National Institute for Health Research Health Informatics Collaborative: development of a pipeline to collate electronic clinical data for viral hepatitis research

Opportunities

• The UK’s unified healthcare system (the NHS) generates millions of clinical datapoints each year, which can be leveraged to improve collection of clinical information, address clinical research questions and improve patient care.

• The automated collection of data from electronic patient record systems can dramatically reduce the time and cost of data collection for research and provide opportunities for collaboration with both academic and industry partners.

• Modern machine learning techniques using neural networks require large datasets to be used effectively,2 the reuse of routinely collected data can provide a cost effective way of collating these datasets.

Challenges

• All NHS trusts are separate organisations, responsible for the protection of the data of their own patients. To enable data to be shared across these separate organisations for research, a governance framework needed to be established.

• Each NHS trust has its own electronic patient record, and its own set of customisations, extensions and variations in data entry practice. Alongside the primary electronic patient record, each trust will also have an extensive collection of departmental systems, again subject to customisation and variations in practice.

• Data definitions are not all standardised.

• Not all data are collected electronically at all sites.

• Large amounts of important data are stored in free text rather than discrete values.

• Clinical practice can differ between sites.

• Data can be produced and collected differently between sites. (eg, different laboratory methods or platforms used for tests).

• Different trusts have different levels of expertise in clinical informatics.

• Projects such as the NIHR HIC require sustained investment before they start to deliver tangible results.

The NIHR HIC aims to overcome these challenges and demonstrate the value of these data for research in key therapeutic areas; the first five areas considered were viral hepatitis, ovarian cancer, critical care, acute coronary syndromes and renal transplantation. This paper focuses on the viral hepatitis theme, which is led by Oxford University Hospitals NHS Foundation Trust.

Viral hepatitis is a global health problem with an estimated 1.35 million people dying from either end-stage liver disease, hepatocellular carcinoma or other viral hepatitis-related diseases in 2015.3 The majority of these deaths are as a result of hepatitis B virus (HBV) and hepatitis C virus (HCV) infections; this is greater than tuberculosis, HIV or malaria. Unlike these other infections, the number of viral hepatitis deaths has increased since 1990.4 International targets arising from the United Nations ‘sustainable development goals’ have set a challenge for the elimination of viral hepatitis as a public health threat by the year 2030.5 6 As part of meeting this goal, leveraging existing clinical data are a cost-effective way to answer vital research questions. The NIHR HIC Viral Hepatitis Theme aims to address key research questions (table 1), to demonstrate the utility of the NIHR HIC methodology.

This paper presents a comprehensive methodology that has been proposed, implemented and validated by the NIHR HIC for the development of a new data collection and management pipeline. This development is under a comprehensive governance framework that allows data to be collated across multiple centres for collaborative research on viral hepatitis. Under this governance framework, a data collaboration involves the generation of a research-ready dataset that is broad enough to support a wide range of investigations in a specific clinical area. The dataset is assembled to the same agreed standards at each centre. The data transformations needed to achieve this, starting from patient records, are documented and shared.

Governance framework

The protocol for the collection and management of the data for the viral hepatitis theme has been reviewed and approved by South Central—Oxford C Research Ethics Committee (Reference Number: 15/SC/0523). In addition, an NIHR HIC data sharing framework, covering a wide range of data and research collaborations has been signed by all participating centres. This document, in conjunction with the rest of the governance framework established by the NIHR HIC, addresses common requirements and considerations regarding data sharing between centres, contractual responsibilities, confidentiality, intellectual property and a publications policy. This general agreement will be used to underpin individual agreements for research collaborations with third party academic and industry partners. Any collaboration with industry partners requires additional agreements, with additional governance checks by participating sites. This process has been simplified by the creation of a ‘framework industry collaboration agreement’ which simplifies the addition of individual participating sites to an industry collaboration. A scientific steering committee has been established which is made up of a representative of each participating site, and reviews and approves data requests from external collaborators.

Development of data model

The development of the NIHR HIC viral hepatitis data model, which outlines the structure of the dataset and the associations between the data fields, began by the clinical leads defining data fields, required to answer the initial academic questions posed by the clinical and scientific leads across centres (table 1). The feasibility of collecting this dataset was tested using the electronic data capture software OpenClinica.7 Each site was required to manually complete case report forms in OpenClinica for a small subset of their patient cohorts; an assessment of the completeness of these data led to a refinement of the dataset and a second version, which was used to generate the Extensible Markup Language (XML) Schema Definition (XSD) used to define the dataset. This XSD was then further refined to remove errors and to provide a more efficient data structure.

