Introduction
Homelessness is an increasing social and public health emergency (World Health Organization, 2018). In the general population, the social gradient is considered a ‘slope’ of association between disadvantage and poor health outcomes (Marmot et al., Reference Marmot, Stansfeld, Patel, North, Head, White, Brunner, Feeney and Smith1991). For homeless populations, health inequity is so severe that this has been conceptualised as a ‘cliff’ (Story, Reference Story2013). People who experience homelessness often experience multiple adversities (Bramley et al., Reference Bramley, Fitzpatrick, Edwards, Ford, Johnsen, Sosenko and Watkins2015), have high levels of secondary care service use (Field et al., Reference Field, Hudson, Hewett and Khan2019) and premature mortality (Aldridge et al., Reference Aldridge, Story, Hwang, Nordentoft, Luchenski, Hartwell, Tweed, Lewer, Katikireddi and Hayward2018).
There are evolving preferences around terminology for multiple health conditions. In policy, clinical practice, and research, multiple co-occurring health conditions are referred to as multimorbidity (Macmahon et al., Reference Macmahon, Calverley, Chaturvedi, Chen, Corner, Davies, Ezzati, Guthrie, Hanson and Jha2018). There are different types of multimorbidity, dependent on different combinations of conditions (Johnston et al., Reference Johnston, Crilly, Black, Prescott and Mercer2019): for example, mental–physical multimorbidity includes co-occurring mental and physical conditions (Taloyan et al., Reference Taloyan, Alinaghizadeh, Wettermark, Hasselström and Bertilson2022). Studies of electronic health record data from patients accessing specialist homelessness primary care services have found high rates of multimorbidity and trimorbidity (co-occurring mental, physical and substance use conditions) (Bowen et al., Reference Bowen, Marwick, Marshall, Saunders, Burwood, Yahyouche, Stewart and Paudyal2019; Queen et al., Reference Queen, Lowrie, Richardson and Williamson2017; Vallesi et al., Reference Vallesi, Tuson, Davies and Wood2021). Similarly high rates of multimorbidity have been found in surveys of people in hostels, shelters, and homelessness outreach services (Rogans-Watson et al., Reference Rogans-Watson, Shulman, Lewer, Armstrong and Hudson2020; Vickery et al., Reference Vickery, Winkelman, Ford, Busch, Robertshaw, Pittman and Gelberg2021). There are indications that rates of multimorbidity are increasing in homeless populations (Vickery et al., Reference Vickery, Winkelman, Ford, Busch, Robertshaw, Pittman and Gelberg2021). People who have several diagnoses are more likely to be excluded from services, compounding barriers to accessing healthcare already faced by people who experience homelessness (Armstrong et al., Reference Armstrong, Shulman, Hudson, Stone and Hewett2021).
Common Mental Disorders (CMDs) include types of depression and anxiety (Lewis et al., Reference Lewis, Pelosi, Araya and Dunn1992). A recent updated systematic review found that, despite a widespread acknowledgement that CMDs and multimorbidity are highly prevalent in homeless populations, studies have rarely assessed for comorbidities with mental disorders (Gutwinski et al., Reference Gutwinski, Schreiter, Deutscher and Fazel2021). There is evidence that people with concurrent mental and physical morbidities have worse health-related quality of life and clinical outcomes, as well as an increased risk of premature mortality, compared to those with physical or mental morbidities only (Macmahon et al., Reference Macmahon, Calverley, Chaturvedi, Chen, Corner, Davies, Ezzati, Guthrie, Hanson and Jha2018).
Concerns have been raised about the poor quality of data on people who have experienced homelessness (Kaushal et al., Reference Kaushal, Jagpal, Khanal, Vohra, Lowrie, Johal, Jenkins, Saunders and Paudyal2021). Studies to date have largely focused on people who experience long-term homelessness and those who are in touch with specialist services. There is some research on formerly homeless populations living in settings such as supported accommodation and Housing First, however very little is known about health outcomes for people who move into private housing after homelessness (Demakakos et al., Reference Demakakos, Lewer, Jackson and Hayward2020). Studies which have examined former homelessness in private households have largely focused on other outcomes, such as adverse experiences, mortality, and overlaps between homelessness and other domains of disadvantage and exclusion (Bramley and Fitzpatrick, Reference Bramley and Fitzpatrick2018; Demakakos et al., Reference Demakakos, Lewer, Jackson and Hayward2020; Sosenko et al., Reference Sosenko, Bramley and Johnsen2020; Toro et al., Reference Toro, Tompsett, Lombardo, Philippot, Nachtergael, Galand, Schlienz, Stammel, Yabar and Blume2007). This is with the exception of two studies in the United States which have examined mental and physical health outcomes for people who previously experienced homelessness using private household survey data (Greenberg and Rosenheck, Reference Greenberg and Rosenheck2010) and electronic health records (Harper Sutherland and Rosenoff, Reference Harper Sutherland and Rosenoff2021). Both studies found marked inequalities, particularly with mental health and substance use conditions, for people who had previously experienced homelessness.
