This study will use a longitudinal, prospective matched cohort design. Research advocates for the use of longitudinal studies to better assess the relationship between mental health and subsidized housing [
31,
32]. This approach is also useful for understanding physical health and healthcare use, as prospective cohort designs are particularly strong when used to relate an outcome (e.g. mental health, physical health and healthcare use) to an event (e.g. receipt of subsidized housing) [
33]. In this case, the study design will allow the research team to associate changes in health to receipt of subsidized housing. Further, any potential cohort effects can be adjusted for by accounting for individual sociodemographic variations within the cohort of housing applicants [
33,
34].
Primary data collection
The sampling frame for this study is all public housing applicants in New Brunswick, which includes approximately 4750 households at the study start date. Each household will receive a letter mailed from the Department of Social Development (DSD), which will provide information about the study, a link to an online survey, an email, and a phone number for the study team. Online participation will be encouraged; however, participants may choose to complete the survey over the phone with a Research Assistant or via mail. New Brunswick is a bilingual province so all study materials will be available in French and English.
Email addresses, mailing addresses, and phone numbers will be recorded during each survey to prevent study attrition. Upon completion of each survey, participants will be mailed or emailed a $10 gift card to Tim Horton’s coffee shop. Their names will also be entered into a draw for one of three $500 VISA gift cards. The draw for the gift cards will take place immediately after data collection concludes.
Study participants will enter the study as control group members while they wait for access to subsidized housing. During this time, participants will be asked to complete a baseline survey which asks questions on demographics, self-reported mental and physical health, and a variety of potentially confounding measures, which are described in detail below. After the baseline survey is complete, control group participants will be provided with shorter follow-up surveys at 6, 12, and 18 months following their initial baseline survey that assess changes to the main outcomes (physical and mental health) and variable factors (e.g., experiences of stigma, residential satisfaction, etc.).
The research team will ask participants for their consent to share their names with the provincial DSD. Those who consent will have their name sent to DSD via WatchDox (
www.watchdox.com), which is used by the Provincial government to transfer confidential information. Program staff with DSD will check the names provided against offers for subsidized housing each month and will provide the research team with updated information and move dates for those who become housed during the study period. Not all participants will consent to sharing their names; therefore, each survey administered to the control group after baseline will ask participants if they have received subsidized housing. Participants who indicate that they have received subsidized housing will be asked when they moved or started to receive a subsidy and will be moved to the intervention group.
The intervention group will receive additional follow-up surveys at six, 12, and 18 months after they begin receiving subsidized housing. Participants who are not subsidized within 24 months of their baseline participation date will not crossover into the intervention group and their study participation will be complete. At the start of the study, many of the households will have already been on the waitlist for months. Therefore, households at the top of the waitlist or those who experience conditions that assign them priority status (e.g. homelessness or intimate partner violence) will move into housing faster than others. Recruiting from the entire waitlist will ensure that households from the top, middle, and bottom of the waitlist are contacted for study participation.
It is possible that control group participants may remove their names from the waitlist during the study period. If this happens, the previous data collected from these participants will be kept and their study participation will be complete. It is also possible that participants in the intervention group may receive and then lose or leave subsidized housing. If this happens, the research team will note this, and their study participation will be complete. Their data prior to exiting subsidized housing will be included in analyses. Should a large enough portion of participants leave the wait list or subsidized housing, their data will be compared with others who either stayed on the wait list or continued to receive subsidized housing to see if any significant differences exist between the groups.
In the absence of any data reporting CESD-10 findings and data from the DAD in intervention studies similar to ours, we will estimate the power to compare pre- vs post-intervention CESD-10 total scores and healthcare use at the end of the study, using Cohen’s d effect sizes for paired samples [
35]. Assuming that there will be 30% attrition by the end of the study, a sample size of 1,138 data pairs achieves 100% power to detect effect sizes ranging from 0.3 (moderate effect size) to 0.8 (large) with a significance level equal to 0.05 using a two-sided paired t-test. As analyses will compare intervention and control periods, the researchers expect that the high power calculated using the paired t-test at the end of the study will approximately hold when we fit mixed models to the data.
Administrative data linking
This study also uses administrative dataset linking to measure differences in physical and mental healthcare use between the intervention and control groups. With each participant’s consent at baseline, their name and date of birth will be used to link their survey results with their matched records in the New Brunswick Institute for Research Data and Training (NB-IRDT) database. The NB-IRDT is an organization that houses and links data with large, provincial administrative databases. It provides individual level data on education, health, social services use, and employment. The primary data collected through this study will be linked with participants’ healthcare use data from the Discharge Abstract Database (DAD), which provides information on patient billing for hospitalizations, walk-in clinic use, and primary care appointments. The research team will use the date that housing subsidies were received to create a time variable that indicates their receipt of the intervention. The DAD and the time variable will then be used to compare individuals’ hospitalizations, walk-in clinic use, and primary care appointments in the 18 months prior to and following their moves into housing. The same analyses will be performed for individuals in the control group to assess differences between the two groups.
