Suchergebnisse

ObjectiveData linkage is increasingly used to generate research-ready datasets to help inform decision makers. Although use of linked data is increasing and accepted as an efficient way to generate new insights, linkage methods and error can have a significant impact on the subsequent analysis and interpretation of the results. Current linkage services overcome this by spending time explaining the linkage to the user. However, as the service demand increases and a generalised approach adopted, ensuring the information is available and understood is a growing challenge. Providing and presenting the information in an accessible way has therefore become a new problem for the data linkage experts.
ApproachUser researchers are working with groups of non-linkage experts using linked data to collate and understand feedback on:

what information researchers expect.
the information researchers find useful.
how researchers interpret and use the linkage information to inform their analysis.

ResultsFindings to date will be presented by exploring how information should be presented and what information researchers are able to interpret and use as part of their analysis.
Conclusions/ImpactFindings will provide guidance as to the information which should be included with a linkage output and the structure most accessible to the research community using linked data. Therefore, the research will help shape how the data linkage service is structured in the future by ensuring researchers can access information they require to interpret linked data and therefore use it appropriately. Ultimately, the feedback will help deliver data linkage services on a large scale.

ObjectiveAn innovative large-scale automated method has been developed to produce de-identified linkable data. The objective is to create a wide pool of ready-to-use data to enable faster and wider collaborative analysis for the public good.
ApproachA configurable automated pipeline prepares data for onward linkage at location, person, business and classification level through:

Big data profiling pre- and post-processing, for overview of variables and characteristics
Flagging potentially sensitive/identifiable variables
Generalisable linkage methods for large-scale data, to enable the addition of unique IDs for onward linkage of de-identified data
De-identification, hashing and redaction mechanisms, to remove and/or obscure sensitive/identifiable variables
Automated production of metadata, capturing linkage quality and transformations across the data journey
Quality assurance checks, including measure of linkage quality, assurance of variable derivations and redactions, and consistency checks on remaining data.

ResultsThe pipeline enables a configurable automated approach to producing de-identified, linkable, ready-to-use data in a traceable and fully documented manner. To achieve this we have: overcome scalability issues in working with big data; implemented automation at various levels; and enabled standardisation across data types and data structures to deliver a consistent recognisable final product.
Conclusions and ImplicationsBuilding an automated pipeline to enable onward linkage of de-identified administrative data was a complex process that has resulted in positive change around how our organisation operates and distributes data. This represents an important step towards future integration of linkage within the platform and the basis of future innovation in the area.

ObjectiveOrdinarily data linkage is approached on a project-by-project basis, tailoring the method to the research need. An increase in demand for data linkage has led to the development of a more efficient strategy, i.e. a generalised approach.
Through data linkage, a series of indexes representing people, businesses, geography and classifications are created using administrative data. Additional data are linked to the indexes via a general method to collect a common identifier.
Adopting a generalised approach requires an evidence base to ensure research aims can be met by the methods. This research will provide the evidence needed to underpin the strategy as well as guidance for implementing a generalised approach using an index.
ApproachThe existing data linkage portfolio provides an opportunity to compare the application of a bespoke to a generalised methodology. Comparisons will collate information on:

Index coverage;
Linkage results and quality including precision and recall estimates;
Analysis of the bias observed within the linked data.

Of particular interest is the linkage of different population sub-groups, particularly those classified as vulnerable or disadvantaged, for example, data for prisoners, refugees, the homeless and third-sector organisations.
ResultsWe will present the findings to date on recommendations for the use of a generalisable method for indexing, by exploring:

Data linkage quality by bespoke methods versus a generalised approach
The coverage strengths/limitations of the indexes
Sub-groups representation in the linked data

Conclusions/Implications Findings will provide evidence for adopting a new data linkage strategy, ensuring the most appropriate data linkage methods and quality assurance have been applied. Therefore, producing accurate and unbiased analysis to form the evidence base for policy decisions.

Objectives Due to the availability of data and the speed of delivery required, NHS number (Health ID) was used during the COVID-19 pandemic to join different health datasets to census 2021. This research sought to assess the quality of linkage done by NHS number, between COVID-19 vaccinations data and census 2021, by comparing to the quality that could be achieved with deterministic linkage using Personal Identifiable Information (PII). Specifically, we wanted to assure against concern that unlinked records may have been misclassified as unvaccinated due to inaccurate recording of NHS number at time of vaccination.
Approach We used deterministic (rules-based) linkage to identify extra links that were not made on NHS number. We also investigated how many links were incongruent between NHS number linkage and PII deterministic linkage.
Results We were able to link 98.5% of census records to vaccines data, with a precision of 98.94% and a recall of 99.71% This identified 1,833,626 additional links made using PII linkage. Furthermore, analysis of the linked data was re-run with the additional links.
Conclusion We were able to assure that the NHS number linkage used to produce COVID-19 vaccinations statistics, was valid. Whilst 1,833,626 more links were made, the findings resulting from the linked data were unchanged.
Implications Whilst this research demonstrates the utility of NHS number-based linkage, acquisition and use of PII on health datasets can increase the link rate and plays a vital role in estimating linkage error.

