Agreement for holding the HES data beyond IQVIA HES-THIN sublicense agreement

The University of Birmingham (UoB) have specialist skills in the extraction and analysis of routinely collected data.

UoB currently host and utilise The Health Improvement Network (THIN) database, and HES linked data supplied by data provider IQVIA (which holds a data sharing agreement with the NHS Digital) in order to conduct a wide variety of clinical epidemiological study design.

GP practices in England that contributed to the The Health Improvement Network (THIN) data collection scheme between 2011 and 2014 have their non-identified patient data linked with Hospital Episode Statistics (HES) data for the purpose of producing enhanced datasets for use in medical research studies. This linkage was carried out in collaboration with an Enhanced Trusted Third Party (eTTP) who had developed a secure encryption technology which enabled the THIN data linkage to take place without the need to export any patient identifiable data from the data providers (THIN practices and NHS Digital).

HES data allows the validating of secondary care diagnoses recorded in THIN data, and provides further information on hospital attendances, procedures, diagnoses, length of stay and outcomes. In order to enable UoB to continue the research, UoB need to be able to retain the HES data that has been provided by IQVIA via a sub-licence agreement. This agreement has been reviewed and approved by NHS Digital. Record level data are not shared with any third party and only used by the University of Birmingham. The funding for the next three years to meet the objectives of processing is provided by the University of Birmingham (sole data controller) with support from the National Institute for Health Research (NIHR) and Medical Research Council (MRC).

The request for HES-linked data in this application to continue the completion of a variety of epidemiological studies would fulfil the GDPR rules for lawful processing. As UoB, in its capacity as an academic institute is a Public authority, lawful processing falls under Article 6 (1)(e) of the GDPR whereby “processing is necessary for the performance of a task carried out in the public interest or in the exercise of official authority vested in the controller.” The task being carried out in the public interest relates to conducting epidemiological research which will provide benefits to public health in terms of; health service research, health care evaluation and epidemiology of diseases to better understand their causal pathways.

This request also fulfils the exemptions noted in Article 9 (2) (i) and Article 9 (2) (j). The data being requested will be used for scientific and research purposes whereby appropriate data protection safeguards will be implemented to allow for the safe processing of the data. In addition, the data being processed will be used for the benefit of public health as mentioned above. Specifically, it will be to evaluate and improve the standards of quality and safety of health care, medicinal products and medicinal devices.

Each project conducted using the proposed linked dataset will undergo a peer-reviewed ethics procedure. This will be to ensure that ethics considerations are addressed prior to the dissemination of research. Confidentiality, security and safety of the data will be paramount throughout processing. Any data given to UoB will be processed on secure encrypted hard-drives and cloud services. The data will remain in an pseudonymised format to ensure no risk of identifying individual patients from the dataset.

Use of the linked THIN-HES data is limited to only the following areas of medical research: epidemiology, pharmacoepidemiology, health services and public health research, including risk factor identification, policy evaluation and outcome studies, and health economics research.

There are several example areas which would demonstrate the use of such data linkage.

1. Delivering a programme of applied health services research under the West Midlands Collaboration for Leadership in Applied Health Research and Care programme (CLAHRC) and its successor: The CLAHRC covers multiple themed research streams to answer research questions involving patient flows through primary and secondary care systems and the resulting patient outcomes. CLARHC sits within UoB and so the sole data controller would remain UoB. To do this UoB requires contextual data on the English acute hospital population to make valid comparisons, using metrics such as length of stay, in-hospital mortality rates and readmission applied to various groups of patients, determined by epidemiology, demography or service use derived from HES and THIN data.

2. Maternal and child health research workstream: Pregnancy and birth events are poorly recorded in primary care. HES linked data enables UoB to improve the current research in maternal and child health by providing missing information surrounding birth events, improving the linkage of anonymised mother and child data and allowing to follow these individuals up over time to identify a range of associations and the impact of family history in a variety of conditions. Another example specifically for conditions relating to abuse in childhood and adulthood is that the exposure of abuse is poorly recorded in primary care data. Therefore, linkage with hospital data allows for a more accurate description and identification of exposure. This enables associations with poor health consequences including cardiovascular, gastrointestinal, respiratory, musculoskeletal, neurological as well as multiple other body systems explored through cohort study designs. Thirdly this allows to link individual family data to explore the consequences of intergenerational abuse. Mothers and babies are linked in primary care using family number and pregnancy codes. By linking these families, it is possible to identify what the outcome of a specific exposure to a mother is on their children.

