Search

Introduction We present a prototype solution for improving transparency and quality assurance of the data linkage process through data provenance tracking designed to assist Data Analysts, researchers and information governance teams in authenticating and auditing data workflows within a Trusted Research Environment (TRE). Methods Using a participatory design process with Data Analysts, researchers and information governance teams, we undertook a contextual inquiry, user requirements interviews, co-design workshops, low-fidelity prototype evaluations. Public Engagement and Involvement activities underpinned the methods to ensure the project and approaches met the public's trust for semi-automating data processing. These helped inform methods for technical implementation, applying the PROV-O ontology to create a derived ontology following the four-step Linked Open Terms methodology and development of automated scripts to collect provenance information for the data processing workflow. Results The resulting Provenance Explorer for Trusted Research Environments (PE-TRE) interactive tool displays the data linkage information extracted from a knowledge graph described using the derived SHP ontology and results of rule-based validation checks. User evaluations confirmed PE-TRE would contribute to better quality data linkage and reduce data processing errors. Conclusion This project demonstrates the next stage in advancing transparency and quality assurance within TREs by semi-automating and systematising data tracking in a single tool throughout the data processing lifecycle, improving transparency, openness and quality assurance.

Synthetic Data has the potential to improve efficiency of data analysis for researchers. However, there is no standard approach to synthetic data governance, access controls or infrastructure requirements, and researchers may face inconsistencies in how they can access or use synthetic data across trusted research environments. We present a federated solution taken by the Scottish Safe Haven Network to address these barriers to facilitate researcher use of synthetic data. We documented and evaluated existing governance pathways, access controls and infrastructure design for non-synthetic data across the Network, recognising uniformity and establishing equivalence using the 5 Safes framework, ISO27001 standards and the SATRE TRE specification. We also interviewed current and potential researchers using our trusted research environments to identify common use cases for accessing synthetic data. We then mapped researcher requirements against the documented equivalencies, validating with current and prospective users. We identified several use cases: to undertake feasibility studies, to understand dataset structure and format and to write analysis code whilst waiting on the project-specific data to be provided. By mapping the use cases onto existing governance and access processes and infrastructure designs, we were able to agree to a standard application process, access control mechanism, and infrastructure platform across the Network to provide a consistent process for researchers. A federated approach to synthetic data access will improve the speed at which research can be conducted as well as improving the transparency and consistency of data governance and access across organisations, ultimately improving the experience for researchers using TREs.

Linking data across local authorities and health boards has the potential to reduce inequalities and improve outcomes. However, cross-agency data sharing leads to complexities around data ownership, legal conditions for processing, organisational controls and risk assessments to ensure privacy and confidentiality of subjects whilst facilitating research and service improvements. We share a governance framework for cross-agency data sharing between a local authority and regional health board in Scotland. A cross-sector Child Neglect working group was established with representation from public sector agencies, the NHS, University researchers and a Trusted Research Environment to support research on children’s mental health prescribing and secondary specialist services provision and linking to the Aberdeen City Council Child Protection Register. Lessons learned were captured throughout the governance, data transfer and data linkage phases. Communication is the most important driver, with key individuals, deputies and routes of escalation agreed and involved throughout for clarity of roles and expectation setting. Key lessons learned include requiring input from all parties at the project initiation stage, particularly IG teams, including agreements on necessary templates. Knowledge exchange is required between the research team, data owners and data processors to ensure research protocol requirements are correctly understood and data transfer is supported. If possible, internal and/or external validation of data should also be performed before data is linked. Cross-agency data sharing is a vital means to understanding and improving outcomes for individuals and populations. Critical to this is establishing an action plan and formalised project process to achieve outcomes.

