Programme - Healtex

The programme features keynote talks, research papers, discussion panels, software demos and poster sessions. The PhD forum will feature talks from early carer researchers presenting their ongoing research.

Programme outline at a glance
Detailed programme

Programme Outline

Wednesday, April 18th 2018

08:30-09:00	Registration

09:00-11:00	PhD forum
11:00-12:00	Career talk: Prof Wendy W. Chapman, University of Utah Navigating the Mountain of My Career: Lessons Learned from Skiing

12:00-13:00	Lunch

13:00-14:00	Keynote: Prof Pierre Zweigenbaum, LIMSI-CNRS ICD-10 coding of death certificates: CLEF e-Health Shared Tasks datasets and methods In many countries, causes of death are collected and coded into the WHO International Classification of Diseases (ICD) to provide key data for monitoring and comparing mortality across time and countries. Death certificates are very short texts that list causes of death. On the one hand, they exhibit the usual challenges of free text for automated processing and coding, including misspellings, ambiguity and variety of expression; and the size of the ICD classification, with well over 10,000 classes in ICD-10, increases the difficulty of the task. On the other hand, their constrained purpose and format may make them more amenable to automatic coding than the average clinical text. Moreover, the routine coding of these texts along the years has created gold standard annotations that are a natural source for creating large-scale training or test datasets. In the context of a collaboration with CépiDc, the team that codes causes of death in France, this led us to organize shared tasks that target ICD-10 coding of death certificates.I will present the shared tasks run over French and English certificates, and currently open on Italian and Hungarian data too. Gold standard annotations provide a set of codes for full certificates or at best for each certificate field, but no entity spans, therefore participants adopted two main strategies. They addressed the task either as an entity recognition and normalization task, using dictionary-based methods, or as a text classification task, using supervised classifiers. I will summarize these methods and their results, that set the current state of the art on this task, as well as our own experiments on these datasets. (Slides)
14:00-15:00	Research paper presentations – session 1
15:00-15:30	Posters and demos – session 2 (with coffee break)
15:30-17:00	Panel: Text mining veterinary clinical records: opportunities for veterinary and human healthcare researchers Large volumes of veterinary electronic patient records (EPR), creating huge corpora (millions of records) of veterinary clinical narratives are being collated by projects including the Small Animal Veterinary Surveillance Network and VetCompass. These initiatives aim to gain insights applicable to surveillance and retrospective/prospective studies used to inform evidence-based medicine approaches. Linking to human healthcare corpora to take a ‘One Health’ approach might produce both methodology development and potentially novel results that could inform understanding of veterinary, environmental and human health issues. Various text mining developments are being applied to veterinary narratives, extracting key clinical variables including symptoms, clinical parameters, suspected diagnoses and potentially offering automated clinical coding. A number of challenges remain and this panel aims to initiate discussions on these and various opportunities surrounding veterinary text mining. Aims of panel: To share current state of veterinary clinical text mining To identify areas of potential collaboration in developing text mining solutions across veterinary, human health care and other corpora. To consider specific, focussed challenges in clinical text mining that might be tested using veterinary data, for instance: extraction of key clinical parameters (heart rate, respiratory rate, weight) held in free text. Panelists: Rachel Dean, PJ Noble, Noel Kennedy, Fabio Rinaldi, Mercedes Arguello Casteleiro
17:00-18:30	HealTAC-2018 Industry Forum Technology awareness, the “rebirth” of AI, and financial pressures are increasingly leading the health sector to engage with the text analytics industry, for the provision of solutions to deliver data for research, administration and for regulatory requirements. The health sector presents the industry with a unique set of organisational, technical, social, and economic problems. The Industry Forum will be lead by a panel of experts from both industry and service procurement, and will explore these issues by presenting real world examples, and through in-depth discussion and debate with the audience. Panelists: Angus Roberts (King’s College London), Simon Thompson (CTO, SAIL Databank), Joseph Connor (Code4Health / NHS Digital), Ed Doshoki (IMO), Luca Toldo (SAP SE), Ed Brown (Lingumatics)

18:30-20:30	Drinks reception and dinner

Thursday, April 19th 2018

09:00-10:30	Research paper presentations – session 3
10:30-12:20	Posters and demos (with coffee) – session 4

