UK-HDAN Research Roadmap
1 Overview
The UK Health Data Analytics Network (UK-HDAN) is an open community of researchers and
practitioners who share an active interest in applications of ‘big data’ methods in health and social
care. It brings together the health informatics and data science communities to exploit synergies,
identify research challenges, and build new partnerships. As of January 2017 the Network has more
than 450 members from over 100 HE, NHS and industry organisations.
Between January and March 2016 the Network organised a series of workshops, attended by over 160
individual participants, with the aim of developing a Health Data Analytics Research Roadmap. This
document – a first draft of the Roadmap – provides a structured summary of the workshop outputs,
with three components: an analysis of the Healthcare Opportunities – the ways in which the
application of data science could transform health and social care; a summary of the Data Science
Research Challenges – the methodological advances that will be necessary to realise the potential for
transformation; and an outline Ethical and Responsible Innovation Framework, which lists non-
technical issues that should be considered by researchers working in this field.
2 Healthcare Opportunities
The roadmap focuses on ways in which data-intensive methods, supported by data analytics could
have a direct impact on the delivery of health and social care. There is also potential for indirect benefit
through applications of data science in biomedical research, but that is out of scope – though there
are areas of overlap. We identify five broad areas of opportunity, which share common assumptions
about the availability and use of personal health and health-related data: New Insights from
Ubiquitous Data, Better Care Through Patient-specific Prediction, Personalised Care, New Models of
Care, and Learning Health Systems; they provide closely related but distinct views of potential for
benefit.
2.1 New Insights from Ubiquitous Data
Fundamental to the transformational potential of data-intensive care is the opportunity to gain a more
complete picture of individuals’ health, lifestyle, exposures and experience than is currently available,
by integrating and analysing data from multiple sources. The idea is to augment formal, intermittently
acquired information, currently stored in health records, with data that provides a much richer, more
continuous description of lived experience. Examples include data from mobile and wearable devices,
environmental data, retail transactions, social media, patient-reported experience, digital footprint,
utility usage and, more generally, internet of things. These data can be used both to underpin
improved care for the individual and provide new insights at the population level. Potential broad
impacts, illustrated with more specific examples in subsequent subsections, include the following.
Better informed care. Currently, healthcare decisions are based, typically, on limited information,
sampled intermittently (and often unreliably) during formal interactions with the healthcare system.
Data science has the potential to provide a more complete description of lived experience and health
outcomes, to inform better-targeted care and to support self-care. This includes understanding
severity and patterns of symptoms, triggers, behavioural determinants, and medication adherence.
Responsive care. Currently, healthcare is provided in a relatively standardised way, based on typical
needs and responses to treatment. Data science has the potential to facilitate a more responsive
approach to care, developing an understanding of what is normal for an individual, and detecting
significant change. This can be used to influence behaviour, engage peer support, and target
professional intervention at the point of need.
2
Population insights. Currently our understanding of disease and response to treatment ‘in the wild’ is
limited. Data science has the potential to build a more complete understanding of the development,
temporal characteristics, outcomes, and response to treatment of disease across populations –
including their determinants and variability. This can be used as a basis for improving patient pathways
and developing new treatments.
2.2 Better Care through Patient-Specific Prediction
Harnessing data analytics to make more effective use of pervasive data has the potential to improve
care at the level of the individual by making patient-specific predictions – providing actionable
information at the point of care. The idea is to develop patient-specific models that draw on both
population and personal data to predict health outcomes – including prognosis and response to
potential treatment – with the aim of deploying the right investigative, preventive or therapeutic
intervention at the right time. Potential impacts include the following.
Precision medicine. Many conditions that have in the past been classified and treated as single
diseases, are now recognised as clusters of diseases with similar symptoms but differing underlying
pathology and, thus, response to treatment. Data science has the potential to discover new
phenotypes (ideally endotypes), and allocate individuals to the most appropriate treatment group.
Given the known influence on outcomes, of cultural and socio-economic factors, this approach should
embrace behavioural as well as bio-pathology phenotypes.
Dynamic management. Existing patient pathways take a one-size-fits-all approach to condition
management, militating against agile response to disease progression, comorbidities, changes in
response to treatment or altered circumstances. Data-intensive methods have the potential to allow
and support dynamic, individualised reconfiguration of pathways, modifying treatment and other
interventions, and instigating investigations, in response to evidence of change in efficacy or risk.
Forestalling acute episodes. A large proportion of current NHS cost is associated with unplanned
hospital admissions due to acute exacerbations of long-term conditions (eg respiratory, mental
health), often accompanied by permanent reduction in quality of life. By building a detailed
understanding of what is normal for a given patient, data analytics can be used to provide alerts for
the individual, their family/friends and health care professionals, allowing early intervention to
forestall acute episodes.
2.3 Personalised Care
Building on existing experience in the retail sector, data-intensive care has the potential to transform
patients’ experience of healthcare, providing services, information and advice relevant to their needs,
empowering them to engage in their own care, and managing it in a way that suits their personal
preferences. Different aspects of personalisation can potentially be subsumed in a virtual personal
health assistant (eg mobile device app) that is fully aware of an individual’s health status, treatment,
preferences, history and context. Potential impacts include the following.
Patient choice. Current practice provides very limited opportunity for patients’ personal preferences
to influence care. Comprehensive deployment of data-intensive care presents an opportunity for
patients to influence their care directly, both by providing explicitly stated preferences regarding
treatments, personal goals, use of data, involvement of family and friends etc, and by inferring them
from their interactions with personalised advice and feedback.
Experience sampling. Patient-reported experience of their condition is arguably the most relevant
measure of impact on quality of life, but currently is recorded infrequently and unreliably – generally
in face-to-face consultations – with little impact on the delivery of care. A data-intensive approach to
care provides the opportunity to collect patient reported experience data more
frequently/systematically and use it to influence care directly.
3
Personalised support. Currently, patients receive generic advice and very limited personalised
feedback. Data-intensive care creates the opportunity to provide personalised feedback, explanation
and advice directly relevant to an individual’s condition(s), level of activation (commitment and skills
to self-care), and context, improving their experience of the health system and empowering them to
take greater control of their own health.
2.4 New Models of Care
There is a pressing need to develop new models of care, particularly for long-term conditions,
delivering high-quality care in a community setting. Data-intensive methods have the potential to
power new models of care in the community, increasing the focus on prevention, transforming care
by supporting patients to manage their own health, and targeting resources intelligently to meet
patient needs. Potential impacts include the following.
Wellness. Currently, healthcare resources are focussed mainly on treating individuals who are ill,
despite the fact that much of the burden of disease is preventable. There is already a consumer market
in wearable devices and associated data-driven methods to support and encourage healthy
behaviours. Given financial pressures and skill shortages, it will become increasingly important for
health and social care providers to extend this approach, building more holistic views of, and
influencing positively, individuals’ health behaviours.
Self- and collaborative-care. As the population ages and individuals live longer with chronic
conditions, there will be increasing pressure for patients to take a more active role in manging their
own long-term conditions, supported by relatively low-cost care in the community. Data-intensive
methods have the potential to power this transformation by providing individuals and their carers
with the information and tools to collaborate in producing and implementing their own care plans.
For example, by identifying links between personal behaviours and health outcomes, data analytics
can be used to drive behaviour-influencing prompts and personal decision-support tools, helping
individuals to live independently for longer.
Targeted support. Currently, resources to support care in the community are spread thi