There is a dearth of evidence on health outcomes for people who have previously experienced homelessness in England. We undertook a study using nationally representative data from people living in private households in England to assess the association of prior experiences of homelessness and co-occurring CMDs, physical health problems and alcohol/substance dependence (multimorbidities). Given evidence which has shown mental and physical health inequalities for currently homeless populations (Aldridge et al., Reference Aldridge, Story, Hwang, Nordentoft, Luchenski, Hartwell, Tweed, Lewer, Katikireddi and Hayward2018; Gutwinski et al., Reference Gutwinski, Schreiter, Deutscher and Fazel2021), we hypothesized that we would observe a higher prevalence of these health outcomes for people who formerly experienced homelessness compared to those who never experienced homelessness.
Methods
Participants
The Adult Psychiatric Morbidity Survey (APMS) is a nationally representative stratified and multi-stage probability survey of adults (aged ≥16 years) living in private households in England. Private households are defined as owned and rented accommodation, including both private and social rented accommodation. The present study utilises data from the 2007 and 2014 waves of the survey (N = 14,949), from the first phases of survey data collection, which were conducted face-to-face by lay interviewers. Further details on the survey methods have been described elsewhere (Mcmanus et al., Reference Mcmanus, Bebbington, Jenkins, Morgan, Brown, Collinson and Brugha2020).
Measures
To identify a prior experience of homelessness, the interviewer presented the participant with a card which included eight problems or events (including ‘being homeless’) and asked: ‘Now looking at this card, could you tell me if you have ever experienced any of these problems or events, at any time in your life?’. If a participant indicated that they had experienced ‘being homeless’ from this list, they were asked when they most recently experienced being homeless: within the last 6 months, over 6 months ago and after the age of 16, or over 6 months ago and before the age of 16.
The presence of current CMDs were ascertained using the Clinical Interview Schedule Revised (CIS-R) scale (Lewis et al., Reference Lewis, Pelosi, Araya and Dunn1992). The CIS-R is a structured validated instrument with high sensitivity and specificity to International Classification for Diseases 10 (ICD-10) CMD diagnoses (Lewis et al., Reference Lewis, Pelosi, Araya and Dunn1992). Lay interviewers ask about CMD symptoms over the last month, and then ask more detailed questions about symptoms in the previous week. A CIS-R score of ≥12 indicates presence of a CMD, and a score of ≥18 indicates severe symptoms (more symptoms and/or a longer duration of symptoms) which would likely warrant intervention (Stansfeld et al., Reference Stansfeld, Clark, Bebbington, King, Jenkins and Hinchliffe2016). Participants’ responses to the CIS-R were mapped to diagnostic categories for CMDs (Stansfeld et al., Reference Stansfeld, Clark, Bebbington, King, Jenkins and Hinchliffe2016), including depression (sub-categorised into mild depression, moderate depression, severe depression and combined depression groups), generalised anxiety disorder (GAD), panic disorder, phobias, obsessive-compulsive disorder, and CMD not otherwise specified (CMD-NOS). Participants who scored ≥12 on the CIS-R but did not meet the criteria for any specific CMD were included in the CMD-NOS group (Stansfeld et al., Reference Stansfeld, Clark, Bebbington, King, Jenkins and Hinchliffe2016).
Participants were shown a list of physical health conditions or problems and reported if they had experienced these in the last 12-months. We included physical health conditions listed in both the 2007 and 2014 surveys (a total of 21 conditions, see Table 3).
