Exploring the feasibility of conducting a multisite randomised controlled trial of the healthy and HomED model of care and its impact on representations among people experiencing homelessness in an Australian Emergency Department

Abstract

Objectives

To determine the feasibility of conducting a multisite randomised controlled trial of the Healthy and HomED model of care to determine its capacity to reduce emergency department (ED) representations among people experiencing homelessness.

Design

Feasibility randomised controlled trial with process evaluation.

Setting

Single site public metropolitan ED.

Participants

Our goal was to recruit 204 adults experiencing homelessness attending ED. In total, 190 participants were enrolled. The final sample comprised n=101 control and n=86 intervention.

Interventions

The Healthy and HomED model of care comprises screening for homelessness, assessing unmet needs via the Homeless Health Access to Care Tool and a Decision Assistance Guide that informs care planning. The intervention was applied by the Assessment Liaison Early Referral Team (ALERT), an ED-based multidisciplinary team specialised in supporting underserved populations.

Main outcome measures

Clinician acceptability of the implementation of the model of care, improved identification of homelessness and a reduction in ED representations within 28 days among people experiencing homelessness.

Results

Implementation of the Healthy and HomED model of care was feasible to the ALERT clinicians and the research team. While the Healthy and HomED did not significantly reduce representations to the ED, it improved the identification of homelessness by a third (35.3%). Qualitative findings suggest that the Homeless Health Access to Care Tool provided helpful standardisation to assessments. The Decision Assistance Guide was seldom added to the plan of care for senior clinicians but was reportedly helpful to junior clinicians with less experience in homeless healthcare.

Conclusions

The study provided assurance that running a multicentre hybrid trial to test the effectiveness and implementation of the Healthy and HomED is feasible. Process evaluation found that intervention adherence could be improved with greater contextualisation to local resources and increased engagement from ED medical and nursing teams. These factors could be addressed through the codesign of a future multisite trial.

Trial registration number

ANZCTR 12622001085763.

Strengths and limitations of this study.

• This is one of the first studies to explore the effectiveness of a model of care designed to address emergency department (ED) access to care for people experiencing homelessness in Australia.

• Participant recruitment was far slower than anticipated, and there were high levels of drop-out from postintervention surveys.

• There was a small but not statistically significant reduction in ED representations (primary outcome) in the intervention group compared with the control.

• High levels of health-related vulnerability existed among participants experiencing homelessness.

• Codesigning a future trial would provide the opportunity for lived and living experience input to the use of surveys and other data collection methods, which may improve design appropriateness and retention throughout a trial.

Introduction

The incidence of homelessness is increasing globally and brings with it serious risks to physical and mental health.¹ A person is defined as experiencing homelessness if they are living in either non-conventional accommodation (eg, squatting and living in parks), sleeping rough or living in short-term or emergency accommodation.² In this study, homelessness is categorised according to Chamberlain and MacKenzie’s definition³: primary—people who are unable to access suitable accommodation, including those sleeping rough and people sheltering in improvised dwellings, such as sheds or abandoned buildings. Secondary—people in temporary accommodation who move regularly. This includes emergency or transitional accommodation, couch surfing and boarding house accommodation for 12 weeks or less. Tertiary—concerns people with a boarding house tenancy of 13 weeks or more.

In the UK, homelessness rose by 14% to an estimated 309 000 from 2022 to 2023.⁴ In the same period in the USA, homelessness increased by 12% to an estimated 653 104 people.⁵ In Australia, the 5 yearly Census data reflected a 5.2% increase in homelessness.⁶ The greatest increase in homelessness in Australia has been seen in women (10.1% increase) and among Aboriginal and Torres Strait Islander peoples (6.4% increase). Aboriginal and Torres Strait Islander peoples are over-represented among people experiencing homelessness, comprising 20.4% of those homeless, yet only 3.8% of the total population.⁷

The circumstances leading to homelessness can be complicated and are not easily resolved. Many social, economic and health-related factors lead to homelessness.⁸ Mission Australia reports domestic and family violence is one of the leading causes of homelessness, especially for women and children.⁹ Other common causes of homelessness include sudden life changes (eg, divorce and unemployment), lack of affordable and social housing and mental health challenges.⁹ Compared with the general population, those experiencing homelessness have poorer health outcomes, including higher mortality rates,^{10 11} and are more likely to suffer from mental health conditions, infections, cardiovascular and respiratory conditions and premature ageing.¹¹ Concomitant issues with substance use, mental and physical health (trimorbidity) are also common.^{12 13} Structural causes of homelessness include poverty, incarceration and availability of affordable accommodation.¹² Individual factors such as physical and mental health conditions, and domestic and family violence and childhood trauma can both cause and exacerbate homelessness.^{12 14}