Scales and measures
The measures proposed for this survey are discussed below. Additional questions may be added into follow-up surveys if deemed necessary by the research team.
Primary outcome measures
The primary outcomes for this study are mental health, physical health and healthcare use. In this study mental health is conceptualized as the presence or absence of depressive, anxious, and distress symptoms. Depressive symptomology will be measured using the Centre for Epidemiological Studies Depression Scale Short Form (CESD-10) [
36‐
38]. The CESD-10 is an abbreviated, validated version of the CESD-R. A scoring algorithm is applied to each of the 10 questions and the values from all the questions are summed to provide a score ranging from 0–30, with 10 points on the scale being the clinical cutoff that is used to indicate the presence of depression. However, the scores are also suitable for use as a continuous variable [
39,
40]. The Kessler 6 (K-6) will be used to measure distress and anxious symptomatology. The K-6 was designed for the U.S. National Health Interview Survey and measures the presence of distress and anxious symptoms using a simple six item scale [
41]. The K-6 is an abbreviated version of the K-10. It is quickly administered and is deemed highly reliable and valid [
42‐
44].
Participants will be asked if they have ever received a mental health diagnosis and will be provided with a list of common psychiatric conditions from which to choose. An option to specify a condition that is not listed will be provided.
To assess physical health, the EQ-5D-5L and EQ-VAS will be administered. The EQ-5D-5L is validated measure comprised of five dimensions of health that relate to quality of life. The EQ-VAS is a visual analog scale to measure reported overall health [
45,
46]. Participants will also be asked to self-report any intellectual, developmental, or physical disabilities.
The DAD, which captures physician billing data on hospitalizations, walk-in clinic use, and primary care appointments, will be used to measure healthcare use. The NB-IRDT has yet to receive data on Emergency Department use, so this measure will not be included in the present study; however, once these data are available, a secondary analysis of Emergency Department use may be conducted.
Demographic and potential confounding variables
Standard demographic information will be collected from each participant (e.g. gender/sexual identity, income, sources of income, work status, marital status, ethnicity, citizenship status, rural or urban residency, and household composition). The NB-IRDT will provide linked data from the Citizen Registry and Vital Stats, which will allow the researchers to account for chronic and comorbid conditions, and movement out of province or death.
New Brunswick’s DSD has indicated that their subsidized housing tenants often feel stigmatized, and this negatively impacts their experiences of mental health and wellbeing. Although there is no current data to confirm this, recent studies from other jurisdictions suggest that public housing tenants experience perceived or actual stigma which negatively impacts wellbeing [
47‐
49]. To measure stigma, the Self-Stigma Short (SSS) will be administered. This is a 9-item validated scale, typically used to measure stigma of mental illness; however, it allows researchers to replace the condition of interest to meet their own research needs [
50]. For the purpose of this study, mental illness will be replaced with public housing applicant (control) and public housing resident (intervention). This will allow the research team to assess whether stigma contributes to mental health in the intervention and control groups.
Data on substance consumption will be collected using six adapted measures selected from the Canadian Tobacco and Drugs Survey [
51]. These questions will measure the frequency of alcohol, tobacco, and cannabis consumption over the six-month period preceding each survey. The research team only tracked use of legal substances, as illicit drug use is often associated with secrecy and stigma and the use of illicit substances was not critical to the study [
52]. This will allow the research team to control for the impacts of any potential changes in substance use on mental and physical wellbeing.
Social support will be measured using the Oslo Social Support Scale (OSS-3). This scale was selected as it is widely used with a variety of populations; further, it is a brief measure of social support which is important to reduce participant fatigue [
53]. The scale consists of three questions which are designed to measure the level of social support that people perceive they have. We will include this measure as social support is highly correlated with physical and mental health [
54‐
57].
Housing and neighbourhood measures
Previous studies indicate that housing and neighbourhood satisfaction and quality contribute to mental health [
58‐
64]. The survey will use an abbreviated version of the Residential Environmental Satisfaction Scale (RESS), which is highly correlated with the total RESS scale (0.96) [
65]. This scale measures both housing and neighbourhood satisfaction. Participants will also be asked to indicate their housing type (e.g. detached, high rise apartment, etc.), housing tenure, and the number of individuals who live at their primary residence, as these are found to impact mental health [
66]. This will allow the research team to determine if potential changes to health and healthcare use can be attributed to perceptions of living environment rather than just the affordability aspect of subsidized housing.