Participants were asked to self-report their use of alcohol and illicit substances using a Computer-Assisted Self-Completion Interview. This section of the survey included the Alcohol Use Disorders Identification Test (AUDIT) (Saunders et al., Reference Saunders, Aasland, Babor, De La Fuente and Grant1993) to assess harmful alcohol use. A score of 8–15 indicated hazardous drinking, 16–19 indicated harmful drinking and/or mild dependence and ≥20 indicated probable dependence (Room et al., Reference Room, Babor and Rehm2005). Substance dependence was ascertained by five questions designed to assess symptoms of drug dependence according to the Diagnostic Interview Schedule (including cannabis; amphetamines; cocaine; crack; ecstasy; heroin, methadone or physeptone; tranquillisers; glue, solvents, gas or aerosols) (McManus et al., Reference McManus, Bebbington, Jenkins and Brugha2016). These questions asked about the past month and year and covered: daily use for 2 weeks or more, sense of need or dependence, inability to abstain, increased tolerance and withdrawal symptoms. A positive indication on any of these items was considered as an indication of possible dependence (McManus et al., Reference McManus, Bebbington, Jenkins and Brugha2016).
Outcomes
The primary outcome of this study was CMD–physical multimorbidity – defined as the co-occurrence of a CMD (ascertained by a score of ≥12 on the CIS-R) and one or more physical health problems (self-reported). We also ascertained the following binary health outcomes and multimorbidity types (Johnston et al., Reference Johnston, Crilly, Black, Prescott and Mercer2019):
• CMDs (CIS-R ≥ 12).
• Physical health problem (≥1).
• Alcohol dependence (AUDIT ≥ 16).
• Substance dependence (≥1 possible dependence symptoms).
• Physical multimorbidity (≥2 physical health problems).
• CMD–substance multimorbidity (CMD and alcohol dependence and/or substance dependence).
• Trimorbidity (CMD, ≥1 physical health condition or problem and an alcohol problem (AUDIT score ≥16) and/or substance use dependence) (Hewett and Halligan, Reference Hewett and Halligan2010).
Covariates
The following sociodemographic variables were self-reported by participants: age (grouped into 10-year bands, with the exceptions of 16–34 and ≥55 due to small sample sizes), sex, ethnicity, marital status, employment status, educational attainment, and current housing tenure. Due to small sample sizes in the formerly homeless group, we aggregated Black ethnic groups (including Black, African, Caribbean and Black British groups), Asian ethnic groups (including Asian and Asian British) and Mixed/Multiple/other ethnic groups. For the same reason, we defined marital status as follows: those who were married, cohabiting, or in a same-sex couple were placed in one category, while those who were divorced, separated, or widowed were placed in another category. We were also interested in other adverse childhood and life-course experiences self-reported by participants in both 2007 and 2014 surveys, including problematic debt, problems with the police involving court appearances, violence in the home, sexual abuse, bullying, running away from home, spending time in an institution in childhood, and being expelled from school.
Statistical methods
The analysis protocol was pre-registered on Open Science Framework (https://osf.io/sfyrq/). Stata version-17 was used (Statacorp, Reference Statacorp2021), and survey weights for the APMS 2007 and 2014 combined surveys were implemented across all analyses using the svy command. Missing data for included variables were examined. Participants who had data missing for the homelessness, sociodemographic, or health variables were excluded, and a complete case analysis was undertaken. Descriptive characteristics of the combined sample were assessed, including raw frequencies and survey-weighted proportions for experiences of the last episode of homelessness, as well as sociodemographic characteristics, adverse experiences, health conditions/problems and multimorbidity by former experiences of homelessness.
Across all analyses, the exposure was a previous experience of homelessness. Multiple separate unadjusted logistic regression models were used to assess associations between prior experiences of homelessness and the health and multimorbidity outcomes. These models were then adjusted for age, sex, ethnicity, marital status, and survey year. Models were additionally adjusted for employment status and educational attainment, as individual-level indicators of socioeconomic position. A third model included additional adjustment for current smoking status. As adverse experiences were identified as potential confounders (Bramley and Fitzpatrick, Reference Bramley and Fitzpatrick2018; Demakakos et al., Reference Demakakos, Lewer, Jackson and Hayward2020), we additionally adjusted for childhood and life-course adverse experiences across multivariable logistic regression models. Confounders were selected a priori. Across all models, interactions by survey year and sex were assessed. As no interactions were evident, these variables remained in the models as putative confounders. Population attributable fractions (PAFs) for adjusted associations between prior homelessness and health outcomes were estimated using the STATA command punaf from the odds ratios obtained from weighted adjusted logistic regression models. PAFs estimate the proportion of outcome (e.g., CMD) risk in the population attributable to homelessness according to the derived model (Levine, Reference Levine2007).