We present a prototype solution for improving transparency and quality assurance of the data linkage process through a data provenance dashboard designed to assist data analysts, researchers and information governance teams in authenticating and auditing data workflows within a trusted research environment (TRE). Building on prior research (Scheliga, et al., 2022), this work describes our first operationalised prototype tested in a real-world setting. The prototype development involved four stages: (1) Co-design of interfaces with end users for provenance data collection and visualisation, producing a low-fidelity design; (2) Extension and refinement of the Safe Haven Provenance (SHP) ontology (https://tre-provenance.github.io/SHP-ontology/releases/v0.2/index-en.html); (3) Design and implementation of mechanisms for semi-automated collection of data linkage provenance using the SHP ontology; (4) Implementation and user evaluation of the dashboard. A participatory design process with data analysts, researchers and information governance teams resulted in a low-fidelity prototype and was validated via public consultations to ensure it met public trust. The resulting prototype dashboard (https://tre-provenance.github.io/) was built as an offline desktop app to match the deployment TRE requirements. The interactive dashboard displays the data linkage information extracted from a knowledge graph described using the SHP ontology (e.g., modifications of datasets, data release) and results of rule-based validation checks (e.g., checking extracted data against researchers' specification). User evaluations confirmed the dashboard would contribute to better quality of data linkage. This project demonstrates the next stage in advancing transparency and quality assurance within TREs by semi-automating and systematising data tracking from ingress to egress in a single tool.

Abstract Background One in eight children in the United Kingdom are estimated to have a mental health condition, and many do not receive support or treatment. The COVID-19 pandemic has negatively impacted mental health and disrupted the delivery of care. Prevalence of poor mental health is not evenly distributed across age groups, by sex or socioeconomic groups. Equity in access to mental health care is a policy priority but detailed socio-demographic trends are relatively under-researched. Methods We analysed records for all mental health prescriptions and referrals to specialist mental health outpatient care between the years of 2015 and 2021 for children aged 2 to 17 years in a single NHS Scotland health board region. We analysed trends in prescribing, referrals, and acceptance to out-patient treatment over time, and measured differences in treatment and service use rates by age, sex, and area deprivation. Results We identified 18,732 children with 178,657 mental health prescriptions and 21,874 referrals to specialist outpatient care. Prescriptions increased by 59% over the study period. Boys received double the prescriptions of girls and the rate of prescribing in the most deprived areas was double that in the least deprived. Mean age at first mental health prescription was almost 1 year younger in the most deprived areas than in the least. Referrals increased 9% overall. Initially, boys and girls both had an annual referral rate of 2.7 per 1000, but this fell 6% for boys and rose 25% for girls. Referral rate for the youngest decreased 67% but increased 21% for the oldest. The proportion of rejected referrals increased steeply since 2020 from 17 to 30%. The proportion of accepted referrals that were for girls rose to 62% and the mean age increased 1.5 years. Conclusions The large increase in mental health prescribing and changes in referrals to specialist outpatient care aligns with emerging evidence of increasing poor mental health, particularly since the start of the COVID-19 pandemic. The static size of the population accepted for specialist treatment amid greater demand, and the changing demographics of those accepted, indicate clinical prioritisation and unmet need. Persistent inequities in mental health prescribing and referrals require urgent action.

There has been a recent emphasis to establish and codify large-scale or national Trusted Research Environments (TREs) in the United Kingdom, with a view to limit smaller, local TREs. The basis for this argument is that it avoids duplication of infrastructure, information governance, privacy risks, monopolies and will promote innovation, particularly with commercial partners. However, the work around establishing TREs in the UK largely ignores the long-established local TRE landscape in Scotland, and the way in which local TREs can actually improve data quality, solve technical architecture challenges, promote information governance and risk minimisation, and encourage innovation and collaboration (both academic and commercial). This data centre profile focuses on the Grampian Data Safe Haven (DaSH),a secure, virtual healthcare data analysis and storage centre located in Aberdeen, Scotland. DaSH was co-established by the NHS Grampian Health Board and University of Aberdeen to allow for the secure processing and linking of health data for the Grampian and Scottish population when it is not practicable to obtain consent from individual patients. As an established trusted research environment now in its 10th operating year, DaSH technology ensures healthcare, social care data and other types of sensitive data, routinely collected and used without individual patient consent, are made accessible for both academic research and clinical service evaluation and improvements whilst protecting individuals' privacy at the local, national and international levels. DaSH has registered almost 600 projects and facilitated over 200 distinct research projects with data hosting, extraction, and novel linkages to completion. Ongoing innovation and collaboration between DaSH and the NHS Grampian Health Board continues to expand researcher access to new types of data and data linkages, introduce new technologies for advanced statistical research methods, and supports interdisciplinary research using population health and social care data for research, clinical and commercial advancements, and real-world practitioner applications. The purpose of this paper is to present DaSH's data population, operating model, architecture and information technology, governance, legislation and management, privacy-by-design principles and data access, data linkage methods, data sources, noteworthy research outputs, and further developments in order to demonstrate the value of local TREs within the data management and access debate.

