cost basis. All cost measures are reported in U.S. dollars. The net cost measures after accounting for cost offsets are presented in 2015 constant dollars. For studies with comparison groups, the committee calculated the differences between the intervention and the control group, showing the net effects of intervention polices (difference-in-difference), and reports these results. Moreover, to align with the framework for conducting CEA for PSH, the committee recategorized various effectiveness and cost measures from those studies to be consistent with the framework outlined above.⁴ The details in transforming various data and numerical results from those reported in the cited studies are provided in Annex 4-1.

Tables 4-2 and 4-3a,b, c show effectiveness and cost measures, respectively, based on the general cost-effectiveness framework outlined above. In most of the studies reviewed, cost measures were incomplete and effectiveness measures scarce. As a result, Table 4-2 includes only two studies and Table 4-3a, b, c contain many empty cells, reflecting the lack of available data to complete these tables.

Effectiveness

As summarized in Table 4-2, only two of the seven studies (Rosenheck et al., 2003; Basu et al., 2012) provide measures of effectiveness that may be incorporated

___________________

⁴ Examples of recategorization include reporting legal costs and incarcerated days separately, and categorizing stable housing days separately from homeless days.

TABLE 4-1 Summary of Select Studies Examining Cost-Effectiveness of PSH

TABLE 4-2 Measures of Effectiveness for Incorporating into Individual Quality of Life Index (QLI)

^a-f See Annex 4-1 for further explanation of computational details used in transforming data from cited studies.

into the individual quality-of-life index. Five studies were excluded because of double-counting (related to changes in emergency department visits, outpatient visits, hospitalized days, and incarcerated days) in the effectiveness measure.

One important quality-of-life measure is the reduction in days spent homeless. Basu et al. (2012) found that homeless chronically ill individuals who were randomly assigned to participate in the Housing First program in Chicago experienced an average reduction of 62.3 days per year spent homeless compared to individuals in the usual-care control group (the usual-care group received only discharge planning services with no follow-up (p < .05). The study conducted by Rosenheck et al. (2003) also found a reduction in average homeless days, albeit a more modest one. Veterans with major psychiatric disorders and/or substance use disorders who were randomized to HUD-VA Supportive Housing (HUD-VASH) compared to standard VA care in San Francisco, San Diego, New Orleans, and Cleveland had an average reduction of 29.6 homeless days per year (p < .001).

The study by Rosenheck et al. (2003) also reported additional effectiveness measures in health scores and quality-of-life scores. No significant differences between the two groups were seen in the health scores along the dimensions of mental health, physical health, and substance abuse. However, the social network size of the treatment group (the HUD-VASH–housed individuals) was significantly greater than the network sizes of the usual-care group and the case management group.

The authors also reported a statistically insignificant reduction in earned income of $126 per person per year in the cost measures, which should be regarded as a reduction in the effectiveness. It may be noted that while health quality is expected to improve under the supportive housing intervention, this statistically insignificant finding may be due to potential problems in measurement of health outcomes, as noted in earlier chapters.

TABLE 4-3a Summary of Program Costs Reported in Key Studies of Cost-Effectiveness of PSH

^a-g See Annex 4-1 for further explanation of computational details used in transforming data from cited studies.

NOTE: Program costs include difference between experimental and control (per person, per year). Figures are before inflation adjustments.

TABLE 4-3b Summary of Average and Induced Cost Changes in Key Studies of Cost-Effectiveness of PSH

^a-kk Costs were calculated as the difference between experimental and control (per person per year). See Annex 4-1 for further explanation of computational details used in transforming data from cited studies.

NOTE: ED = emergency department; ES = emergency services.

Unfortunately, none of the seven identified studies took into account benefits over the life course or aggregated benefits to individuals based on their individual characteristics (such as age, gender, race, education, and employment status) to produce an appropriate average measure. Furthermore, none of the studies considered incorporated net societal benefits resulting from externalities in public health, legal, and neighborhood environmental perspectives. It is most likely that the reported measures of effectiveness are biased downward due to these omissions.