Public and patient involvement
Nine people who had lived experience of homelessness were consulted throughout the research process for this study. Three people who advised on the study were connected with Rethink Mental Illness, a charity that works with people severely impacted by mental illness; six people were members of the Pathway Lived Experience Group, which is a homelessness healthcare charity. Advisors provided feedback at the study design stage to inform analyses – for example, their feedback on the analysis plan led to us distinguishing between CMD–physical multimorbidity and trimorbidity outcomes. People with lived experience were also consulted on which adverse experiences should be included as relevant confounders. Results were discussed with the two groups, and these discussions contributed towards interpretations of the data which have been highlighted in this paper.
Role of the funding source
The funders were not involved in any aspect of the study design; in the collection, analysis and interpretation of data; in the writing of the report, or in the decision to submit this paper for publication.
Results
Figure 1 shows how the sample was derived. Missing data was low (only 7.3% across the sample, more detail in Appendix 1). Out of the complete 2007 and 2014 combined case sample (n = 13,859), 535 participants reported a previous experience of homelessness. The prevalence estimate for at least one prior experience of homelessness was 3.6% (95% CI 3.3–4.0). Out of those who had experienced homelessness, 444 (82.0%) last experienced homelessness more than 6 months ago aged 16 or older, 76 (14.9%) last experienced homelessness more than 6 months ago before the age of 16 and 15 (3.1%) experienced homelessness in the 6 months before the survey interview.
Figure 1. STROBE flow diagram.
Table 1 shows the sociodemographic characteristics of the sample and adverse experiences, by prior experiences of homelessness. Men and women were evenly split in both the formerly homeless and never homeless groups. In total, 69.2% of people who had never experienced homelessness owned their current property; conversely, 68.3% of the formerly homeless group lived in rented accommodation, of which the majority lived in social housing (47.0%). People who had experienced homelessness had experienced more adverse childhood and life-course experiences, compared to people who had never experienced homelessness, with large differences observed between groups. Over half of the formerly homeless group (52.2%) had experienced bullying, and over a third (40.0%) had experienced violence in their homes.
Table 1. Sociodemographic characteristics and adverse experiences of respondents to the Adult Psychiatric Morbidity Surveys, by experience of homelessnessa
a Data are shown as raw sample n, survey weighted proportions % (with 95% Confidence Intervals). Pearson’s χ2-test p-values < 0.0001, unless indicated otherwise.
b p = 0.338.
c p = 0.011.
d At any time in life, unless indicated otherwise.
Table 2 shows the frequencies and prevalence estimates for health conditions and multimorbidities, by prior experiences of homelessness. The majority of both the formerly homeless and never homeless groups self-reported at least one physical health condition or problem. Higher prevalence estimates were observed for the formerly homeless group for the following physical health conditions: migraines or frequent headaches, asthma, allergies, stomach ulcer/digestive problems, liver problems, back/bone/joint/muscle problems, infectious disease and skin problems.
Table 2. Survey-weighted prevalence estimates for health conditions and problems, by experience of homelessness
a Clinial Interview Schedule-Revised.
b Estimates for dementia or Alzheimer’s disease and stroke not presented due to small cell counts (n < 5 for formerly homeless group).
c Alcohol Use Disorders Identification Test- groups are mutually exclusive.
d Physical multimorbidity: ≥2 physical health conditions/problems.
e CMD–physical multimorbidity: CMD (CIS-R ≥12) & ≥1 physical health problem.
f CMD–substance multimorbidity: CMD (CIS-R ≥12) and alcohol dependence (AUDIT ≥16) and/or substance dependence.
g Trimorbidity: CMD (CIS-R ≥12) and ≥1 physical health problem & alcohol dependence (AUDIT ≥16) and/or substance dependence.
