Premature mortality in people affected by co-occurring homelessness, justice involvement, opioid dependence, and psychosis: a retrospective cohort study using linked administrative data
[ad_1]
Summary
Background
Homelessness, opioid dependence, justice involvement, and psychosis are each associated with an increased risk of poor health and commonly co-occur in the same individuals. Most existing studies of mortality associated with this co-occurrence rely on active follow-up methods prone to selection and retention bias, and focus on a limited set of specific exposures rather than taking a population-based approach. To address these limitations, we did a retrospective cohort study using linked administrative data.
Methods
In this retrospective cohort study, we linked a population register of adults resident in Glasgow, UK, to administrative datasets from homelessness and criminal justice services; community pharmacies; and a clinical psychosis registry with data from April 1, 2010 to March 31, 2014. Linkage to death registrations from April 1, 2014 to March 31, 2019 provided follow-up data on premature mortality (age <75 years) from all causes, non-communicable diseases, and causes considered potentially avoidable through health-care or public health intervention. We estimated hazard ratios (HR) using Poisson regression, adjusting for age, gender, socioeconomic deprivation, and calendar time.
Findings
Of 536 653 cohort members, 11 484 (2·1%) died during follow-up. All-cause premature mortality was significantly higher among people with multiple exposures than among people with single exposures, and among people with any exposure than among people with none (eg, homelessness plus other exposures vs no exposures: HR 8·4 [95% CI 7·3–9·5]; homelessness alone vs no exposures: HR 2·2 [1·9–2·5]). Avoidable premature mortality was highest among those with multiple exposures (eg, imprisonment plus other exposures vs no exposures: HR 10·5 [9·1–12·3]; imprisonment alone vs no exposures: HR 3·8 [3·0–4·8]). Premature mortality from non-communicable disease was higher among those with any exposures than among those with none, despite accounting for a lower proportion of deaths in the exposed group; although in some cases there was little difference between estimates for single versus multiple exposures.
Interpretation
The co-occurrence of at least two of homelessness, opioid dependence, justice involvement, or psychosis is associated with very high rates of premature mortality, particularly from avoidable causes of death, including non-communicable disease. Responding to these findings demands wide-ranging efforts across health-care provision, public health, and social policy. Future work should examine the timing and sequencing of exposures to better understand the causal pathways underlying excess mortality.
Funding
Chief Scientist Office, Medical Research Council, NHS Research Scotland.
Introduction
,
,
,
Evidence suggests that these experiences frequently co-occur, although the extent of this overlap varies by context.
,
,
These experiences and their co-occurrence might influence health through multiple and complex pathways. For instance, their harmful effects might combine (or even synergise) or have no additional impact over the baseline risk conferred by each independently. Alternatively, multiple disadvantages might have a paradoxical beneficial effect by conferring additional entitlements or access to services (eg, where based on a threshold of need).
Evidence before this study
We did a systematic review of morbidity and mortality among people with overlapping experiences of homelessness, substance use, imprisonment, severe mental illness, or sex work. We did a further search of MEDLINE, Embase, and PsycINFO on Dec 14, 2021, to identify more recent studies published in English since June 11, 2018 (the limit of the previous searches) and extend the scope to include people in contact with community justice services. We found that there was a scarcity of evidence on the health of people affected by these co-occurring experiences from longitudinal studies that used population-based samples, included multiple exposure combinations, and examined outcomes other than infectious disease, mental illness, and external causes (such as overdose, accidents, and assault). Evidence on mortality from non-communicable diseases and other potentially avoidable causes was especially scarce.
Added value of this study
This study draws on a novel, population-based cohort of more than half a million people, created by linking records between administrative datasets from multiple sectors. The study shows that much of the extremely high rate of premature mortality among people with co-occurring experiences of homelessness, justice involvement, opioid dependence, or psychosis is accounted for by causes that are avoidable through timely access to high-quality health-care and public health interventions, including a substantial burden of non-communicable diseases.
Implications of all the available evidence
Experiences of homelessness, justice involvement, opioid dependence, and psychosis commonly co-occur. In many countries, the population of people affected by these experiences is growing, and growing older. The intersection between these experiences is associated with extremely poor health outcomes, yet there appears to be substantial scope for prevention and mitigation through health-care and public health services. To date, service and policy responses have often been fragmented and uncoordinated, focusing on single issues in isolation and on a narrow range of health conditions. Efforts by health systems and other policy sectors to address these experiences must be reoriented to recognise intersecting forms of disadvantage and to better reflect those conditions causing the greatest burden of ill health.
There was also a scarcity of longitudinal studies from countries outside North America, Scandinavia, and Australia and studies on exposure combinations other than imprisonment and substance use, or severe mental illness and substance use.