Health-seeking behaviour differs between those experiencing homelessness and those in stable housing. Reported barriers include affording transport to an appointment, balancing healthcare with competing demands of finding somewhere to sleep and obtaining food and accessing social supports, and the stigma and subsequent shame often experienced when accessing care.14,16 Australian data have shown that rather than seeking primary healthcare, people experiencing homelessness are more likely to attend an emergency department (ED), often at a later stage of ill health.14,21 People experiencing homelessness account for between 8% and 10% of ED attendances,^{19 22} and their presentations are more likely to lead to a hospital admission, compared with those in stable housing.²³ People experiencing homelessness are also more likely to reattend an ED. A recent Australian study identified that 43% of people experiencing homelessness reattended the same ED within 28 days.²²

Ensuring health equity for people experiencing homelessness is extraordinarily challenging, even in countries with access to universal healthcare. The patterns of healthcare access among those experiencing homelessness mean that EDs have a crucial role in helping because they are sometimes a person’s initial and sole point of contact with a healthcare system.^{15 16 22} While EDs are not necessarily responsible or resourced to resolve the multifaceted health and social challenges faced by people experiencing homelessness, they can serve an important role in identifying homelessness and potentially delivering or directing interventions.2224,26 The prevalence of homelessness and the high level of representations create an urgent need for ED interventions that are tailored to address the unique needs of patients experiencing homelessness. Such interventions hold the potential to reduce the mortality and morbidity associated with homelessness, enhance cost-effectiveness and may alleviate the strain of overcrowding in EDs.^{15 16}

Healthy and HomED model of care

To address the identification of unmet need and improve access to care of people experiencing homelessness, a team of researchers and clinicians since 2019 has developed an assessment tool called the Homeless Health Access to Care Tool (HHACT) (online supplemental file 1) and a Decision Assistance Guide (online supplemental file 2). Collectively, these were originally named the ‘Homeless Health Response Bundle’ and have since been renamed through a codesign process as the ‘Healthy and HomED’ to reflect a strengths-based approach. The HHACT is a 24-item screening tool that identifies a person’s level of health and social-related vulnerability, which is defined as a person’s burden of injury and illness combined with their ability to access healthcare and social supports. The tool is underpinned by two literature reviews,^{27 28} a Delphi study²⁹ and pilot psychometric testing.³⁰ Depending on the level of health and social-related vulnerability (high>21, moderate 11–20 and low 10 or less) determined by applying the tool, a site-specific decision assistance guide is used to support shared decision-making between clinicians and people experiencing homelessness.

Many clinicians perceive that they are underprepared to provide a holistic response to people experiencing homelessness. Findings from our recent survey and interviews with nurses in Australia suggest that they perceive a lack of skills, knowledge and resources to respond effectively to the often complex presentations of people experiencing homelessness.^{31 32} As one of the nurse participants stated, ‘…what’s the point of me asking ‘Are you homeless?’…I can’t do anything about it’ but desire a clearer way to identify and respond, ‘it would be really nice if you could follow-up with something else.’³² The decision assistance guide is designed to overcome this underpreparedness by providing supporting prompts to evidence-based investigations, treatments and referrals that link people with existing hospital and community services. We hypothesise that a systematic and inclusive model of care for screening, assessing and responding to the needs of people experiencing homelessness in ED will improve their access and will optimise the use of existing hospital and community services and build ED clinician capability. The purpose of this study is to explore the feasibility of implementing the Healthy and HomED and testing its effectiveness using a randomised controlled trial design.

Methods

Trial design

With the future goal to undertake a large multisite hybrid randomised controlled trial of the effectiveness of the Healthy and HomED, a feasibility parallel randomised controlled trial was developed to test this design in one centre. The study was conducted at a metropolitan ED in Melbourne that receives, approximately, 50 000 presentations per year.³³ The study is reported here using the Consolidated Standards of Reporting Trials extension for pilot trials guideline.³⁴ The study was registered with the Australia New Zealand Clinical Trials Registry (12622001085763).

A key component of assessing feasibility was whether the ED clinicians would actively engage with the Healthy and HomED and view it as an acceptable, feasible and appropriate model of care. We chose the Integrated Promoting Action Research In Health Sciences (iPARIHS) framework to guide implementation of the Healthy and HomED because of its strong focus on facilitation of interventions and appreciation of context.^{35 36} To provide support to both the ED staff and research assistants, a site facilitator was employed. Implementation was undertaken in three stages, as shown in figure 1. Of particular value was the site readiness assessment, which was conducted during preimplementation using the iPARIHS preimplementation facilitation checklist. The facilitation checklist led to the identification of barriers and facilitators and assisted in identifying the key thought leaders within the ED, with whom we subsequently met to discuss the trial and smooth the path for implementation.

