Evaluation of Provision of Permanent Supportive Housing for Chronically Homeless Individuals Through Pay for Success
Evaluation of Provision of Permanent Supportive Housing for Chronically Homeless Individuals Through Pay for Success
Santa Clara County has partnered with Abode Services to provide 112 units of permanent supportive housing (PSH) for up to 6 years for chronically homeless individuals with multiple co-morbidities who are high cost users of services in the County. Because the target population of high cost users who are chronically homeless greatly exceeds the supply of permanent supportive housing to be made available via this mechanism, the County plans to use this opportunity to examine the effectiveness of PSH compared to treatment as usual by ethically allocating a limited resource using a lottery type system for eligible individuals who consent to participate in the Pay For Success (PFS) initiative. Specifically, while there are estimated to be over 2,500 people in Santa Clara who would be eligible for participation on any given night, the PFS program will only provide 112 PSH units at any given time. As the study duration will last six years, the investigators anticipate that approximately 150-200 people will be housed via this mechanism. Outcomes will be reported to the Institutional Review Board (IRB) on a quarterly basis and will be based on scheduled quarterly reports provided to the county.
This evaluation will be a randomized controlled trial comparing the outcomes for chronically homeless individuals randomized to receive PSH to those who are randomized to receive usual care. The aims of this study are to:
All data utilized for this evaluation will be County administrative data. The investigators hypothesize that some costly and acute health services utilization will decrease over time for the intervention group, while other more sustaining health services will increase.
Recruitment: Screening Tool
Eligible homeless individuals will be identified using a Screening Tool developed by the University of California, San Francisco (UCSF) evaluators that uses County administrative data to identify the most costly users of health and social services in the County based on their medical and behavioral health conditions, length of homelessness, and County services utilization history ("Eligible Homeless Individuals").
Abode Services (Abode) is the housing service provider selected by the County for this PFS project. Abode will obtain 112 units of permanent supportive housing over the course of the first year of this study and place referred clients into those units.
Eligibility Confirmation/Assessment/Consent
Individuals identified by the Screening Tool who have been located by the Study Team will have their eligibility confirmed through administrative data and be screened in-person to determine key inclusion and exclusion criteria. Following eligibility confirmation, a designated Study Team member who has been trained in the process of informed consent will ask the individual if he/she is interested in learning about the project.
The explanation of the study prior to administration of informed consent will be standardized so that all potentially eligible participants are approached in the same manner and receive the same information about the study prior to undergoing informed consent. This process will be conducted in the potential participants' language of choice. The investigators anticipate that the most common language choice for those with limited English proficiency is Spanish and Vietnamese. Study Team members will be fluent in those languages, as well as provide consent forms in those languages. For languages other than Spanish and Vietnamese, the investigators will use trained medical interpreters, who will interpret the consent process and attest to their interpretation on the consent form. The Study Team will use the "teach back" method to ensure comprehension of the study goals and risks as part of their informed consent procedure. Any individual who is unable to give informed consent, as expressed by inability to repeat back the risks and benefits of the study after three successive teach backs, will be deemed ineligible.
Upon granting informed consent, eligible participants will be randomized to either the intervention or control groups.
RCT "Intervention" and "Usual Care" Groups: Intervention Group
Individuals randomized to the intervention group will be referred to Abode Services for the opportunity to be placed in permanent supportive housing. Specifically, Abode will provide individuals with access to a housing unit, along with a broad array of services designed to achieve sustained residential stability and wellness. These services are designed to end the participants' homelessness, increase income (primarily through enrollment in Supplemental Security Income), and provide increased access to ongoing physical and behavioral health services. Abode will use a modified Assertive Community Treatment (ACT) team model of service delivery. Abode's modified ACT model will provide community-based clinical services, integrated with a flexible array of housing options delivered through a Housing First approach to provide housing and supportive services for the most vulnerable members of the chronically homeless population who most frequently use costly emergency services.