,
Descriptive epidemiology can also provide a baseline picture against which efforts to address these forms of adversity—and their health consequences—can be evaluated.
Administrative data typically provide extensive (or even complete) population coverage; are of low cost to obtain; and have high external validity and policy relevance. Record linkage between such datasets across different sectors can be uniquely powerful in helping understand the social and structural determinants of health and identify opportunities for intervention on cross-cutting policy issues.
This method is especially valuable in understanding the experiences and needs of marginalised populations who might be poorly represented in primary research, for instance due to ascertainment difficulties or participation burdens that affect recruitment and retention, but who often have high levels of need for and use of public services.
In this study, we aimed to use cross-sectoral linkage of administrative data to investigate premature mortality among a population cohort containing data on exposure to homelessness, justice involvement, opioid dependence, and psychosis, with a particular focus on mortality from potentially avoidable causes, mortality from non-communicable diseases, and years of potential life lost.
Methods
Study design and participants
Table 1Data sources used in cohort creation to ascertain exposures
NHS=National Health Service.
Permission to access and link the relevant datasets was provided by the following organisations: the Local Privacy Advisory Committee of the West of Scotland Safe Haven (National Health Service [NHS] Greater Glasgow and Clyde population register, prescribing records, Psychosis Clinical Information System register, and death records); the Data Protection Officer and relevant Head of Service of Glasgow City Health and Social Care Partnership (HL1 and Criminal Justice Social Work Report datasets); and the Scottish Government Statistics Public Benefit and Privacy Panel and the Scottish Prison Service Research Access and Ethics Committee (PR2 dataset). Following approval from these organisations, a letter of comfort was issued by the research ethics committee of the University of Glasgow College of Medical, Veterinary, and Life Sciences. To minimise the risk of potential identification of individuals by deductive disclosure, on some occasions categories have been combined or results suppressed.
Procedures
We defined each exposure as the presence of at least one episode in the relevant dataset during this 4-year period: as such, exposure definitions reflect the cumulative experience across this period. For justice involvement, we assigned individuals to one of two exposure categories using the combination of prison and court records: custodial (ie, any imprisonment during the study period, regardless of whether a court report was made) and community (ie, court report without imprisonment). Those without any episodes recorded in the administrative datasets used for exposure ascertainment were classified as unexposed.
To ensure sufficient size in each exposure group, and in light of our interest in premature mortality associated with multiple co-occurring exposures, the primary exposure categories used in mortality analyses classified exposed individuals into those with a given exposure in isolation (eg, homelessness alone) versus that exposure in combination with others (eg, homelessness plus opioid dependence), on the basis of their cumulative history during the exposure period.
Data on deaths among the cohort were obtained from death registrations collected by National Records of Scotland and provided to the West of Scotland Safe Haven. The follow-up period for mortality outcomes was defined as April 1, 2014 to March 31, 2019, with follow-up ceasing (ie, censoring) on the date of the earliest of the following four events: death; migration out of the area covered by NHS Greater Glasgow and Clyde (the area in which migration and mortality data were available); turning 75 years of age; or end of follow-up on March 31, 2019.
To account for the possibility of death on the first day of follow-up, 0·5 days of survival time were added for everyone in the cohort, except for those who did not die, migrate out, or turn age 75 years during the study period and therefore completed the full 1825 days of follow-up.
Outcomes
,
we used the internationally harmonised Organisation for Economic Co-operation and Development (OECD)-Eurostat 2019 definition, which further subdivides avoidable causes into preventable causes (those which can be mainly avoided by effective public health and primary prevention activity, such as deaths from vaccine-preventable diseases or skin cancer) and treatable causes (those which can be mainly avoided through timely access to high-quality health care, such as deaths from appendicitis or asthma).
For non-communicable disease mortality, we used the NCD4 definition employed by the WHO NCD Global Monitoring Framework and UN Sustainable Development Goals, comprising four main types of non-communicable disease: cancer, cardiovascular disease, diabetes, and chronic respiratory disease.
ICD-10 code lists for cause-specific mortality definitions used in the Article are listed in the appendix (pp 5–11). Finally, we defined years of potential life lost as the difference between age at death and an age threshold of 75 years, in keeping with the definition of premature mortality used elsewhere in this analysis, the constraints of our data with regard to historical death records for older age groups, and the definition of this indicator used in OECD health statistics.
