Programme

Given the situation with the corona virus, HealTAC 2020 is postponed until 2021. This will also include the pre-conference workshop.

While we are postponing the face-to-face conference, we recognise that it is important that the community stays in touch and the work that has been prepared for this year should be made available to the wider community. In collaboration with Healtex and Health Data Research UK, we are therefore organising a half-day virtual event on April 23 this year for the community catch-up and discussions, focusing on translational activities related to COVID-19.

Please register here – the registration is free, but only registered participants will be provided links to join the meeting and break-out sessions.

Joining the meeting

The details as how to join the event have been sent through Eventbrite/Healtex (Subject: “Message to attendees of HealTAC 2020 – virtual meeting”) – if you can’t find these, please email contact_at_healtex.org and we will re-send the details as soon as possible.

Please use the main zoom URL for all plenary sessions and different zooms as specified in the email for different breakout sessions.

10:30	Introductions Sumithra Velupillai, Goran Nenadic (Healtex) Caroline Cake (HDR UK)
10:40	Keynote: Prof Patrick Ruch (HEG/HESSO Geneva) Literature triage services to support biocuration: from neXtProt to COVID-19 slides
11:10	PhD forum Contributors ... Mat Rawsthorne: Prototyping an unobtrusive measure of online psychological flexibility in a moderated mental health peer support forum Task sharing in the form of patients supporting each other via the internet is seen as a potential solution to the rising unmatched global demand for professional psychological help. However, there are outstanding research questions about how mental health conditions can affect engagement with digital technologies, the potential risks and the therapeutic nature of online relationships. The REBOOT study of a moderated mood disorders forum (RCT of the Effectiveness of Big White Wall Compared to Other Online SupporT) was designed to explore these [4]. This embedded doctoral study examines the language used by participants, categorising the flexibility of their responses as a way of providing insight to their thinking patterns and the influence of peer interactions. (slides) Evangelia Gogoulou: Language Models for early prediction of treatment failure in Internet based Cognitive Behavioral Therapy A Contextualized Language Modeling architecture, targeting the task of early prediction of treatment failure in Internet based Cognitive Behavioral Therapy, is presented in a conceptual level. The application dataset consists of data for more than 8000 patients in Stockholm. Given text written by patients during the treatment, the model predicts the probability of treatment failure, defined as getting a post-treatment score under clinical cutoff. Our hypothesis is that the statistical properties of patient’s language, learned by our model, are strong predictors of treatment success. The goal is to achieve sufficient classification accuracy as early as possible, using the minimum amount of patient’s text. The proposed model is going to support clinicians in their work, as part of an operational intelligent system for Internet based Cognitive Behavioral Therapy. (slides) Julia Walsh: How and for what purposes is spontaneously generated online patient experience (SGOPE) data currently being used within health research As interest grows in developing methods to utilise spontaneously generated online patient experience (SGOPE) data, the aim of this study is to review the current literature to explore both why and how spontaneously generated online patient experience data is currently being used within health research. We used an umbrella scoping review method, to incorporate the interdisciplinary nature of the research surrounding it. Search terms based on keyword combinations to cover setting, analysis, usage and methods. After searching and screening, 40 review papers from 2015-2019 were included. Most reviews from Yearbook of Medical Informatics (7) and JMIR (6). All others only one per journal. Data extraction table created to tabulate key features and findings. Reviews were classified as to their purpose, the health conditions they studied, and the methods used. Common concerns revolved around the lack of explanation around methods, the lack of credible evaluation and the need for shared datasets. (slides) Panelists: Hercules Dalianis, Patrich Ruch, Rob Stewart
