Healtex: UK Healthcare Text Analytics Research Network

Healthcare is a prime example of "big data science" with a number of challenges and successful stories where actionable information extracted from data has improved and saved lives [1]. The majority of concerted efforts focused on real-time processing and integration of structured data streams coming from clinical coding, diagnostic tests, sensor measurements, questionnaires, etc. to support timely clinical interventions and facilitate patients' self-management. Nonetheless, natural language remains the main means of communication within healthcare with its written accounts becoming increasingly available in an electronic form, thus giving rise to big text data. Prominent examples include text data embedded within electronic health records (e.g. referral letters, case notes, pathology reports, hospital discharge summaries, etc.), patient-reported outcome measures (e.g. questionnaires, diaries, etc.) or unsolicited informal feedback shared openly on the Web 2.0 (e.g. social media, fora, etc.). Unfortunately, the capacity to effectively utilise information from unstructured text data on a big scale is lagging behind its structured counterpart. The fact that the majority of actionable information in healthcare is contained within text data (some estimates shows as much as 85%) clearly indicates a potential to dramatically transform community health and care by the ability to process and integrate such information in real time. However, automated and large-scale "understanding" of diverse healthcare sublanguages is still largely unsolved research challenge due to their dynamics, idiosyncrasy, ambiguity and variability.

The aim of this proposal is to build a UK-wide multi-disciplinary research network in order to explore the barriers to effectively utilising healthcare narrative text data, road-map research efforts and principles for sharing text data and text analytics methods between academia, NHS and industry. The network will directly address the "Transforming Community Health and Care" grand challenge by enabling research that will deploy healthcare narratives as real-time sensors and integrate them with the structured data streams into a patient-focused collaborative ecosystem, which will involve healthcare professionals, patients, carers and researchers. Such systemic network of healthcare activities will facilitate informed decision making, timely interventions, deeper digital phenotyping for clinical epidemiology and population-based modelling. On the other hand, by processing patient-generated narratives, which are often a preferred and likely means to provide patient responses (e.g. text messages) to complement structured healthcare data (e.g. signals from wearable devices), we will "use real-time information to support self-management of health and wellbeing".

The main outcome of the network will be a strong, sustainable community that will continue its mission after the initial 3 years of support. Other outcomes will include (1) reports describing the state-of-the-art and challenges for key barriers in harnessing text narratives and making sense from them; (2) a research roadmap for healthcare text analytics; (3) an enlarged membership and expanded collaborations within the network, in particular with early career researchers and internationally; (4) a series of focused pilot/feasibility projects that will inform further developments and kick-start collaborative projects; (5) a collection of research papers at conferences and journals, improving the UK competitiveness in this growing area; (6) several project proposals scoped during the project and prepared for submission; (7) proposals for discipline-bridging personal fellowships, and (8) an interactive registry of healthcare text analytics expertise, resources and tools so that the users and collaborators can identify existing resources and initiate new collaboration.

The network will create societal impact in:

a. Health impact. Research into improving our health relies mostly on structured variables that are collected routing in clinical practice. Clinical narratives convey other valuable diagnostic and contextual information that is predictive of the prognosis and biological behaviour of a disease process, but are rarely used for secondary purposes, e.g. to improve our understanding of patients clinical pathways. Currently 85% of actionable information remains locked within such narratives, i.e. available only for sporadic manual inspection and therefore often overlooked. Using such data to understand and improve the health of our nation is therefore a key area that will benefit from our network. Building on a range of previous studies that have proven the feasibility of NLP for structuring clinical narratives, we will pilot new systems that focus specifically on the types of narratives used within the NHS (e.g. referral letters, pathology and imaging reports, accident reports, etc.) in order to extract accurate data on patient clinical journey and service provision. This will facilitate actionable healthcare analytics, which aims to improve quality and efficiency of care, thus bringing direct health benefits both at the patient level (timely interventions) and to the entire population. For the latter, information extracted from clinical narratives will lend itself to large-scale epidemiology studies describing fine-grained findings. This will help identify previously unknown risk factors, drug interactions, drug's unintended targets, etc. Feeding these research results back into the healthcare service will open another avenue for improving patient health outcomes, public health and quality of life.

b. Impact on healthcare services: With patient groups becoming more adept at lobbying for their consumer "rights" on social media, we will be be able to identify globally unmet patient needs and concerns by mining patient narratives on the web. Information extracted from clinical narratives can be used to ensure that patient pathways are designed to optimise quality, patient outcomes and cost effectiveness. When integrated with other clinical data, narrative can be then used to support decision making for individual patients, and deeper digital phenotyping for clinical epidemiology and population-based modelling. Such findings can impact professional standards within the NHS and inform NICE guidance, which will in turn improve the quality of healthcare, and consequently patient health outcomes, public health and quality of life.

c. Impact on public policy. While healthcare stands to benefit from automated processing of clinical narratives, there are ethical and legal issues associated with their use as they contain private and confidential information. Patient data cannot be used for text analytics purposes without consent. In the EU, Data Protection Directive (95/46/EC) specifies rules on using information about individuals. Unfortunately, even when all reasonable measures to protect privacy and confidentiality are taken (e.g. de-identification and data use agreements), in most cases only researchers with local affiliation are allowed access. This remains the main bottleneck for progress in healthcare applications of text analytics. We will provide research evidence that will stimulate and inform data privacy policy debate with the ultimate aim of adjusting legislation, regulations and guidelines for data sharing that fully protects personal privacy whilst maximising benefits to society.

d. Impact on industry. The outcomes from the network will also provide benefits to pharmaceutical and healthcare businesses, buy making free-text data part of their data space. For example, in the world of drug safety, much of the rich information about the plausibility of a causal association between drug and side effect is held within the narrative, and this can be used by both sectors.