STEADFAST Modelling the associations between wider health and social characteristics and diabetes-related health - Flows into DHCW and SAIL for linkage in SAIL

This Data Sharing Agreement permits the retention and further processing of the data provided under previous iterations of this Agreement for an interim period.

Cardiff University requires National Diabetes Audit Data for use in a research project, ‘Investigating interrelationship between diabetes and children’s educational achievement’, which aims to model the association of the relationship between wider health and social characteristics and diabetes related health and better understanding the effects of educational trajectories, experiences and outcomes on diabetes management. The background to this study is the evidence that most of the costs arising from Type 1 diabetes come not from the day to day care and medications but from the complications arising from sub-optimal management over the life course, yet the causes of sub-optimal management are poorly understood. An aim of this study is to broaden the search beyond the purely clinical setting to investigate the wider health and social determinants of diabetes related health, using linked administrative data and focusing on education.

The project aims to quantify the links between diabetes related health trajectories and educational trajectories. Diabetes related health is measured using HbA1c, a proxy for blood glucose management control which is measured regularly as part of diabetes clinical management and recorded in the diabetes audit data. Educational outcomes are recorded from the time a child enters school until they leave university, including measures of attendance, attainment, and broader characteristics of the educational experience such as special educational needs and school exclusions.

The causality in this relationship is complex. It is hoped that three potential causal relationships will be identified. Firstly, diabetes related health may affect educational outcomes, for example through biological mechanisms including the effects of excess glucose on the brain and social mechanisms such as adjusting management routines in order to fit in with a university lifestyle. Secondly education may affect diabetes related health, for example staying on in education beyond compulsory schooling might provide structure and support that facilitate better management. Thirdly there may be individual specific characteristics that directly affect both education and diabetes related health, for example motivation and intelligence.

The processing of the data under this Data Sharing Agreement involves the following organisations:

- Cardiff University is the data controller and will perform analysis of the linked data and publish the findings;

- Digital Health and Care Wales (DHCW) is a data processor for Cardiff University and will receive, link and pseudonymise the data through the creation of a hashed linkage field before passing it on to SAIL;

- The Secure Anonymised Information Linkage Centre (SAIL) at the University of Swansea is providing secure facilities to store the datasets as linked datasets using the hashed identifier provided by DHCW so that it can be remotely accessed by an individual from Cardiff University for the purpose of analysis.

The Welsh Government (as the data controller for the schools and college data) and HESA (Higher Education Statistics Agency) will supply educational data to DHCW which will be linked with the data from NHS England (formerly NHS Digital). Neither organisation (Welsh Government or HESA) will process data supplied under this Agreement.

Cardiff University is the sole data controller for the data under this Agreement and for data from other sources to which the data will be linked for the purpose of this study. The data will be processed by Digital Health and Care Wales (DHCW) and Swansea University Secure Anonymised Information Linkage (SAIL) Databank, under the direction of Cardiff University. The Medical Research Council (MRC) are funding the work to undertake research such as this but cannot access data nor has a role in analysis or interpretation.

Data may be accessed by individuals holding an honorary contract under the supervision of a substantive employee of Cardiff University for the purposes described in this DSA only. Cardiff University must maintain records in a single location that cover the following details of each individual given access under an honorary contract:

o Their substantive employer;

o Their role in respect of the purpose for the processing specified in the DSA;

o The start date and end date of the duration in which the Data will be accessed by the individual under an honorary contract;

o The necessity for the Data to be accessed by the person(s) holding an honorary contract, instead of a substantive employee of an organisation named as controller or a processor in this DSA;

o Confirmation that an appropriate contract is in place which follows the relevant guidance and is countersigned by the substantive employer of the honorary contract holder.

This work is research in the public interest as it aims to improve care for all children with Type 1 Diabetes – informing clinicians and commissioners of variation and outcomes and complications to improve and standardise support in clinical and education settings.

Under this Agreement, the data may be securely stored and processed. No new data will be provided by NHS England (formerly NHS Digital) under this Agreement.

The request involves the disclosure of confidential patient information from the National Diabetes Audit (Adults – England), National Diabetes Audit (Adults – Wales) (both held by NHS England) and the National Paediatrics Diabetes Audit (held by the Royal College of Paediatrics and Child Health) to Digital Health and Care Wales (DHCW).