Data architecture for collection and integration of data

Each site within the collaboration provided a data product containing the data items outlined in the agreed dataset. While each site had pre-existing data warehouses and systems for collecting, storing and using patient data, these systems were designed to be used for patient care, administrative and financial purposes, and were not always suitable for generating the required data product. The variety of electronic patient record systems and laboratory information management systems used across the participating sites meant that data were often stored in different formats. This meant that at each site a data architecture outlying the flow of data from electronic patient record systems had to be developed to allow the effective flow of data from operational systems to research systems. This required the development of data warehousing or data management infrastructure at all sites.

To allow the compilation of the full dataset, each site generated a data product that was integrated in the NIHR HIC Viral Hepatitis Central Data Repository (figure 1). In some cases, these data were already structured and could be transferred directly into the data repository. Where data were stored in an unstructured (free text) format, the data had to be either manually entered or extracted (figure 2). The data were then anonymised by removing patient identifiers and a data product was created using the XML format. To avoid duplication of records, each site was responsible for maintaining a link between the patient’s local clinical identifiers and the identifier used in the database. The data product was securely transferred to the NIHR HIC Viral Hepatitis Central Data Repository, where the data were put into a database for queries and analysis. As each site has different infrastructure in place to produce the data product, updates are submitted on request and there is no fixed schedule for new data submissions from sites. When a dataset is requested by a research group, the request is first reviewed and approved by the scientific steering committee, and following internal governances process, an extract of the integrated dataset is provided to the research group for their analysis.

Request permission

Request permission

NIHR HIC Viral Hepatitis Central Data Repository

Anonymised data from the providing centres were transmitted to the lead centre in XML format via secure email or submitted directly to the NIHR HIC Data Acquisition Management (HICDAM) system via a secure web-based front end, again in XML format. The primary service inside HICDAM is the Message Receiving, Curating, and Understanding Repository (MeRCURy), which performs the validation, processing and storing of the submitted data.

The MeRCURy system supports two types of data validation: basic, automatic integrity checks, which must be satisfied before the data are loaded into the database, and more sophisticated, manual checks of data consistency, which are performed after the data have been loaded. The basic integrity checks involve validation against the agreed XSD, confirming that the data are correctly formatted, together with logical checks on the type or range of values submitted: for example, a check that any value given for the date of death is strictly later than those given for dates of treatment.

The manual checks reflect working assumptions regarding the relationships between the values of different data items submitted, for example, that a certain combination of treatments would never be used in practice, or where there are clear duplications of data. In each case, some additional information may be needed to determine whether data is incorrect, or whether the assumptions are invalid.

The system will inform data providers of the outcome of any submission. If the data submitted fail a basic integrity check, they are rejected, and a report is generated containing appropriate diagnostic information. If the data are accepted, a confirmation message is sent, and the data may then be reviewed using a secure interface provided by the LabKey8 application. Data can be then be explored and/or exported from the system in a variety of formats, including .xls, .xlsx, .tsv and .csv.

Chronic HBV cohort description

Data have been extracted from the NIHR HIC Viral Hepatitis Research Database by using HBV lab test information. Patients included in the extract meet at least one of the following criteria:

1. Two positive hepatitis B surface antigen tests, at least 6 months apart.

2. Positive hepatitis B surface antigen and positive HBV DNA, at least 6 months apart.

The characteristics of the chronic HBV cohort of 960 patients are described in table 2, including age, gender, race, HCV coinfection, HIV coinfection, comorbidities, the severity of liver disease and treatment information. Among these patients, 254 patients (26.5%) have previously received medications or are currently under treatment. There are 17 patients coinfected with HDV, with data extracted based on detectable HDV viremia or anti-HDV antibody. However, there were 475 patients who had no HDV test data. Missing data for each variable is reported as ‘unknown’ in table 2.

Chronic HCV cohort description

Data have been extracted for patients with a positive HCV RNA test or HCV genotype from the NIHR HIC Viral Hepatitis Research Database. The characteristics of HCV cohort of 1404 patients are described in table 3, including age, gender, ethnicity, HCV genotype, HBV coinfection, HIV coinfection, comorbidities, the severity of liver disease and treatment information. Among these patients, 914 patients (65.1%) had received interferon-free direct-acting antiviral treatment by February 2018. Similarly, to the HBV cohort, there were various degrees of missing data for each variable, which is reported as ‘unknown’ in table 3.

HEV cohort description

Data have been extracted from the NIHR HIC Viral Hepatitis Research Database by using HEV IgM and HEV IgG lab test information. HEV usually causes an acute and self-limiting infection but can occasionally cause severe disease and/or chronic infection; our data collection approach sets out to include both categories. Patients included in the extract meet at least one of the following criteria:

1. Patients with anti-HEV IgM and anti-HEV IgG were both positive.

2. patients with HEV RNA detected in serum and/or stool.

There are 14 patients with acute HEV infection in our current database. Among them, 4 patients are male, and gender information was missing for the other 10. The median age at the time of patients first positive HEV test was 61.5 years (range: 21–86).