As part of the PSH service model and Abode's overall service delivery approach, Abode will offer a range of supportive services to participants, including, but not limited to:
RCT "Intervention" and "Usual Care" Groups: Usual Care Group
Chronically homeless individuals who are in the control group will have access to all existing safety-net services that are available to homeless individuals in the County. In addition, since chronically homeless individuals often have complex conditions that make them high users of safety-net services, chronically homeless individuals who are not in the treatment group may be eligible for the array of other programs that are designed for high users of specific systems. These programs include Full Service Partnerships for individuals with a serious mental illness including those who are involved in the criminal justice system, and the upcoming Acute Mental Health Pay of Success project. Finally, members of the control group will also have access to other all other permanent supportive housing programs for chronically homeless persons in the County.
Enrollment and Randomization Procedures
Enrollment will occur in "batches" in order to provide Abode with sufficient participants to fill the units at a pace that will maximize both occupancy of available units and efficient use of Abode resources. The number of eligible individuals who are ultimately randomized and the pace of randomization will be based on
Each time new housing units become available, the County will enable the screening tool to be applied to a batch of incoming individuals. Potentially eligible homeless individuals whom have not been enrolled in the study will be offered study enrollment.
Individuals randomized to the treatment group who are referred to and then located by the housing provider will be offered enrollment in the permanent supportive housing (PSH) intervention program. If an individual agrees to enroll in the PSH intervention, the individual will begin to receive case management services from the housing provider immediately, even if a housing unit is not yet available, and will be offered a housing unit once it becomes available. The intervention group will continue to receive case management services throughout the study, whether or not the individuals remain housed successfully.
Data Collected
Individuals will be identified using data from multiple data sets that track health and criminal justice system use within Santa Clara County. UCSF will receive identifiable data from the following data sources from the County based on the provisions of the data-sharing Memorandum Of Understanding (MOU) that exists between UCSF and Santa Clara County. If the data are not linked by the County, UCSF will link the data using the participants' unique county identifier. UCSF will discard these identifiers once data are linked for the purpose of the analysis, and will retain only the unique study identifier (ID). The unique study ID can only be linked back to participating individuals' names via a master document that links name to study ID, which will be kept under password protection by the evaluators and will only be available to participating study staff. This master document will be destroyed at the conclusion of the study period or after a pre-determined period of time put forth by the IRB after the study's conclusion. The main anticipated data sources are as follows:
Tenancy in permanent supportive housing will be reported by the housing service provider and verified by the investigator.
Analysis Plan
For the randomized controlled trial (RCT) framework used to evaluate health utilization outcomes, the study team will use an intention-to-treat (ITT) analysis, in which all participants randomly assigned to one of the groups are analyzed together, regardless of whether or not the participants completed or received that treatment. In this case, participants who are randomized to the permanent supportive housing intervention will be retained in the treatment group even if the participants are not able to be located, if the participants are located but never enter housing, or are located, enrolled, and enter housing, but are not retained in housing long-term. Participants randomized to the control group will be followed administratively using data provided by the County in order to track the outcomes outlined above in comparison to those in the intervention group.
The randomization will be evaluated by comparing the characteristics (e.g., demographics, chronic health conditions) of the intervention and control groups using chi-square and non-parametric tests. The associations of baseline characteristics, medical conditions, and social factors with stable tenancy will be evaluated using logistic or Cox regression analyses. The health services use outcomes (i.e., emergency department and hospital services, social services and criminal justice system utilization) and their associations with stable tenancy will be evaluated using logistic, ordinal, or linear regression, as appropriate. Comparisons of rates of service utilization will be performed at α = 0.95 with one-tailed tests of the outcomes in direction hypothesized (e.g., emergency department (ED) visits in the treatment group will be ≤ ED visits in the control group). However, the investigators will also test for trends over time, since it may be likely that health services utilization increases in the treatment group in the short term as services are more readily available and participants are more accessible, and then decreases in the longer term as the participant's health improves and stabilizes.