Statistical analysis
We calculated crude and age-stratified absolute mortality rates for each of the outcomes of interest. We did Poisson regression to obtain hazard ratios (HRs) for each mortality outcome adjusted for age, gender, socioeconomic position (using quintiles of the area-based Scottish Index of Multiple Deprivation measure), and year of follow-up. Due to the presence of an interaction between exposure and year of follow-up, these primary results should be interpreted as the weighted average of the HRs over the 5-year follow-up. To account for this interaction, we did secondary analyses in which HRs were estimated separately for each exposure category and year of follow-up. We also did a secondary analysis to explore the potential differential survival by exposure combination during the exposure period, which might affect results observed during the outcome period (eg, causing bias towards the null if those at highest risk were to die soon after exposure).
We estimated mean years of potential life lost per 100 000 people by exposure combination, based on all-cause mortality before the age of 75 years.
Role of the funding source
The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report.
Results
Figure 1Linkage process for creation of cohort
CHI=Community Health Index. CJSWR=Criminal Justice Social Work Report. PsyCIS=Psychosis Clinical Information System.
Table 2Data sources used in cohort creation to ascertain exposures
Exposure combinations are ordered by frequency of the overall (any) category. SIMD=Scottish Index of Multiple Deprivation.
There were 2 502 096 person-years of follow-up, with a mean of 4·7 person-years per individual (SD 1·0). 11 484 individuals died during follow-up (2·1% of cohort), with a further 37 302 individuals (7·0%) leaving the cohort due to migration out of the study area and 21 576 (4·0%) due to turning 75 years of age.
Table 3All-cause mortality among the cohort, by exposure status
Exposure combinations are ordered by frequency of the overall (any) category. HR=hazard ratio. SIMD=Scottish Index of Multiple Deprivation.
Figure 2UpSet plot showing frequency of exposure combinations, adjusted HRs with 95% CIs for all-cause premature mortality, and frequency of any exposure
Each column corresponds to a specific exposure combination, indicated by the coloured boxes under the X-axis: for example, the left-most (most frequent) exposure combination is homelessness only, whereas the right-most (least frequent) exposure combination is opioid dependence, community justice, and psychosis. For each exposure combination, the vertical bar shows the number of people affected (left-hand Y-axis) and the circle and line show the adjusted HR and 95% CI for premature mortality (right-hand Y-axis). The exposure combinations shown in this graph are ordered by frequency and are mutually exclusive (ie, all individuals in the cohort with any exposure feature in only one category, with no double-counting). The small horizontal bar plot at the bottom left shows the total size of each set—ie, how many individuals had any exposure to that specific experience. This bar is colour-coded according to the number of individuals who had that exposure only (dark shading) versus in combination with other exposures (light shading). HRs are omitted for exposure combinations in which less than three deaths occurred during follow-up: HL, PSY, and COMM; ODep, PSY, and COMM; HL, PSY, and CUST. COMM=community justice involvement. CUST=custodial justice involvement. HL=homelessness and housing insecurity. HR=hazard ratio. ODep=opioid dependence. PSY=psychosis.
Figure 3Age-stratified mortality rates per 100 000 person-years, by exposure combination and cause
Premature mortality due to avoidable causes, comprising preventable and treatable causes (A), and non-communicable diseases, comprising cancer, cardiovascular disease, chronic respiratory disease, and diabetes (B). Exposure combinations are ordered by frequency of any flag for that exposure. COMM=community justice involvement. CUST=custodial justice involvement. HL=homelessness and housing insecurity. NCD=non-communicable disease. ODep=opioid dependence. PSY=psychosis.
Discussion
Using cross-sectoral administrative data linkage from local authority, health-care, justice, and vital registration systems, we found that people with co-occurring experiences of homelessness, justice involvement, opioid dependence, and psychosis had high rates of premature mortality compared with individuals with one or none of these experiences. However, the impact of multiple disadvantage varied; for instance, people with opioid dependence had high premature mortality regardless of whether they had co-occurring experiences. The rate and proportion of deaths from avoidable causes among people with multiple disadvantages was higher than among people with only one for almost all age groups, which was in turn higher than among those with none; most of these avoidable deaths were accounted for by conditions preventable through public health and primary prevention. Although non-communicable disease accounted for a lower proportion of deaths among people with one or multiple disadvantages than among people with none, absolute rates and relative hazards of mortality from non-communicable disease were higher for any versus no exposures of interest, and for most combinations of multiple versus single exposures.
,
and might facilitate the identification and support of subgroups at particularly high risk of poor outcomes. However, it is notable that mortality was substantially increased even among those only exposed to one form of disadvantage. For instance, individuals with homelessness alone accounted for almost 2% of the Glasgow population yet had a 2·4-times greater hazard of premature death than their unexposed peers. Similarly, among people with a history of imprisonment, HRs for premature mortality were 3·4-times greater than those for the unaffected population, even in the absence of other well established risk factors such as opioid dependence or homelessness.