Objectives Scotland has an established Trusted Research Environment (TREs) through the Scottish Safe Haven Network. These Safe Havens traditionally service structured datasets, however, researchers increasingly require access to large multi-modal linked datasets that include medical imaging. We therefore introduced an equivalent to access anonymised NHS images and reports. Approach A pan-Scotland collaboration of 15 partners from industry, NHS, and academia, collaborated to design and deploy a Safe Haven equivalent for SMEs and researchers interested in accessing anonymised NHS imaging and reports to allow them to develop, test, and validate AI algorithms for greater patient benefit. Two Safe Haven sites, worked with a leading medical software research and development team, to deliver a secure analytical platform for the research and development of AI for medical applications. Results The Safe Haven Artificial Platform (SHAIP) is designed to support development of sophisticated AI components and has been used by several SME’s to undertake exemplar projects in stroke, breast screening, and x-ray interpretation for limb fracture and chest. These researchers were supported by the Safe Haven’s governance processes. SHAIP became the first research environment to be directly linked to the Scottish National PACS archive. Access was provided to project specific deidentified images using ‘Hidden in plain sight’; a privacy and data-structure preserving technique. The researchers were able to upload virtual-machines, or ‘dockers’, to bring their toolsets to their research workspaces. An annotation capability was provided to support ground-truth development for machine learning. Conclusion The project has shown we can pull large scale medical images and reports into a TRE in such a way NHS Boards are reassured by our methods, creating a safe analytical platform for AI development that will benefit patient care by providing faster, cheaper and more accurate solutions.

Objectives Research using unconsented healthcare data has the potential to have significant public benefit in terms of improved health outcomes. However, due to the sensitive nature of this data, appropriate governance, approvals, and mitigations are crucial to ensure that patient confidentiality and public trust are maintained. Approach As the permissions process can be a barrier to research, we established a local Governance Forum with the aim to streamline processes, increase transparency, and document approval requirements. Governance Forum membership includes representatives from the relevant health board and university departments (e.g., the regional Safe Haven, Information Governance, Ethics Committee, Data Protection, Research and Development). To achieve the Forum’s objectives, the regional Safe Haven raises current and potential project scenarios at quarterly meetings to confirm requirements and help develop a clear understanding of the approval pathway, with specific topics discussed with the relevant members out with these meetings as required. Results The Forum has strengthened communication and collaboration with Forum members which has enabled the Safe Haven to better support researchers. Outputs have included a permission pathway guidance document for researchers. This initial guidance document outlines approval requirements and provides relevant advice and support, with work currently ongoing to build on and improve this. In addition, as channels have been established for onboarding new project types and novel datasets, the diversity of datasets available for research is increasing and the route to approvals is now clearer for both researchers and internal teams. The Governance Forum activity has the potential to facilitate future research which may lead to academic publications and research funding, as well as improved health outcomes and policy change in the longer-term. Conclusion Although the progress so far has had significant impacts, the approval processes remain challenging for certain projects. Therefore, it will be imperative that sufficient resources are available in the future to ensure that further progress can be made via the Governance Forum as well as national initiatives.

Objectives Building upon the Scottish Safe Haven Network’s (SSHN) collaborative experience to develop an exemplar dataset on Scottish national level laboratory tests within the SSHN. To test the feasibility of a federated safe haven network in Scotland for clinical laboratory data using existing federation solutions. Approach Considerations were given to the clinical tests that are being commonly used across Scotland. An investigation of laboratory data structure was conducted using SQL with the assistance of White Rabbit which captures the metadata information from each safe haven. Data heterogeneity or commonalities were investigated in detail in R. We examined the inter-relationships among clinical laboratories from different regions of NHS Scotland. A common data structure was proposed to facilitate the sharing of clinical test results. We investigated multiple existing federation solutions to streamline access to data across the SSHN. Results A dataset of all laboratory tests for patients registered with the Scottish Health Research Register and Biobank (SHARE) from laboratory data sources within the network (Grampian Data Safe Haven, Health Informatics Center, Glasgow Safe Haven & Lothian/DataLoch Safe Haven) were developed. An open-source toolkit that includes SQL scripts to harmonise laboratory data from multiple safe haven and R scripts to conduct heterogeneity investigation to streamline the sharing and the analysis was created. We have shown a working model for federation within Scotland for the first time with the view to expand into more routine projects. The project has developed a systematic concept mapping of data available across the SSHN to aid cohort discovery. Conclusions The exemplar national level laboratory data together with the toolkit will help provide researchers with information such as which data are available to access. It will also support and improve national-level research studies at a faster pace. The work has demonstrated the feasibility of a common process for federated data discovery and sharing across the SSHN.