12:20-13:00	Lunch

13:00-14:00	Keynote: Prof Wendy W. Chapman, University of Utah NLP Out of the Box: Is it Possible? Natural language processing is a critical ingredient in the transformation of health care. For decades we have been developing NLP tools to extract knowledge from clinical reports. In the last decade many open source tools have been made available so that in theory all hospitals can apply NLP for decision support, cohort creation, and quality improvement. Yet NLP remains elusive. Will NLP ever be out of the box? How close are we? What are the challenges we face when trying to scale NLP beyond the task for which an original system was created? In this talk, I will provide examples of these challenges and offer thoughts on scalability and portability.
14:00-14:30	Open discussions
14:30-15:10	Posters and demos (with coffee) – session 5
15:10-16:40	Panel: Understanding how to gain public trust in healthcare text analytics One of the biggest barriers to progressing the field of healthcare text analytics is gaining access to patient free text data from medical records and letters. Data custodians may be wary of allowing this data to be shared for research because of increased privacy risks and because no evidence exists about public opinions of medical text data being shared for research. In this panel session we will present the public engagement and involvement work being undertaken by Healtex members in this area, and hear from a patient representative, a clinician and a philosopher about their views on sharing medical text data. This will facilitate a discussion on special considerations that may be necessary for gaining public trust in data sharing for healthcare text analytics. Panelists: Elizabeth Ford, Lamiece Hassan, Malcolm Oswald, Jessica Stockdale, Benjamin Brown, Grace-anne McCorkle

16:40-17:00	Close

Keynote speakers

Prof Wendy W. Chapman, University of Utah

Dr. Chapman earned her Bachelor’s degree in Linguistics and her PhD in Medical Informatics from the University of Utah in 2000. From 2000-2010 she was a National Library of Medicine postdoctoral fellow and then a faculty member at the University of Pittsburgh. She joined the Division of Biomedical Informatics at the University of California, San Diego in 2010. In 2013, Dr. Chapman became the chair of the University of Utah, Department of Biomedical Informatics. Dr. Chapman’s research focuses on developing and disseminating resources for modeling and understanding information described in narrative clinical reports. She is interested not only in better algorithms for extracting information out of clinical text through natural language processing (NLP) but also in generating resources for improving the NLP development process (such as shareable annotations and open source toolkits) and in developing user applications to help non-NLP experts apply NLP in informatics-based tasks like clinical research and decision support.She has been a principal investigator on several NIH grants from the National Library of Medicine, National Institute for Dental and Craniofacial Research, and National Institute for General Medical Sciences. In addition, she has collaborated on multi-center grants including the ONC SHARP Secondary Use of Clinical Data and the iDASH National Center for Biomedical Computing. Dr. Chapman is a PI and a co-investigator on a number of VA Health Services Research and Development (HSRD) grant proposals extending the development and application of NLP within the VA.A tenured Professor at the University of Utah, Dr. Chapman continues her research in addition to leading the Department of Biomedical Informatics. Dr. Chapman is an elected fellow of the American College of Medical Informatics and currently serves as Treasurer, was the previous chair of the American Medical Informatics Association (AMIA) Natural Language Processing Working Group, and is the chair of the AMIA Student Paper Awards Committee.

Prof Pierre Zweigenbaum, LIMSI-CNRS

Dr. Zweigenbaum has pioneered the use of natural language processing in biomedicine, particularly in the field of biomedical information extraction and automatic linguistic knowledge extraction from free text in multiple languages. Dr. Zweigenbaum’s research focus is Natural Language Processing (NLP), with medicine as a main application domain – what is now known as “BioNLP”, of which he is one of the pioneers and a recognized authority.

Through his work on information extraction in multilingual settings, he has developed innovative methods (and companion tools) to detect various types of medical entities, expand abbreviations, resolve co-references, and detect relations. He has also designed novel methods for the automatic acquisition of linguistic knowledge from corpora and thesauri, as well as methods to help extend lexicons and terminologies, including in multiple languages.

