Household Tenure and Its Associations with Multiple Long-Term Conditions amongst Working-Age Adults in East London: A Cross-Sectional Analysis Using Linked Primary Care and Local Government Records

Multiple long-term conditions (MLTCs) are influenced in extent and nature by social determinants of health. Few studies have explored associations between household tenure and different definitions of MLTCs. This study aimed to examine associations between household tenure and MLTCs amongst working-age adults (16 to 64 years old, inclusive). This cross-sectional study used the 2019–2020 wave of an innovative dataset that links administrative data across health and local government for residents of a deprived borough in East London. Three definitions of MLTCs were operationalised based on a list of 38 conditions. Multilevel logistic regression models were built for each outcome and adjusted for a range of health and sociodemographic factors. Compared to working-age owner-occupiers, odds of basic MLTCs were 36% higher for social housing tenants and 19% lower for private renters (OR 1.36; 95% CI 1.30–1.42; p < 0.001 and OR 0.81, 95% CI 0.77–0.84, p < 0.001, respectively). Results were consistent across different definitions of MLTCs, although associations were stronger for social housing tenants with physical-mental MLTCs. This study finds strong evidence that household tenure is associated with MLTCs, emphasising the importance of understanding household-level determinants of health. Resources to prevent and tackle MLTCs among working-age adults could be differentially targeted by tenure type.


Introduction
The co-occurrence of multiple long-term conditions (MLTCs) within a single individual is a major public health challenge both globally and in the UK. The nature and extent of MLTCs is influenced by social determinants of health (SDoH) [1]. The role of individualand area-level social determinants has been widely reported-prevalence and incidence of MLTCs are greater with increasing age, for women, for ethnic minorities, and those living with greater socioeconomic deprivation [1][2][3][4][5][6]. Yet recent evidence suggests that household-level SDoH (such as household tenure) are often overlooked as determinants of MLTCs despite comparatively large effect sizes for household compared to area-level SDoH [7]. In their landmark report, the Academy of Medical Sciences (AMS) concluded that most evidence focuses on "population or individual-level" determinants and that "it will be valuable to consider whether factors that operate at the household-level can also influence MLTCs" [1]. In addition, exploring these relationships amongst working-age 1.
Basic MLTCs, the co-occurrence of two or more long-term conditions within a single individual; 2.
Physical-mental MLTCs, the co-occurrence of two or more long-term conditions within a single individual, one of which must be depression or anxiety and one of which must be a physical condition; 3.
Complex MLTCs, the co-occurrence of three or more long-term conditions affecting three or more different bodily systems within a single individual [24].
The third definition was operationalised as conditions originating from different bodily systems are thought to be harder to treat due to different origins and/or treatment plans [1,24]. See Table A2 for the 38 conditions, how these conditions were grouped by bodily system and their distribution across the study cohort. Binary variables were created to indicate the presence or absence of each MLTCs outcome for each participant.

Main Exposure: Household Tenure
Individuals were defined as "owner-occupiers" if living in an owner-occupied household (outright or with a mortgage), "private renters" if living in a privately rented property, or "social housing tenants" if living in a socially rented household (from local government or a housing association). A fourth "unknown" category was created to account for missing data. Data on tenure were extracted from the council's housing data systems.

Covariates
Data on age and sex were extracted from primary care records. Eight categories were created to code individuals' ages in years (<16, 16-29, 30-44, 45-54, 55-64, 65-74, 75-84, and 85+). Sex was coded as male or female. Data on ethnicity were extracted from council records and coded into five categories: "White", "Black", "Asian", "Other" and "Unknown". Data on BMI and smoking status were extracted from primary care records. BMI was coded into five categories defined by the NHS as follows: underweight (below 18.5), healthy (between 18.5 and 24.9), overweight (between 25 and 29.9), obese (between 30 and 39.9) and morbidly obese (over 40), with a sixth "unknown" category to account for missing data. Smoking status data were coded into four categories: non-smoker, smoker, ex-smoker, or "unknown".
Data on household welfare benefits, occupancy and household type were extracted from council housing records. Households receiving welfare benefits to support rental payments ('housing benefit') were classified by whether eligibility was based on receipt of other welfare benefits and, if so, the type: Employment Support and Allowance (ESA), Pension Credit, Income Support or Job Seeker's Allowance (JSA). Two further categories reflecting households solely in receipt of housing benefit or in receipt of no benefits were created. Occupancy data were recorded into four categories to reflect 1-2, 3-5, 6-10 and 11 or more people within a household. Data on household type captured households as six types: adults with children, adults with no children, single adult with children, single adult, older adults with no children, and three generations.
To provide a marker of overall deprivation in each participants' residential area relative to other areas in the borough, borough-specific Index of Multiple Deprivation (IMD) quintiles were calculated for each small geographical area (Lower Super Output Area; LSOA) using 2019 IMD scores [25]. Each LSOA comprised a maximum of 3000 residents and 1200 households [26].