Objective To protect privacy, routinely-collected data are processed and anonymised by third parties before being used for research. However, the methods used to do this are rarely shared, leaving the resulting research difficult to evaluate and liable to undetected errors. Here, we present a provenance-based approach for documenting and auditing such methods. Approach We designed the Safe Haven Provenance (SHP) ontology for representing provenance information about data, users, and activities within high-security environments as knowledge graphs. The work was based on a case study of the Grampian Data Safe Haven (DASH) which holds and processes medical records for 600,000 people in Scotland. The SHP ontology was designed as an extension to the standard W3C PROV-O ontology. The auditing capabilities of our approach were evaluated against a set of transparency requirements through a prototype interactive dashboard. Results We demonstrated the ability of the SHP ontology to document the workflow within DASH: capturing the extraction and anonymisation process using a structured vocabulary of entities (e.g. datasets), activities (e.g. linkage, anonymisation) and agents (e.g. analysts, data owners). Two provenance reporting templates were designed following interviews with DASH staff and clinical researchers: 1) a detailed report for use within DASH for quality assurance, and 2) a summary report for researchers that was safe for public release. Using a prototype data-linkage project, we formalised queries for report generation, and demonstrated use of automated rules for error detection (e.g., data discrepancies) using the structure of the SHP knowledge graphs. All of the project outputs are available under an open-source license. Conclusions This project lays a foundation for more transparent high-quality research using public data for health care and innovation. The SHP ontology is extendible for different domains and potentially represents a key component for further automation of provenance capture and reporting in high-security research environments.

Objective In March 2020, Scottish government identified people clinically extremely vulnerable to COVID due to pre-existing health conditions. These people were advised to strictly self-isolate (shield) at home. We examined who was identified as clinically extremely vulnerable, how their healthcare changed during isolation, and whether this process exacerbated healthcare inequalities. Approach We linked all individuals on the shielding register in NHS Grampian to their in-patient and out-patient healthcare records from 2015 through 2020. We analysed the method of patients’ identification as clinically extremely vulnerable (via an algorithmic NHS record scan or designated ad hoc by their care-providers). We measured out-patient, in-patient, and emergency healthcare attendances, and compared use rates between two 3-month periods before and during the first strict isolation period. We evaluated changes in care use between those shielding and the general non-shielding population, and differences between shielding sub-populations (by clinical reason for shielding, age, sex, and socio-economic deprivation). Results The shielding register included 16,092 people (3% of the population). 42% of people on the register were not identified by national healthcare record screening, including the majority of cancer and immunocompromised patients. People added to the register by their care-providers were more likely to be young and less economically-deprived. Shielders’ healthcare use decreased during isolation (rate compared to pre-isolation: 0.65 out-patient, 0.54 scheduled in-patient; 0.75 emergency in-patient; 0.71 A&E). However, people shielding had better maintained care than the non-shielding population (e.g. RR 2.9 for scheduled in-patient care). There were inequalities in whose scheduled care was maintained while shielding: younger people and those with cancer had significantly higher visit rates. However, there were no differences in care-preservation between men and women or between socioeconomic deprivation levels. Conclusions The reliance on emergency care while shielding indicates that, overall, continuity of care for existing conditions was not optimal. However, there was notable success in maintaining care for cancer. We suggest that integrating demographic and primary care data would improve identification of the clinically vulnerable and help equitably prioritise care.