12:00	Lunch break & breakout sessions (each starting at 12:30) Clinical text mining 1 Contributors ... Hang Dong, Honghan Wu and Cathie Sudlow: Initialising Label Embedding for Automated Medical Coding of Clinical Notes Diagnostic or procedural coding of medical free-text documents (e.g. discharge summaries) aims to derive a coded summary of disease-related information about patients. In UK hospitals, such coding is usually done manually but could potentially be automated. In this study, preliminary results of automated medical coding of discharge summaries from the MIMIC-3 intensive care dataset showed that a novel, label embedding initialisation approach, which leverages label correlation and improves the state-of-the-art deep-learning-based multi-label classification model. (slides) Rebecca Bendayan, Honghan Wu, Zejko Kraljevic, Robert Stewart, Tom Searle, Jaya Chaturvedi, Jayati Das-Munshi, Zina Ibrahim, Aurelie Mascio, Angus Roberts, Daniel Bean and Richard Dobson: Identifying physical health comorbidities in a cohort of individuals with severe mental illness: An application of SemEHR Multimorbidity research in mental health services requires data from physical health conditions which is traditionally limited in mental health care electronic health records. In this study, we aimed to extract this data from physical health conditions from clinical notes using natural language processing (NLP). Specifically we use SemEHR, an open source toolkit that integrates text mining and semantic computing to identifying mentions of Unified Medical Language System (UMLS) concepts in clinical documents. Data were extracted from Clinical Record Interactive Search (CRIS) system at South London and Maudsley Biomedical Research Centre (SLaM BRC) and the cohort consisted of all individuals who had received a primary or secondary diagnosis of severe mental illness between 2007 and 2018. Three pairs of annotators annotated 2403 documents with an average Cohen’s κ of 0.757. Results show that the NLP performance (F1 score) varies across different diseases areas from 0.601 – 0.954 suggesting that the language patterns or terminologies of different condition groups entail different technical challenges to the same NLP task. (slides) Dmitri Roussinov, Andrew Conkie, Andrew Patterson and Christopher Sainsbury: Predicting Clinical Events Based on Raw Text: from Bag of Words to Attention-Based Transformers Predicting hospital readmission or a patient fatality often happens to be resource- and life- saving, thus is a very important and challenging task for NLP/IR and machine learning applications in e-Health domain. While many successful approaches exist to predict such clinical events based on categorical and numerical variables, most of health records consist of raw text clinical notes. However, the models taking advantage of the free from natural language found in those notes rarely reach the accuracy level acceptable for the clinicians. In addition, in spite of their success in other domains, the superiority of deep neural approaches over classical bags of words has not yet been convincingly demonstrated for this task. Using a publicly available dataset with clinical notes, we have explored several text classification models to predict patient re-admission or a fatality and established that 1) The performance of our deep neural models exceed those based on bag of words by several percentage points. 2) This allows to achieve the accuracy typically acceptable for the clinicians as of practical use (area under the ROC curve above .70). 3) Our model based on averaging n-gram embeddings works the best, exceeding more specialized ones such as recurrent, convolutional, and attention-based transformer models. 4) Our modifications in the attention-based transformer model suggested here to overcome its input size limit are crucial to achieve the top performance. Clinical text mining 2 Contributors ... Natasha Chilman, Xingyi Song, Angus Roberts, Robert Stewart, Anna Kolliakou and Jayati Das-Munshi: Text-Mining Occupations from the Electronic Health Record: The Development and Evaluation of a Natural Language Processing Application to Extract Occupations from the Clinical Record Interactive Search (CRIS) Case Register Patient occupations are not well captured or characterized in electronic health record systems. Currently there are no existing methods for the reliable extraction of occupations from psychiatric healthcare narrative text, which is a major barrier for research investigating socioeconomic position, employment and health. This paper traces the development and testing of a novel application using natural language processing methods to extract occupations from the free-text of de-identified clinical records from a major mental health Trust in the UK. The application was developed and tested on extracts of psychiatric assessments which detail the patient’s personal history and was also tested on 13 other free-text fields in the clinical record. Results from testing indicate that the application is reaching high levels of performance, particularly when applied to personal history sections of clinical text (precision = 0.96). This application represents an internationally unique advance in our ability to text-mine occupations from healthcare text. Thomas Rowlands and Tim Beck: GWAS