Main Data Analysis
Multilevel logistic regression modelling was used to explore associations between household tenure and MLTCs prevalence amongst working-age residents with complete data (see Table 1 and Figure A1). To assess the relative impact of adjusting for individual compared to household-level covariates on the association between tenure and MLTCs prevalence, we built three distinct models for each outcome. First, an unadjusted model with no covariates included. Second, a model adjusted for individual-level sociodemographic characteristics available in the dataset and found to be associated with both MLTCs prevalence and household tenure in previous literature [17,[27][28][29]. These covariates were age, sex, ethnicity, BMI and smoking. The third and final model for each outcome additionally adjusted for household benefits receipt, occupancy and type to control for potential household-level factors correlated with both household tenure and MLTCs (see covariates above). We chose to adjust for household benefits receipt as it was the best proxy measure available in the dataset for other important covariates such as employment. We chose to adjust for household occupancy and type as a previous systematic review examining household-and area-level social determinants of MLTCs found these factors were associated with MLTCs prevalence in some contexts [7]. Model fit was assessed using Akaike's Information Criteria (AIC). We considered multilevel models to account for the potential clustering of individuals within geographical areas, as individuals are likely to be more similar in terms of individual, household-and area-level factors if residing in the same

Subgroup and Sensitivity Analyses
Three interaction terms were separately added to the final model for each outcome to evaluate potential interactions between household tenure and other household factors. We assessed interactions with receipt of benefits, household occupancy and type (see covariates above) as these are most likely to modify the association between housing tenure and MLTCs, and they also act at the household-level. Any differences in these household-level characteristics by tenure type can be found in Table A3.

Participant Characteristics
Of the 232,671 participants whose primary care and local government records were successfully linked, 132,296 participants were eligible for inclusion in this study. A total of 78,379 records (33.7%) were excluded as individuals were not of working age, 21,847 records (9.39%) were excluded due to unconfirmed resident status and 95 were excluded due to living in a residential home (0.04%) (see Figure A1).

Subgroup Analyses
Our subgroup analyses suggest subgroup effects according to household benefits receipt, occupancy and household type (see Tables A4-A6). The odds of MLTCs for private renters (compared to owner-occupiers) were considerably stronger for households in receipt of benefits compared to those not receiving benefits. For example, odds of basic MLTCs were 76% greater for privately rented households where someone was in receipt of ESA compared to households not receiving ESA (OR 1.76, 95% CI 1.35-2.29). There was no evidence of an interaction between living in social housing and household benefits receipt (see Table A4). The odds of MLTCs for both social housing tenants and private renters (compared to owner-occupiers) were higher for single-adult households compared to households with adults and children. For example, the odds of basic MLTCs for social housing tenants compared to owner-occupiers were 31% greater for single-adult households (OR 1.31, 95% CI 1.15-1.50). Evidence for subgroup effects for other household types were weaker, with most interactions not statistically significant (see Table A6).

Summary of Study Findings
Risk of MLTCs amongst working-age residents of a deprived East London borough was greater for social housing tenants and lower for privately renters, when compared to owner-occupiers. These associations remained significant after adjusting for a range of individual-and household-level characteristics and were consistent across different definitions of MLTCs. Other household-level variables-household benefits receipt, occupancy, and type-were important modifying factors, with associations between tenure and MLTCs greater for individuals in single-adult households and households in receipt of certain benefits.