Induced Cost Changes

All studies except Larimer et al. (2009) reported average cost changes (Larimer et al., 2009, reported median). Most studies focused on induced changes in hospital costs and legal costs. As shown in Table 4-3b, Basu et al. (2012) considered changes in alternative nursing home and/or shelter stay costs (compared to costs of days spent in permanent supportive housing) and found a nearly $900 cost saving per person per year in the intervention group compared to the control group. This cost offset was somewhat higher in McLaughlin (2011), but as noted earlier, this study did not have a comparison group. Rosenheck et al. (2003) and Basu et al. (2012) also reported a cost saving of comparable magnitude ($897 and $671, respectively) for residential treatment of substance abuse, while McLaughlin (2011) found small increases in prescription drug costs and community support costs (although its magnitude cannot be compared to the other two studies because of the lack of a comparison group).

Almost all studies showed a reduction in legal costs. For example, a reduction of more than $700 was seen in New York (Culhane et al., 2002), Chicago (Basu et al., 2012), and Seattle (Larimer et al., 2009; Srebnik et al., 2013). The cost offset is smaller in San Diego (less than $300 as shown by Gilmer et al., 2009). However, a small legal cost increase of $101 was reported by Rosenheck et al. (2003) in the HUD-VASH study.

The most important component of induced cost changes—hospital costs in the form of outpatient, inpatient, and emergency department services—was also examined. In all but one study (Rosenheck et al., 2003), there were sizable cost offsets, ranging from $2,708 (Gilmer et al., 2009) to $42,828 (Larimer et al., 2009). The larger offset figures may be due to selection bias associated with quasi-experimental designs or with the high health care utilization of some study subjects (such as the Chicago Housing First study). The cost offsets are particularly large in the two Seattle programs (Larimer et al., 2009; Srebnik et al., 2013) with on-site medical care. However, these findings may not be generalizable to typical supportive housing programs. In the studies with hospital cost offsets, there is an across-the-board cost reduction in emergency department visits and inpatient services, ranging from $3,051 (Gilmer et al., 2009) to $34,603 (Srebnik et al., 2013). Outpatient costs rose in all but one study (Srebnik et al., 2013), where there was a clinic on-site. The increase in outpatient costs is to be expected, however, given the reduction in emergency room and inpatient costs.

In the case of the HUD-VASH study (Rosenheck et al., 2003), the inpatient costs rose by almost $300 (with non-VA hospital service costs adjusted based on VA cost shares). As a result, the overall inpatient hospital services costs in this study rose by about $3,000. In addition, emergency services costs were not included in this study, which is likely to underestimate the possible cost offset. To test whether the VA subpopulation was different from the full sample, the hospital cost offset of the VA subsample was compared with the average hospital cost offset of the full sample in the NY/NY study by Culhane et al. (2002). This comparison shows that the hospital cost offset for the VA subsample is dramatically lower than the full sample in Culhane et al. (2002) (an offset of $595 compared to $15,534).

In summary, some studies suggest that PSH interventions may induce cost offsets or yield modest net cost increases for those with persistent patterns of homelessness and serious mental illness. The offsets result primarily from reductions in emergency department and inpatient services costs and from reductions in alternative nursing home and/or shelter stay costs.

Net Costs

The program cost minus the induced cost offsets yields a measure of the net cost. Table 4-3c shows the raw net cost figures along with their inflation-adjusted counterparts in 2015 constant dollars. The Basu et al. (2012) study was the only one to specify unit cost measures in 2010 dollars, and Culhane et al. (2002) documented their measures in 1999 dollars. For those studies specifying the last year of the follow-up of the study, the end year of study was used as the base of the dollar measures. In the study conducted by McLaughlin (2011), such information is not provided, so a Consumer Price Index (CPI) of 2006 (5 years prior to the publication date) was used as the base of the dollar measures. The CPI ratios of the respective base years to 2015 are reported in Table 4-3c.

