Improving homes, improving health
Housing plays an important role in health and wellbeing. Factors such as insulation, heating efficiency and energy performance can influence the quality of the home environment and, in turn, affect health outcomes. Understanding these relationships is vital to improving public health in Wales, but to do so, researchers need access to detailed data about the energy performance of our homes. In this blog Rhodri Johnson, Research Officer for the ADR Wales Health and Care theme, discusses the work that he has been involved in to safely unlock the potential of home energy efficiency data, so that this data can be a valuable resource for health research.
Energy Performance Certificates or EPCs are a single source of information on a property’s energy efficiency that include information such as standards of insulation, heating systems used, and many other factors related to the property. EPC data is open-source and freely downloadable data (https://epc.opendatacommunities.org/) provided by the UK Government (Department for Levelling Up, Housing and Communities). But while EPC data is rich in detail, it also comes with a challenge; it carries a high re-identification risk (RIR) if imported directly into a Trusted Research Environment (TRE), so appropriate steps need to be taken to reduce this risk.
In EPC data, each record for a property also contains standard address data along with a Unique Property Reference Number (UPRN). Within various governance and legal frameworks (including the GDPR) there is ambiguity as to whether UPRNs are classed as personal identifiable data or not. Regardless of the legal status, once a UPRN is linked to restricted individual-level data i.e. to allow selection of individuals who live at the property, there is an increased risk of reidentification of individuals and potential legal change in the status of a UPRN from open data to personal data. Governance and data privacy protections are key to the success and trust which allow TRE’s such as the SAIL Databank to operate. So what have we done to ensure that this data will never be identifiable?
The challenge: safeguarding sensitive geospatial and address data
- EPC data contains:
– Full addresses and UPRNs.
– data variables relating to the property including energy rating, housing tenure, heating types and many more. - Some variables appear to be free text fields, depending on how assessors input data.
- Very precise numerical values can be present (e.g. floor space measurements to multiple decimal places).
These features make EPCs particularly vulnerable to linkage with open data sources. For example, floor area can be recorded with highly specific values, such as 90.899m² which is easily matched with the same value in an open dataset, undermining anonymisation and risking disclosure.Finding a solution
As part of the ADR Wales Health & Care theme, and in collaboration with colleagues from the Environmental and Health Research centre (ENVHE) at Swansea University, we set out to understand the risks posed by linkage with the EPC as open data and to design a data ingest strategy which satisfies SAIL governance framework.
Following a data reduction strategy to identify which variables were required to answer the research question, we first tested the more automated, statistically based method of k-anonymisation, but quickly found this reduced the data’s utility to the point where it would no longer serve research effectively. Instead, we used a more pragmatic risk-based approach using various masking techniques that better balanced risk with usefulness. This was a more manual, but also more controlled approach. Methods used to reduce RIR included:
- Rounding and/or categorisation of numerical data
- Re-categorisation of variables into meaningful but less granular bands.
- Masking sensitive variables.
- Selecting only the key data items necessary for the specific research project.
- Restricting the number of records per property.
The overall aim was to reduce RIR to a point where any reidentification would require specific additional actions which would entail malicious intent by a researcher. Any applicant to SAIL must pass strict governance rules prior to accessing data (https://saildatabank.com/governance/ ).
Collaboration in practice
The work to allow EPC to be safely used within SAIL had been under consideration for ADR specific projects, but was accelerated following an application from a researcher aiming to use EPC data in SAIL to complete a PhD on links between housing energy efficiency and ill health, particularly respiratory conditions. Our partnership approach was completed in collaboration with Isobel Braithwaite at University College London (UCL); the work was a genuinely collegiate process – working together to understand what variables, and what level of detail were needed for analysis, while ensuring the data remained safe.
The data which covers the period from 2009 to 2024 prepared through this process has since been used in a number of projects, with Isobel’s PhD analysis using the linked EPC data having been submitted, along with a number of conference presentations, and academic journal submissions from Isobel and the wider study team at UCL, co-authored with Rhodri, in preparation. The work to safely link the EPC data enabled this study to leverage electronic health record data alongside detailed information about the energy efficiency characteristics of individuals’ dwellings, and to follow the cohort up longitudinally. This has provided valuable new insights about associations of energy efficiency with respiratory disease incidence, which can inform housing policy decisions.
Innovating governance: a triage-based model
Alongside the practical data preparation, this project also helped trial a new triage-based approach to environmental and spatial data governance in TREs, using SAIL as a testbed. This model considers the nature of the dataset and its source to assess risk, along with a basic application form.
We initially setup a risk-based triage approach where, for example, area-level data, or data not in the public domain would be deemed low risk and would not require any data modification prior to importing into SAIL. Open-source data such as the highly granular EPC would be deemed high risk and require further work prior to allowing data into SAIL alongside standard governance procedures and approvals.
An application process involved applicants specifying which variables were required along with a narrative to explain why each variable would be required in line with the approved research project.
By initiating this triage model, we have successfully allowed the import of EPC data into SAIL in line with existing SAIL governance framework; work is ongoing towards appropriate processes to handle environmental and spatial data with differing disclosure risks.
Looking forward
The work carried out through the ADR Wales Health and Care theme and the Environment and Health research centre, demonstrates how data about how and where people live can be adapted for safe use alongside sensitive personal data in research – protecting privacy while enabling important health and social research to progress.
Our approach shows that it is possible to unlock the potential of EPC data without compromising anonymity, ensuring that future studies can continue to explore the links between housing, energy efficiency, and health outcomes in Wales. This work has allowed to improvement to policy and processes surrounding the use of such data which will benefit other researchers wishing to use geospatial and address level data safely within the SAIL Databank.
Acknowledgements
This work was supported by the ADR Wales Health & Care thematic research area. Collaboration included colleagues from Rebecca Pedrick-Case and Professor Richard Fry in the Environmental & Health Research team at Swansea University, with special thanks to Isobel Braithwaite (University College London and Tower Hamlets Public Health) for co-developing the research requirements and applying the data in her PhD project, and the National Institute for Health Research which funded her PhD Fellowship (award no. 302316). We also acknowledge the UK Government (Department for Levelling Up, Housing and Communities) for making EPC data available for use within research.