Figure 1. Implementation approach.

Interventions

The trial involved implementation of the three components of the Healthy and HomED: (1) screening for homelessness; (2) applying the HHACT to identify the degree of vulnerability and unmet need and (3) using the decision assistance guide to develop a plan of care. Hospitals in Australia do not routinely screen for housing status; therefore, the first step in our trial was to proactively screen people attending the ED for homelessness. Screening was conducted by research assistants using a standardised script developed by Lee et al²² (online supplemental file 3). The script asked about a patient’s accommodation status over the last 6 months. Once homelessness was identified and participant consent obtained, the intervention group received the HHACT, administered by a member of the Assessment Liaison Early Referral Team (ALERT). The ALERT is an expert multidisciplinary team of care coordinators comprising allied health, a peer support worker and nursing staff, who are based in the ED. They provide comprehensive psychosocial assessments and develop holistic plans of care for people attending the ED who are at risk of representation.

Once administered, the vulnerability level derived from the HHACT (low, moderate and high) was responded to using the decision assistance guide to inform a participant’s plan of care. The decision assistance guide was developed during preimplementation, through discussions with key ED stakeholders, including the ALERT members, ED Medical Director, ED nursing staff members and the Director of Mental Health and Alcohol and other Drug. The decision assistance guide was intended as a prompt for the ALERT staff and the ED health professionals to consider undertaking specific physical and cognitive assessments, fast-tracking access to a bed in the ED, Mental Health and Alcohol and Other Drug Unit or short stay unit, or referrals. Prior to implementation, two 1-hour online education sessions were delivered by the authors (JC, AC, JH and CD) to the ALERT staff on how to use the HHACT and the decision assistance guide.

Study population

Eligibility

In order to determine the times of day to conduct the study, we reviewed heat maps from routinely collected ED data to understand the peak times that people experiencing homelessness attended the study site. People experiencing homelessness attended most often during weekdays; therefore, participants who attended between 7:30 and 19:30 Monday to Friday were invited to be screened for homelessness and enrol in the study. We originally stipulated that participants were eligible if they had not yet been treated by a medical or nurse practitioner. During the first few days of data collection, this proved problematic as many patients eligible for enrolment during the research assistants shifts had already been medically reviewed. To increase enrolment, we changed this criteria so that participants were eligible to participate in the study once triaged and prior to being reviewed by a member of the ALERT. Of those experiencing homelessness, written consent was sought for their participation in the study. Patients were ineligible if they were under 18 years of age, experienced cognitive impairments or other conditions that rendered them unable to provide informed consent (eg, critically unwell, intoxicated and distressed) and those not experiencing homelessness presently or within the past 6 months. Participants were eligible to enrol only once in the study.

Enrolment was undertaken in the ED by research assistants. We employed research assistants with a health background (eg, nursing, social work and paramedicine) and who were familiar with the ED environment. The site facilitator provided specific training and support in administering the homelessness screening script and enrolling participants to the study. Participants were recruited from all areas of the ED, including the waiting room, subacute and acute area and the ED-situated ward area. Prior to obtaining consent, a participant information sheet was provided and discussed in detail, including a plain English explanation of the project appropriate to each participant.

Data collection

Data were collected at four distinct time points (figure 2). Baseline (T1): after obtaining consent, a research assistant administered the satisfaction with life domains scale.³⁷ This scale had previously been employed to assess the quality of life in people experiencing homelessness.^{38 39} Once each participant had completed their episode of care in the ED, the Australian Hospital Patient Experience Question Set was administered (T2).⁴⁰ At 28 days after each participant’s initial ED attendance, electronic hospital records were used to identify any representations (T3). At 30 days after each participant’s initial ED attendance, the satisfaction with life domains scale was administered for a second time via telephone by a member of the research team (AV-H). To thank them for their time, participants were provided with a $20 voucher at the completion of their episode of care, and another $20 voucher on completion of the 35-day follow-up.

Figure 2. CONSORT diagram. CONSORT, Consolidated Standards of Reporting Trials.

At the end of the study, the implementation of the Healthy and HomED was assessed using the acceptability, appropriateness and feasibility survey.⁴¹ This survey was distributed by an email from one of the authors (CD) to the ALERT staff members and doctors and nurses in the ED (n=230). A reminder email was sent 2 weeks later. Three focus groups were conducted by two authors (JC and AV) to explore the ALERT members’ perspectives on using the Healthy and HomED. One focus group was conducted halfway through the trial, and two focus groups were conducted 1 week after completing the trial.