Information Extraction using Natural Language Processing The GWAS Central resource enables broad and convenient integrative visualisation of, and access to, summary-level GWAS data. GWAS reported in the biomedical literature are manually interpreted, curated and imported into the database. To accelerate the import process, and further expand the database content, we are using NLP to extract GWAS data and metadata from publication full texts. The accuracy of the phenotype and trait descriptions extracted is enhanced by incorporating an ontology-driven text mining process. (poster). Jaya Chaturvedi and Angus Roberts: Comparison of Drug Resources to Build a Dictionary of Medications Recognition and extraction of medications from electronic health record (EHR) documents is an important sub-task of clinical information extraction, as medications and their prescriptions are central to understanding the course of patient treatment, both in clinical care and in research. Medication recognition is usually considered to be a type of named entity recognition (NER). As with any NER task, lists of terms known to describe the entity of interest are of value in the recognition step. Additional value may be provided if these terms are derived from existing domain knowledge resources, allowing the text to be linked to these resources. However, such resources are not generally designed with NLP in mind. We compare different available medication resources and use these to create gazetteers of medication names for NLP purposes. We compare the terms in these gazetteers to both an existing gazetteer and a set of medication terms manually extracted from EHR documents, and draw conclusions about the suitability of different resources, and the way in which their terms are processed for inclusion in the gazetteers. (poster) Veterinary text mining Contributors ... Oluwatosin Dairo, Mercedes Arguello, David Singleton, Goran Nenadic, PJ Noble and Glen Martin: Automating anti-microbial usage compliance monitoring from veterinary clinical text Tracking anti-microbial usage compliance in small animal practice is not well structured as most of the data is stored in free-text prescription notes. We performed a case study on Ronaxan, a broad spectrum anti-microbial drug, using the SAVSNET database to demonstrate the potential of large-scale processing of veterinary clinical notes. (poster, slides) Mercedes Arguello-Casteleiro, Pj Noble, Oluwatosin Dairo, Alan D. Radford and Goran Nenadic: Real-time monitoring of veterinary clinical symptoms: a case study on temperature and vomiting With recent examples of fast-developing potential outbreaks of diseases in small companion animals (e.g. prolific vomiting in February 2020), it is important to develop methods for real-time monitoring of key clinical symptoms. We here demonstrate a case study on monitoring changes in vomiting and temperature patterns using the SAVSNET data. Peter Noble and Alan Radford: Identification of disease outbreak in UK dogs using topic modelling The University of Liverpool Small Animal Surveillance Network (SAVSNET) collects clinical notes from a sentinel network of ~250 UK veterinary practices. Clinical narratives are submitted in real time (~6000 records/day) alongside patient signalment and post-code. Veterinary surgeons apply a ‘main presenting complaint’ (mpc) label to all records chosen from: gastroenteric, respiratory, pruritus, tumour, kidney disease, other unwell, post-op check, vaccination, other healthy. From 01/11/2019 there was a national 2-fold increase in gastroenteric mpc (gi_mpc) records coinciding with local vets reporting an acute vomiting syndrome. Here we evaluate unsupervised record classification using latent Dirichlet allocation (LDA) to identify this outbreak. Topic modelling provides a means to generate unsupervised record annotation that can identify disease outbreaks in clinical records with potential to recognise unanticipated disease syndromes. (slides) Imogen Schofield, Noel Kennedy, David Brodbelt, Stijn Niessen, David Church, Rebecca Geddes and Dan O’Neill: Incorporation of Natural Language Processing into Veterinary Electronic Health Records for Machine Learning Analysis Veterinary Electronic Health Records (EHR) collected within the VetCompassTM programme are largely formed as unstructured free text. A pilot study on one condition in dogs (Cushing’s syndrome) was undertaken. Cases of Cushing’s typically show non-specific clinical signs and the time from initial suspicion to diagnosis can be long. This study aimed to apply natural language processing (NLP) methods to free text veterinary EHR for earlier disease prediction. Methods: NLP was applied to free text notes drawn from the VetCompassTM corpus. The input were the clinical notes written up to 30 days prior to initial suspicion in time order (longitudinal). A Transformer was applied to the individual notes and a long short-term memory (LSTM) processed