NHS England will produce an extract of relevant variables and cohorts (based on education cohort availability) for patients born after 31/08/1984 from the NDA data. A new ‘system ID’ (sometimes referred to as project ID or study ID) will be generated for each case in this extract and used exclusively for re-joining the two ‘split files’, the split file process is used to avoid flowing identifiers alongside substantive data.

NHS England will split the NDA extract into two parts i.e. the split files:

The first split file is the ‘identifiers file’ and will contain the system ID plus the identifying information, in the case of the NDA extract this will be the patient name, NHS number, date of birth, gender and postcode. This file will have no substantive diabetes data.

The second split file is the ‘substantive file’, and will contain the system ID, plus the substantive diabetes related clinical data. This file will have no identifying information other than the system ID.

The identifiers file will be sent by NHS England to DHCW. DHCW will create a hashed version of NHS Number, referred to as the Anonymised Linkage Field (ALF), then destroy the identifying information (NHS number, date of birth to nearest day, postcode). After this process, the identifiers file will contain only the system ID and linkage field (ALF), along with less disclosive versions of the demographic data (gender, week of birth, lower super output area). DHCW will then flow the modified identifiers file to SAIL.

The substantive file is sent directly to SAIL.

Once the substantive file and linkage field file are within SAIL, the substantive file is rejoined to the linkage field file using the system ID, this is so that each case of the substantive data gets a linkage field (ALF). It is the linkage field (ALF) that is used to link data within SAIL.

The same process as outlined above for NDA data, also happens for the other datasets that are placed in SAIL. Datasets can be linked in SAIL using the linkage field (ALF). No identifiable information is ever put into SAIL. Other datasets that will be brought into SAIL and linked to NDA data are National Paediatric Diabetes Audit, Higher Educational Statistics Agency dataset (HESA), and Welsh Dataset Education Records (WED, AKA National Pupil Database NPD). Data already held centrally by the SAIL databank including measure of family structure and additional measures of child health outcomes will also be linked.

Alternatives to these data flows are not viable to achieve the amount of linkage required for this study. Single datasets could be flowed, for example HESA data could be shared with NHS England to look at the effect of higher education on adult diabetes trajectories, however single flows would only enable exploration of one substantive area, whereas the objective of the study is to follow the entire life course trajectories of both diabetes related health and education through the combined linkage of both the paediatric and adult audits and education data from primary school to university.

Data is flowed using the split file technique to minimise the risk of disclosure, which means there are two streams of data, one for the identifiers and a second for the substantive data. The level of data is individual for all flows of data leading to the research dataset used for analysis in SAIL. All outputs arising from the research in the databank for publication are aggregated and small number suppressed.

The Cardiff University research team will be accessing the linked dataset in SAIL. The research team may include students and individuals on a honorary contracts (typically one or two per year). No data will be accessed outside the UK. Data will not be shared with third parties beyond those required for the creation of the pseudonymised linked dataset used for analysis in the SAIL databank (i.e. only DHCW & SAIL). The data will be used for statistical modelling of education and health outcomes.

There should be no NDA cases in the requested extract that will not meet the conditions of the study. It would be expected all cases to have attended school, and those that do not attend university are still important as controls for comparison with those that did.

Data minimisation has been discussed at length with NHS England’s National Diabetes Audit Team and the request has been restricted to only the essential clinical measures and associated metadata such as date of measure, location of measure (clinic). The team have also restricted the participants to only those birth cohorts that would also be included in the linked administrative education data.

The NDA data extract has been designed to only include data of individuals that meet the conditions of the study. This excludes:

• Individuals born prior to 1979 since their education data is not currently available;

• Individuals born from 1979 to 1983 since the data on their first year of university is not available;

• Individuals born after 2002 since they would be too young to be in the NDA. As part of the study, other data will be used for individuals born between 2003 and 2013, but not from NHSE, since those individuals should still be in paediatric care and recorded in the NPDA.

NDA cases born 1984 – 1996 have higher education data for the most common ages at university, in addition the later of these cohorts have some schools data, including key stage 4 and key stage 3 assessments, these cohorts will form the core of the analysis, all of these cohorts are requested for analysis.

NDA cases born 1997 – 2002 have sufficient university or schools data to make inference on the associations between diabetes related health and education, though unlike the core cohorts there will not be a record of whether they completed university.

