Keynote 1
9.55 - 10.35
How Data and Analytics Shapes Policy
The UK response to the COVID-19 pandemic both highlighted the need for rapid and accurate data analysis to inform decision making, plus established new ways of working to support this. This presentation will reflect on the experience from within the Joint Biosecurity Centre (JBC) and UK Health Security Agency (UKHSA) of using data and analytics, for COVID and beyond.
It will examine the institutional, analytical and policy developments over the last few years. The vast amounts of data available during the pandemic posed challenges as well as opportunities for those at the heart of government. Harnessing those data, and presenting key insights in the right way, was key to many of the public health interventions the UK successfully deployed. Moreover, many of the lessons from Covid-19 have left a lasting impact on the UK’s approach to public health, including the response to Mpox in 2022 and Avian Flu in 2023.
As well as the contextual public health background, the presentation will address all aspects of data and analysis for policy such as data acquisition; methodological approaches; and effective outputs. Challenges and opportunities will be discussed.
Lightning talks
10.35 - 11.15
Modelling nosocomial transmission routes of SARS-CoV-2
Jess Bridgen, Lancaster University
Impact of COVID-19 on broad-spectrum antibiotic prescribing for common infections in primary care in England
Xiaomin Zhong, University of Manchester
Validating digital mobility outcomes from wearables as clinical trial endpoints: insights from the Mobilise-D consortium.
Silvia Del Din, Newcastle University
Exploring the value of routinely collected patient-reported outcome measures as prognostic factors in adults with advanced non-small cell lung cancer receiving immunotherapy.
Kuan Liao, University of Manchester
Research Project Presentations
11.15 - 12.15
Safety issues stemming from training data of uncertain quality for Machine Learning based decision support
Philippa Ryan, Research Fellow in Assured Responsibility for AI, University of York
Machine Learning (ML) offers the potential to provide predictors for clinical conditions by identifying patterns in vast sets of clinical data which have not been identified before. However, training data shortfalls can cause varying safety performance in ML. In this talk I'll present a case study of an ML-based clinical decision support system for Type II diabetes-related co-morbidity prediction (DCP). The DCP ML component has been trained using real patient data, but the data was taken from a large database of clinical information gathered over many years, with limited checks for errors, missing information or bias towards certain groups of patients. All of these issues can lead to uncertain performance and latent errors in the final system. I'll discuss how performing systematic safety analysis of the issues can support effective identification and mitigation of the risks they pose.
Augmenting the clinician’s eye through computer vision: detecting and quantifying signs of neurodegenerative diseases using smartphones
David Wong, Associate Professor of Health Data Science and Health Informatics, University of Leeds
The practice of neurology relies upon expert judgement during clinical examination, for both diagnosis and monitoring. Much of that judgement is based upon a visual and auditory assessment of the patient. However, this practice depends upon a small group of experienced experts, of which there is a global shortage. Furthermore, even experts are limited by human sensory judgement, which cannot reliably or precisely detect and measure subtle changes.
A promising solution to these problems is to assess patient movement and voice using smartphone videos. Videos are non-contact and may therefore be suitable in situations with higher risk of infection. Videos can also be recorded remotely, without the presence of a trained clinician or technician and may be useful within mobile health applications.
Here, I will show some our recent work in trying to measure and classify some signs of neurodegenerative diseases such as bradykinesia and tremor. I will talk about our on-going NIHR project on quantifying hand tremor and discuss some of the practical challenges that we face as we try to translate into clinical practice.
Predicting emergency admissions in Scotland
James Liley, Co-Director of Durham Biostatistics Unit, and Ioanna Thoma, University of Durham
Pre-empting emergency admission (EA) is advantageous both to individual health and healthcare system efficiency. Prediction of emergency admission risk can contribute to reduction of EA both through primary care and through public health planning. We developed a population-scale predictive score for EA risk in Scotland using predictors derived from electronic health records (EHRs).