the output in time-series order to track disease progression through time. The algorithm predicted whether or not the canine patient will ultimately get diagnosed with Cushing’s, at the moment the clinician first suspected Cushing’s. The model was validated on a held-out test set. Results: A population of 905,544 dogs under veterinary care in 2016, were screened by a domain expert to identify 10,141 dogs with a suspicion of Cushing’s syndrome in the EHR. Of these, 1,066 (10.5%) dogs went on to be diagnosed with Cushing’s but most did not (9,075; 89.5%). The model was trained to predict diagnosed/not diagnosed dogs prior to this point in time. Following NLP-analysis of the clinical notes and the LSTM algorithm training, the model demonstrated good discrimination AUROC = 0.74 (95% CI 0.73 – 0.74) indicating the NLP algorithm could identify dogs which truly have Cushing’s prior to them receiving a clinical diagnosis. Conclusion: This pilot study demonstrates the applicability of NLP methods to facilitate diagnosis of disease earlier in the clinical trajectory. A future prospective intervention study could evaluate the clinical impact of incorporating such methods into the VetCompassTM programme. (slides) Heather Davies, Peter-John Noble, Gina Pinchbeck and David Killick: Dictionary Creation for the Identification of Suspected Adverse Drug Reactions in Veterinary Free Text Clinical Narratives Electronic health records potentially contain information pertaining to adverse drug reactions observed by veterinary surgeons or pet owners, which have not been reported to regulatory authorities. We developed a method to identify dogs suffering from convulsions following use of lokivetmab (Cytopoint) in veterinary free text narratives from first opinion veterinary practices participating in the SAVSNET project in the UK. (slides) Patient-focused text mining, engagement and governance Contributors ... Elizabeth Ford and Keegan Curlewis: The Benefits to Patients of Research using Clinical Free Text: A Systematic Review of the Literature Engagement activities with patients and regulators, in which we have discussed their views on the sharing of clinical free text data for research, have identified that stakeholders would like to understand the potential clinical benefits that could be achieved if access to free text for clinical research were improved. We set out to systematically identify and analyse all UK research studies which used clinical free text and which reported direct or potential benefits to patients, in terms of informing or improving the quality of their care. We identified 17 papers and derived four themes of benefits: health-care quality improvement, observational research, case-finding for clinical trials, and development of clinical decision support. This group of studies can be used to form a public-facing repository of research case-studies which will help the health-care text analysis research community better communicate the impact of their work. (slides) Lama Alsudias and Paul Rayson: Analyzing the Spread of Influenza in Arabic Twitter Twitter is a real time messaging platform used by people and organisations to share information on different topics like health. It could potentially be useful to analyse tweets for health surveillance purposes in order to reduce reporting lag time compared to the traditional approaches for monitoring infectious diseases. Here, we collect around 2,000 tweets in Arabic related to Influenza and clean them. We label the tweets depending on our Arabic Infectious Diseases Ontology which includes non-standard terminology. We apply Multi-label classification techniques: Binary Relevance, Classifier Chains, Label Powerset, and Adapted Algorithm (MLKNN). We find that F1-score is around (80-88)\%. (poster) Hanna Berg, Aron Henriksson, Uno Fors and Hercules Dalianis: De-identification of Clinical Text for Secondary Use: Research Issues Privacy is challenged by both recent advances in AI-related technologies and recent legal regulations, making it important to solve issues concerning it. The problem of privacy has been extensively studied within the privacy community, but has largely focused on methods for protecting and assessing the privacy of structured data. Research to protect the integrity of patients based on clinical text has primarily referred to US law and relied on automatically recognising predetermined, direct and indirect, identifiers. This article discusses the various challenges concerning the re-use of unstructured clinical data, in particular in the form of clinical text, and focuses on ambiguous and vague terminology, how different legislation affects the requirements on de-identification, differences between methods for unstructured and structured data, the impact of using only named entity recognition (NER) and replacing sensitive data with surrogates, and the lack of measures for usability and risk of re-identification. (slides)
