Blog

A third of all homeless deaths are due to preventable health conditions. In this blog, author, Robert Aldridge, (RA) University College London, discusses his research article published on Wellcome Open Research with reviewer, Glen Bramley, (GB) Heriot-Watt University, covering the unmet healthcare needs of the homeless and why they used a linked data approach for their investigation.

Rob Aldridge is a Wellcome Trust Clinical Research Career Development Fellow at the Institute of Health Informatics. He qualified in medicine from University College London in 2007, gained an MSc in Epidemiology at The London School of Hygiene and Tropical Medicine in 2010 and completed his PhD at University College London in 2015. Prior to studying medicine, he studied Mechanical Engineering at The University of Nottingham and worked in management consultancy. 

Glen Bramley is one of Britain’s leading academics in housing and urban economics having been engaged in academic and applied policy research in this field since the mid-1970s. Glen has been Professor of Urban Studies at Heriot-Watt University since August 2002 and was Professor of Planning & Housing, Heriot-Watt University/Edinburgh College of Art from 1994-2002. 

Estimating the risk of death

RA: In 2012 there was a study conducted by a group of charities that found 70% of people who were experiencing homelessness and admitted to hospital were subsequently discharged to the street without having their health and care needs assessed.

A year later, the Department of Health launched the ‘Homeless Hospital Discharge Fund’, allocating £10 million to 52 projects in the voluntary or not-for-profit sector to test different models of specialist integrated homeless health and care (SIHHC).

Our study aimed to look at the health outcomes of people experiencing homelessness that had been seen by one of these SIHHC schemes, and the focus of this particular research article was to describe their underlying causes of death.

A linked data approach  

RA: To carry out the study we worked with 17 SIHHC sites who provided us with the names, dates of birth, sex and NHS number (a unique ten-digit numeric identifier for patients in the healthcare system assigned at first encounter) of all patients they had seen between 1st November 2013 and 30th November 2016.

We used this information to link to subsequent hospital and death records for all of these people. Linking records in this way allowed us to follow up large numbers of people across the whole of England regardless of which SIHHC site they were originally seen at. This is a particularly important part of the study design as this group is very mobile and difficult to follow up using traditional research studies.

However, to be able to use this information, which was given to us without the explicit consent of each individual, we had to apply for special permission to use it for our research. Before undertaking the study, we worked with people who had experience of homelessness and found that using data without consent was acceptable to them so long as adequate data security measures were used and studies are ethically and legally approved.

We had to get several specific approvals, including from the Secretary of State for Health through the Confidentiality Advisory Group (CAG); Health Research Authority Research Ethics Committee approval; and local Research and Development approvals were set up prior to local data collection at each of the SIHHC sites. Because of the strict permissions placed upon us in collecting and analysing this dataset we are unfortunately unable to share it for reuse, however, we would happily share all of our analytical code which we used to create, clean and analyse the dataset.

Linking records

GB: The ability to link data between different administrative systems does in general offer a wide range of opportunities to research questions of pressing societal, policy and academic interest which cannot easily be addressed by looking at individual administrative systems in isolation, or by conventional sample surveys of households or populations.

The ability to link data between different administrative systems does in general offer a wide range of opportunities to research questions of pressing societal, policy and academic interest which cannot easily be addressed by looking at individual administrative systems in isolation

Glen Bramley, Heriot-watt University

In the case of key outcomes, the health outcome of mortality (by certain avoidable causes), the mortality record itself has limited information although proximate cause of death will be recorded. However, other health records which could be linked to it, for example hospital episodes or prescriptions data, can provide a lot more evidence/indicators of relevant conditions and treatments experienced prior to the mortality event. Conversely, linking mortality to health episodes provides more evidence on longer term outcomes from those episodes.

The ability to link to records in other policy/service domains, while perhaps hedged with more obstacles in terms of governance and data protection, opens up a much more powerful tool to explore the background factors associated with, and potentially influencing or determining, the health outcome of interest. This could include for example information about income, unemployment, disabilities (from DWP systems) which are all recognised as potentially significant elements in the socio-economic influences on health outcomes and inequalities.

The ability to link records at the individual level is potentially more powerful than the reliance (traditional over the last 30 years) to use the socio-economic profile of small neighbourhoods as a proxy for individual experiences or characteristics.

A strong role to play

GB: Clearly, when modelling and looking to try to detect significant effects or associations, the ability to look at a whole population (within the parameters of the study) has much more power than a sample survey. The administrative records may also be closer to a truer record than imperfectly remembered and biased recall in retrospective survey questions (particularly around more sensitive or stigmatized circumstances, such as homelessness or addictions).

This approach may have a particularly strong role to play in the field of Inclusion Health, or what is sometimes referred to as studying ‘Complex needs’ or ‘Severe and Multiple Disadvantage’ (SMD). Homelessness is often involved in combination with other needs – substance dependency, offending, mental ill-health or being a victim of violence or abuse – which to varying degrees are characterised by a degree of stigma.

A very important finding was that nearly one in three of the deaths in this group were caused by treatable conditions such as tuberculosis and gastric ulcers, if the people that have them are given timely and appropriate healthcare.

Robert Aldridge, University College London

They are also often characterised by problems of not being adequately represented in conventional social survey research instruments, such as household surveys, partly because people affected are very often non-responders and partly because a significant proportion of them are not staying in conventional household settings (especially those who are homeless). This approach can be useful in profiling groups with combinations of such disadvantages, especially when and where there are admin data systems which record information about people’s experience/involvement in these and other domains.