a. Automated infant mother linked studies

b. Long term impact of pre-incision antibiotics born by caesarean sections

c. Long term impact of recurrent caesarean sections

d. Novel risk factors and long-term health effects of postpartum haemorrhage

e. Women with adverse metabolic profiles and their offspring outcomes

f. Long term conditions, disability and complications in pregnancy with maternal and child health outcomes.

g. Abuse (child and adulthood) and ill health outcomes

h. Reproductive factors and risk of cardiometabolic outcomes

3. Non-communicable and communicable disease epidemiology stream: Often incidence and prevalence studies to assess the burden of the disease are particularly challenging due to limitation in recording in primary care alone. Therefore, linkage with HEs data allows for capture-recapture methods which can estimate the burden of disease comparing with hospital attendance, improving the disease burden estimates. In addition, linkage to hospital data provides invaluable information for the hospital admission burden related to specific disease. In addition, it provides additional information about specialist treatments that occur in secondary care, of which the effects can be investigated in long term primary care data. This can provide the opportunity to conduct healthcare evaluation using such methods.

a. Epidemiology of cardiovascular disease

b. Epidemiology of diabetes and endocrinological conditions (such as hyperthyroidism treatment modalities, polycycstic ovarian syndrome outcome studies)

c. The impact of Ramadan fasting on health

d. Epidemiology of rare diseases (e.g. Idiopathic intracranial hypertension)

e. Epidemiology of gastrointestinal disease

f. Infectious disease workstream

4. Overdiagnosis, overtreatment and health services research workstream: Similarly, to the above, linkage to HES data provides useful information about the delivery of certain procedures and therapies that are delivered in secondary care. When combined with primary care data it provides the opportunity to evaluate whether these services are being given to the correct patient population. This will provide significant benefit to the population as it will give an indication as to whether health professionals are suggesting unnecessary treatments for patients. Therefore, the output of the research will be able to provide guidance to health professionals as to the dangers of overtreatment to their patient cohort. An example study being the exploration of indication for tonsillectomies. It has been previously reported using primary care by the research team that tonsillectomies may be inappropriately delivered, and so possibly exposing patients to risks which may not weigh up with the possible benefit.

5. Wider determinants of health workstream: A specific example relates to the West midlands air quality improvement programme WM-Air will provide improved understanding of pollution sources and levels in the region, and new capability to predict air quality, health and economic impacts of potential policy measures. The project runs for a five-year period from 2018. In this project UoB will link present day and future air quality metrics, with literature derived air quality health multipliers, to identify disease-specific exposure estimates for the West Midlands population (or subsets thereof). Geocoded air pollution exposure will be linked to demographic and health information (HES, THIN) sources to develop impact multipliers for disease conditions and for susceptible population groups. The linkage of data will allow for a more complete dataset when exploring certain demographics such as ethnicity, where there is currently much missing data in primary care alone.

Specific projects listed above have received funding already. For example, workstream 1 (CLAHRC) is an NIHR funded programme. Projects within workstream 2 have also received funding from NIHR. Workstream 3 have received funding from organisations from the British heart foundation, and workstream 5 from the natural environment research council. In addition, UoB anticipate funding from charities such as the British heart foundation and also from industry partners such as Sanofi. HES data will not be used for any commercial purposes. Funders will be not be involved in data processing, study design or interpretation of the results. UoB determines which data are processed for each project/research question and ensures the legal basis for requesting, storing and processing these data are met. UoB will remain the sole data controller therefore it will be the responsibility of UoB to ensure that the data uses are within the scope listed within this agreement. For each project, the data controller UoB will ensure that the project meets the relevant GDPR exemption and be in line with the stated purposes within the scope of this agreement. If any project does not meet this exemption, the project will not be completed using data set out in this agreement.

In the studies mentioned above and planned work using the THIN-HES linked data, UoB anticipate utilising multiple different observational study methods. These will include undertaking cross-sectional, case-control and cohort studies. In addition, UoB look to explore the possibility of time-series methods to evaluate healthcare interventions and policy. As a result of the variety of methods, the exposed and control groups will change dependent on the question being asked. A simple example being the outcomes of cardiovascular disease following a diagnosis of hypothyroidism. In this example, the expose cohort group would consist of women who have experience intimate partner violence identified through Read codes and ICD 10 codes. They will be matched to suitable groups who do not have an exposure code present. Then both sets of patients will be followed up in the dataset until an outcome of interest occurs (cardiovascular disease).