,
,
These findings suggest the need for wide-ranging policy and service efforts across the population to prevent these experiences and mitigate associated poor health outcomes.
Our results extend previous research showing that each of these experiences in isolation is associated with a substantial increase in the risk of death from avoidable causes.
,
,
,
Together, these findings suggest that current public health and health-care provision is failing to benefit many of those with the experiences of interest, creating unjust inequalities in risk of death.
,
Other studies have found that access to prevention and treatment for common physical health conditions among people experiencing social marginalisation and exclusion is often poor.
,
,
Current priorities for service delivery and research activity with these populations tend to be dominated by the prevention and management of infections and external causes: our findings suggest that this does not adequately reflect their true burden of ill health, to which non-communicable diseases make a substantial and probably increasing contribution, and that greater attention must be paid to the prevention and treatment of common long-term conditions.
Strengths of this study include its population-based approach, which enables us to assess the associations between diverse exposure combinations and mortality in comparison with an unexposed population, rather than assessing risk factors for mortality among people selected on the basis of an index exposure. Although the exposures under investigation in this study do not represent an exhaustive set of identities or experiences associated with social exclusion, their occurrence and outcomes are heavily influenced by policy choices (eg, relating to the housing market, poverty and social security, or justice and sentencing policy), and they are tractable to study through existing datasets collected routinely in Scotland, making them an ideal focus for social epidemiology.
The use of linked administrative and registry data maximises coverage and ascertainment, and reduces the risk of threats to validity from participation and attrition biases, which are common in traditional cohort studies with people experiencing social disadvantage and exclusion.
Needle Exchange Surveillance Initiative (NESI): prevalence of blood-borne viruses and injecting risk behaviours among people who inject drugs (PWID) attending injecting equipment provision (IEP) services in Scotland, 2008–09 to 2017–18.
whereas data from the Scottish Household Survey around the same time found that 60–65% of people reporting a history of homelessness had approached their local authority for help during the most recent episode (although this survey is potentially biased by being restricted to those individuals now living in private households). Ascertainment might be strengthened in future work through triangulation with additional administrative datasets (eg, from third sector as well as statutory services) or data from primary research.
However, a further consideration is that the extent to which records from non-health sources (including the PR2 dataset) could be assigned a CHI number—and therefore included in the linked cohort—varied between datasets. Failure to identify a CHI number might be explained by migration out of the NHS Greater Glasgow and Clyde area following exposure (as the CHI register is a live database, in contrast to the retrospective exposure datasets) or incorrect identifiers in one or multiple datasets, but also reflects the use of a fairly stringent matching algorithm likely to prioritise specificity over sensitivity. The CHI register is updated automatically when individuals register with health services in other areas and is subject to regular checks on residence, although there might still be some degree of under-ascertainment of migration (eg, due to individual delays in re-registration). Therefore, exclusion of records for which a CHI could not be assigned is an important limitation that could be addressed in future work using national (rather than regional) population registers, prospective rather than retrospective linkages, and threshold-based approaches to probabilistic linkage permitting sensitivity analyses. Developments in data access, to enable use of national-level data for exposures and outcomes, would also enhance the generalisability of our findings.
moreover, many instances of these experiences are themselves avoidable through wider social policy measures in the realms of welfare, employment, housing, and justice.
,
However, previous modelling suggests the overall impact of this is likely to be small
and, in our study, the median in-prison time during the exposure period among those imprisoned was only 9% (equivalent to 130 of 1460 days [IQR 44–334]; Tweed, unpublished). We did not have data on heterogeneity within our exposure categories—for instance, polysubstance use or treatment access among people with opioid dependence—which might have affected mortality risk.
Nonetheless, our findings show the value of administrative data linkage in understanding the health experiences of people for whom participation in primary research can be challenging. Future work should extend these methods to better understand the causal pathways underlying excess mortality, for instance by examining the timing and sequencing of exposures; investigating effect modification by factors such as gender and socioeconomic position; and evaluating natural experiments that affect exposure to the experiences of interest.
At present, such cross-sectoral linkage is often resource-intensive and time-consuming, hindering efforts to monitor trends and evaluate interventions at the population level, and to support joined-up care provision and multi-agency working at the individual level. Therefore, investment in routine and responsive linkage across multiple sectors might facilitate service and policy responses that are not only better informed by evidence but also more holistic in their approach.
The study was conceptualised and planned by EJT with input from other authors. EJT liaised with relevant agencies to gain access to the data and did all analyses, with input from other authors. EJT drafted the manuscript. EJT and AHL accessed and verified the data. All authors contributed to interpreting the results of analyses, critically revised the manuscript, had full access to the data in the study, and had final responsibility for the decision to submit for publication.
[ad_2]
Source link