In the six studies with a comparison group, three show large net cost savings, ranging from $6,875 (Basu et al., 2012) to $33,502 (Larimer et al., 2009) per person per year in 2015 constant dollars. The remaining three studies suggest net cost increases of $250 (Gilmer et al., 2009), $1,414 (Culhane et al., 2002), and $3,093 (Rosenheck et al., 2003). While the small net cost figure in the first case could be viewed as negligible, the net cost in the other two studies may require more context. The increased net cost obtained in the NY/NY Housing program studied by Culhane et al. (2002) could be viewed as comparatively modest, especially because the program is in a location with a high cost of living. The high net cost obtained in the HUD-VASH study by Rosenheck et al. (2003) is likely specific to the veteran subpopulation with major psychiatric disorders or substance abuse disorders, as discussed above.

TABLE 4-3c Raw Net Costs for Select Studies of Cost-Effectiveness of PSH

NOTE: Program cost minus induced cost offsets yield a measure of net cost.

CONCLUSIONS

The committee examined studies that purported to assess the cost-effectiveness of PSH and found that, at present, there is insufficient evidence to demonstrate that the PSH model saves health care costs or is cost-effective. Unfortunately, the literature on cost effectiveness of PSH is sparse; few randomized controlled studies have been conducted. Most studies in this regard use a quasi-experimental design. Further, the available studies have not been conducted in a manner that is methodologically aligned with generally accepted health care cost-effectiveness research design. In principle, the most robust scientific evidence to answer the question would come from studies using a randomized design and that cover a comprehensive array of cost and effectiveness measures. Ideally such studies would allow for constructing the cost-effectiveness ratio to compute the net cost required per unit of QALYs or, at a minimum, provide information on the net cost required for increasing one stably housed day. Unfortunately, there were very few randomized studies, and among these, cost measures were incomplete and effectiveness measures scarce.

The committee found mixed results on the cost-effectiveness of PSH from the two randomized studies it identified. There was a sizable cost saving of $6,875 in the Chicago study (Basu et al., 2012) of homeless adults with chronic medical conditions, but a net cost increase of about $3,000 in the HUD-VASH program (Rosenheck et al., 2003). In the latter study, however, the net cost led to an increase of more than 1.5 months of stable housing. These studies also did not incorporate net societal benefits resulting from externalities in public health, legal, and neighborhood environmental perspectives, so it is likely that these studies

underestimated the cost-effectiveness. Future CEAs should make a stronger effort to include societal benefits, as outlined by the 2016 recommended standards (Sanders et al., 2016).

The committee also considered studies using quasi-experimental, pre/post study designs. Generally, these studies suggest that most PSH programs would induce cost offsets that either exceeded the program cost or yielded relatively modest net costs on balance, although this cannot be assumed. Primary cost offsets arise from reductions in emergency department and inpatient services and reductions in alternative nursing home and/or shelter stays. As described in this chapter, McLaughlin’s (2011) results indicated a 57 percent cost savings due to a decline in mental health costs from before housing to after housing. In other words, as the author stated, “aggregate data demonstrate that permanent supportive housing both saves money and provides homeless people with mental illness significantly more efficient and appropriate housing and service delivery” (McLaughlin, 2011, p. 409).⁵ Although the evidence is not strong, in the aggregate, it suggests that PSH is cost-effective for those with persistent patterns of homelessness and serious mental illness.

The committee notes that a common question embedded in the evaluation of PSH programs and other health interventions is whether these programs result in a monetary return on investment such as cost savings (Keyes and Galea, 2016). However, PSH was designed with the primary goal of preventing and ending chronic homelessness and not for the purpose of accruing cost savings (USICH, 2015). The committee believes that evaluations of these programs should a priori be expected to show broad benefits of health and well-being, including keeping individuals experiencing homelessness stably housed. The committee does not believe policy makers and others should expect that PSH programs would yield net cost savings, although some cost savings could be identified in specific studies such as those that exclusively focus on persons who are persistently high utilizers of emergency medical services systems.