Randomisation

Our intention was to use a computer-generated block randomisation schedule. The block sequence of randomisation was prepared by an administrative assistant who was not involved in the study and the sequence was emailed to the study site. The randomisation sequence was placed in individual envelopes stored in a locked cupboard in the ED. Once consented to participate, a research assistant informed an ALERT staff member, who then opened the next preprepared envelope by number order, thereby randomising participants to either control or intervention. The research assistants and statistician were blinded to participant allocation. Due to an administrative oversight, the envelopes were not labelled with their designated numbers on the outside. The envelopes were, therefore, opened in a non-prescribed sequence that equated to a simple randomisation method.

Outcomes

The study’s primary outcome was a 20% reduction of ED representations within 28 days. Other outcomes were (1) feasibility, acceptability and appropriateness of the Healthy and HomED to staff members, (2) patient satisfaction with the healthcare received, (3) patient’s perceived quality of life, (4) impact on access to care measured by: did not wait to be seen, disposition (admission/discharge), length of time waiting to be seen, triage category and length of time in the ED and (5) feasibility of conducting a multisite trial through the analysis of recruitment rate, drop-out rate and effect size.

Sample size

To understand the required sample size, we consulted the literature. A study by Lee et al²² showed ED 28-day representations among people experiencing homelessness to be 43%. To reduce hospital reattendances by 20% between the intervention and usual care groups, 86 participants were required per group to achieve 80% power with a significance level of 5%, 104 after accounting for 20% dropout. Therefore, our goal was a total sample size of 208 participants. Sample size calculations were produced using PASS (https://www.ncss.com/software/pass/). Based on the estimates from other studies^{19 22} and routinely collected ED data from the site, we anticipated that, approximately, five participants per day would be recruited. For various reasons, the recruitment was slower than anticipated and took 8 months in total (2–3 per day). It took the research assistants a few weeks to gain confidence in their roles, which required one-on-one support from the site facilitator. The fact that participants could only enrol once in the study, and not multiple times, also slowed the rate of recruitment.

Analysis

All analyses were performed on an intention-to-treat basis. Categorical variables were described using frequencies and percentages and analysed using χ². Fisher’s exact test was used when the expected count was low. Logistic regression was used to provide an effect size for hospital representation with ORs and 95% CIs reported. Continuous variables were described using means and SD or median and IQR. Quality of life was measured pre and postintervention, with paired t-tests used to examine within-group differences. General linear models were used to explore group differences with postscores adjusted by baseline quality of life scores. Bootstrapped 95% CIs and p values were reported for all QOL data due to being on a seven-point ordinal scale. Missing data for the QOL data were examined for patterns of missing using χ² and t-tests.

The qualitative focus group data were transcribed by one author (AV) and analysed using the thematic approach of Braun and Clarke⁴² by two authors independently (JC and AV). The themes and codes were then discussed between the two authors and a consensus was reached.

Patient and public involvement

Patients and public were not involved in the design or conduct or reporting or dissemination plans of our research.

Identification of shortcomings

Issues related to the study were monitored through a progress log that was maintained by the site facilitator and reviewed at weekly research team meetings that were attended by all researchers, including the research assistants.

Results

Sample

Baseline characteristics of participants were similar in terms of age, gender, homelessness type and Aboriginal and Torres Strait Islander identification (table 1). The sample included a high proportion of people experiencing primary homelessness (n=82, 44.6%). The study site has a system for recording housing status using a ‘homeless flag’, which is available through a dashboard using data from the electronic medical records (Power-BI report). The homeless flag is applied after a person’s ED presentation based on their recorded address and discharge administrative codes. A further n=66 (35.3%) people were identified as homeless in addition to those that had a homeless flag on the housing report.

Table 1. Demographic variables.

Characteristic	N	Overall, N=187*	Group		P value†
			Control, N=101*	Intervention, N=86*
Age	187	43.1 (12.0)	44.6 (12.2)	41.5 (11.6)	0.078
Age category	187				0.260
<26		18 (9.6%)	7 (6.9%)	11 (12.8%)
26–35		33 (17.6%)	18 (17.8%)	15 (17.4%)
36–45		59 (31.6%)	28 (27.7%)	31 (36.0%)
46–55		48 (25.7%)	32 (31.7%)	16 (18.6%)
56–65		25 (13.4%)	13 (12.9%)	12 (14.0%)
>65		4 (2.1%)	3 (3.0%)	1 (1.2%)
Homelessness_type	184				0.763
Primary		82 (44.6%)	47 (46.5%)	35 (42.2%)
Secondary		81 (44.0%)	42 (41.6%)	39 (47.0%)
Tertiary		21 (11.4%)	12 (11.9%)	9 (10.8%)
Homeless_flag‡	187				0.914
Homeless		121 (64.7%)	65 (64.4%)	56 (65.1%)
Not homeless		66 (35.3%)	36 (35.6%)	30 (34.9%)
Aboriginality	184				0.395
Aboriginal and Torres Strait Islander		28 (15.2%)	13 (13.1%)	15 (17.6%)
Non-indigenous		156 (84.8%)	86 (86.9%)	70 (82.4%)
Gender	187				0.865
Female		44 (23.5%)	25 (24.8%)	19 (22.1%)
Male		141 (75.4%)	75 (74.3%)	66 (76.7%)
Non-binary		2 (1.1%)	1 (1.0%)	1 (1.2%)
Marital_status	186				0.264
Married or defacto		15 (8.1%)	6 (6.0%)	9 (10.5%)
Divorced/separated/widowed		25 (13.4%)	11 (11.0%)	14 (16.3%)
Never married		146 (78.5%)	83 (83.0%)	63 (73.3%)