13:00	Keynote: Prof Kalina Bontcheva (University of Sheffield) COVID-19 Disinformation: Why it matters and why should you get involved slides
13:30	Panel discussion: Text analytics for Covid-19 Panelists: Rob Stewart, KCL (chair); Luke Slater, University of Birmingham (slides); Beatrice Alex, University of Edinburgh (slides), Richard Dobson, KCL; Will Dixon, University of Manchester (slides)
14:30	Demo break-outs Linguamatics (starts at 14:30 and again at 15:00) Details... This session will cover a software demo and an overview of Linguamatics NLP solutions in healthcare. NLP can be used to extract information hidden in unstructured data to provide better insight into patients’ conditions and to generate better patient outcomes. In addition to extracting information reliably and accurately, other factors such as agility, transparency and scalability are important. Considering the nature of healthcare data, the approach should also be able to address data confidentiality and provide real-time analysis. As a company with over 15 years of experience in life sciences and healthcare, Linguamatics aims at providing easy-to-use NLP platforms that enables non-NLP experts to extract information. IE-Viz (starts 14:45 and again at 15:00) Details... We present EdIE-Viz, a web interface that enables the comparison of our rule based and neural network models for entity recognition and negation detection in brain imaging reports. Our rule based model, EdIE-R, outperforms our neural network model, EdIE-N, on both tasks evaluated using F-score. Introspection of our models sketches a more nuanced picture, with both models differing on how they are affected by spelling errors, corpus statistics and artefacts such as annotation ambiguities. We leverage EdIE-Viz to highlight the strengths and weaknesses of both models for the aforementioned cases, report our findings, and point towards promising directions for future work. We release the EdIE-viz demo and make the corresponding code for the demo and both systems available on GitHub. MedCATTrainer:2.0 (starts 14:45 and again at 15:00) Details... We present MedCATTrainer:2.0 an assisted annotation interface for building, modifying and further training by domain experts of a supplied ‘MedCAT’ model, a state-of-the-art library for training biomedical Named Entity Recognition and Linking models. MedCATTrainer:1.0 has been entirely re-written adding features such as: configurable individual ‘user’ and ‘project’ settings, multi-document concept annotation, common biomedical ontology support (SNOMED-CT, UMLS and ICD-10, OPCS-4), and a streamlined user interface combining the concept annotation and research question specific annotations into one interface. We have used the interface for annotating epilepsy clinical notes with two clinical staff for symptoms, diseases and medications, and have a further ongoing project with 16 clinical coders across two hospital sites at King’s College Hospital NHS Foundation Trust, UK. (slides) SemantEx (starts at 14:30 and again at 15:00) Details... Statistical natural language processing (NLP) is widely used in healthcare text-analytics and can achieve useful results in sentiment analysis and short-text question answering. However, in applications where data is sensitive and decisions need to be defensible, it faces significant problems. By contrast the SemantEx system uses a deep rules-based approach to extract and graph, entity relationships across long (or indeed multiple) un-structured texts. We will demonstrate actual and potential advantages of this approach in a variety of applications. These range from redaction in General Data Protection Regulation (GDPR) subject access requests, through a variety of human-resources (for instance care-giver/care-user interaction) monitoring tasks, to multiple subject timeline extraction. 2Dsearch (starts at 14:30 and again 14:45) Details... Knowledge workers (such as healthcare information professionals, patent agents and media monitoring professionals) need to create and execute search strategies that are accurate, repeatable and transparent. The traditional solution is to use the command line ‘query builders’ offered by proprietary database vendors. However, these offer limited support for error checking or query optimization, and their output is often compromised by errors and inefficiencies. In this demo, we present a new approach to query formulation in which concepts are expressed as objects on a two-dimensional canvas and relationships are articulated by direct manipulation. Interactive search term suggestions are provided using a combination of machine learning NLP techniques and ontological resources. This approach has the potential to eliminate many sources of inefficiency, make the query semantics more transparent, and offers unified approach to structured searching which promotes transparent methods and reproducible results.
15:15	Open session and Close