Objective 1 in 8 young people in the United Kingdom are estimated to have a diagnosable mental health condition. Prevalence is increasing over time, many are untreated, and need is not evenly distributed across the population. We aimed to investigate trends in children’s mental health prescribing and referrals to specialist outpatient services. Approach We linked individual-level healthcare administrative records on community prescribing and referrals to outpatient Child and Adolescent Mental Health Services (CAMHS). The study cohort included all children aged 2 through 17 in the NHS Grampian Health Board region from 2015 to 2021 (average annual population circa 100,000) with a mental health prescription or CAMHS referral. We measured prevalence of mental health prescribing and referrals to CAMHS over time. We investigated demographic and socioeconomic differences, including comparison of rates by age, sex, and residential area deprivation. We also investigated socioeconomic and demographic differences in referral acceptance and rejection. Results Prescriptions for mental health drugs have risen 40%: from 5,000 per month in 2015 to 7,000 in 2021. 75% of prescriptions to primary schoolers are to boys, mostly for attention deficit hyperactivity disorder medications. Prescriptions to girls rise during secondary school, mostly for anti-depressants. Prescribing rates are 2.6-fold higher in the most versus least deprived areas. Referrals to CAMHS have risen 20% over the study period, and the proportion of referrals rejected has increased from 18% to 31% – leaving the number of children accepted to specialist care stable. Boys are referred twice as often at younger ages, while girls’ referrals spike during puberty. Since 2015, boys have been referred less and rejected more, with girls now making up 61% of those treated. Referral rates are two-fold higher in the most versus least deprived areas. Conclusions Both mental health prescribing and referrals to CAMHS have increased in this population, but the CAMHS service size remained fixed. Presentation and treatment patterns vary dramatically by age and sex, and socioeconomic inequalities are clear and persistent.

Objectives To determine the risk of misidentification when using a “Hidden In Plain Sight (HIPS)” Named Entity Recognition (NER) de-identification methodology applied to Scottish healthcare data within The Industrial Centre for Artificial Intelligence Research in Digital Diagnostics (iCAIRD) Safe Haven Artificial Intelligence Platform (SHAIP). Approach Rather than the traditional redaction of potential identifiable information in routinely collected healthcare data, our HIPS methodology utilises an NER “find and replace” approach to de-identification that keeps the structure of text intact. This ensures that context is maintained, key to the interpretation of free text information and potential Artificial Intelligence applications. To our knowledge these methods have been previously untested on Scottish healthcare data. We therefore performed assessment of this approach in terms of potential risk of misidentification using HIPS on structured Scottish data deployed in SHAIP as part of the iCAIRD programme. Results Five individual cohorts, with a total of 169,964 patients were included. For each cohort the HIPS approach was applied, and then compared to actual patient information from within the same region, in order to determine the risk of misidentification. The following fields were included: Forename, Surname, Previous Name, Gender, Date of Birth (DOB), and Postcode. Across the five cohorts and varying combinations of identifiable data fields there were a total of 94 instances of potential misidentification (0.06%). 85/94 (90.4%) of these were for the combination of Gender, Date of Birth and Postcode. Across the five cohorts there were only 3 instances (0.002%) of Forename/Surname/DOB, and 5 instances (0.003%) of Forename/Surname/Postcode potential misidentification amongst the 169,964 patients. Conclusions The iCAIRD NER HIPS Methodology provides an acceptably low misidentification rate. Further work is now required to determine the recall and precision rates. Benefits of this approach include retaining the structure of free text, as well as reducing the ability to detect any potential leaked identifiable data.

For over a decade, Scotland has implemented and operationalized a system of Safe Havens, which provides secure analytics platforms for researchers to access linked, deidentified electronic health records (EHRs) while managing the risk of unauthorized reidentification. In this paper, a perspective is provided on the state-of-the-art Scottish Safe Haven network, including its evolution, to define the key activities required to scale the Scottish Safe Haven network’s capability to facilitate research and health care improvement initiatives. A set of processes related to EHR data and their delivery in Scotland have been discussed. An interview with each Safe Haven was conducted to understand their services in detail, as well as their commonalities. The results show how Safe Havens in Scotland have protected privacy while facilitating the reuse of the EHR data. This study provides a common definition of a Safe Haven and promotes a consistent understanding among the Scottish Safe Haven network and the clinical and academic research community. We conclude by identifying areas where efficiencies across the network can be made to meet the needs of population-level studies at scale.