^*Mean (SD); n (%).

^†Welch two-sample t-test; Fisher’s exact test and Pearson’s χ2 test.

^‡Homeless flag applied retrospectively based on the postcode of 1000, no fixed abode and diagnostic ICD-10 code Z59.

Recruitment and consent

A total of 6765 individuals attending the ED between 15 November 2022 and 28 July 2023 were screened for homelessness. Of these, n=241 were eligible to participate, of which n=51 were unwilling and n=190 were deemed eligible for randomisation (figure 2). Funding constraints meant that we ceased recruitment before reaching our goal of n=208.

The Australian Hospital Patient Experience Question Set was completed by n=47 in the control group and n=31 in the intervention. There was a high level of drop-out for the postintervention surveys. In some instances, this was because the patient had completed treatment and left the ED before the research assistant had the opportunity to undertake the survey with them, and patients enrolled towards the end of the day and were discharged after 19:30 were missed. There were no significant differences in the responses between the two groups (online supplemental file 4). The pre–postcompletion of the satisfaction with life domains scale also had a large reduction in the number of completions via telephone follow-up at T4 since many were not contactable via the phone number they had provided (online supplemental file 5). Informal feedback from the research assistants suggested that the satisfaction with life domains scale was difficult for participants to comprehend.

Shown in table 2, the number of participants that did not wait to be seen in ED was low. Over half of the samples were admitted to the short stay unit in the ED or the mental health/alcohol and other drug hub with the intention of discharging them within 24 hours, and a small proportion was admitted to hospital (n=4, 2.1%). Regarding the primary outcome, there was no difference in representation to ED between groups (p=0.883) with approximately 40% of patients with an average of 0.8 representations (SD=1.3).

Table 2. Clinical variables.

Characteristic	N	Overall, N=187*	Group		P value†
			Control, N=101*	Intervention, N=86*
Length_of_stay_minutes	187	226.0 (136.5, 358.5)	227.0 (132.0, 366.0)	223.0 (158.8, 354.8)	0.939
Mode_of_arrival	187				0.937
Ambulance—road		58 (31.0%)	33 (32.7%)	25 (29.1%)
Other		48 (25.7%)	25 (24.8%)	23 (26.7%)
Public transport		42 (22.5%)	23 (22.8%)	19 (22.1%)
Own transport		33 (17.6%)	16 (15.8%)	17 (19.8%)
Police vehicle		4 (2.1%)	3 (3.0%)	1 (1.2%)
Priv amb/mas/rav		2 (1.1%)	1 (1.0%)	1 (1.2%)
Departure_status	187				0.905
Short stay admission		88 (47.1%)	50 (49.5%)	38 (44.2%)
Home		55 (29.4%)	28 (27.7%)	27 (31.4%)
Ed Mhaod hub		32 (17.1%)	16 (15.8%)	16 (18.6%)
Left—not seen		5 (2.7%)	2 (2.0%)	3 (3.5%)
Admit to ward		4 (2.1%)	3 (3.0%)	1 (1.2%)
Left—part Rx only		2 (1.1%)	1 (1.0%)	1 (1.2%)
Another hospital		1 (0.5%)	1 (1.0%)	0 (0.0%)
Triage_category	187				0.276
1–3		108 (57.8%)	62 (61.4%)	46 (53.5%)
4–5		79 (42.2%)	39 (38.6%)	40 (46.5%)

^*Median (IQR); n (%).

^†Wilcoxon rank sum test; Fisher’s exact test and Pearson’s χ2 test.

Determining a primary outcome for the multicentre trial

There was a small but not statistically significant reduction of representations in the intervention group (OR=0.96, 95% CI 0.53, 1.72; p=0.883) compared with control (table 3). Data extracted from the completed HHACTs reflected extremely high levels of health and social vulnerability, particularly in relation to the presence of mental health conditions and disability among people experiencing homelessness and extremely high levels of hospital use (table 4).

Table 3. Percentage of representations over 28 days by group.