Detailed Programme

Wednesday, April 18th 2018

08:30-09:00	Registration

09:00-11:00	PhD forum (Chair: Sumithra Velupillai) 9:00- 9:20 Shanu Sadhwani: Understanding under-diagnosis: The potential contribution of free-text to dementia research Early diagnosis of dementia has proven benefits to patients, yet, estimates suggest almost 50% of people living with dementia do not have a formal diagnosis. This study investigated how GPs record consultations when presented with a typical case of cognitive decline in an elderly patient. A cross-sectional within-subjects, experimental design using an online questionnaire, presented fictional clinical vignettes to 60 GPs in the UK. Participants were asked how they would record each consultation in Read codes and free-text. 68% of participants suspected dementia, with 58.5% recording this in the electronic health record (EHR). 16.7% of participants recorded this information within the free-text with no indication of dementia in Read codes. A lack of diagnostic dementia Read code in an EHR does not necessarily indicate that dementia is unidentified or that care is substandard. EHR researchers should be mindful that clinicians may use symptom codes and free-text when recording dementia. 9:20-9:40 Ghada Alfattni: Integrating Structured and Unstructured Sources for Temporal Representation of Patients’ Histories Temporal information (e.g. dates, duration, frequency) is key for making sense of healthcare data as health events (e.g. symptoms, treatments, etc.) all happen in a particular order and have a particular duration and frequency. Such information is both available in structured data (e.g. time-stamps for wearables or dates of examinations) and unstructured data (clinical narrative, patient reported data). Some of temporal Information is redundant, some is contradicting. The main purpose of this project is to develop a methodology to extract structure and integrate temporal information from a variety of healthcare data sources. It will involve designing text-mining methods to capture main expressions and relations in unstructured narrative and their aggregation to identify patterns together with structured data. The methodology will be applied to a number of real-world case studies in the clinical domain. 9:40-10:00 Fernando Santos Sanchez: Knowledge capture for a novel text analytics tool: analysis of Public Comments on Patient Information Leaflets to Measure Quality Perception and Patient Understanding The objective of this preliminary study is to assess which public comments on Patient Information Leaflets (PILs) for Randomised Controlled Trials (RCTs) measure the information quality and patient understanding of essential features of the trial. We collected comments from reviewers (n=24) on PILs for 4 trials of low-risk interventions that had low recruitment. Regression analysis of the data showed that the education level and number of comments made on specific sections of the leaflet have significant correlation with the participant understanding of the information (p<.0001 in both cases). On the other hand, the number of general comments on the PILs did not correlate with the patient understanding or the perception of document quality. The data collected by this study is being used to develop a novel text analytics Web platform to help PIL authors improve their documents. 10:00-10:20 Coffee break 10:20-10:40 Anastasia Zunic: A deep learning approach to sentiment analysis of patient narratives The proliferation of user-generated data on the Internet provides many practical applications of opinion analysis tasks that require customer opinion as an alternative to the more traditional research methods such as surveys, questionnaires, interviews and focus groups. Text mining has emerged as a solution to the problem of information overload associated with reading vast amounts of text from various sources. In particular, sentiment analysis or opinion mining is applied to the patient narratives in order to analyze and evaluate the progress and the outcome of a health treatment which the patient has received. We will conduct a systematic literature review in the domain of sentiment analysis of patient narratives in order to identify main issues associated with poor performance of sentiment analysis on this type of text data. (slides) 10:40-11:00 Maksim Belousov: Mining adverse drug reactions and their impact on patients’ quality of life in social media TBA
11:00-12:00	Career talk: Prof Wendy W. Chapman, University of Utah Navigating the Mountain of My Career: Lessons Learned from Skiing