This study is an example of a health-based study which is trying to flag homelessness in the analysis, through the fortunate ‘natural experiment’ of having a number of special hospital units which focus on the homeless group. Other recent studies have sought to profile the homeless through overlaying address information about homeless hostels and other accommodation on the health records, as well as using administrative flags for ‘No Fixed Abode’ where these are recorded on the health data (which are said to be inconsistently recorded at best).

None of these are very satisfactory, partly due to the disparate nature of homelessness. What is really needed here is a better admin system for recording people’s episodes of homelessness, which can then be linked to health and other records. Such a system is just being introduced in England (H-CLIC); a similar system (HL1) has existed in Scotland for nearly two decades, and was recently subject to a major data linkage to health records study led by Andrew Waugh.

A third of all homeless deaths

RA: We collected data on 3,882 individual homeless hospital admissions that were linked to 600 deaths. The median age of death was 51.6 years and after adjusting for age and sex differences in this group compared to the general population, the top three underlying causes of death were heart disease, cancer and lung disease.

A very important finding was that nearly one in three of the deaths in this group were caused by treatable conditions such as tuberculosis and gastric ulcers if the people that have them are given timely and appropriate healthcare.

Unmet healthcare needs

RA: In 2019, the United Kingdom’s Office for National Statistics (ONS) found that the most common underlying causes of death in people experiencing homelessness were accidents (including drug poisoning), suicides and diseases of the liver. However, a large number of the deaths in this analysis were identified through coroners reports, which are often undertaken when there is an unexpected or accidental death.

It is therefore not surprising that the ONS methodology so many of these types of death. Conversely, because of the way we collected our data, our results are likely to include higher numbers of people dying from chronic diseases after being admitted to hospital for treatment.

Overall, it’s very hard to imagine the “perfect” study for this excluded population, and we think that the two studies both provide helpful evidence that should be considered in the context of where the data came from.

Tailoring health services

RA: Our results show that existing specialist NHS hospital discharge schemes and community homeless health services are engaging with a population that has extremely high levels of unmet healthcare needs.

We believe our study illustrates why we urgently need to identify individuals experiencing homelessness before the onset of these diseases, and develop models of care that enable them to access and engage with healthcare.

Robert Aldridge, University College London

We believe our study illustrates why we urgently need to identify individuals experiencing homelessness before the onset of these diseases, and develop models of care that enable them to access and engage with healthcare. We believe there is a need for specialist residential intermediate care facilities that are designed to encourage safe, timely transfers out of hospital and, support self-management, recovery and recuperation.

Equally important will be the creation of tailored approaches that take account of the lived reality of people experiencing homelessness in order to find ways to support adherence to essential medicines such as those that can lower blood pressure or cholesterol as these are plausibly linked to the high number of deaths from heart disease that we found.

GB: I suspect that there is a lot to learn from the wider study of which this is a part, as it is essentially a comprehensive assessment of a large-scale initiative to tailor acute health services towards the particular needs of homeless people.

As I argued in my review, we do need to recognise the wide spectrum of homelessness and the differences between definitions and concepts involved, and certainly not assume that homelessness simply means sleeping rough or in shelters.

Altering perceptions

GB: The key relevant conclusion of the Scottish study referred to is that although ‘external events’ (drugs, alcohol, suicide) were disproportionately high for the homeless group, this group also suffered a significantly enhanced risk of death from heart disease, respiratory illnesses of cancer.

it is essentially a comprehensive assessment of a large-scale initiative to tailor acute health services towards the particular needs of homeless people.

Glen Bramley, Heriot-watt university

This is consistent with the findings from Aldridge et al.’s study, where it is claimed that a third of homeless deaths were from conditions amenable to health care. So, the notion that this group are mainly killing themselves through reckless lifestyle choices is modified; they are also suffering the diseases associated with poverty, but to a significantly greater extent than, for example, the people living in IMD 5 who are just poor. Yet despite this higher need, they are not receiving a commensurate response from the health service.

Clearly the balance between these types of case does seem to vary from study to study, which likely reflects the particular subgroups within the broader ‘homeless’ population who are selected by the different research approaches.

I would say, from looking at the recent Scottish study and other evidence, that both things are true. The relationship between substance use and homelessness is very strong for a substantial chunk of the homeless population, particularly measured in terms of presentations; such use is very rare indeed in populations drawn from relatively affluent areas.

But it is also true that the broader range of diseases leading to mortality, do affect homeless people more, and that should not be a surprise, given the strong relationship of homelessness with current and past poverty, and the conditions under which many homeless people are living from day to day.

Constructive reviews

RA: I think the open peer review model is excellent. We believed there was a strong case for our study results to be published rapidly and the open peer review model facilitated this enormously by allowing the initial results of our study to be made public during the peer review process.

hopefully creating a more constructive and less combative type approach to peer review – see the reviews of our article for a very nice illustration of this constructive approach!

Robert Aldridge, University College London

It seems to me that the open peer review model probably creates a different mindset in the reviewers as their names are attached with the feedback they write, hopefully creating a more constructive and less combative type approach to peer review – see the reviews of our article for a very nice illustration of this constructive approach!

One concern I have is whether open peer review will reduce the number of early career researchers willing to undertake reviews, particularly if they are looking at the papers of senior colleagues as they could be worried about what might happen if they provide feedback that these senior academics do not agree with.

GB: It seems to be an interesting process. I think it is more difficult to get away with making snide and dismissive comments about other’s research without having some grounds for making those criticisms and being able to show that you have something to bring to the debate. But I suppose it also highlights issues which may identify limitations of papers which general readers might not have spotted.