As mentioned above, in order to fulfil the aims of the workstreams above, UoB will require a THIN-HES linked dataset key code. In order to achieve the aims requested, the dataset can be pseudonymised as there is no need for patient identifiable data such as name and address. In order to maximise the usefulness of the dataset, UoB will require the number of years of HES data to match the THIN database. HES data is from 1997 to 2018. There are many scientific and medical reasons why so many years of data are required. In order for real world evidence studies in patient data to be scientifically sound, all information relating to a patient’s past medical events should be considered as this will influence their doctor’s decision and affect their current care. Historical data on patient contact with secondary care is important because the lead up to diagnosis of many conditions, can be complex and lengthy. Real world data can play a very important role in understanding disease incidence in prevalence. It is paramount to have a patient’s entire history to understand incidence of disease, especially within real world data sources. If the amount of available data was reduced this would decrease the number of patients forming cohorts which could impact understanding of disease incidence and progression. As THIN is a national database, in order to link the datasets, HES data such equally match the THIN data geographically. As there are no other national hospital datasets available, there is no other method of joining national primary care and secondary care data.

There are no other organisations involved in the processing of data. However, it is likely that many of the projects utilising the dataset will be connected to funding. Funders will be not be involved in data processing, study design or interpretation of the results.

Should UoB want to do any additional projects they would be required to make an application to amend this agreement and seek approval from NHS Digital.

The UoB will not to attempt to try and re-identify patients taken from the linked dataset. The linkage will be between THIN and HES. A previous method of data linkage conducted by the THIN data providers IQVIA is as follows; a) the patient identifier was encrypted using NHS number in the both provider data sources to create an encryption key; b) the keys were uploaded to a secure website along with the pseudonymised THIN and HES patient IDs; c) the matching keys were linked using only the pseudonymised THIN and HES patient IDs; d) the pseudonymised THIN and HES ID is then used to extract the data for the linked patients. The same approach will be adopted by the University of Birmingham for new datasets if they are to be provided by NHS Digital in the future.

Data processing will only be conducted by substantive employees of the organisation (which also includes the PhD students working on the data) and they will have undertaken appropriate data protection and confidentiality training. Once linked, data delivered to UoB will be accessible using secure servers onsite at the University. In order to access the secure environment, the organisation will ensure that each substantive employee has a secure password to enter the secure environment to view the pseudonymised data and facilitate analysis to happen within this secure environment. Data will be stored on the UoB premises and managed by UoB.

All findings will be disseminated to relevant stakeholders. They will include health care professionals who directly provide care to patients, policy makers, health service administrators and patients.

All projects listed in the agreement have yet to publish any outputs due to delays in provision of data. The initial projects currently underway using this dataset include; “Long term impact of pre-incision antibiotics born by caesarean sections,” “Risk of postpartum haemorrhage among users of SSRI in a cohort of pregnant women with anxiety or depression”, “Postpartum haemorrhage and the incidence of cardiovascular diseases”, “Postpartum haemorrhage and the incidence of mental health conditions”, “Use of progestogen only and combined oral contraceptive pill and risk of postpartum haemorrhage” etc. Once these projects are published, the expected outputs will include several of the following.

Outputs will include but not limited to;

• Reports

• Submissions to peer reviewed journals

• Presentations

• Conferences

• Dashboards

• Government white papers

• Policy documents

Outputs will contain only aggregate level data with small numbers suppressed in line with HES analysis guide.

UoB will aim to facilitate the dissemination of the research to stakeholders during the project and after its completion. The dissemination activities will target an audience of researchers, scientists, innovative technology-focused organisations, research participants. UoB will also aim to make contact with policy makers as a result of the research conducted using the dataset. The nature of the stakeholder groups will be dependent on the project conducted. For example, the project titled “Long term impact of pre-incision antibiotics born by caesarean sections” mentioned above includes clinical and non-clinical stakeholders. These include individuals such as; maternal health clinical staff, the patient and public involvement group involved in the study, professional organisations with an interest in maternal health and also government led maternal and baby units.

The aim of the dissemination activities shall be two-fold: to enable the engagement with the scientific and policy-making communities / to ensure that knowledge developed by the research can benefit these communities. There will be multiple channels of dissemination to this group including journals, websites and social media.