Comparisons with Existing Literature
Our prevalence estimates are in keeping with previous estimates for this age group [6,9,32,33]. Prevalence of MLTCs was greater with increasing age and for females, consistent with previous literature [1,6]. However, prevalence was lower for ethnic minority compared with White participants, which contradicts many studies and may be an age-related effect [1,27,32]. In this study, participants lived in a deprived borough in East London where older and younger individuals tend to be White and ethnic minorities, respectively.
We found that social housing tenants exhibited greater risk of MLTCs compared to owner-occupiers, aligning with findings from Northern Ireland yet contradicting those from a Hong Kong-based study [34,35]. This supports the idea that associations between household-level SDoH and MLTCs may be context specific, influenced by housing policy, supply and conditions of social housing, stigma and other household circumstances such as benefits receipt (see Table A3) [7]. In the UK specifically, social housing tenants may be exposed to various "hard" (material) and "soft" (psychological) factors that interact to cause or exacerbate MLTCs [14]. Evidence suggests social housing tenants in the UK have higher levels of C-reactive protein, a biomarker of inflammation associated with various long-term conditions [17,36]. In addition, social housing tenants have less control over the condition of their property and their built environment, and are less able to leave their property, whilst owner-occupying affords ontological security-the sense of security and control afforded when owning your home [37,38]. On top of this, the UK Housing Act (1998) requires social housing to be allocated based on certain criteria, one of which is ill health. As such, MLTCs may be a qualifying characteristic for eligibility for social housing, which may explain our estimated associations.
The lower risk of MLTCs found for private renters compared to owner-occupiers contradicts previous research from the US and Northern Ireland [32,35]. Our analyses adjusted for variables not adjusted for in these studies-household benefits receipt, occupancy, and household type. Our findings suggest these were important explanatory factors for the association between tenure and MLTCs, but they did not explain all of the additional risk experienced by social housing tenants, nor the decreased risk for private renters. In the UK, the private rental market is expanding considerably, and private renters are an increasingly heterogenous group in terms of their demographic, social and economic circumstances [11]. As such, more longitudinal, causal analyses are needed to unpick the complex relationships between different tenure types and MLTCs, taking into account the influence of other household characteristics.
We found that the association between tenure and MLTCs was greater for individuals in single-adult households and households with one or two occupants when compared to higher numbers of occupants. However, previous research examining associations between living alone and MLTC prevalence presents mixed results [7]. In our context, a deprived borough of East London, single-adult households may have less social support and be more financially uncertain than households with multiple occupants, increasing their vulnerability to any adverse effects imposed by their tenure [39]. We also found that the association between tenure and MLTCs was greater for individuals in households where someone was in receipt of certain benefits. Only one previous study has explored subgroup effects in the relationship between tenure and MLTCs and they similarly found that household financial burden mediated this relationship, albeit with a small effect [34]. Our findings support this work, and, again, suggest further research should capture data on, and account for, other household-level characteristics when examining relationships between tenure and MLTCs.
Differences in the risk of MLTCs with tenure type were not explained by commonly used area-level deprivation measures as most areas in our study are amongst the most deprived nationally [7]. These findings further demonstrate the importance of capturing data on, and understanding, household-level SDoH as this information could support service planning when area-level deprivation measures are unable to capture enough variation to model socioeconomic inequalities in MLTCs. In addition, our findings were consistent across different definitions of MLTCs, illustrating the importance of household tenure as a risk factor for MLTCs.

Strengths and Limitations
This is the first study to explore associations between household tenure and MLTCs in England. Our findings add to the current literature, and our analyses would not have been possible without the innovative linkage of primary care and local government data. We operationalised three definitions of MLTCs that captured different types of MLTCs with different degrees of complexity. We used publicly available code lists to determine the presence of each condition.
Our study was conducted in one deprived borough in East London and, whilst our findings could be generalisable to other urban areas, they may not hold in contexts that are less deprived, more rural and have different tenure profiles [7,40]. We restricted our analyses to complete cases, which assumes that any differences between individuals with missing and complete data are explained by differences in observed individual and household characteristics included in the regression models. We recognise that there may be other variables associated with the missing data that we have not adjusted for. However, this is unlikely to have significantly changed the results due to the limited role that BMI and smoking status have in the association between tenure and MLTCs prevalence [41]. We did not account for disease severity or symptom burden on the patient, or other dimensions of MLTCs such as frailty. We may have misclassified households where owner-occupiers privately rented rooms, which may have biased estimates towards the null if private renters who co-resided with their owner-occupying landlords differed systematically in their health compared to private renters who did not. In addition, our measure of household benefits receipt did not capture eligibility for benefits, and we could not adjust for other important factors such as education. The cross-sectional study design did not allow us to explore temporal relationships between tenure and MLTCs. We adjusted for household benefits receipt, occupancy, and household type as potential confounders, but also demonstrated important subgroup effects according to some of these characteristics. It is possible these variables may modify the relationship between tenure and MLTCs. More longitudinal analyses are needed to determine how these factors interact over time to impact MLTCs.

Implications for Practice and Policy
Most interventions for MLTCs focus on retired, older adults, yet our findings indicate that working-age adults are an important population to consider when aiming to address MLTCs. There is currently a gap in models of care or interventions aimed at working-age adults, for whom there may be greater opportunity for prevention of MLTCs through addressing SDoH than amongst older adults [1]. Initiatives that target preventative resources at working-age adults with MLTCs who live in social housing could slow the progression of MLTCs and improve health outcomes, ultimately saving future costs [8].

Conclusions
This study finds strong evidence that risk of MLTCs amongst working-age residents of a deprived East London borough was greater for social housing tenants and lower for privately renters when compared to owner-occupiers. Associations were consistent across different definitions of MLTCs, which emphasises the importance of understanding and addressing household-level determinants of health. Our findings suggest that resources to prevent and tackle MLTCs could be differentially targeted by tenure type and that workingage adults are an important population to consider in preventative strategies. Further research should employ longitudinal research methods to assess temporal relationships between household social determinants and MLTCs.