Characteristic	Overall, N=187*	Group		P value†
		Control, N=101*	Intervention, N=86*
Representation_28_days	75 (40.1%)	41 (40.6%)	34 (39.5%)	0.883
count_of_representations				0.999
0	112 (59.9%)	60 (59.4%)	52 (60.5%)
1	38 (20.3%)	20 (19.8%)	18 (20.9%)
2	21 (11.2%)	12 (11.9%)	9 (10.5%)
3	6 (3.2%)	3 (3.0%)	3 (3.5%)
4	5 (2.7%)	3 (3.0%)	2 (2.3%)
5	1 (0.5%)	1 (1.0%)	0 (0.0%)
6	4 (2.1%)	2 (2.0%)	2 (2.3%)

^*n (%).

^†Pearson’s χ2 test and Fisher’s exact test.

Table 4. HHACT scale by representation (HHACT screening tool available in online supplemental file 1).

			Representation		P value†
Characteristic	N	Overall, N=86*	No, N=52*	Yes, N=34*
Aged>55 years	71	11 (15.5%)	8 (72.7%)	3 (27.3%)	0.506
Interpreter required	71	1 (1.4%)	0 (0.0%)	1 (100.0%)	0.408
Aboriginal/Torres Strait Islander	71	18 (25.4%)	7 (38.9%)	11 (61.1%)	0.043
Part of the stolen generation	71	15 (21.1%)	7 (46.7%)	8 (53.3%)	0.268
Non-binary	71	3 (4.2%)	2 (66.7%)	1 (33.3%)	>0.999
Gender changed since birth	71	2 (2.8%)	1 (50.0%)	1 (50.0%)	>0.999
Identifies as LGBTIQ+	71	18 (25.4%)	10 (55.6%)	8 (44.4%)	0.719
Pregnant	71	2 (2.8%)	0 (0.0%)	2 (100.0%)	0.163
Primary homelessness‡	71	31 (43.7%)	17 (54.8%)	14 (45.2%)	0.515
Homeless >6 months	71	42 (59.2%)	23 (54.8%)	19 (45.2%)	0.365
Has an emergency contact	71	28 (39.4%)	18 (64.3%)	10 (35.7%)	0.229
Medicare access	71	9 (12.7%)	8 (88.9%)	1 (11.1%)	0.073
Refugee or seeking asylum	71	6 (8.5%)	3 (50.0%)	3 (50.0%)	0.683
Avoid care when unwell	71	35 (49.3%)	16 (45.7%)	19 (54.3%)	0.023
In past 6 months received ED care/taken to hospital by ambulance or police/hospital admission/ police custody?	71				0.603
None of the above		5 (5.8%)	4 (80.0%)	1 (20.0%)
1 of the above		14 (19.7%)	9 (64.3%)	5 (35.7%)
2 of the above		52 (73.2%)	29 (55.8%)	23 (44.2%)
Able to self-care own needs	71	14 (19.7%)	7 (50.0%)	7 (50.0%)	0.437
Observed to be unable to self-care	71	3 (4.2%)	0 (0.0%)	3 (100.0%)	0.064
No income	71	10 (14.1%)	8 (80.0%)	2 (20.0%)	0.183
Number of medical conditions	71				0.310
0		20 (28.2%)	13 (65.0%)	7 (35.0%)
1		21 (29.6%)	10 (47.6%)	11 (52.4%)
2		16 (22.5%)	12 (75.0%)	4 (25.0%)
3 or more		14 (19.7%)	7 (50.0%)	7 (50.0%)
Advised regular medications	71	48 (67.6%)	26 (54.2%)	22 (45.8%)	0.217
Not taking medications as advised	71	19 (26.8%)	8 (42.1%)	11 (57.9%)	0.043
Consumed alcohol or drugs	70	46 (65.7%)	24 (52.2%)	22 (47.8%)	0.064
Appears under the influence/withdrawing	69	21 (30.4%)	9 (42.9%)	12 (57.1%)	0.064
Worried about mental health	69	49 (71.0%)	29 (59.2%)	20 (40.8%)	0.950
Mental health condition	70				0.501
0 condition		17 (24.3%)	12 (70.6%)	5 (29.4%)
1 condition		11 (15.7%)	6 (54.5%)	5 (45.5%)
2 conditions		16 (22.9%)	11 (68.8%)	5 (31.3%)
3 conditions		26 (37.1%)	13 (50.0%)	13 (50.0%)
Self-harm thoughts	69	28 (40.6%)	15 (53.6%)	13 (46.4%)	0.414
Thoughts of suicide	69	29 (42.0%)	16 (55.2%)	13 (44.8%)	0.541
Signs/symptoms mental health	70	32 (45.7%)	21 (65.6%)	11 (34.4%)	0.378
Has a disability	70				0.122
0 disability		42 (60.0%)	29 (69.0%)	13 (31.0%)
1 disability		21 (30.0%)	9 (42.9%)	12 (57.1%)
2 disabilities		7 (10.0%)	4 (57.1%)	3 (42.9%)
Observed disability	70	15 (21.4%)	6 (40.0%)	9 (60.0%)	0.143
Feel unsafe/threatened/harmed	70	19 (27.1%)	9 (47.4%)	10 (52.6%)	0.188

^*n (%).