Our data comprise acute hospital records and community prescriptions for all individuals living in Scotland with a recorded interaction with the healthcare system. We predict EA in the year following prediction, and deploy predictions monthly to general practitioners for patients in their care. We introduce a novel protocol allowing safe model updating. We demonstrate strong predictive power demonstrate model stability over a 3 year timeframe, and show that machine-learning predictive methods lead to substantial improvements over linear models. We found particular strength in predicting respiratory and metabolic/endocrine related admissions and a difference in risk contribution across deprivation deciles equivalent to up to 40 years of age.
EAs in Scotland can be accurately predicted from EHRs, and contemporary machine learning methods can meaningfully improve accuracy, also giving insight into EA epidemiology. Our work constitutes a direct application of machine learning tools in medical practice.
The NIHR DynAIRx Project: Artificial Intelligence for dynamic prescribing optimisation and care integration in multimorbidity
Lauren Walker, Senior Clinical Lecturer in Clinical Pharmacology & Therapeutics and Internal Medicine, University of Liverpool
Funded by the NIHR, DynAIRx (Artificial Intelligence for dynamic prescribing optimisation and care integration in multimorbidity) aims to develop new, easy to use, artificial intelligence (AI) tools that support General Practitioners (GPs) and pharmacists to find patients living with multimorbidity (two or more long-term health conditions) who might be offered a better combination of medicines.The NHS introduced Structured Medication Reviews by GPs and pharmacists to reduce the number of people taking potentially harmful combinations of drugs. However, there is no easy way of predicting which patients are most likely to benefit from a medication review and prioritising them. DynAIRx will develop tools to combine information from electronic health and social care records, clinical guidelines and risk-prediction models to ensure that clinicians and patients have the best information to prioritise and support Structured Medication Reviews.
Lunch and posters
12.15 - 1.30
Keynote 2
1.30 - 2.10
Addressing barriers in health data research
Health Data Research (HDR) UK is the national institute for health data science with a mission to “Unite the UK’s health data to enable discoveries that improve people’s lives”. Our long-term vision is to “enable large-scale data and advanced analytics to benefit patient interactions and public health”.
One critical step to achieve this vision is reducing the time and effort for researchers to find and access data within a fragmented, complex landscape. Health data is highly sensitive and confidential and often accessed within Trusted Research Environments, of which there are many.
The level of maturity in the health data sector is low due to lack of consistency in information governance, policies, standards, and interoperable technology.
Researchers are often required to approach multiple data custodians when seeking to understand the available health datasets and navigate complex processes for data access for research. The researcher’s journey from finding to accessing data can take many months, and in many instances years, delaying time to new discoveries that can improve people’s lives.
HDR UK vision is to work with Health data sector in identifying solutions that will address these barriers.
This talk will cover HDR work towards providing innovative technical solutions to streamline the researcher journey, whilst adhering to data governance requirements. It will also discuss opportunities to co-create solutions and our work to convene a Technology Ecosystem Community.
Lightning talks
2.10 - 2.50
Key Fundamentals of AI/ML including challenges and opportunities for Health Data Community with a use case of DynAIRx for Optimising Prescriptions
Asra Aslam, University of Leeds
Classification of phonocardiogram signals to detect heart murmurs
Nicola Rennie, Lancaster University
Living beyond colorectal cancer: A digital tool to deliver personalised evidence-based advice on diet and physical activity
Anna Fretwell, Newcastle University
Research Project Presentations
3.20 - 4.20
The South Yorkshire Digital Health Hub
Tim Chico, Professor of Cardiovascular Medicine, University of Sheffield
Too many digital health projects, products, and services fail to improve patient outcomes. The South Yorkshire Digital Health Hub was established to address the barriers to successful translation of early-TRL research.
The Hub brings together technical experts, clinicians, industrial partners, patients and investors to identify the unmet clinical needs of greatest importance in our region and co-create digital innovations that address these.
A wide range of online and in-person training will be provided both at scale remotely and to cohorts of digital innovators in a range of remote and in-person events. Selected innovators will receive tailored support, mentorship and development with seed funding provided to those with the most clinical and commercial potential.