Institutional Review Board Statement:
This study was conducted in accordance with the Declaration of Helsinki. The study protocol was approved on 13th March 2020 by Care City's formal process for data access (no project identification code provided).

Informed Consent Statement:
Patient consent was waived as this work uses data provided by patients and collected by the NHS as part of their care and support. Only anonymised data were released.

Data Availability Statement:
Restrictions apply to the availability of these data. Data were obtained from Care City and no applicable data are available without their permission. The study protocol is available on request.

Acknowledgments:
The authors would like to thank Jenny Shand, Simon Lam and Phil Canham for their support with data access and their help with understanding the origins of the data. We would also like to thank Melvyn Jones, the Care City Community Board and the NIHR ARC North Thames Research Advisory Panel for their advice and expertise when developing our definitions of multiple long-term conditions.

Conflicts of Interest:
The authors declare conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

Appendix A. Overview of the Care City Cohort and Data Linkage Steps
In 2017, the leaders of Barking and Dagenham Council, North East London NHS Foundation Trust (NELFT) and Barking and Dagenham, Havering and Redbridge Clinical Commissioning Group (BHR CCG), and their Caldicott guardians (a senior person within each organisation who is responsible for protecting the confidentiality of people's health and care information and making sure it is used properly), signed data sharing agreements to create a dataset that linked administrative data for the population of Barking and Dagenham (B&D) between 1st April 2011 and 31st March 2017. Since its creation, the dataset has been updated on an annual basis. It is hosted in the Barking and Dagenham, Havering, and Redbridge NHS Accredited Data Safe Haven, with governance and oversight provided by the Barking and Dagenham, Havering, and Redbridge Information Governance Steering Committee.
The dataset was created as part a larger research programme of work [18]. It contains routinely collected administrative health and social data across local government services, health providers, and health commissioners. Data are linked at the individual and household levels using linkage keys (replacing NHS numbers and Unique Property Reference Numbers; UPRNs). The data are pseudonymised and include information on sociodemographic characteristics, health variables, household variables and data on health and social care service utilisation. Data on all sociodemographic and health variables for each cross-section are taken as a snapshot on 1st April 2019 to account for in-year changes in variables. The dataset is not currently publicly available but was made available to the wider research community in Autumn 2020.
More information on the dataset can be found here [42] and here [43]. More information on the codes and algorithms used to classify variables as part of the creation of the Care City Cohort can be found at this reference [18].
This study used data from the 2019/20 cross-section of the Care City Cohort. We requested access to pseudonymised sociodemographic and health variables extracted from primary care data, and resident data extracted from local government data. We did not have access to other data available within the Care City Cohort, such as data on health and care service utilisation.
Data were provided unlinked with linkage keys, i.e., with the identification codes generated to replace NHS numbers and UPRNs. We used these to link the data at the individual and household levels. First, we linked the individual-and household-level local government data on Household_ID (the household-level identification code created by Care City to replace UPRNs). Second, we linked the individual-level primary care data to the linked local government data on Patient_ID (the individual-level identification code created by Care City to replace NHS numbers). Third, we linked a fourth dataset provided by Care City that detailed care homes in Barking and Dagenham and their Household_IDs. We linked this to the cohort data on Household_ID. Finally, we linked a fifth dataset from ONS that contained area-level deprivation data from 2019. We linked this dataset to the data on LSOA code (a unique number identifying each small area/LSOA in England). All linkages were conducted in R software using the merge function from the R base package. Figure A1 illustrates the results of the linkages of the separate primary care and local government datasets. A total of 232,671 individuals were linked across primary care and local government datasets (84.0% of the original primary care records).
To assess whether there were any potential selection biases in the linkage results, we calculated standardised differences in key variables for matched and unmatched primary care records [44]. Standardised differences of 0.2, 0.5, and 0.8 indicate small, medium and large effect sizes, respectively [44]. We were not able to assess potential biases in social variables extracted from local government records (i.e., in the household tenure variable and other household variables) as, by definition, unmatched primary care records did not have corresponding local government data. However, the number of unmatched local government records was considerably low (N = 369). Table A1 presents the results of analyses conducted to assess potential biases in the linkage results for matched and unmatched primary care records. These results indicate that selection biases were not introduced in selected variables originating from primary care records as a result of the success of data linkages, which is in keeping with previous analyses of this data [18].
unmatched primary care records. These results indicate that selection biases were not introduced in selected variables originating from primary care records as a result of the success of data linkages, which is in keeping with previous analyses of this data [18]. Figure A1. Results of data linkages. * Number of participants with missing data on each variable sum to greater than 29,866 (132,296 minus 102,430) as some participants had missing data across more than one variable. Figure A1. Results of data linkages. * Number of participants with missing data on each variable sum to greater than 29,866 (132,296 minus 102,430) as some participants had missing data across more than one variable.