HealTAC 20202 contributions

Oral presentations (long)

Samuel Dobbie, Huw Strafford, Owen Pickrell, Beata Fonferko-Shadrach, Ashley Akbari, Simon Thompson and Arron Lacey	Markup: A Web-based Clinical Annotation Tool with Enhanced Ontology Mapping
Denis Newman-Griffis, Jonathan Camacho Maldonado, Pei-Shu Ho, Albert Lai and Eric Fosler-Lussier	Diving for Pearls: Indexing Mobility Information in US Social Security Administration Clinical Records with a Neural Relevance Tagger
Ghada Alfattni, Natalia Viani, Sumithra Velupillai and Goran Nenadic	Investigating the Challenges in Medication Information Extraction from Biomedical Free-text Data

Oral presentations (short)

Abeed Sarker, Yuan-Chi Yang and Mohammed Al-Garadi	A Light-weight Text Summarizer for Fast Access to Medical Evidence
Hanna Berg, Aron Henriksson, Uno Fors and Hercules Dalianis	De-identification of Clinical Text for Secondary Use: Research Issues
Natasha Chilman, Xingyi Song, Angus Roberts, Robert Stewart, Anna Kolliakou and Jayati Das-Munshi	Text-Mining Occupations from the Electronic Health Record: The Development and Evaluation of a Natural Language Processing Application to Extract Occupations from the Clinical Record Interactive Search (CRIS) Case Register
Elizabeth Ford and Keegan Curlewis	The Benefits to Patients of Research using Clinical Free Text: A Systematic Review of the Literature
Dmitri Roussinov, Andrew Conkie, Andrew Patterson and Christopher Sainsbury	Predicting Clinical Events Based On Raw Text: From Bag of Words to Attention-Based Transformers
Jaya Chaturvedi and Angus Roberts	Comparison of Drug Resources to Build a Dictionary of Medications
Anastazia Zunic, Padraig Corcoran and Irena Spasic	Improving the performance of sentiment analysis in health and wellbeing using domain knowledge
Rebecca Bendayan, Honghan Wu, Zejko Kraljevic, Robert Stewart, Tom Searle, Jaya Chaturvedi, Jayati Das-Munshi, Zina Ibrahim, Aurelie Mascio, Angus Roberts, Daniel Bean and Richard Dobson	Identifying physical health comorbidities in a cohort of individuals with severe mental illness: An application of SemEHR

PhD forum

Mat Rawsthorne	Prototyping an unobtrusive measure of online psychological flexibility in a moderated mental health peer support forum
Julia Walsh	How and for what purposes is spontaneously generated online patient experience (SGOPE) data currently being used within health research
Evangelia Gogoulou	Language Models for early prediction of treatment failure in Internet based Cognitive Behavioral Therapy