12:00-13:00	Lunch

13:00-14:00	Keynote: Prof Pierre Zweigenbaum, LIMSI-CNRS ICD-10 coding of death certificates: CLEF e-Health Shared Tasks datasets and methods In many countries, causes of death are collected and coded into the WHO International Classification of Diseases (ICD) to provide key data for monitoring and comparing mortality across time and countries. Death certificates are very short texts that list causes of death. On the one hand, they exhibit the usual challenges of free text for automated processing and coding, including misspellings, ambiguity and variety of expression; and the size of the ICD classification, with well over 10,000 classes in ICD-10, increases the difficulty of the task. On the other hand, their constrained purpose and format may make them more amenable to automatic coding than the average clinical text. Moreover, the routine coding of these texts along the years has created gold standard annotations that are a natural source for creating large-scale training or test datasets. In the context of a collaboration with CépiDc, the team that codes causes of death in France, this led us to organize shared tasks that target ICD-10 coding of death certificates.I will present the shared tasks run over French and English certificates, and currently open on Italian and Hungarian data too. Gold standard annotations provide a set of codes for full certificates or at best for each certificate field, but no entity spans, therefore participants adopted two main strategies. They addressed the task either as an entity recognition and normalization task, using dictionary-based methods, or as a text classification task, using supervised classifiers. I will summarize these methods and their results, that set the current state of the art on this task, as well as our own experiments on these datasets.
14:00-15:00	Research paper presentations – session 1 (Chair Goran Nenadic) 14:05-14:25 Claire Grover, Richard Tobin, Beatrice Alex, Catherine Sudlow, Grant Mair and William Whiteley: Text Mining Brain Imaging Reports We describe a text mining system for classifying radiologists’ reports of CT and MRI brain scans, assigning labels indicating occurrence and type of stroke, as well as other observations. Our system, the Edinburgh Information Extraction for Radiology reports (EdIE-R) system, was developed and tested on a collection of 1,168 reports from the Edinburgh Stroke Study (ESS), a hospital-based register of stroke and transient ischaemic attack patients. Automated reading of reports such as these opens up avenues for population health monitoring and audit, and can provide a resource for epidemiological studies. Here we describe the EdIE-R system and report on its development and evaluation on annotated data from ESS. We aim to make the development and testing annotations for the ESS collection available for research. 14:25-14:45 Anoop D. Shah, Emily Garrett, Tim Williams, Spiros Denaxas and Harry Hemingway: Natural language processing of primary care records for patients with myocardial infarction: a pilot study Much clinical information in electronic health records (EHR) is unstructured and not easily accessible for research. We carried out a pilot study on 2000 patients with myocardial infarction (MI) in the CALIBER resource, which includes primary care data from the Clinical Practice Research Datalink. We previously developed rule-based natural language processing (NLP) software for primary care data (the Freetext Matching Algorithm), and applied it to text recorded within 90 days before or after the MI. We extracted 10,927 diagnoses, 3,658 test results, 3,313 negated statements negation, and 850 suspected diagnoses. Including free text increased the recorded proportion of patients with chest pain in the week prior to MI from 19% to 27%, and differentiated between MI subtypes in a quarter more patients than structured data alone. In summary, NLP can potentially unlock useful information from free text, but has not previously been used at scale in UK primary care EHR. (Slides) 14:45-14:55 Arron Lacey, Beata Fonferko-Shadrach, Ashley Akbari, Simon Thompson, David Ford, Ronan Lyons, Mark Rees and Owen Pickrell: Using natural language processing to extract clinical data from epilepsy clinic letters Electronic Health Records (EHR) are a powerful resource in enabling large-scale healthcare research. EHRs often lack detailed disease-specific information that is collected in free text within a clinical setting. We present an automated clinical text extraction application capable of accurately extracting several categories of epilepsy-specific information from unstructured clinic letters. We used the General Architecture for Text Engineering (GATE) framework to build custom rule sets for nine categories of epilepsy information as well as clinic date and date of birth across 200 clinic letters. We compared the results of our algorithm to a manual review by a clinician. We identified 1939 items of information with overall precision, recall and F1-score of 92.7%, 77.7% and 85.6%. Precision and recall for epilepsy specific categories were: epilepsy diagnosis (85.3%,92.4%), epilepsy type (93.7%,83.2%), focal seizures (99.0%,68.3%), generalised seizures (92.5%,57.0%), seizure frequency (92.0%,52.3%), medication (96.1%,94.0%), CT (66.7%,47.1%), MRI (96.6%,51.4%) and EEG (95.8%,40.6%). Our results show that Natural Language Processing techniques can be used to accurately extract rich details from clinic letters that often missing from routinely collected data. Capturing these new data types opens up the possibility of conducting novel epilepsy research, as well as showing the potential to expand to other disease areas.
15:00-15:30	Coffee break (with demos and posters) – session 2 15:10 Aji Barot (HealthUnlocked): Using word embeddings to link the language of patients with standardised terminology (demo) Analysis of user-generated content from HealthUnlocked’s communities has enabled us to build a patient-friendly ontology. In this project we used word embeddings to represent the most frequently used colloquialisms as well as formal medical terms on the platform – then computed similarities between them.
15:30-17:00	Panel: Text mining veterinary clinical records: opportunities for veterinary and human healthcare researchers Large volumes of veterinary electronic patient records (EPR), creating huge corpora (millions of records) of veterinary clinical narratives are being collated by projects including the Small Animal Veterinary Surveillance Network and VetCompass. These initiatives aim to gain insights applicable to surveillance and retrospective/prospective studies used to inform evidence-based medicine approaches. Linking to human healthcare corpora to take a ‘One Health’ approach might produce both methodology development and potentially novel results that could inform understanding of veterinary, environmental and human health issues. Various text mining developments are being applied to veterinary narratives, extracting key clinical variables including symptoms, clinical parameters, suspected diagnoses and potentially offering automated clinical coding. A number of challenges remain and this panel aims to initiate discussions on these and various opportunities surrounding veterinary text mining. Panelists: PJ Noble, Hannah Doit, Noel Kennedy, Fabio Rinaldi, Mercedes Arguello Casteleiro
17:00-18:30	HealTAC-2018 Industry Forum Technology awareness, the “rebirth” of AI, and financial pressures are increasingly leading the health sector to engage with the text analytics industry, for the provision of solutions to deliver data for research, administration and for regulatory requirements. The health sector presents the industry with a unique set of organisational, technical, social, and economic problems. The Industry Forum will be lead by a panel of experts from both industry and service procurement, and will explore these issues by presenting real world examples, and through in-depth discussion and debate with the audience. Panelists: Angus Roberts (King’s College London), Simon Thompson (CTO, SAIL Databank), Joseph Connor (Code4Health / NHS Digital), Ed Doshoki (IMO), Luca Toldo (SAP SE)