Almost half of the formerly homeless group met criteria for current CMDs (44.8%, 95% CI 40.2–49.5) compared to 15.0% of the never homeless group (95% CI 14.3–15.7). A high proportion of the formerly homeless group experienced severe CMD symptoms (29.7%, 95% CI 25.4–34.3) (Table 2). People who had experienced homelessness were more likely to screen positive for all types of CMDs, compared to those who never experienced homelessness (Appendix 2). Depressive disorders, GAD and CMD-NOS, appeared to account for most of the excess of CMDs. People who formerly experienced homelessness had almost six times the prevalence of current substance dependence compared to people who had never been homeless (Table 3). The formerly homeless group had over four times the prevalence of probable alcohol dependence (Table 3).
Table 3. Stepped adjusted logistic regression models examining associations between former homelessness and multimorbiditiesa
a All p-values <0.0001 except where otherwise indicated.
b Adjusted for age, sex, ethnicity, marital status, survey year.
c Adjusted for age, sex, ethnicity, marital status, survey year, educational attainment, employment status.
d Adjusted for age, sex, ethnicity, marital status, survey year, educational attainment, employment status, smoking status.
e Adjusted for age, sex, ethnicity, marital status, survey year, educational attainment, employment status, smoking status, debt in past year, problems with police involving court appearance, violence in the home, sexual abuse, bullying, living in an institution up until age 16, expelled from school, running away from home.
f Physical multimorbidity: ≥2 physical health conditions/problems.
g CMD–physical multimorbidity: Common Mental Disorder (CIS-R ≥12) and ≥1 physical health condition/problem.
h CMD–alcohol/substance multimorbidity: Common Mental Disorder (CIS-R ≥12) and alcohol dependence (AUDIT ≥16) and/or substance dependence.
i Trimorbidity: Common Mental Disorder (CIS-R ≥12) and ≥1 physical health condition/problem & alcohol dependence (AUDIT ≥16) and/or substance dependence.
j p = 0.024.
The formerly homeless group also had substantially higher prevalence estimates for all types of multimorbidity. The most common type of multimorbidity experienced by both formerly homeless and never homeless groups was physical multimorbidity, which was present in over half of the formerly homeless group (54.0%, 95% CI 49.0–58.9), compared to under half of those who had never experienced homelessness (40.2%, 95% CI 39.2–41.2). The largest absolute difference in prevalence between the formerly homeless and never homeless groups was observed for CMD–physical multimorbidity (formerly homeless = 37.7%, 95% CI 33.2–42.5; never homeless = 12.1%, 95% CI 11.5–12.8). Appendix 3 includes quasi-proportional Euler diagrams which illustrate the mutually exclusive overlaps between CMDs, physical and alcohol/substance dependence domains for the two groups. People who had formerly experienced homelessness rarely experienced alcohol/substance dependence or CMDs in isolation. The prevalence of trimorbidity was almost seven times higher for the formerly homeless group.
Table 3 includes the crude and adjusted odds ratios (aORs) and 95% confidence intervals for the logistic regression models examining the associations between former homelessness and multimorbidities. People who formerly experienced homelessness were more likely to experience all morbidities and multimorbidities. Particularly, people who were formerly homeless were over sevenfold more likely to experience CMD–alcohol/substance multimorbidity (OR 7.89, 95% CI 5.64–11.03) and trimorbidity (OR 7.16, 95% CI 5.00–10.26). Associations between prior experience of homelessness and all health outcomes persisted even after adjustment for demographic characteristics, survey year, indicators of socioeconomic position and smoking. The largest aORs were observed for CMD–alcohol/substance multimorbidity (aOR 3.54, 95% CI 2.49–5.04), CMDs (aOR 3.45, 95% CI 2.81–4.23) and CMD–physical multimorbidity (aOR 3.44, 95% CI 2.77–4.27).
In subsequent analyses, we further adjusted for other adverse experiences (Table 3). We found that these attenuated some of the associations with health outcomes to an extent. This was particularly the case for multimorbidities including substance or alcohol dependence, where associations were no longer evident. However, associations between homelessness and CMDs, physical health problems, physical multimorbidity and CMD–physical multimorbidity remained after adjustment for adverse experiences.
PAFs are presented in Table 4. While 3.6% of the private household population had experienced homelessness, the largest PAFs were observed for CMD–alcohol/substance multimorbidity (17%), trimorbidity (16%) and substance dependence (13%).