^†Fisher’s exact test and Pearson’s χ2 test.

^‡Sleeping on the street, in a car and on a train.

ED, emergency department; HHACT, Homeless Health Access to Care Tool.

The mean representations by group were the same (mean=0.8, SD=1.3). A gradient was observed for the HHACT score, with a larger percentage of patients representing to the hospital as vulnerability increases (p=0.026) (table 5).

Table 5. HHACT vulnerability categorisation by representation.

Characteristic	N	Overall, N=86*	Representation		P value†
			No, N=52*	Yes, N=34*
HHACT_cat	68				0.026
Slightly vulnerable		11 (16.2%)	10 (90.9%)	1 (9.1%)
Moderately vulnerable		36 (52.9%)	21 (58.3%)	15 (41.7%)
Highly vulnerable		21 (30.9%)	9 (42.9%)	12 (57.1%)

^*n (%).

^†Fisher’s exact test.

HHACT, Homeless Health Access to Care Tool.

Acceptability of the model of care

Five ED clinicians completed the acceptability, feasibility and appropriateness survey. Using the scale of 1–5, mean responses were as follows: acceptability (3.3, SD 0.8), appropriateness (3.6, SD 0.8) and feasibility (3.3, SD 0.7). Through thematic analysis of the qualitative data, three themes were identified: (1) identification of vulnerability; (2) impact on access to care and (3) continuity of the Healthy and HomED (online supplemental file 6). The Healthy and HomED was considered helpful in its capacity to facilitate the identification of homelessness and identify vulnerabilities, especially those relating to the social determinants of health. The ALERT members perceived that the HHACT and Decision Assistance Guide were easy to use and were a useful prompt but seldom altered their plans of care. In relation to access to care, participants perceived that the Healthy and HomED would be helpful to newer and less experienced staff members and could be used to advocate for people experiencing homelessness in relation to admissions and referrals.

“As a structured tool for our staff to use when they first step into that space, it could be really useful for them to have that structured guide and the kinds of referrals or interventions that they could be thinking about” (FG2)

In relation to continuing to use the Healthy and HomED, there were mixed perceptions on the best way for it to be used, with some concerns regarding the capacity of ED clinicians to use it. Others recommended using the Healthy and HomED in case management, and as a part of a workplace orientation, and as a part of after-hours ED discharge. One participant even suggested, “It could become a standard language across the homeless sector…” (FG3).

Discussion

There are very few studies that examine ED access for people experiencing homelessness, even fewer randomised controlled trials. It is our intention that findings from our feasibility trial will inform the design and conduct of other trials focused on ED access for people experiencing homelessness. Our recruitment strategy using research assistants with a health professional background was effective, and the majority of eligible participants were willing to participate, which led to adequate participation numbers. Our sample was reflective of the three types of homelessness as follows: rough sleeping (primary), couch surfing (secondary), boarding house (tertiary) and a broad range of ages and included a proportion of Aboriginal and Torres Strait Islander people that reflects national statistics. We learnt that recruitment was far slower than anticipated, and it took double the length of time than initially planned, which is an important consideration in designing future trials.

In terms of data collection, the drop-out rate at T2 and T4 was high. The drop-out at T2 was partly a reflection of the fast pace of ED care, which meant it was extremely challenging for the research assistants to administer a patient satisfaction survey before each of the participants was discharged from the ED. It was also reflective of episodes of care not being finalised prior to the completion of a research assistants shift. The drop-out at T4 reflected the difficulties associated with contacting participants via telephone a month following their enrolment to the study. In designing a definitive multicentre trial, we would recommend using as few surveys as possible and instead relying on data that forms a part of the intervention and care provided, in this case the HHACT, and the administrative data that form a part of usual care, for example, representation data, triage and length of stay. We also concluded that codesigning a future trial would provide the opportunity for lived and living experience input to the use of surveys and the questions included in surveys, which may improve design appropriateness and retention throughout a trial.

Practical implementation of the healthy and HomED

The screening of homelessness conducted by the research assistants was extremely meaningful in identifying people experiencing homelessness that would, otherwise, have not been identified. In designing future trials, we intend to partner with leaders in national health policy and hospital accreditation to ensure that outcomes related to proactive screening can be rapidly translated into practice.