18:30-20:30	Drinks reception and dinner

Thursday, April 19th 2018

09:00-10:30	Research paper presentations – session 3 (chair Rob Stewart) 9:00-9:20 M. Arguello Casteleiro, R. Stevens, D. Maseda-Fernandez, M.J. Fernandez-Prieto, J.J. Des-Diz, C. Wroe, G. Demetriou, D.A. Singleton, E. Arsevska, P.J. Noble, P.H. Jones, J. Dukes-Mcewan, A.D. Radford, J. Keane and G. Nenadic: Can a Semantic Deep Learning approach gain insights for One Health? A case study on heart failure In the UK, the Small Animal Veterinary Surveillance Network (SAVSNET) has recently collected a 2.5M free-text de-identified clinical veterinary narratives dataset (VetCN for short) from approximately 500 veterinary clinics across the UK. This study applies Semantic Deep Learning (Deep Learning combined with Semantic Web technologies/resources) to 300K PubMed systematic review articles (PMSB for short) and VetCN from SAVSNET. The aim is gaining knowledge useful for medical and veterinary healthcare professionals, i.e. “One Health” knowledge. We implemented a text-processing pipeline that derives concepts with explicitly defined and machine-processable semantics (i.e. an ontology serves as a metadata schema) from the unstructured text by exploiting word embeddings created with CBOW and Skip-gram. To validate the effectiveness of the method, we focus on Heart Failure (HF) – a leading cause of human deaths worldwide – and utilise the BMJ Best Practice, as its content both supports frontline clinicians and promotes evidence-based medicine. Our results show that there is very little overlap among the concepts derived from both corpora (the PMSB and the VetCN) for HF, although all of them appear within The BMJ Best Practice for HF. 9:20-9:30 Hegler Tissot and Richard Dobson: Identifying misspelt names of drugs in medical records written in Portuguese There is an increasing amount of unstructured medical data that can be analysed for different purposes. However, information extraction from free text data may be particularly inefficient in the presence of spelling errors. Existing approaches use string similarity methods to search for valid words within a text, coupled with a supporting dictionary. However, they are not rich enough to encode both typing and phonetic misspellings. In this paper, we describe a hybrid similarity approach to efficiently perform a joint string and language-dependent phonetic similarity match in order to identify misspelt names of drugs in a set of medical records written in Portuguese. (Slides) 9:30-9:40 Josiah Poon, Yiu-Ming Ng, Kong-Ming Leung, Simon Poon, Justin Wu, Vincent Mok and Alexander Lau: Text mining of stroke outcomes from the electronic health records in an integrative stroke unit Stroke is one of top killers in the world and a leading cause of disability. Clinical data analysis can help reveal the effectiveness of treatments, which was usually supported by structured data stored in a database. However, there are valuable information kept in the discharge notes and the follow-up consultation notes, that are currently not analyzable because of its free-text nature. The aim of this pilot project was to create an enabler to extract this un-utilised digital information to become usable data for analysis. The most interesting pattern identified was the creation of character-based graphs by clinician to represent the location and strength of limb power. A rule-based system was created to apply to a set of clinical notes from an integrative stroke clinic. The system has achieved the performance of 0.84, 0.92, and 0.88 for recall, precision and F1 accordingly in the testing. This paper documented the experience and challenges experienced by the team. 9:40-9:50 Break 9:50-10:00 Alicja Piotrkowicz, Owen Johnson and Geoff Hall: Large scale healthcare text analytics with IBM Watson: a Pilot Study Significant amounts of health data are stored as free-text documents, producing large-scale corpora. Processing these massive datasets in order to identify and extract relevant clinical information is a considerable research challenge. One software solution that has been proposed in industry is the IBM Watson analytics suite which includes artificial intelligence based analytics capable of processing large document collections at scale. In this paper we present an overview of IBM Watson Content Analytics and reflect on an exploratory feasibility study we conducted using Content Analytics with large-scale corpora of clinical free-text reports from one British and one American hospital. 