Overall, the committee found few studies evaluating the cost-effectiveness of PSH programs, and the studies that have been done provide incomplete data that do not fully capture the health benefits of PSH. To address these problems, the committee recommends:

___________________

⁵ The recent RAND evaluation of the Housing for Health (HFH) program providing PSH in Los Angeles County (Hunter et al., 2017) considered whether HFH participants used fewer public services after receiving housing than before and whether these changes in service utilization resulted in cost savings for the county. The RAND study was not included in the present study because it was published after the committee completed its analysis of the research.

ANNEX 4-1: COMPUTATIONAL DETAILS

In this Annex, the committee provides further explanation of computational details used in transforming data from cited studies in Tables 4-2, 4-3a, and 4-3b.

Table 4-2: Effectiveness Measures (notes a-f)

Table 4-3a: Program Costs

a. For Basu et al. (2012), all figures are taken directly from the paper except the change in incarcerated days, −3.67 = 4.06 − 7.73.

b. For Rosenheck et al. (2003), figures were recomputed based on difference between experimental and control, recategorized by cost weight and adjusted to 1 year.

c. 318 = (380 − 389 + 967 − 4)/3 Annual administrative costs of transfer payments and of Section 8 vouchers for the HUD-VASH group (380 + 967) minus the costs for the control standard-care group (389 + 4). Data reported in Rosenheck et al. (2003) are for the 3-year follow-up study; thus here it is adjusted for a 1-year period (i.e., divided by 3).

d. For Gilmer et al. (2009), figures were recomputed based on difference-indifference (experimental minus control, post-intervention minus pre-intervention) and adjusted to 1 year.

e. 3,201 = [(7,423 − 412) − (1,038 − 429)]/2 Case management cost difference between post- and pre-intervention, and between experimental and control groups, were divided by 2 because the data reported in Gilmer et al. (2009) correspond to a 2-year period.

f. For McLaughlin (2011), values were recomputed in post-intervention minus pre-intervention, recategorized, and adjusted to per person.

g. 2,945 = (1,349,355 − 560,045)/268 Cost comparison is per person for shelter night use ($1,349,355) and housing program ($560,045).

Table 4-3b: Induced Cost Changes (notes a-kk)

a. For Basu et al. (2012), all figures are taken directly from the paper except the change in incarcerated days, −3.67 = 4.06 − 7.73.

b. For Rosenheck et al. (2003), values were recomputed based on difference between experimental and control, recategorized by cost weight, and adjusted to 1 year.

c. 286 = {[(12,023 + 4,043) * (46,249/36,524)] − [(9,318 + 4,824)*(39,287/28,515)]}/3 Annual inpatient and residential care costs (mental health care [$12,023] plus medical-surgical care [$4,043]) are multiplied by the weight of VA health costs for the HUD-VASH group minus the same cost calculations associated with the control standard-care group. Data reported in Rosenheck et al. (2003) are for the 3-year follow-

up study; thus here data are adjusted for a 1-year period (i.e., divided by 3).

d. 2,748 = [(17,267 * (46,249/36,524)] − [9,886 * (39,287/28,515)]}/3 Annual outpatient care costs including mental health care, medical-surgical care, and homeless case management ($17,267) are multiplied by the weight of VA health costs for the HUD-VASH group minus the same cost calculations associated with the control standard-care group. Data reported in Rosenheck et al. (2003) are for the 3-year follow-up study; thus here data are adjusted for a 1-year period (i.e., divided by 3).

