Predictive analytics for clinical decision support with application to Cardiovascular management

Lead Supervisor
Dr Abdel Douiri
Reader
School of Population Health & Environmental Sciences, King’s College London
abdel.douiri@kcl.ac.uk

Co-supervisor
Dr Vasa Curcin (King’s College London)

Project Details

There are exciting opportunities for health data research to improve understanding and management of vascular diseases and to use data as a communicator. Prediction models help to design clinical decision support (CDS) systems that could aid clinicians to identify and manage cardiovascular disease (CVD) risks. CDS will also provide tailored assessments and treatment recommendations based on individual patient data. 

By taking advantages of the existing integrated data sources to better phenotype patients with cardiovascular disease, myocardial infarction (MI) and Stroke, the aim is to develop an accurate risk prediction of subsequent vascular events and build a decision-making tool to support preventive strategies. Through these analyses, this will enable the identification of subgroups of the population at increased risk of further vascular events, estimate missed opportunities for effective prevention, acute and longer-term management that will inform the design and evaluation of more efficient, effective pathways of care. 

Building on our previous predictive modelling we will use the linked data (primary and secondary care) to develop predictive analytic models for patients for a range of vascular events (initial and subsequent vascular events (e.g. recurrent stroke/MI, disease transitions (MI/Stroke) or CVD death).  Common risk stratifying factors for stroke and MI will be analysed. Selected variables and features will be analysed using random forest approach to build prognostic risk models of vascular subsequent events (and outcomes). A machine learning approached based on decision trees will be used in building the patient-specific score and risk predicted trajectories. Predictive analytic models developed will form the basis for sets of decision rules to be combined with clinical guidelines to develop the clinical decision support. These will be embedded into the electronic health record (HER) that will require us to adapt and apply the auditable decision support technical and usability framework previously developed by us. The resulting CDS will be transparent to patients, carers, clinicians, and their usability into a primary care EHR and effectiveness will be investigated.

This project will be a resource for research and facilitate improved vascular care particularly in primary care. Improvements in identification of at-risk groups, prescribing, joint decision making, and monitoring outcomes are anticipated to improve the efficiency and quality of care. 

Datasets

Data sources: 

We will harness longitudinal population-based patient data collected systematically in the South London Stroke Register (SLSR), based on 22 electoral wards in Lambeth and Southwark (since 1995 with 22 yrs annual follow up), with a denominator population of 357,308 made up of 56% White, 25% Black, and 19% of other ethnic Groups (2011 Census). The SLSR comprises approximately over 6,600 first ever stroke patients and are assessed at onset, 3 months, and annually up to 23 years after stroke, with information collected on risk factors, subsequent vascular events, management and structured outcome assessments (13). Regarding cardiovascular health, we will use data from National Institute Cardiovascular Outcomes Research (NICOR) which collects data to monitor and improve the quality of care and outcomes of cardiovascular patients. NICOR manages six cardiovascular clinical audits and two clinical registers, including Myocardial Ischaemia National Audit Project (MINAP) and percutaneous angioplasty since 2007(14, 15). Another sources of notification and information will be the London heart attack audit (LHA), Hospital Episode Statistics (HES) and the Office for National Statistics (ONS) mortality register (cause specific mortality data), which are all linked to NICOR. Lambeth datanet (LDN) is a database of electronic health records (EHR) of 390,000 patients from all general practices in the London Borough of Lambeth since 2006. There are approximately 2503 patients with stroke and 3673 patients with MI recorded in LDN. LDN provides information in the domains of: socio-demography, diagnoses, morbidities, appointments prescriptions and procedures. LDN is already linked to the Clinical Records Interactive Search (CRIS) system at South London and Maudsley (SLAM) NHS Trust, and this will allow further detailed information of mental health outcomes. The project will test the feasibility of integrating components of each of these sources of data to tackle the research questions in the longer-term.

Data governance and ethics: 

Linkage between LDN data and SLSR was established in 2015 (in 2018 for longitudinal follow up) following approval by the Information Governance team, SE London Commissioning Support Unit (CSU), the Lambeth Information Governance Steering Group (IGSG) and the Lambeth CCG Caldicott Guardian. NICOR already support under section 251 of the National Health Service (NHS) Act, 2006. But since the research uses‚ confidential patient information without consent, application will be sought to the Confidentiality Advisory Group of the Health Research Authority or/and to the designated Caldicott Guardian for their Trust/hospital which done at project bases. These consents also cover information collected through linkage with primary and secondary care data. The SLSR has HRA approval for ongoing data collection

Keywords

Risk prediction, clinical decision support, data linkage, electronic health record, cardiovascular disease, machine learning, Statistics, Health informatics.