10:00-10:10 Danielle Mowery, Albert Park and Mike Conway: Applying the Transtheoretical Model of Behavioral Change to Reddit Data: A Pilot Study of Cessation Strategies and Outcomes among Tobacco Users Despite the existence of numerous studies demonstrating that tobacco use is strongly associated with cancer, heart disease, stroke, and other health problems, tobacco continues to be a major public health threat worldwide. Social media provides a rich resource for population-level, health-based monitoring and understanding tobacco cessation efforts and trends. In this pilot study, we identified common and emerging tactics used by Reddit users for tobacco cessation and demonstrated that the Transtheoretical Model – an established and widely used behavioural change theory – could be leveraged to track and identify successful cessation of tobacco usage within this population. (Slides) 10:10-10:30 Irena Spasic, David Owen, Andrew Smith and Kate Button: Closing in on open–ended patient questionnaires with text mining Knee injury and Osteoarthritis Outcome Score (KOOS) is an instrument used to quantify patient’s perception about their knee condition and associated problems. It is administered as a 42-item closed-ended questionnaire in which patients are asked to self-assess five outcomes: pain, other symptoms, activities of daily living, sport and recreation activities, and quality of life. We developed KLOG as a 10-item open-ended version of the KOOS questionnaire in an attempt order to obtain deeper insight into patients’ opinions including their unmet needs. However, open–ended nature of the questionnaire incurs analytical overhead associated with the interpretation of responses. The goal of this study was to automate such analysis. To that end, we implemented KLOSURE as a system for mining free–text responses to the KLOG questionnaire. The precision of the system varied between 64.8% and 95.3%, whereas the recall varied from 61.3% to 87.8% across the 10 questions. (Slides)
10:30-12:20	Posters and demos (with coffee) – session 4 10:40 Michael Upshall (UNSILO): UNSILO Classify for Medical Literature search (demo) UNSILO, in conjunction with medical publisher Karger Publishing of Basel, Switzerland, has built a tool to improve the speed and quality of literature searches for medical topics in Karger journals. The tool is based on UNSILO’s unsupervised concept extraction engine, but innovatively provides tools by which a human an edit and refine the search. Using the UNSILO Classify tool, users have dramatically reduced the time required to carry out Boolean searches of the medical literature. The tool requires no specialist information retrieval skills. (demo slides) 11:10 Simon Baker (University of Cambridge): Introducing the LION literature-based discovery system for cancer biology (demo) We introduce LION LBD, a literature-based discovery (LBD) system, enabling researchers to navigate published biomedical information, and supports hypothesis generation and testing. The system is built with a focus on the molecular biology of cancer using state-of-the-art machine learning and natural language processing (NLP) methods, including named entity recognition and grounding to domain ontologies. Covering a wide range of entity types and a novel approach for detecting references to the hallmarks of cancer in text. LION LBD implements a broad range of co-occurrence based metrics for analyzing the strength of entity associations, and its design allows real-time search to discover indirect associations between entities in a database of tens of millions of publications while preserving the ability of users to explore each mention in its original context in the literature. 11:40 Azad Dehghan (DeepCognito), Cathal Gallagher (d-Wise Technologies, Inc.): Blur/AITAP – Free Text De-identification (demo) Blur has the ability to find and anonymise data in documents whilst maintaining high levels of data utility. Subject identifiers are changed consistently and dates can be offset by a consistent value that is unique to each patient and protects patients’ privacy whilst maintaining the integrity of the data. Redaction is bare bones compliance, but anonymised documents with high data utility are what has real-world practical value. 12:00 Simon Thompson, Jeffrey Peng (Swansea University Medical School): MedGATE (demo) Swansea University Medical school has recently taken a step change in its capabilities within the NLP field and this demo will show the architecture that will be used to deliver a federated network of NHS nodes opening research access and deliver new dataset to the SAIL databank and others. The deployable solution, which uses GATE, also addresses issues of scaling and performance as well as data governance and other concerns common to NHS data projects.