e. −671 = [(3,291 * (46,249/36,524)] − [4,486 * (39,287/28,515)]}/3 Annual inpatient residential care treatment cost ($3,291) is multiplied by the weight of VA health costs for the HUD-VASH group minus the same cost calculations associated with the control standard-care group. Data reported in Rosenheck et al. (2003) are for the 3-year follow-up study; thus here they are adjusted for a 1-year period (i.e., divided by 3).

f. −800 = (2,375 − 4,774)/3 Annual shelter stay cost for the HUD-VASH group ($2,375) is minus the cost for the control standard-care group ($4,774). Data reported in Rosenheck et al. (2003) are for the 3-year follow-up study; thus here they are adjusted for a 1-year period (i.e., divided by 3).

g. −17.56 = (17.25 − 21.64) * 4 Number of days spent in nursing home per year for the HUD-VASH group is minus days spent for the control standard-care group. Data provided in Rosenheck et al. (2003) correspond to a 90-day period, and so here they are multiplied by 4 to get the annual calculation.

h. 101 = (1,062 − 758)/3 Annual legal/incarceration costs for the HUD-VASH group is minus the same cost associated with the control standard-care group. Data reported in Rosenheck et al. (2003) are for the 3-year follow-up study; thus here they are adjusted for a 1-year period (i.e., divided by 3).

i. For Culhane et al. (2002), values were recomputed based on difference-in-difference (experimental minus control, post minus pre), population weighted, recategorized by cost weight, and adjusted to per person per year.

j. −15.84 = {[(59 − 137.3) − (131.4 − 138.5)] * 570 + [(7.5 −34.2) − (15.6 − 33.4)] * 791 + [(26 − 52.1) − (41.8 − 51.5)] * 294}/[(570 + 791 + 294)/2] Annual hospitalized days are calculated as follows: Total mean days in OMH hospital (relative difference between post- and pre-intervention, and

between experimental and control groups, over the total service users) plus same calculation corresponding to HHC hospital (non-Medicaid) plus the one corresponding to the VA inpatient days, divided by the grand total of users (570 + 791 + 294), and then divided by 2 because the data reported in Culhane et al. (2002) correspond to a 2-year period.

k. 38.85 = [(177.1 − 93.3) − (87.8 − 81.7)]/2 Mean outpatient visit difference between post- and pre-intervention and between experimental and control groups are divided by 2 because the data reported in Culhane et al. (2002) correspond to a 2-year period.

l. −77.35 = [(27.5 − 190.6) − (122.5 − 130.9)]/2 Mean shelter days difference used by persons between post- and pre-intervention, and between experimental and control groups, is divided by 2 because the data reported in Culhane et al. (2002) correspond to a 2-year period.

m. −747 = − (624 + 490)/2 * (16,281/12,145) Estimated cost reductions from reduction in the Department of Corrections at the state ($624) and city ($490) levels are adjusted by the cost weight (ratio between total cost reductions per housing unit [$16,281] and the total cost reductions per placement [$12,145]), and divided by 2 to get the annual cost.

n. −6.00 = {[(3 − 11.2) − (11.5 − 11)] + [(6.6 − 11) − (11.5 − 12.6)]}/2 Annual incarcerated days are calculated as the difference between days per total persons incarcerated between post- and pre-intervention, and between experimental and control groups, and divided by 2 because the data reported in Culhane et al.(2002) correspond to a 2-year period.

o. −3,787 = −5,650/2 * (16,281/12,145) Cost reduction for Medicaid inpatients is adjusted by the cost weight (ratio between total cost reductions per housing unit [$16,281] and the total cost reductions per placement [$12,145]) and divided by 2 to get the annual cost.

p. −8.3 = [(29.6 − 46.3) − (41.9 − 42)]/2 Mean hospitalized days reimbursed by Medicaid is calculated as the difference per person between post- and pre-intervention and between experimental and control groups and is divided by 2 because the data reported in Culhane et al. (2002) correspond to a 2-year period.