To overcome the absence of contemporaneous screening at the ED study site and to enable participant enrolment into the trial, screening for housing status was conducted by research assistants. Screening for homelessness at ED reception is an important component of the full implementation and sustainment of the Healthy and HomED model of care. In future studies, it will be important to test the effectiveness of screening conducted by ED reception staff.

In completing the HHACT survey and the decision assistance guide, the site insisted on using paper-based forms. This meant that summing the survey response scores was undertaken manually. On analysing the data, it was evident that errors in the summing of the vulnerability scores occurred. For six participants, their incorrect scores were clinically significant because they led to the use of an incorrect DAG pathway; for example, the moderately vulnerable pathway was used when highly vulnerable was the correct pathway. Future use of electronic data collection that automatically sums the overall vulnerability score will be critical.

With the exception of one incident, the trial did not cause harm or distress to the participants. During one of the postimplementation focus groups, an ALERT member reflected that asking the mental health questions in the HHACT had led to one of their patients becoming distressed because the patient did not wish to discuss their mental health conditions. Based on this feedback, we have since codesigned the HHACT with people with lived and living experience of homelessness, which has led to changes in the order of the questions and the addition of a question on smoking and nicotine replacement therapy, but there were no changes recommended for the mental health questions. The occurrence described above might be an isolated incident; however, education to clinicians in future trials about applying the HHACT will include this example as a point of discussion and reflection.

Qualitative data from the ALERT suggested that implementation of the Healthy and HomED was feasible. Feedback on the HHACT indicated that it standardised patient assessment and was a useful prompt that facilitated a holistic and comprehensive assessment approach. The decision assistance guide was perceived as being more useful to less experienced and junior members of staff, who were less familiar with the best-evidence approaches to planning care for underserved populations. Developing the decision assistance guide with the site was important. In a definitive trial, wider ED medical and nursing participation in developing the decision assistance guide, as well as lived experience input and community services input, would provide even further contextualisation to local resources.

Although not identified by study participants, it is possible that shortages in the availability of housing and community services will reduce the capacity of the Healthy and HomED to impact ED reattendances. For example, shortages of social housing and supportive models of housing may leave ED clinicians with limited discharge options, despite their best efforts. In Australia, the government funds Specialist Homeless Services (SHS), which are community-based charities and not-for-profit organisations that provide health, housing and social support services. While SHSs provided services to over 280 100 people between 2023 and 2024, they are also limited in the resources they can offer.⁴³

Another key lesson from the Healthy and HomED study is the difficulty of avoiding contamination between the intervention and control group when randomising at the patient level. We believe there was potential contamination in the trial design because the ALERT was providing support to participants in both the intervention and control groups, although they were not using the HHACT and the decision assistance guide with participants in the control group. This could be avoided by randomising by site using a cluster step-wedge or a cluster randomised controlled trial.

Primary outcome measure for future trials

The primary outcome for this study was representations to ED, and although the trial did not identify a statistically significant difference between control and intervention groups, we believe that this was due to other factors rather than the choice of primary outcome. The percentage of representations in this trial (40%) was similar to the study by Lee et al,²² also conducted in Melbourne, that reported a 28-day representation rate of 43%. Using anecdotal data from our trial site, in the 8 months prior to conducting this feasibility study, the representation rate of people with a homeless flag was 48%, and this figure increased to 49% during the trial period and 50% post-trial. Future studies are needed to understand the effectiveness of the Healthy and HomED. There may also be utility in exploring reattendances in different time periods (eg, 48 hours and 7 days), especially longer periods that could indicate the sustainability of the impact of the Healthy and HomED model of care. The identification of a representation gradient for the HHACT score is a clinically valuable finding in terms of planning care for those identified as moderately or highly vulnerable, for whom short admissions and linking to specialist community services for case management might be impactful.

Implications

According to the literature, the Healthy and HomED is the first ED-based model of care to be subjected to a randomised controlled trial in Australia. The National Health Service England has developed an ED pathway to support people experiencing homelessness.⁴⁴ The pathway comprises a series of questions and prompts on providing practical support in terms of referrals to housing and social support, and offering clothing and bedding, but a formal evaluation has not been reported.

Conclusion

This feasibility study has been extremely valuable in informing a multicentre study protocol to test the implementation and effectiveness of the Healthy and HomED. Our most important learnings are that randomisation at the patient level was not possible and, therefore, our future approach will be a cluster step-wedge or cluster randomised controlled trial. Codesign of future trials would ensure that all data collection approaches are appropriate and acceptable to people with living or lived experience of homelessness and ED clinicians. Including community service providers in the codesign of the site-specific decisions assistance guide would further enhance the local contextualisation of the Healthy and HomED.

Data availability statement

All data relevant to the study are included in the article or uploaded as supplementary information.