SPANS
In order to account for a larger amount of the data the investigators have collected for individuals enrolled in the evaluation, the investigators grouped data into one-year spans of time for each individual in the treatment and control group. For example, if an individual was enrolled for 4 years, the individual would have four separate one-year spans in the regression analysis. The use of spans allows the investigators to include the most available data for each individual in the study.
For participants who had potential spans that lasted ≥ 6 but <12 months, the investigators prorated utilization counts to a one year. To account for outliers in the data, the investigators top coded all span-level counts to the 99th percentile. The investigators included indicators in the regression analysis to signify the year in the program in order to account for patterns of use that may decrease or increase over time.
The investigators censored spans at the time of death. To account for the possibility that participants moved out of County, the investigators censored data 6 months after the last point of contact. For example, if the last time an individual used any services measured in our study was January 1, 2017, the investigators censored their data after June 30, 2017, and constructed spans with the data that remained.
The investigators used negative binomial regression analysis on count data outcomes using an intent-to- treat framework where treatment status was based on assignment to the treatment group. The investigators present results as incident rate ratios. The investigators clustered standard errors at the individual level to account for individuals with multiple spans.
In sensitivity analyses, the investigators recoded outcome variables to a binary indicator for whether an individual used any of a given service within the one-year span. The investigators explored sensitivity to alternative modeling choices such as ordinary least squares and present sensitivity analyses. The investigators also explored allowing the treatment effect to vary by how long the individual was enrolled in the program by including interaction terms for treatment status and year indicators.
EXPLORATORY LOCAL AVERAGE TREATMENT EFFECT (LATE) ESTIMATE:
The investigators also present negative binomial (count model) regression that include a treatment indicator and controls for time "spans," as noted above, as well as an exploratory analysis using linear regression models that incorporate instrumental variables to estimate the local average treatment effect (LATE) by using assignment to the treatment group as an instrument for housing. LATE represents the effect of treatment on those induced to take up housing because of assignment to treatment group.
Based on program data and housing data from Santa Clara County, 85% of those assigned to the treatment group were ever housed as compared to 37% in the control group. The LATE estimates represent the causal effect of treatment on the roughly 50% of the treatment group that was housed because of the intervention. This type of model effectively scales the impact of being in the treatment group by accounting for how assignment to the treatment group is correlated with take-up of housing. It is not attenuated by imperfect take-up of housing in the treatment group and by the fact that there is "contamination" of the control group whereby some individuals find housing despite being assigned to the usual care group.
The investigators are also working on two additional sets of analyses:
Impact of housing on those who received it: Because of our ITT evaluation framework, the impact of housing on individuals who receive it may be diluted. The investigators are working with the County on two additional data elements to examine this in more depth. First, the investigators are working to obtain complete data from the County regarding housing placements among individuals randomized to the control group and for individuals randomized to the intervention group who were never housed by Abode. These data are currently being reviewed for completeness and quality. Assuming the data are valid, the investigators will estimate the average treatment effect on the treated using instrumental variables regression to better account for the impact of the Abode PSH "treatment" on those who actually received it. The average treatment effect on the treated (ATET) estimate takes into account the fact that adherence with treatment assignment was not perfect, i.e. some assigned to housing were not housed and some assigned to the control group were housed. The investigators will also report descriptive statistics on housing entry and exit dates, and the number of moves for each group as a measure of housing stability that can be examined for both treatment and control groups.
Participant deaths: Individuals enrolled in Project Welcome Home had higher rates of death than the investigators anticipated. To explore this further, the investigators have obtained death certificates from the County for all individuals enrolled in the study who have died. The investigators have begun to review these data and to categorize reasons for deaths. Once the investigators receive and validate the housing data outlined above, the investigators will examine the temporal relationship between death and housing. The investigators also plan to examine utilization after enrollment and up to the time of death to uncover possible predictors for death and to better understand why the death rate in this population is higher than expected.
Inclusion Criteria:
Exclusion Criteria: