NHS Digital Data Release Register - reformatted

Moorfields Eye Hospital NHS Foundation Trust projects

83 data files in total were disseminated unsafely (information about files used safely is missing for TRE/"system access" projects).

Detecting Dementia in the Retina: a Big Data Machine Learning Approach — DARS-NIC-116883-L8W9Q

Type of data: information not disclosed for TRE projects

Opt outs honoured: Yes - patient objections upheld, Anonymised - ICO Code Compliant, Yes (Section 251 NHS Act 2006)

Legal basis: Health and Social Care Act 2012 – s261(1) and s261(2)(b)(ii), Health and Social Care Act 2012 – s261(1) and s261(2)(b)(ii), Health and Social Care Act 2012 – s261(2)(b)(ii), Health and Social Care Act 2012 - s261 - 'Other dissemination of information'

Purposes: No (NHS Trust)

Sensitive: Non Sensitive, and Non-Sensitive

When:DSA runs 2019-01-14 — 2022-01-13 2019.06 — 2019.11. breached contract — audit report.

Access method: One-Off

Data-controller type: MOORFIELDS EYE HOSPITAL NHS FOUNDATION TRUST, UNIVERSITY COLLEGE LONDON (UCL)

Sublicensing allowed: No

Datasets:

  1. Hospital Episode Statistics Outpatients
  2. Hospital Episode Statistics Accident and Emergency
  3. Hospital Episode Statistics Admitted Patient Care
  4. Hospital Episode Statistics Accident and Emergency (HES A and E)
  5. Hospital Episode Statistics Outpatients (HES OP)
  6. Hospital Episode Statistics Admitted Patient Care (HES APC)

Objectives:

The Moorfields Eye Hospital NHS Foundation Trust (MEH) and University College London (UCL) require HES data to investigate the association between changes of the retina, as measured using retinal photography and scans, with the onset of dementia. The only purpose is for medical research.

Moorfields Eye Hospital NHS Foundation Trust and The University College London (UCL) are joint Data Controllers. The University College London School of Life and Medical Sciences is the Data Processor for the purpose of this research.

A study working group, termed the AlzEye working group, consisting of the Chief Investigator, co-investigators, the Trust Caldicott Guardian and representatives from information technology and information governance will convene quarterly to evaluate the management and ongoing progress of the trial. The study working group will not control the purpose for or manner in which the data are processed. The role of the working group is for study monitoring and advising on interpretation of the results.

The proposed data processing is in line with Article 6(1)(e) ‘processing is necessary for the performance of a task carried out in the public interest’. Dementia affects more than 800,000 people in the UK alone and with the progressive shift in the age of the population, these numbers are likely to increase over the next decade. As deemed by the Confidential Advisory Group, there is strong public interest in the activity proceeding as ‘any advancement in detecting the onset of dementia had the potential for significant wider benefit’.

Public interest is in line with Article 9(2)(j) ‘processing is necessary for scientific or historical research purposes’.

Risks to participants' confidentiality arise through the transfer of identifying data to NHS Digital. This will be mitigated by closely consulting with NHS Digital and adhering to all data privacy standard operating procedures. In the worst case scenario, this would identify an individual as having had a retinal scan at some point in the last 10 years at Moorfields Eye Hospital. Subsequent data processing will take place on pseudonymised records and it is not envisaged that this would pose a risk to the individual participants.

The focus of this newly undertaken project is the development of a screening tool for the early detection of dementia. The most common form of dementia, Alzheimer’s Disease (AD), affects 26 million people globally, a figure expected to quadruple by 2050.

The most significant features of AD are the accumulation of plaques and tangled proteins in the central nervous system. The optic nerve and retina develop from the same embryonic tissue as the brain, and are thus a sensory extension of the central nervous system. As the only structure of the central nervous system not covered by bone, the retina provides unique access to direct imaging, which can be achieved using optical coherence tomography (OCT).

A number of small studies have identified morphological changes in the retinas of patients who have developed AD. However, plaque deposits in the brain can occur 15 years before the onset of clinical symptoms, and there is evidence to suggest that these plaques can be seen to form in the internal layers of the retina even before this.

The project objective is to characterise the changes in retinal parameters associated with the diagnosis and development of dementia, particularly Alzheimer’s Disease. The data required includes HES labels of neurodegenerative disease from the HES outpatients and A&E database. Additional covariate labels of diabetes mellitus and cardiovascular disease are requested for cohort matching as these other conditions can affect retinal structure. Pseudonymised patient record level HES data is required over a 10 year period in the London area to allow longitudinal analysis of scans in individuals developing neurodegenerative disease. Due to the large number of patients included, the historical nature of the data, advanced age, risk of selection bias and difficulty in contacting patients, it would not be feasible to obtain consent from individuals. As a result, approval has been obtained from the Health Research Authority Confidential Advisory Group to process confidential information without consent under section 251 of the NHS Act 2006.

The purpose for processing the data pertains to a stand-alone project and the data will allow the largest and most in-depth analysis of retinal changes in dementia to be carried out. Cases will be defined as those with a HES label of Alzheimer’s disease (AD) and will be compared with controls. In addition, longitudinal analysis of cases will be analysed to identify retinal changes associated with an eventual diagnosis of AD.

Information is requested on individuals over the age of 40, who have attended Moorfields Eye Hospital and had a retinal scan between 01/01/2008 and 01/06/2018. In collaboration with machine learning partners within UCL, the Moorfields Eye Hospital propose to analyse their repository of more than 2 million retinal scans performed regularly on patients since 2008. By pseudonymously linking these scans, at a patient level to data from the Hospital Episode Statistics database, to identify those patients who went on to develop AD, an algorithm can be trained to identify the patterns of retinal changes which are associated with the development of AD. Additionally, by linking the retinal images to a set of confounding variables, associated with the development of AD, a closely matched cohort of control scans which do not correspond to an AD diagnosis can be constructed.

MEH's and UCL's aim is for UCL to perform a pseudonymised linkage of retinal imaging data from Moorfields Eye Hospital to the HES database, in order to create a pseudonymised dataset of retinal images. This will be a database of retinal images linked at an individual image level to corresponding diagnoses of neurodegenerative disease. Cohorts of images will then be matched on the study ID.

This pseudonymised database will be maintained by the University College London Institute of Ophthalmology, a department within UCL's School of Life and Medical Sciences.

The initial outcome of this project will be a substantial research image database with relevant labels of neurodegenerative disease – 2-3 orders of magnitude greater in size than that which currently exists. The primary outcome from Moorfields Eye Hospital will be a comprehensive description of the morphological features of neurodegenerative disease in the retina. The ultimate outcome of this project will be a machine learning derived algorithm capable of identifying features suggestive of the development of dementia on the basis of an OCT scan of the patient’s retina. This system can then be evaluated with a prospective observational study to assess its applicability outside of Moorfield Eye Hospital's dataset.

The value of a reliable screening tool for AD would be vast; by facilitating intervention early in the disease process it would lead to improvements in terms of quality adjusted life years, and reduce mortality and the economic burden of caring for functionally impaired patients.

Funding comes from a small grant award by Fight for Sight UK and Alzheimer’s Research UK. The project methodology has been peer-reviewed by their respective Grant Committees and deemed worthy of award. The funders will neither have access to the requested data nor be involved in the processing of data but are likely to be involved in facilitating dissemination of outputs from this project. The funders will not influence of suppress the findings of this research.

Expected Benefits:

Dementia affects more than 800,000 people in the UK alone. The most common form, Alzheimer’s disease, affects 26 million people globally, a figure expected to quadruple by 2050. While there are currently no cures for most types of dementia, early diagnosis can help patients receive the appropriate treatment and support to help maintain mental function. By characterising changes in the retina associated with dementia, this project seeks to identify early biomarkers of this disease in patients. This would lead to improvement in quality-adjusted life years and reduce mortality, and the economic burden of caring for functionally impaired patients.

In 2016, dementia overtook cardiovascular disease as the leading cause of death in the UK. With the progressive shift in the age of the population, these numbers are likely to increase over the coming decades. At the same time, it is noted that 50-80% of people with the most common form of dementia, Alzheimer’s disease, are not diagnosed in the developed world.

Through research to identify retinal biomarkers of dementia, there is potential to improve the diagnosis rates of a condition, where early recognition can significantly improve quality of life and disease progression. Moreover, in an era where novel therapies are being developed for many forms of dementia, the use of structural biomarkers using a non-invasive scan would be of significant benefit in evaluating responses to therapy.

While there has been considerable interest in studying retinal changes in people with dementia, most large-scale studies have thus far focused on the correlation between retinal thickness and cognitive function, as measured by a catalogue of tests incorporating memory assessment and numerical reasoning, rather than a specific diagnosis of dementia. The smaller number of studies with specific AD labels have predominantly been cross-sectional studies with descriptive outcomes. Moreover, few are longitudinal with the only such report to demonstrate the increased risk of developing AD with thinner retinal nerve fibre layer, the Rotterdam Study, including only 86 positive cases. This study will therefore address an important question with potentially substantial public impact. Dissemination of results in the scientific community will inform the research of other academic groups exploring the association between dementia and the retina. Moreover, retinal biomarkers of other neurological conditions, for example multiple sclerosis, are now being considered as trial outcome measures for novel interventions. A similar situation is anticipated for the results of the proposed study. Depending on the results, prospective evaluation of retinal scans in patients developing dementia would be sought to establish a potential screening test.

The implications of this research will have substantial public benefit but furthering understanding of retinal manifestations of dementia and the utility in using these parameters in predictive modelling of disease development.

Moorfields Eye Hospital and UCL have substantial experience in disseminating findings to patients, the public and health policy makers. Moreover, the involvement of charitable organisations, such as Fight for Sight UK and Alzheimer’s Research UK, will further facilitate result dissemination to members of the public.

In expanding understanding of retinal signs indicative of neurodegenerative disease, there is potential to improve risk stratification of those likely to develop these conditions as well as identify novel biomarkers, which can be used to structurally define disease progression. The magnitude is significant – dementia has become the leading cause of death in the UK and numbers are only likely to increase with the ageing demographic of the population. Moreover, dementia is under diagnosed with estimates of 50-80% of patients being missed in developed countries.

Given the global impact of the results of this study, other research groups, such as the Department of Epidemiology Team at the University Medical Centre, Rotterdam, will benefit from the outputs of this study by informing their research.

Benefit will be measured through scientific output of peer-reviewed articles and conference presentations.

The value of a reliable screening tool for Alzheimer’s disease would be vast; by facilitating intervention early in the disease process it would lead to improvements in terms of quality adjusted life years, and reduce mortality and the economic burden of caring for functionally impaired patients.

It should be technologically feasible to begin the implementation of such a screening tool in clinical trials by 2021.

Outputs:

The initial outcome of this project will be a substantial research image database, containing NHS Digital data, with relevant neurodegenerative disease labels. The database will be completed by March 2019 and access to this database will be limited to UCL School of Life and Medical Sciences.

The primary outcome will be a comprehensive description of the morphological features of dementia in the retina, including details on how retinal morphology evolves throughout the disease course. These will form the basis for a series of submission to peer reviewed journals (e.g. NEJM & Nature) from early 2020 onwards.

Scientific findings from this project will be communicated through presentations at both national and international conferences, which extend beyond ophthalmology into cardiology and neurology. Results will be submitted to peer-reviewed journals across a range of fields including ophthalmology, neuroscience, epidemiology and cardiology. As the study has been peer-reviewed and received funding from Fight for Sight UK and Alzheimer’s Research UK, results may also be presented at forums supported by these charities. Presentations will be by members of the research team but may also come from two members of the public, who will sit on the Working Group of the study.

Engagement with a wide range of stakeholders will be sought. These goals are facilitated by the Chief Investigator, who has extensive experience in communicating research findings with the public across a range of settings as well as the Patient and Public Involvement and Engagement (PPIE) team at the NIHR Biomedical Research Centre at Moorfields Eye Hospital-University College London. Significant PPIE has already contributed to the establishment of the AlzEye database and been complimented by the UK Confidential Advisory Group (CAG) of the NHS Health Research Authority (HRA). In addition to surveying nearly 500 interested lay members at Moorfields Eye Hospital into the methodology of AlzEye, two members of the public will sit on the working group and dissemination of results is planned through both medical and public forums.

The conclusions of this study will also be of interest to non-medical research groups exploring linked health research, predictive modelling and implications of real-world studies on shaping public policy. The group also plans to present study findings at pertinent meetings for public health researchers.

To engage with members of the public, results of the study will be presented at research open days at Moorfields Eye Hospital, seminars and symposia within University College London and public meetings. The methodology of this study will be presented to the open public science initiative, Café Scientifique, which facilitates interactions between researchers and the public in Spring 2019. To further promote public interest and communication, two lay members from the AlzEye working group will also be offered the opportunity to present study results at meetings, such as those hosted by Fight for Sight UK. Digital settings will also be explored with lay summaries uploaded to websites - details of the study are already available on the Moorfields Eye Hospital and Fight for Sight UK websites as well as the public database, clinicaltrials.gov. These organisations additionally have well-established social media departments frequently used for dissemination of results.

The outputs from this study will be under the ownership of the Sponsor, Moorfields Eye hospital.

Initial results from this study are anticipated to be disseminated 1 year following receipt of data from NHS Digital, therefore February 2020.