For these cohorts the study is primarily interested in the effect of diabetes during education and thus restricting cases to those diagnosed with any form of diabetes prior to age 24 or younger.

The predicted ages of school students includes early years (nursery and reception) aged 3-5, compulsory schooling aged 5-16, and key stage 5 ages 16-18. Some students start school early or leave later than normal, however there will be no request for additional cohorts for this contingency and will instead make this an amendment if it transpires to be an issue. University cohorts are typically 18-21, however due to the high prevalence of delays in starting time at University (gap year, changing university, changing course etc.) and the varying length of courses (sandwich degrees, placements, conversion courses) and will include all ages up to government definition of the end of youth education, i.e. up to age 24.

The following items of confidential patient information are required as part of the NDA extract to be supplied by NHS England for the purposes of creating the linkage field:

> NHS number: Used to create Anonymised Linkage Field (ALF).

> Date of birth: Linkage validation (also translated to week of birth for analysis).

> Gender: Linkage validation (also used for analysis).

> Postcode. Linkage validation (also translated to LSOA for analysis).

The project has Section 251 approval for the NDA identifiers file to be linked to the patients' name to facilitate linkage with the wider data sources in the study. NDA does not hold the patient name so NHS England will trace the patient name once the NDA data has been prepared.

Wider clinical information will be provided from the diabetes audits for inclusion in the analysis dataset.

The Data will only be used for the purposes described in this Agreement.

This Agreement permits the secure retention and processing of the data only.

The expected outputs will be academic journal papers modelling the associations between education outcomes and trajectories of diabetes related health. The nature of data-based outputs would be descriptive statistics and regression coefficients, all of which would be aggregate data with small numbers suppressed, this is a requirement of taking any results out of the SAIL databank and is rigorously checked with any outputs from analysis in the secure gateway having to be screened by a SAIL analyst.

The findings will be shared with clinicians both through academic publications such as BMJ, Lancet Diabetes Endocrinology, Diabetes Care and through presentations at national forums such as the Diabetes UK & Brecon Scientific Meeting.

There will be a plain English version available to the audit for sharing more broadly and the applicant will present results at patient forums in Cardiff and Swansea.

The outcomes of the study will provide clinicians with a robust quantification evidence for the links between education factors and diabetes-related health (defined using HbA1c measures from the diabetes audit data) for both children, young people and adults. The findings will be shared with clinicians both through academic publications such as BMJ, Lancet Diabetes Endocrinology, Diabetes Care and through presentations at national forums, Diabetes UK & Brecon Scientific Meeting. The team have already presented the planned work to a national meeting of diabetes clinicians and the response was that such evidence would be useful to help them include educational factors in the way they differentiate practice.

The work on children in schools will focus on how rates of absence and attainment trends are associated with diabetes management. Whilst it is expected those children, who have always had lower rates of attendance or attainment (both before and after diagnosis) to have worse diabetes management, the study will additionally differentiate between those who experience larger changes in attendance or attainment following diagnosis. The applicant will also consider if the characteristics of schools are associated with management, and whether there are changes during the transitions between primary and secondary schooling. Due to changes in legislation and the improvement in communication between diabetes nurses and school staff, there are robust links that can be exploited to seek feedback from teachers where issues over attendance or attainment that can be used in clinical care.

The work around the period of transitions from compulsory schooling to A-levels, vocational courses, or the labour force will help inform how these choices subsequently impact the trajectories of HbA1c. For example, the study may find that those who choose not to continue formal education beyond the age of 16 have less structure and see a worsening of diabetes management. Clinicians could provide extra support or delay transition from paediatric to adult diabetes care until after the transition to employment is complete.

The work focusing on universities will look for changes in trajectories of diabetes management, for example, comparing those that live at home with those who move away to attend university, if there is a significant change in management and the timings of that change, perhaps in the first term as students try to reconcile good management routine with university social life or perhaps it may not alter until the stress of the final year with impending high stakes exams. Clinicians maybe more proactive in supporting the transition between home and university care and arrange follow-up appointments both at home and university at times that fit with any sensitive periods and academic terms.

The work on adult outcomes identifies characteristics of the educational history that predict sub-optimal adult diabetes management and propensity to earlier complications of diabetes. The team expect those with more years of schooling to have better control and seek to identify whether the wider characteristics of the school experience are associated with worse control later in the life course, these might include school factors such as special educational needs or behavioural problems, or factors that interact with education such as socioeconomic status. The benefits to practice are on better identification of the important non-clinical risk factors for sub-optimal adult diabetes management to support interventions to prevent costly complications.