Preliminary data analysis
Random effects regression has the advantage of allowing researchers to explicitly account for within-person changes or unmeasured heterogeneity within individuals across time [
67]. Unmeasured heterogeneity can be described as the unmeasured consistencies in individuals that might influence mental health and healthcare use within each wave of data collection. The research team will first explore the longitudinal changes in primary and secondary outcomes using descriptive statistics pre- and post-intervention, as well as spaghetti plots. To take advantage of the longitudinal nature of our data, we will estimate generalized linear mixed effects models that we predict will take the following form:
$$G\left({Y}_{i,t}\right)={X}_{i,t}\beta +{Z}_{i}u+{\epsilon }_{i,t}$$
\({Y}_{i,t}\) is our outcome variable (see main and secondary outcomes above) and \(G\) is an appropriate link function (i.e. logistic for dichotomous variables and identity for continuous variables). \({X}_{i,t}\) is a vector of variables that we will treat as having fixed effects (β), \({Z}_{i}\) is a vector of variables and their estimated random effects (u), and \({\epsilon }_{i,t}\) is the remaining error \({X}_{i,t}\), which will include variables that can influence mental health or healthcare use and might not be orthogonal to housing status, like time on waitlist, age, etc. We will also explore whether seasonality (month) or interview wave (baseline, six month, 12 month, 18 month) are appropriate to include in our model. \({Z}_{i}\) is a vector of random effects. We will start by including random intercepts in \({Z}_{i}\) and their estimated coefficients (u), designed to consider whether individual-specific factors can influence outcomes over time, and potentially include random-slope estimates for variables (like sex) if our summary statistics indicate important differences by covariates.
We will explore the effects of gender, age, housing status and chronic disease morbidity at study entry, and interactions of selected key variables. Without observing the data, the research team cannot commit to more sophisticated modeling approaches, but we have a flexible estimation strategy that allows us to take advantage of the longitudinal nature of the data. Interim analyses will be performed as data are collected.
Study retention
New Brunswick’s DSD will partner with the research team to provide access to the study population, recruitment assistance, and monthly updates on receipt of subsidized housing for participants who consent. Prior to obtaining consent at six months, and for individuals who do not consent to share their name with DSD for monthly updates, a screening tool will be used at regular survey intervals to assess whether a participant has received subsidized housing and should be transferred into the intervention group. DSD is committed to using the results of this study to improve the wellbeing of residents who are waiting for and receiving subsidized housing. This study will provide descriptive information on the wellbeing of those waiting for subsidized housing, which may point to the need for additional health supports.
Using a longitudinal study design is advantageous as it allows us to relate any observed mental and physical health effects to exposure to housing affordability concerns. Further, investigating change over time allows us to determine the impact of housing on mental health, physical health and healthcare use when participants move and as they become more settled in subsidized housing. However, a concern with longitudinal cohort studies is study retention.
Some attrition is expected in a longitudinal cohort study. To reduce attrition, Scott’s Engagement, Verification, Maintenance and Confirmation (EVMC) Protocol will be used [
48]. Scott’s use of this protocol resulted in a 95% retention rate in their study of individuals who experience high residential instability. The ECVM Protocol involves training research assistants to properly motivate study participants by informing them of the social benefits of their research participation; collecting and updating contact information; scheduling follow-up surveys at the end of each survey; and providing reminder cards with a number for the participants to call should they need to update their contact information.
The social benefits of study participation will be clearly conveyed to participants by research assistants who administer phone surveys or in text through the electronic and mailed surveys. All participants will be asked to provide a mailing address, email address, and phone number each time they participate. Participants who are unhoused while waiting for public housing will be asked permission to contact them at a shelter, agency, or through another mechanism of their choice. All participants will be reminded at the end of each survey that they will be contacted in approximately six months for their next survey. If contact methods are not up to date at their follow-up dates (e.g. phone number is out of service or email bounce back), a reminder card will be mailed to let them know that it is time for their next survey. This letter will provide the research team’s contact information and a request to contact the study team to update their information. DSD will update contact information monthly for all unreachable participants who agreed to have their information shared for the research.
Participation will be incentivized with a draw at the end of the study and a gift card following each survey, which may motivate some participants to maintain up-to-date contact information. A systematic review of study retention methods finds that offering incentives is an optimal practice to increase study retention [
68].