Objective and ApproachThis paper discusses the Data Linkage Hub (DLH) which is a centre for data linkage excellence and aims to produce and support the delivery of high quality, linked data assets for researchers in the UK. Here, we will give an overview of the DLH's activity as a Community of Practice for Data Linkage.
ResultsWe will discuss the different project types conducted in the DLH; the methods used ranging from bespoke linkage development to the adoption of generalised methodologies; and the customers of the DLH.
We will also discuss the tools the DLH have developed, and make available to others, to standardise, streamline and improve quality assessment of linkage. This will include clerical matching tools known as CROW and the CMS, BAT, a bias analysis tool, and a package of PySpark code specifically developed for data linkage.
In addition, we will explore the data linkage training materials we offer for different audiences and our involvement in a Data Linkage Champions Network, both increasing the awareness of data linkage and promoting best practise.
Conclusion and ImplicationsThe impact of the DLH's work is far reaching, underpinning statistical outputs, research and public policy. This success is down to the creation of a community of practice bringing together experts to regularly update and discuss linkage.

ObjectivesADR-UK supplied the Wage and Employment Dynamics (WED) research team with Pay as You Earn (PAYE) data linked to Annual Survey of Hours and Earnings (ASHE) data. This presentation explains the methodology taken to link PAYE and ASHE databases together, along with quality assurance and analysis of the linked data.
MethodsThe data were primarily linked using deterministic linkage methods on an encrypted National Insurance Number (NINo) variable. A total of four Pay as You Earn tables were linked to the ASHE data.
Where a PAYE table contained an encrypted NINo variable, the PAYE table was linked to the ASHE database using exact matching on NINo. Where a PAYE table did not contain an encrypted NINo variable, the table was linked to another PAYE table with an existing NINo variable first and then linked to the ASHE data.
ResultsAs multiple HMRC tables were linked to the ASHE database, the linkage rate for each table varies. However, at least 0.93% of each PAYE table linked to the ASHE database. As the ASHE data is based on a 1% sample of employee jobs taken from HMRC's PAYE records, these linkage rates are considered relatively high.
To compensate for loss of variables needed to calculate linkage quality, the data was filtered into categories on the 'sex' and 'age' variables and the linkage rates analysed. Bias analysis associated to records with a missing NINo value was also investigated. When the records with a missing NINo value were excluded from calculations, the linkage rate for each table rose to 1%.
ConclusionThis data will be used to assist research outputs that focus on low-pay labour markets and wage progression, wage inequalities and employment. The presentation will also discuss challenges overcome and lessons learnt from this project, with suggestion of how this information can be used to improve ADR-UK projects going forward.

ObjectivesClerical review is an integral part of linkage and is fundamental to quality assurance. CROW is a low dependency, adaptable, clerical matching tool that has been deployed across 40 linkage projects. Due to limitations of the interface, with the initial version (including the limited ability to view large clusters); this project sought to re-develop the tool as a flask application.
ApproachOur approach was to transform the existing CROW desktop application into a Flask (web based) application. Flask was chosen because it is a flexible web framework that enabled the development of a more dynamic interface, as well as better integration with the infrastructure of secure environments. We also engaged with accessibility experts to ensure the inclusivity of our product. Once the development was completed, testing was carried out by a range of users and stakeholders, to identify further issues and bugs for resolution.
ResultsThe successful redevelopment of CROW into a Flask application improved functionality, usability, parametrisation, and accessibility of the tool. Feedback from users indicates that because of using CROW, the clerical matching experience has improved vastly. Users reported that CROW increased accessibility, reducing eye strain and fatigue. CROW also increases efficiency and leads to better decision making.
ConclusionThe successful redevelopment of CROW is a significant improvement to the clerical review process for data linkage projects.
ImplicationsCROW is open-source and low dependency so could be adapted to work in most environments. Therefore, it could be adapted and used by the wider data linkage community.

ObjectivesDevelop administrative linkages within a national Trusted Research Environment (TRE) that hosts Longitudinal Population Study (LPS) data for over 20 LPS. We will describe the methodological development carried out to enable linkage to administrative datasets. These linked administrative data will support research for public good, informing policy and practice.
MethodsThe first sets of administrative data under consideration in this Feasibility Study are from the Department of Work and Pensions (DWP), the Department for Education (DfE) and HM Revenue and Customs (HMRC). Working with UK Government departments through a Task & Finish group we have gathered input from DWP, HMRC and DfE and Office for National Statistics (ONS) data sharing experts. The Task & Finish group identified three pragmatic data linkage and data sharing models, that would enable data to be linked via a newly designed secure data pipeline in a legal, secure, and trustworthy manner for all stakeholders.
ResultsTo encourage sustainability and acceptability, a model designed to be maintained over a long period is based on the re-use of Departmental Personal Identifiable Information (PII) – i.e., name, date of birth, gender, National Insurance number - and attribute data already deposited by the Departments into ONS. ONS will develop for the linkage and extraction of ONS Data into the TRE a system which conducts, and quality assesses the linkage; minimises the Departmental data to participants within the TRE only and the variables specified in the agreements; and, de-identifies the data to their DEA processing standards. The minimised and functionally anonymous data extract will be securely transferred for ingest and integration into the TRE enabling researchers to address a wider range of questions for public benefit.
ConclusionThis is a model for efficient and low-burden linkages to inform cross cutting research. It will form part of a responsive UK data science capability which can inform government research needs and be used to meet future crisis e.g. new pandemics, the impacts of climate change or economic shocks.