Posters

Mathieu Guidere, Christian Fluhr, Aurélie Rossi and Zhen Wang	NLP Applied to Online Suicide Intention Detection (poster)
Sally Giles, Lamiece Hassan, Will Dixon, Goran Nenadic, Stephen Campbell and Caroline Sanders	Automated surveillance of discussions relating to adverse drug reactions on social media: what do patients think?
Luke T Slater, William Bradlow, Robert Hoehndorf, Simon Ball and Georgios V Gkoutos	Komenti: Semantic Text Annotation with Synonym Expansion and Negation Detection
Lama Alsudias and Paul Rayson	Analyzing the Spread of Influenza in Arabic Twitter (poster)
Daphné Chopard, Matthias Treder and Irena Spasić	Automated Data Augmentation for Natural Language Processing
Eleanor Walsh, Avon Huxor, Emma Turner, Raul Santos-Rodriguez and Chris McWilliams	Elements of free text used in decision making: an exemplar from death reviews in prostate cancer and learning disabilities.
Nemanja Vaci, Ivan Koychev, Chi-Hun Kim, Andrey Kormilitzin, Qiang Liu, Christopher Lucas, Azad Dehghan, Goran Nenadic and Alejo Nevado-Holgado	Real-World Effectiveness, Predictors and Onset of Cholinesterase Inhibitors and Memantine in Dementia: A Retrospective Health Record Study of 7400 Individuals using the UK CRIS Platform
Anastazia Zunic, Padraig Corcoran and Irena Spasic	Sentiment analysis in health and well-being
Thomas Rowlands and Tim Beck	GWAS Information Extraction using Natural Language Processing
Hang Dong, Honghan Wu and Cathie Sudlow	Initialising Label Embedding for Automated Medical Coding of Clinical Notes
Aurelie Mascio, Zeljko Kraljevic, Angus Roberts, Robert Stewart and Richard Dobson	Extracting symptoms from medical records: a comparative analysis
Risha Govind, Richard Dobson, Cathryn Lewis and James Maccabe	When did the patient start taking the medication?
Imogen Schofield, Noel Kennedy, David Brodbelt, Stijn Niessen, David Church, Rebecca Geddes and Dan O’Neill	Incorporation of Natural Language Processing into Veterinary Electronic Health Records for Machine Learning Analysis
Oluwatosin Dairo, Mercedes Arguello, David Singleton, Goran Nenadic, Pj Noble and Glen Martin	Automating anti-microbial usage compliance monitoring from veterinary clinical text (poster)
Mercedes Arguello-Casteleiro, Pj Noble, Oluwatosin Dairo, Alan D. Radford and Goran Nenadic	Real-time monitoring of veterinary clinical symptoms: a case study on temperature and vomiting
Lama Alqurashi, Angus Roberts and Rina Dutta	Content bias – a new perspective on linguistic bias
Eva-Lisa Meldau, Sara Hedfors Vidlin, Lucie Gattepaille, Henric Taavola, Lovisa Sandberg, Yasunori Aoki and G. Niklas Norén	Deep learning for de-identification of case narratives in reports of suspected adverse drug reactions
Xi Yang, Goran Nenadic and Chengkun Wu	Mining a Stroke Knowledge Graph from Biomedical & TCM literature
Peter Noble and Alan Radford	Identification of disease outbreak in UK dogs using topic modelling.
Denis Newman-Griffis, Guy Divita, Bart Desmet, Ayah Zirikly, Carolyn Rosé and Eric Fosler-Lussier	Ambiguity in Medical Concept Normalization: An Analysis of Types and Coverage in Electronic Health Record Datasets
Heather Davies, Peter-John Noble, Gina Pinchbeck and David Killick	Dictionary Creation for the Identification of Suspected Adverse Drug Reactions in Veterinary Free Text Clinical Narratives
Adam Lavertu, Russ B. Altman, Amit Kaushal and Jonathan Chen	Using BERT to extract hidden value from clinical pager data

Demos

Hui Feng, Simon Beaulah and David Milward	Linguamatics Natural Language Processing Software Demo
Michael Barclay and Alan Rosenbach	Knowledge Acquisition AI for Healthcare Text-Analytics
Andreas Grivas, Claire Grover, Richard Tobin and Beatrice Alex	EdIE-Viz: An interactive demo for text mining brain imaging reports
Dominic Cassidy, Arron Lacey, Stuart Hendren and Owen Pickrell	Using ARGA to extract structured epilepsy data from unstructured clinic letters – A software demonstration
Thomas Searle, Zeljko Kraljevic, Anthony Shek, James Teo, Mark Richardson, Zina Ibrahim and Richard Dobson	MedCATTrainer:2.0 An Assisted Free-Text Annotation Interface to Train a Named Entity Recognition and Linking Model for Clinical Terminologies
Tony Russell-Rose and Jon Chamberlain	The use of text analytics in transparent and reproducible structured searching