The Northern Health Futures (NortHFutures) Digital Health Hub
Abigail Durrant, Professor of Interaction Design, Newcastle University
Northern Health Futures (NortHFutures) is envisioned as world-leading innovation ecosystem that will facilitate the research, development and acceleration of responsibly designed, human-centred, and data-rich health-technologies (health-tech) to cultivate an entrepreneurial and vibrant community promoting research leadership in Digital Health.
Based in the North East and North Cumbria (NENC) region, with national and global reach, NortHFutures brings together a consortium of partners in Health and Care, Academia, Industry, Public Sector, and Voluntary, Community & Social Enterprise. The hub will address unmet health needs and inequalities in NENC by supporting: multi-directional skills training and sharing in Digital Health; and Health-tech research, innovation and entrepreneurship, enabled by Design. NortHFutures will develop a supportive community infrastructure that will further stimulate socio-economic and cultural growth for all, and will create enabling mechanisms for multiple stakeholders in Digital Health to live, work and age well.
Design and analysis of prevalence surveys for NTDs: how to increase model uptake?
Claudio Fronterre, Lecturer in Biostatistics (Global Health), University of Lancaster
Neglected tropical diseases (NTDs) are a group of parasitic infections that primarily affect impoverished populations in tropical regions, posing significant challenges to global health and development.
Spatial analysis has increasingly become a valuable tool in the field of parasitic diseases. Particularly when disease prevalence is highly spatially heterogeneous, quantifying the spatial variability of disease risk and its uncertainty is crucial to inform disease control and elimination programmes.
In this talk I will show how model-based approaches can render the design and analysis of NTDs prevalence surveys more efficient and cost-effective, an essential aspect when dealing with low-resource settings.
Although these models provide valuable insights for the decision-making process conducted by NTD control program managers, their widespread accessibility and implementation remain challenging, requiring the resolution of existing barriers.
CONNECT - Can we use information from electronic devices (e.g. phones) to predict if someone will have a relapse of psychosis?
Sian Bladon, Research Associate, Division of Informatics, Imaging & Data Sciences, University of Manchester
Relapses amongst people with psychosis are common and come with high associated costs, both in terms of debilitating effects for the individual and treatment costs to the health services. Identifying and responding to early warning signs of relapse can be beneficial, however, this is difficult when contact with mental health services is infrequent and relies on individual’s recall of symptoms.
Previous studies have demonstrated the feasibility of using digital health solutions to monitor people remotely and provide real-time data to mental health care teams, either through active symptom monitoring or passive data collection. The aim of the CONNECT study is to use remotely collected data from participants with psychosis to develop a clinical prediction model to predict an individual’s risk of relapse.
The study will recruit 1,100 individuals from six sites in England, Wales & Scotland and provide them with access to a smartphone app and a smartwatch. Participants will be prompted daily through the app to answer questions around their current mental state. Passive data collected through the smartphone and smartwatch will include physiological measurements and information about phone use e.g. number of incoming/outgoing text messages. This data, along with baseline demographics, will be used to develop the prediction model in the first 800 participants recruited to the study. The model will then be temporally validated in the last 300 participants recruited.
In this talk I will describe the CONNECT study in more detail, including the data collection and how we will develop the clinical prediction model.
AI-MULTIPLY: Untangling the dynamic inter-relationships between multiple long-term conditions and polypharmacy
Paolo Missier, Professor of Big Data Analytics, University of Newcastle
The AI-MULTIPLY project was funded in late 2022 by the NIHR to explore the complex inter-relationships between chronic multimorbid states and the way those are managed over time, often through combinations of disjoint long-term prescriptions, along with personal and social factor. Our hypothesis is that through a better understanding of this entanglement we can develop data-driven methods to optimise treatment for individual patients.
Since the start of the project, two interconnected, multidisciplinary teams at Newcastle University and Queen Mary London have been approaching the problem from multiple directions, using UK Biobank and CPRD data, respectively.
In this short talk we will review some of our data-driven approaches, which are all still in progress, but we also emphasise the importance and cost of performing necessary data engineering on each of the datasets, before any significant analysis can even be attempted.