Table 4. Adjusted population attributable fractions for logistic regression models adjusted for age, sex, ethnicity, marital status, survey year, educational attainment, employment status, smoking status
a Physical multimorbidity: ≥2 physical health conditions/problems.
b CMD–physical multimorbidity: CMD (CIS-R ≥12) and ≥1 physical health condition/problem.
c CMD–substance/alcohol multimorbidity: CMD (CIS-R ≥12) and alcohol dependence (AUDIT ≥16) and/or substance dependence.
d Trimorbidity: CMD (CIS-R ≥12) and ≥1 physical health condition/problem and alcohol dependence (AUDIT ≥16) and/or substance dependence.
Discussion
In a nationally representative sample of adults living in private households, people with a prior experience of homelessness had a substantially increased likelihood of CMDs, physical health problems, alcohol dependence, substance dependence, multimorbidity and trimorbidity. For people who experienced prior homelessness, the odds of CMD–physical multimorbidity were over fourfold compared with those who had never experienced homelessness. Adjusting for adverse life experiences fully attenuated associations with alcohol and substance dependence morbidities and multimorbidities (including trimorbidity), however associations remained evident for CMDs, physical multimorbidity and CMD–physical multimorbidity. Our estimates suggest that prior experiences of homelessness may have significant deleterious associations with adverse mental and physical health, which persist even after people have been re-housed, and after adjustment for other adverse experiences. Furthermore, despite homelessness affecting 3.6% of people in private households, 17% of mental-substance multimorbidity and 16% of trimorbidity were attributable to prior experiences of homelessness in a nationally representative sample. This indicates the public health significance of preventing homelessness for multimorbidities.
Our study complements and extends the inclusion health literature in several ways. To our knowledge, no other study has assessed the prevalence of CMDs and multimorbidity using nationally representative data from people in private households in England who have formerly experienced homelessness. A systematic review and meta-analysis found a 12.6% prevalence of major depression in currently homeless populations (Gutwinski et al., Reference Gutwinski, Schreiter, Deutscher and Fazel2021). Similarly, we found a 13.2% prevalence of depression for people who had formerly experienced homelessness. Taken together, this evidence indicates that an elevated prevalence of CMDs persist beyond the initial experience of homelessness, even when people have moved into private housing and after adjusting for adverse experiences. We also found that people who previously experienced homelessness were more likely to experience more severe symptoms of CMDs, which indicates a higher likelihood of warranting treatment. Previous research has found that more severe CMD symptoms are associated with a worse quality of life, higher rates of unemployment, more days off sick and higher suicidality (Das-Munshi et al., Reference Das-Munshi, Goldberg, Bebbington, Bhugra, Brugha, Dewey, Jenkins, Stewart and Prince2008). Furthermore, a recent systematic review and meta-analysis has found that housing status is an important determinant of depression treatment prognosis (Buckman et al., Reference Buckman, Saunders, Stott, Cohen, Arundell, Eley, Hollon, Kendrick, Ambler and Watkins2022). The granular detail on CMD severity in the present study provides novel insights into the mental health of people following their experience of homelessness.
The prevalence of self-reported physical health problems reflected patterns found in previous research with people currently experiencing homelessness. For example, we found a higher prevalence of asthma among people who had previously experienced homelessness. This has similarly been found in a previous study with people who were rough sleeping in England (Lewer et al., Reference Lewer, Aldridge, Menezes, Sawyer, Zaninotto, Dedicoat, Ahmed, Luchenski, Hayward and Story2019). Participants who formerly experienced homelessness had a lower prevalence of diabetes compared to never homeless participants, which has also been found in previous studies of rough sleeper populations (Bernstein et al., Reference Bernstein, Meurer, Plumb and Jackson2015; Lewer et al., Reference Lewer, Aldridge, Menezes, Sawyer, Zaninotto, Dedicoat, Ahmed, Luchenski, Hayward and Story2019). It has been suggested that this could be due to under-diagnosis (Bernstein et al., Reference Bernstein, Meurer, Plumb and Jackson2015).
We found a higher prevalence of alcohol and substance dependence in the formerly homeless group, which is consistent with other studies for currently homeless populations (Gutwinski et al., Reference Gutwinski, Schreiter, Deutscher and Fazel2021). We found that associations with alcohol and substance dependence fully attenuated after adjusting for other life-course adverse experiences, indicating a possible relationship between multiple disadvantage and substance use. Our PAF analysis further highlights the potential contribution of prior homelessness to alcohol and substance dependence and multimorbidities.