12:20-13:00	Lunch

13:00-14:00	Keynote: Prof Wendy W. Chapman, University of Utah NLP Out of the Box: Is it Possible? Natural language processing is a critical ingredient in the transformation of health care. For decades we have been developing NLP tools to extract knowledge from clinical reports. In the last decade many open source tools have been made available so that in theory all hospitals can apply NLP for decision support, cohort creation, and quality improvement. Yet NLP remains elusive. Will NLP ever be out of the box? How close are we? What are the challenges we face when trying to scale NLP beyond the task for which an original system was created? In this talk, I will provide examples of these challenges and offer thoughts on scalability and portability.
14:00-14:30	Open discussions
14:30-15:10	Posters and demos – session 5 (with coffee break) 14:30 Angus Roberts, Matthew Broadbent, Anna Kolliakou, Jyoti Sanyal, Ian Roberts and Robert Stewart: Sharing NLP and Compute Services: Cloud Based Natural Language Processing at the NIHR Biomedical Research Centre, South London and Maudsley NHS Foundation Trust This demo session will showcase the cloud platform used at the UK’s largest end user of electronic health record natural language processing, the South London and Maudsley NHS Foundation Trust. The platform makes available over 100 NLP applications for analysts and end users to manage, and to routinely run over a corpus of 30 million documents, in support of epidemiological research. The platform is a model for single-site NLP service deployment, and opens up the possibility of service-based collaborative sharing of NLP applications and compute resources. In addition to a live and interactive demonstration of the platform, the session will provide time to discuss and explore the wider questions raised by NLP deployment, and by cross-site, collaborative NLP services. ( Demo slides ) 14:50 Edward Brown (Linguamatics): Extracting ICD-O Coded Pathology Diagnoses from Electronic Health Records Cancer hospitals and institutes hold large quantities of vital diagnosis information in free-text documents. Extracting this information would typically involve domain knowledge and thousands of hours of manual document review. By way of NLP and Machine Learning, we can free that information with a fraction of the time and expense. (Demo slides)
15:10-16:40	Panel: Understanding how to gain public trust in healthcare text analytics One of the biggest barriers to progressing the field of healthcare text analytics is gaining access to patient free text data from medical records and letters. Data custodians may be wary of allowing this data to be shared for research because of increased privacy risks and because no evidence exists about public opinions of medical text data being shared for research. In this panel session we will present the public engagement and involvement work being undertaken by Healtex members in this area, and hear from a patient representative, a clinician and a philosopher about their views on sharing medical text data. This will facilitate a discussion on special considerations that may be necessary for gaining public trust in data sharing for healthcare text analytics. Panelists: Elizabeth Ford, Lamiece Hassan, Malcolm Oswald, Jessica Stockdale, Benjamin Brown, Grace-anne McCorkle

16:40-17:00	Close

Accepted papers (long)

Text Mining Brain Imaging Reports
Claire Grover, Richard Tobin, Beatrice Alex, Catherine Sudlow, Grant Mair and William Whiteley
Closing in on open–ended patient questionnaires with text mining
Irena Spasic, David Owen, Andrew Smith and Kate Button
Natural language processing of primary care records for patients with myocardial infarction: a pilot study
Anoop D. Shah, Emily Garrett, Tim Williams, Spiros Denaxas and Harry Hemingway
Can a Semantic Deep Learning approach gain insights for One Health? A case study on heart failure
M. Arguello Casteleiro, R. Stevens, D. Maseda-Fernandez, M.J. Fernandez-Prieto, J.J. Des-Diz, C. Wroe, G. Demetriou, D.A. Singleton, E. Arsevska, P.J. Noble, P.H. Jones, J. Dukes-Mcewan, A.D. Radford, J. Keane and G. Nenadic

Accepted papers (short presentations)

Identifying misspelt names of drugs in medical records written in Portuguese
Hegler Tissot and and Richard Dobson
Using natural language processing to extract clinical data from epilepsy clinic letters
Arron Lacey, Beata Fonferko-Shadrach, Ashley Akbari, Simon Thompson, David Ford, Ronan Lyons, Mark Rees and Owen Pickrell
Text mining of stroke outcomes from the electronic health records in an integrative stroke unit
Josiah Poon, Yiu-Ming Ng, Kong-Ming Leung, Simon Poon, Justin Wu, Vincent Mok and Alexander Lau
Large scale healthcare text analytics with IBM Watson: a Pilot Study
Alicja Piotrkowicz, Owen Johnson and Geoff Hall
Applying the Transtheoretical Model of Behavioral Change to Reddit Data: A Pilot Study of Cessation Strategies and Outcomes among Tobacco Users
Danielle Mowery, Albert Park and Mike Conway

Posters

Session 2:

Linguistics Derived Rule-Based Methods in Chinese Biomedical and Clinical Text Mining
Josiah Poon, Kun-Jun Lee, Nick Enfield, Zevari Hung, Wai-Ling Lin, Wendy Wong, Yan-Li Ju, Simon Poon, Justin Wu and Alexander Lau
Machine Learning and NLP in EMRs: Identify CHF from Free Text Clinical Notes
Margot Yann, Therese Stukel, Karen Tu and Liisa Jaakkimainen.
KneeQApp: Supporting self-management of knee conditions with question answering
David Owen, Kate Button and Irena Spasic

Session 4:

Large-scale Text Mining of Veterinary Narrative in SAVSNET for Research and Surveillance
A.D. Radford, P.J. Noble, P.H. Jones, F. Sánchez-Vizcaíno, B. Brant, S. Smyth, E. Arsevska, J. Newman, D.A. Singleton, Mercedes Arguello Casteleiro and G. Nenadic
Identification of Time Expressions in Mental Health Records
Natalia Viani, André Bittar, Joyce Kam, Ayunni Alawi, Lucia Yin, Rina Dutta, Robert Stewart and Sumithra Velupillai
Early prediction of dementia with Lewy bodies with natural language processing of electronic health records
Julia Ive, Sumithra Velupillai and Robert Stewart
Temporal Distribution of Suicidality Mentions in Electronic Health Records
André Bittar, Sumithra Velupillai, Johnny Downs, Catherine Polling, Natalia Viani, Robert Stewart and Rina Dutta
Dementia-related information extraction from electronic patient records of a secondary care mental health NHS trust using the UK-CRIS platform
Kim Chi-Hun, Azad Dehghan, Luka Gligic, Nevado-Holgado Alejo, Simon Lovestone and Goran Nenadic
Mining free-text patient feedback comments
Azad Dehghan, Humayun Kayesh, Gavin Daker-White, Evan Kontopantelis, Will Dixon, Caroline Sanders and Goran Nenadic
Mining Free-Text Medical Notes for Suicide Risk Assessment
Marios Adamou, Grigoris Antoniou, Elissavet Greasidou, Vincenzo Lagani, Paulos Charonyktakis and Ioannis Tsamardinos
Ontology Learning from Hand Crafted, Paediatric MCQs
Ghader Kurdi, Bijan Parsia and Uli Sattler
Extracting adverse drug reactions and their context using sequence labelling ensembles
Maksim Belousov, Nikola Milosevic, William Dixon and Goran Nenadic

Session 5:

Improving the Identification of Diagnostic Test Accuracy Evidence using Lexical Statistics
Amal Alharbi and Mark Stevenson
Exploring research domains for author disambiguation from MEDLINE to ClinicalTrials.gov and back
Dina Vishnyakova, Raul Rodriguez-Esteban and Fabio Rinaldi
Using the Big Picture – A Deep and Distant Approach to Classification of Suicidal Ideation in Autism Spectrum Disorder
Xingyi Song, Johnny Downs, Sumithra Velupillai, Rachel Holden, Maxim Kikoler, Rina Dutta and Angus Roberts

Panels

HealTAC-2018 Industry Forum
(organisers: Angus Roberts, Clare Grover and David Milward)
Understanding how to gain public trust in healthcare text analytics
(organisers: Elizabeth Ford, Lamiece Hassan, Malcolm Oswald, Jessica Stockdale, Benjamin Brown, Grace-anne McCorkle)
Text mining veterinary clinical records – opportunities for veterinary and human healthcare researchers
(organisers: Rachel Dean, PJ Noble, Noel Kennedy, Fabio Rinaldi, Mercedes Arguello Casteleiro)

Demos

Sharing NLP and Compute Services: Cloud Based Natural Language Processing at the NIHR Biomedical Research Centre, South London and Maudsley NHS Foundation Trust
Angus Roberts, Matthew Broadbent, Anna Kolliakou, Jyoti Sanyal, Ian Roberts and Robert Stewart
UNSILO Classify for Medical Literature search
Michael Upshall (UNISLO)
Introducing the LION literature-based discovery system for cancer biology
Simon Baker (University of Cambridge)
Blur/AITAP – Free Text De-identification
Azad Dehghan (DeepCognito), Cathal Gallagher (d-Wise Technologies, Inc.)
Using word embeddings to link the language of patients with standardised terminology
Aji Barot (HealthUnlocked)
Extracting ICD-O Coded Pathology Diagnoses, and Quality Measures, from Electronic Health Records
Edward Brown (Linguamatics)
MedGATE
Simon Thompson, Jeffrey Peng (Swansea University Medical School)

PhD forum

[see also details here.]

Understanding underdiagnosis: The potential contribution of free-text to dementia research
Shanu Sadhwani
Knowledge capture for a novel text analytics tool: analysis of Public Comments on Patient Information Leaflets to Measure Quality Perception and Patient Understanding
Fernando Santos Sanchez
Integrating Structured and Unstructured Sources for Temporal Representation of Patients’ Histories
Ghada Alfattni
A deep learning approach to sentiment analysis of patient narratives
Anastasia Zunic
Mining adverse drug reactions and their impact on patients’ quality of life in social media
Maksim Belousov