UoB will ensure active communication of results and outputs to any of the interested parties and groups such as the general public or lay members of society. This will occur through channels such as press releases, websites and newsletters to name a few. Results of the work conducted using the joint database will be disseminated in a similar manner to how the results of the THIN dataset work have been shared. These results have been relayed to the UoB press office, whom then advertise the work on the UoB main website. Via the press office, UoB have been able to liaise with journalists to relay information to the general public with multiple studies gaining national news coverage. A recent example has been a study undertaken exploring the association between domestic abuse and mental illness. The results of this study were relayed to the general public via; UoB website, social media, national news channels: BBC, Sky news, ITV, print newspaper: guardian, independent, express, blogs, and even international news channels.

Research taken from linked data will also be used to lead to the development of algorithms, the testing and development of tools and new technologies. This may involve: the creation of evaluation environments for the assessment and validation of processes, tools and technologies; the development of a service offering; the creation of commercial exploitation pathways of outputs when these are tools, new processes and new technologies (during the lifetime of the project or beyond the project’s completion); further research and development in the same/similar context. These future projects can only be undertaken following an amendment to this agreement and subsequent approval from NHS Digital. To reiterate - these future projects are not covered as a processing activity under this version of the agreement.

THIN database has been used extensively to undertake population based medical research studies with the publications in high ranking peer-reviewed journals including BMJ and PLOS Medicine advancing medical knowledge and understanding in both disease management and in public health, capturing the attention of key opinion leaders within the medical communities, practitioners and patients. For example, the study utilising THIN-HES linked data in the development and validation of a frailty index resulted in the index being recommended for use by NICE in the NICE NG56 guideline (Multimorbidity: clinical assessment and management). This work was carried out by the University of Birmingham in a collaboration with the University of Leeds and University of Bradford (and has now been finalised and completed). Therefore, UoB anticipate that the outputs of the linked dataset will lead to numerous high impact publications which will go onto change practice. Examples of high impact journals which the team have been able to publish large dataset work in include; BMJ, JAMA neurology, European respiratory journal, diabetes care and Gut. Therefore, the anticipated work using this combined dataset will continue to be submitted to high impact journals including the BMJ, Lancet series of journals, JAMA and other specialist journals. It is clear that examples of future work using this dataset will provide information that can change practice. For example, the outcomes of the study titled “Long term impact of pre-incision antibiotics born by caesarean sections” will be able to provide insight and information for expectant mothers and clinicians as to the risks/benefits of providing antibiotics at C-section. Each of the streams of work are public health orientated affecting a sizeable proportion of the population. Therefore, if risks are identified or propose mitigating interventions, this can lead to population wide suggestions for health improvement and promotion. Dissemination of results will be in the public interest, as this will act as another method to promote behaviour change to individual patient groups who may benefit directly from the research. As highlighted in the outputs section, information will be delivered to the public with the assistance of the UoB media and communications team, who will assist in providing lay summaries of the work conducted only. This will then be promoted via various media channels such as online news reports and social media.

There are multiple different areas within the proposed workstreams which could lead to measurable improvements and change to name a few:

1) Programme of research with CLAHRC: Service stakeholders to obtain information to improve patient pathways and outcomes across a number of different services, from small changes to reduce delays and improve access to appropriate care, up to and including complete service re-design. Knowledge transfer is a critical part of the CLAHRC philosophy in which capacity in analysis of clinical data is enhanced by service partners, across all areas of the collaboration;

2) Maternal and child health: It is anticipated that the funded caesarean section project will lead to a NIHR Health Technology Assessment report and lay information resource summarising the key findings with respect to the benefits and potential harms on pre-incision antibiotics will provide balanced information to help with decision making regarding the timing of antibiotic prophylaxis. UoB anticipate that the findings will inform the next revision of the NICE Clinical Guideline 132 on Caesarean Section, and hospital policies regarding prophylactic antibiotic administration for Caesarean section. In addition, it is clear that work focusing on abuse such as domestic abuse, may play a role in shaping future government bills such as the upcoming domestic abuse bill, if the project identifies a need for a change in practice.

3) Non communicable disease: This will for example help planning of existing services for cardiovascular health or inform the Public Health England health protection practice relating to infectious diseases.

4) Overdiagnosis, overtreatment and health services research: This workstream will inform the service providers and NHS commissioners of unwarranted variation (e.g. Tonsillectomies), and potential cost savings;

5) Wider determinants of health: An example being the WM-air work will help predict air quality, and will test the health and economic impacts of potential policy measures.

The use of the dataset will also be crucial and integral to multiple post graduate research streams. UoB anticipates that for each of the projects listed, there will be a minimum of 1 post-graduate student assisting with the delivery of each of the research projects - all of the PhD students are employees of the University of Birmingham.

Yielded Benefits is not a requirement for new applications.