The ultimate outcome of this project will be a machine learning derived algorithm capable of diagnosing pre-symptomatic dementia on the basis of an OCT scan of the patient’s retina. This is expected to be completed in 2020-2021. Thereafter this screening tool can be evaluated with a prospective observational study to assess its applicability outside of MEH's dataset, with a view to incorporation into future clinical trials. Details of this data analysis tool will be publicly published in peer-reviewed journals.

All published data will be in an aggregated form with small numbers suppressed in line with HES analysis guide.

Processing:

This research project requires access to HES data in order to distinguish those retinal scans associated with a diagnosis of dementia from retinal scans of unaffected patients. In addition, for the purposes of cohort matching, HES labels of potential confounding variables such as cerebrovascular disease and diabetes mellitus are requested. This stratification aims to minimise the effect of the most significant confounding variables on the analysis, while also making efforts to minimise the quantity of data being requested for each patient.

Moorfields Eye Hospital will send the identifying details of a cohort of patients who had a retinal scan between 01/01/2008 and 01/06/2018 to NHS Digital. The cohort will be restricted to patients who were over 40 at the time of their scan and identifying details will include:
- NHS numbers;
- Dates of Birth;
- Gender, and
- a Study ID

These details will be transferred to NHS Digital from Moorfields Eye Hospital (MEH), corresponding to each of the patients within their imaging dataset, for whom they are requesting data. This data is transferred with support under section 251 of the National Health Service Act 2006. NHS Digital will extract the requested HES data labels for the individuals whose details were provided.

NHS Digital will remove identifying details and send pseudonymised data including the Study ID, gender and the corresponding HES data (health data) to the UCL School of Life and Medical Sciences Data Safe Haven, which has no access to the original identifying details sent to NHS Digital.

MEH will export a pseudonymised image dataset containing the unique study ID and ophthalmic data to the UCL Data Safe Haven. UCL, acting as a data processor, will link the pseudonymised imaging dataset from MEH with the pseudonymised diagnostic data from NHS Digital using the unique study ID. UCL creates two cohorts of patients, differentiated by the presence of a diagnosis of neurodegenerative disease. These cohorts will be matched on the basis of age, gender, and risk factors for the development of dementia.

Retinal scans are non-identifiable and the members of the study team at UCL will only have access to pseudonymised data. There is no anticipated risk of re-identification. MEH will not have access to the data received by NHS Digital. The encryption key will not be retained by MEH so it would not be possible to backtrack from the Study ID to re-identify an individual. Theoretically, rare forms of neurodegenerative disease, which affect few individuals might allow re-identification if one had access to another database listing patients in London with these conditions. Although this is not possible as the members of the study team at UCL will not have access to these other databases, the study team has mitigated this risk even further by limiting analysis of the received data only to cohorts, categorised by ICD10 codes, exceeding 50 patients. Data will be removed where the cohort includes ICD10 codes affecting less than 50 patients

No data will be accessed outside of the UK.

Data is being requested for a long timescale (2007/018 to 2017/18) in order to capture all of the patients within MEH's existing imaging database. By studying patients over several years it will be possible to observe how changes in the retinal scans evolve prior to the development of dementia.

UCL will perform a hierarchical series of statistical methods, to compare the retinal scans of patients with dementia to a matched cohort of those with no dementia diagnosis. Automatic OCT segmentation will be used to detect degenerative lesions manifesting in retinal nerve fibre layer thinning, macular volume loss and retinal ganglion cell layer thinning. Image analysis will then be undertaken which will focus on the detection of plaque deposits, and the localisation of these deposits and other degenerative lesions. Finally, a supervised machine learning algorithm will be developed. This will be employed with dementia as a training label, in order to construct a combined dementia risk score on the basis of an array of OCT changes.

Data processing will only be carried out by members of the research team at UCL - all of whom are substantive employees of UCL - and will be limited to processing within the UCL's firewall. All such members are required to have appropriate training in data protection, confidentiality and information governance. Moreover, access to the data will be regularly audited by the Study Working Group.

Data will be accessed through the UCL School of Life and Medical Sciences Data Safe Haven. Access is via a remote desktop arrangement. Access is controlled. The Data Safe Haven is subject to external professional penetrating testing on an ongoing basis. Failed logon attempts are recorded in the Data Safe Haven system and are managed by the Data Safe Haven Service Operation Manager. Intrusion attempts and port scans are detected and reported to the UCL security function for investigation as necessary.

The data will not be linked with any data other than as described in this Agreement.