q. 2,658 = 3,965/2 * (16,281/12,145) Cost reduction for Medicaid outpatient visits is adjusted by the cost weight (ratio between total cost reductions per housing unit [$16,281] and the total cost reductions per placement [$12,145]) and is divided by 2 to get the annual cost.

r. 38.85 = [(177.1 − 93.3) − (87.8 − 81.7)]/2 Mean Medicaid outpatient visits are calculated as the difference between post- and pre-intervention and between experimental and control groups are divided by 2 because the data reported in Culhane et al. (2002) correspond to a 2-year period.

s. −8.2 = [(26 − 52.1) − (41.8 − 51.5)]/2 Mean hospitalized days for VA inpatients are calculated as the difference per person between post- and pre-intervention and between experimental and control groups and are divided by 2 because the data reported in Culhane et al. (2002) correspond to a 2-year period.

t. For Gilmer et al. (2009), values were recomputed based on difference-indifference (experimental minus control, post-intervention minus pre-intervention) and adjusted to 1 year. Costs for emergency department, inpatient, and outpatient services totaled $2,708.

u. −3,051 = [(7,249 − 12,291) − (13,756 − 12,696)]/2 Inpatient and emergency costs were calculated as the difference between post- and pre-intervention and between experimental and control groups are divided by 2 because the data reported in Gilmer et al. (2009) correspond to a 2-year period.

v. 344 = [(5,430 − 1,917) − (4,488 − 1662)]/2 Outpatient cost was calculated as the difference between post- and pre-intervention and between experimental and control groups and divided by 2 because the data reported in Gilmer et al. (2009) correspond to a 2-year period.

w. −287 = [(137 − 713) − (428 − 431))/2 Legal/criminal justice system cost was calculated as the difference between post- and pre-intervention and between experimental and control groups and divided by 2 because the data reported in Gilmer et al. (2009) correspond to a 2-year period.

x. For Larimer et al. (2009), values are adjusted from month to year (multiplied by 12).

y. For Srebnik et al. (2013), values were recomputed based on difference-indifference (experimental minus control, post intervention minus pre-intervention), recategorized and adjusted to per person. Costs for emergency department and inpatient services totaled $34,603.

z. −4.39 = [(60 − 234)/29] − [(139 − 189)/31] Difference is between number of visits to the ER before and after the intervention, between the experimental and control groups, per person.

aa.−7.06 = [(123 − 441)/29] − [(110 − 231)/31] Difference is between number of days hospitalized before and after the intervention, between the experimental and control groups, per person.

bb. –672 = −(23,856/29) + 4,656/31 Total cost reduction per person is calculated from outpatient costs for experimental and control groups.

cc. −14.01 = [(36 − 533)/29] − [(272 − 369)/31] Decrease in outpatient visits for substance treatment is calculated as the difference before and after the intervention, between the experimental and control groups, per person.

dd. –1,304 = −10,228/29 − 29,495/31 Decrease in legal costs for participants is calculated as the difference before and after the intervention, between the experimental and control groups, per person.

ee. −10.37 = [(126 − 206)/29] − [(444 − 208)/31] Decrease in days incarcerated is calculated for participants as the difference before and after the intervention, between the experimental and control groups, per person.

ff. For McLaughlin (2011), values were recomputed in post-intervention minus pre-intervention, recategorized, and adjusted to per person.

gg. –478=((198466-299092)+(34582-62071))/268 Emergency department and ambulance use cost difference before and after intervention, per person.

hh. 341 = (397,769 − 306,448)/268 Cost comparison per person for prescription drug use is calculated for before and after intervention, per person.

ii. −1,382 = (2,764 − 373,243)/268 Cost comparison per person for shelter night stay cost is calculated for before and after intervention, per person.

jj. −229 = [(17,189 − 72,085) + (10,043 − 16,511))/268 Legal/jail night stay cost was calculated as difference before and after intervention, per person.

kk. 135 = (581,694 − 545,633)/268 Cost comparison per person for community support was calculated as difference before and after intervention, per person.