Since NHSE (and the other data providers) shared the extracts of data in 2019/20, the study has uncovered an issue with the linked data, specifically the NPDA component: the IDs for 2012/13 are mostly missing, thus breaking the trajectory of HbA1c values wished to be modelled. RCPCH reported that NHSE had fixed this issue, though it appears this is not correct, or at least not correct for most pupils at Welsh schools. This gap in the data can be overcome for the first strand of the modelling (how measures of diabetes-related health influence educational trajectories); however, it is not resolvable for the second strand (how education settings affect HbA1c trajectories) or third strand (how external factors affect both diabetes-related health trajectories and educational trajectories). The team have prepared a research paper based on the first strand, currently under review at the journal ‘Diabetes Care’, which shows (i) children with diabetes have higher absence rates, but similar attainment and progression to university as their peers and (ii) HbA1c tracks with educational outcomes, the team provide tentative evidence that this link is likely to be the results of external factors affecting both education and health outcomes.

To support public involvement and engagement for using this extract, the project team has given presentations of the results obtained so far to practitioners (e.g., the Royal College of Paediatrics and Child Health annual conference, the Brecon Annual Meeting of T1D practitioners in Wales), patients (e.g., the Yorkshire and Humber Young People’s Diabetes Network Annual Meeting), and researchers (The Type 1 Diabetes Consortium at the Diabetes UK conference). They have also improved the Cardiff University project website, with a more user-friendly- page on the current funder's website (Welsh version here), which also links to a blog at the start of the latest phase of this work. Information about how the project consults young people are provided on the webpage for the first 2021 PPIE workshop, the workshop report, and the blog from that event by Charlotte Austin, the Diabetes UK lead for Type 1 diabetes. The team have also recorded the presentations given in the patient engagement sessions, Including Alex Bailey from the Medical Research Council Regulatory Support Centre speaking about using health data for research, Lucie Burgess speaking about the Diabetes UK perspectives on data, Rob French speaking about the specific data usage in the research study, and also feedback from three of the participants. In the second workshop, the young people produced a video explaining how researchers use data relating to young people with diabetes. Although the video and report produced were not formally released, there is a version on YouTube. One can also view the video for the paediatric diabetes audit from that session. Several of the representatives from those PPIE sessions last year have now joined the study coproduction group, helping steer the project and the next iteration of the public engagement. In 2022 the project started a new public engagement workstream funded by DARE (HDRUK, ADRUK, and MRC), which includes discussions with young people on the use of large-scale unconsented data, with a focus on differentiating between the views of children and young people of different ages, and ensuring there is greater ethnic and socioeconomic diversity (target 50% of participants). This study has already has expressions of interest from over 400 young people and consented over 80 of these into the focus groups. [*hyperlinks have been shared but not recognised by the DARS system - can these be inserted manually?]

*2024 update: In February 2024 young people were brought from each health board in Wales together to meet with the Children's Commissioner to discuss the issues raised in our research and the data. This has led to follow up from the Commissioner on behalf of the children with key policymakers including the Education Minister and the Additional Learning Needs implementation working group.

Cardiff University produced a research study in the world's leading clinical diabetes journal outlining the variation in outcomes for children with type 1 diabetes. This included additional analysis for children by age of diagnosis, different levels of diabetes management. Through this study, evidence was also provided based on analysis of the Welsh data, to the new Additional Learning Needs Bill for Wales and the Special Educational Needs and Disabilities Green Paper in England, though these are yet to report their findings so have not yet resulted in any change in practice.

To support public involvement and engagement, the study team has given presentations of the results obtained so far to practitioners (e.g., the Royal College of Paediatrics and Child Health annual conference, the Brecon Annual Meeting of type 1 diabetes practitioners in Wales), patients (e.g., the Yorkshire and Humber Young People’s Diabetes Network Annual Meeting), and researchers (The Type 1 Diabetes Consortium at the Diabetes UK conference). The findings have been well received by clinicians, who report that they feed back some of the findings, particularly the ‘good news story’ that overall, children with diabetes attain in school as well as their peers without diabetes. Practitioners have also reported that the quantification of expected absence (average and range) has also been useful to provide a benchmark for children with diabetes.