Women are often under-represented in homelessness research. In the present study, the formerly homeless group was evenly split by sex. This contrasts with previous research on multimorbidity and homelessness in the United Kingdom, which focus on sub-populations of homelessness in specialist services where men make up >75% of samples (Bowen et al., Reference Bowen, Marwick, Marshall, Saunders, Burwood, Yahyouche, Stewart and Paudyal2019; Queen et al., Reference Queen, Lowrie, Richardson and Williamson2017; Rogans-Watson et al., Reference Rogans-Watson, Shulman, Lewer, Armstrong and Hudson2020). Our study therefore extends the evidence-base by including a higher proportion of women who have experienced homelessness. This study addresses an important gap in the literature by assessing health inequalities in multimorbidity for people who have previously experienced homelessness. A key strength of this dataset is that it is nationally representative, and therefore generalisable to people living in private households in England. The APMS does not include the same biases as healthcare service use datasets, as the latter only includes people in contact with health services, which has tended to form the basis of most previous research in this area. By using structured validated instruments, we could identify potentially undiagnosed and untreated CMDs and alcohol/substance dependence (Mcmanus et al., Reference Mcmanus, Bebbington, Jenkins, Morgan, Brown, Collinson and Brugha2020). The self-report module for alcohol and substance use disorders may have enabled participants to report current use without feeling stigmatised, as these questions were asked via a computer rather than face-to-face with an interviewer (Byron et al., Reference Byron, Morgan, Bridges, Papp, Cabrera-Alvarez, Purdon, Tyrer, Smith, Gill, Brugha, McManus, McManus, Bebbington, Jenkins and Brugha2016). Our study design enabled comparisons with people who had never experienced homelessness, while much of the previous work in this area have lacked population controls. We were also able to pool samples over consecutive years of survey, which enhanced study power to be able to detect differences. Our modelling approaches included a range of putative confounders, including socioeconomic position indicators, smoking and adverse experiences, which have not been adjusted for in previous studies. We involved people with lived experience of homelessness in the research process, which led to changes in the analyses including investigating trimorbidity as an outcome and the selection of relevant confounders. The research process was therefore enhanced and enriched by people who the research applies to.
However, our study has several limitations. First, our analyses relied on cross-sectional data, which limits our ability to draw conclusions about causality. Although we looked at previous experiences of homelessness and current health conditions, we cannot make inferences on the temporality between homelessness and health using this data. There is some evidence that mental health and substance use problems often pre-date homelessness (Fitzpatrick et al., Reference Fitzpatrick, Bramley and Johnsen2013), and the relationship between health conditions and homelessness is complex and likely to be bidirectional. While we were able to adjust for multiple potential confounders, there may have also been unmeasured confounders in the present observational study. We did not have data on the duration and type of homelessness experienced by participants. The broad definition of homelessness means that the generalizability of this research to people with specific experiences (e.g., temporary accommodation) is unclear. Furthermore, the time between experiencing homelessness and becoming housed may be highly variable between individuals. It is a limitation that these desirable details on experiences of homelessness were not included in the survey. Furthermore, while education and employment are indicative of social opportunities and income (Galobardes et al., Reference Galobardes, Lynch and Smith2007), these variables are unlikely to provide a complete reflection of socioeconomic status on their own. Other socioeconomic indicators which were not included may account for some of these associations between homelessness and health outcomes, such as financial circumstances including direct measures of income and wealth (Galobardes et al., Reference Galobardes, Lynch and Smith2007; Shavers, Reference Shavers2007).
Information which is self-reported, including homelessness, may be subject to social desirability bias, and as such may have been under-reported in our sample. It is a limitation that homelessness was not defined during data collection. Some people may not have considered themselves to have experienced homelessness and may not have self-identified, despite experiences which would meet the criteria under homelessness frameworks and definitions (e.g., overcrowding) (Edgar, Reference Edgar2012). However, our 3.6% estimated prevalence of homelessness is comparable to other private household survey findings from the United Kingdom, where estimates of former homelessness range between 1.6% and 7.7% (Demakakos et al., Reference Demakakos, Lewer, Jackson and Hayward2020, Bramley and Fitzpatrick, Reference Bramley and Fitzpatrick2018, Toro et al., Reference Toro, Tompsett, Lombardo, Philippot, Nachtergael, Galand, Schlienz, Stammel, Yabar and Blume2007).
