Research Questions:

1. Where are homes created through PD located in England?

2. What are the rates of all-cause and cause-specific morbidity and mortality in individuals living in PD homes, compared to other properties in areas with a similar level of socioeconomic deprivation?

Approach

Assessment and preparation of permitted development housing data group

We will compile a dataset of PD housing using data from Glenigan (a company with a housing permission database) which will in turn be compared to Ordnance Survey AddressBase data to identify postal addresses of houses in PD buildings. Individuals will be identified as living in PD housing if they are residing in an PD housing address during a hospital admission.

Control group - housed individuals not living in PD homes

A comparison group of other housing created through full planning permission will be similarly prepared. These datasets require pre-processing to detect and eliminate duplications and errors. Comparison with local authority planning databases will assist in ensuring data accuracy. Using these data, we will undertake a descriptive analysis (including mapping) of homes created through PD located in England.

Linkage, storage and information governance of combined dataset

Hospital Episode Statistics (HES) and mortality datasets will be linked to housing data (subject to additional approval from the NHS Research Ethics Committee in addition to the UCL Research Ethics Committee approval covering the rest of the project). First, we will create a master address list, containing a unique pseudonymised variable, for linkage to housing data. This list will be sent to NHS Digital and linked to HES and mortality data. The address information will then be destroyed and the health and unique pseudonymised variable will be securely transferred back to the UCL Data Safe Haven (where all study data will be stored) and linked to housing (using the unique pseudonymised variable) as done in Lewer et al. (2021).

We know that the HES datasets are definitely available. However, we also understand that given lag times between exposure and reactions, many effects of PDR housing will not require hospital admissions and so whilst a comparison of hospital admissions between PDR residents and our control group could demonstrate differences, using alternative data sources for matching to PDR addresses OpenSAFELY and GPES may also be useful. At present we await confirmation that the OpenSAFELY data set would be available to further linkage work beyond what is allowed under COPI notice for COVID whilst previously GPES data costs have been prohibitive. We would, however, be open to utilising these data sets as alternatives if these appear to be feasible at the commencement of the study. The approach would remain the same in terms of identifying PDR housing as a first step in all cases, and then undertaking data linkage and consequent analysis with the chosen data set, albeit the outcomes would then be updated appropriately for the data set(s) actually being utilised. The choice of data set for the linkage can be confirmed having explored all three options whilst the PDR address identification is ongoing and with appropriate engagement of the steering committee.

Sample size:

We will have statistical power (90%) to detect small changes (e.g. hazard ratio of 0.95) for hospital admission rates at the 5% significance level when comparing individuals living in PD homes to other properties.

Analytical plan:

Our analytical plan will build upon our approach that we have conducted previously using linked hospital and death data for a cohort of people experiencing homelessness (Aldridge et al., 2019). We will construct a cohort study of individuals living in PD homes, compared to other properties. For each individual identified as living in a PD home, we will select a housed patient who lived in an area with a similar level of socioeconomic deprivation, matched on age, sex, hospital, and year of discharge. We will undertake the analysis in two phases. In the first phase, we will analyse baseline characteristics of all participants to describe the characteristics of PD intervention and control groups. With the exception of ethnicity, all baseline characteristics are anticipated to be fully observed (chronic disease is presumed to be absent unless recorded). Missing values of ethnicity will be analysed grouped as ‘not recorded’.

In the second phase, we will identify evidence of differences in the baseline characteristics including age, sex, chronic disease and reason for hospital admission at the time of their index admission. We will link to HES data going back to 2010 and up to 2023. We will use hospital admissions prior to an admission in a PD housing (and for controls for the matched time period) to identify prior comorbidities as we have carried out previously (Lewer et al). Individuals will enter the cohort during the first admission when living in PD housing (or matched time periods for controls) and will be followed up until the first of either death or the end of the study period (2023). We will estimate the crude association between each of the primary and secondary outcomes and the study population groups. We will then re-estimate the association between each of the outcomes and the study population group after adjusting for characteristics at the time of admission: age, sex, chronic disease and reason for hospital admission.

An appropriate statistical model (selected on the basis of meeting assumptions such as proportional hazards for Cox regression) will be used to analyse the relationship between the study comparison group and each of the study outcomes. Crude models will be fitted prior to adjustment for ‘baseline’ measurements at or before the index admission. We will write the analysis in accordance with the Reporting of studies Conducted using Observational Routinely-collected Data statement. We anticipate that we will compare our primary outcome of death, analysed in subgroups of 10th version international classification of disease (ICD-10), deaths from specific deaths; and deaths from amenable causes. We will compare all-cause hospital admissions (emergency readmissions, planned readmissions, and Accident and Emergency (A&E) visits) and cause specific admissions in the PD and control group. Particularly relevant health outcomes would include cause-specific mortality and hospital admissions due to Chronic Obstructive Pulmonary Disease, Asthma, Lower respiratory tract infections, Malignant neoplasms, Diseases of the circulatory system and Covid-19.

Sensitivity analyses:

There are several challenges when examining differences in the outcomes for individuals living in PD homes and our sensitivity analyses aim to explore these issues further. First, we will repeat analysis only for those individuals for whom it has been possible to generate confounding variables that can be used for adjustment within the statistical models. Second, we will conduct separate analyses according to the type of index admission (elective or emergency). Third, we will investigate only rates of planned elective readmissions (admission method). Fourth, we will restrict our analysis to admissions in London (and other regions as appropriate) to explore any bias in regional comparisons of PD homes that may exist.