The health conditions included in our study are not exhaustive, for example this does not include severe mental health conditions. Furthermore, our measure for physical health was based on self-report. It is possible that in our study this led to an under-reporting of some physical health conditions/problems (e.g., for diabetes and high blood pressure), particularly if not known to the participant or if managed without medications. Despite this, our prevalence estimates for physical health conditions are broadly comparable to other studies which have used clinician diagnoses or other methods to ascertain physical health (Aldridge et al., Reference Aldridge, Story, Hwang, Nordentoft, Luchenski, Hartwell, Tweed, Lewer, Katikireddi and Hayward2018), suggesting that this was not a particularly prominent source of bias in our study. People who had missing data differed from people who had complete data by sociodemographic characteristics and health outcomes (Appendix 1); however, due to the low levels of missingness, this was unlikely to have impacted estimates.
Lastly, this APMS study used data from surveys in 2007 and 2014. There is some evidence to suggest that inequalities in health outcomes for people who experience homelessness may have widened since 2014 (Vickery et al., Reference Vickery, Winkelman, Ford, Busch, Robertshaw, Pittman and Gelberg2021), and conditions of homelessness may also have changed over time. Therefore, the prevalence estimates and associations in the APMS data may be an underestimate compared to present-day estimates.
In conclusion, this research shows that prior experiences of homelessness have associations for adverse mental health, physical health, and alcohol/substance dependence, even after people have been rehoused in private households. This could account for stark reductions in life expectancy known to impact this group (Demakakos et al., Reference Demakakos, Lewer, Jackson and Hayward2020). While policies and clinical practice guidelines are largely focused on currently homeless populations, the present study highlights that there is a need for longer-term support for people after their experience of homelessness. The National Institute for Health and Clinical Excellence (NICE) guidelines for treating people who experience homelessness advises trauma-informed approaches to care (Nice, 2022). Given the findings on the high prevalence of depression, anxiety, and adverse experiences in the formerly homeless group, the present research indicates the potential value of applying these recommendations for trauma-informed care for people post-homelessness. Interventions which address alcohol and substance-related multimorbidity and address multiple adversities could have the biggest impact on population health. The findings on multimorbidity demonstrate the need for the integration, collaboration and co-ordination between multiple health and social care services to support this population. In clinical practice, assessments and recordings of prior homelessness are likely to be beneficial. Lastly, the findings from this study underscore the value of the life-course prevention of homelessness to reduce health inequalities.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S2045796024000659.
Availability of data and materials
The dataset was de-identified prior to researcher access. The 2007 dataset is open access. The 2014 APMS dataset was available under a Special License, which required a Data Sharing Agreement (DSA) with NHS Digital. A Data Access Request for the present study was approved by NHS Digital in 2019 and extended in 2023 (reference: DARS-NIC-320217-X8P0W-V0.3). The data are held securely on the King’s Centre for Military Health Research (KCMHR) server, which is connected to the department where this research took place (Department of Psychological Medicine), complying with a KCMHR NHS Data Security Protection Toolkit.
Acknowledgements
We would like to thank the advisors with lived experience of homelessness, including Ian Kennedy, Dr Sarah Markham (King’s College London) and the Pathway Lived Experience Group Members, for their invaluable feedback throughout the project. We would also like to thank Dr Richard McManus (Rethink Mental Illness) for his support with NC’s PhD research, including the present study. This research used the King’s Centre for Military Health Research systems for secure APMS data storage.
Author contributions
The study was conceived by all authors (NC, PS, SM, AR, AS, JDM). NC, SM, and JDM were responsible for data curation. All authors contributed to the methodology (NC, PS, SM, AR, AS, JDM). Formal analysis was conducted by NC, under supervision from JDM and PS. Funding was acquired by NC, PS and JDM. The paper draft was led by NC, PS, and JDM, and was critically reviewed and edited by all authors (SM, AR, AS).
Financial support
NC was funded by the Economic and Social Research Council under the London Interdisciplinary Social Science Doctoral Training Partnership (ES/P000703/1). SM is supported by the UK Prevention Research Partnership (Violence, Health and Society; MR-VO49879/1). AR is supported by an MQ Mental Health Research Fellowship (MQF22/12). JDM is supported by the ESRC (ES/S012567/1) and by the NIHR Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and the NIHR Applied Research Collaboration South London (NIHR ARC South London).
Competing interests
None.
Open access
