Search

Summary Objectives: We survey recent work in biomedical NLP on building more adaptable or generalizable models, with a focus on work dealing with electronic health record (EHR) texts, to better understand recent trends in this area and identify opportunities for future research. Methods: We searched PubMed, the Institute of Electrical and Electronics Engineers (IEEE), the Association for Computational Linguistics (ACL) anthology, the Association for the Advancement of Artificial Intelligence (AAAI) proceedings, and Google Scholar for the years 2018-2020. We reviewed abstracts to identify the most relevant and impactful work, and manually extracted data points from each of these papers to characterize the types of methods and tasks that were studied, in which clinical domains, and current state-of-the-art results. Results: The ubiquity of pre-trained transformers in clinical NLP research has contributed to an increase in domain adaptation and generalization-focused work that uses these models as the key component. Most recently, work has started to train biomedical transformers and to extend the fine-tuning process with additional domain adaptation techniques. We also highlight recent research in cross-lingual adaptation, as a special case of adaptation. Conclusions: While pre-trained transformer models have led to some large performance improvements, general domain pre-training does not always transfer adequately to the clinical domain due to its highly specialized language. There is also much work to be done in showing that the gains obtained by pre-trained transformers are beneficial in real world use cases. The amount of work in domain adaptation and transfer learning is limited by dataset availability and creating datasets for new domains is challenging. The growing body of research in languages other than English is encouraging, and more collaboration between researchers across the language divide would likely accelerate progress in non-English clinical NLP.

Background Rates of suicide attempts and deaths are highest on Mondays and these occur more frequently in the morning or early afternoon, suggesting weekly temporal and diurnal variation in suicidal behaviour. It is unknown whether there are similar time trends on social media, of posts relevant to suicide. We aimed to determine temporal and diurnal variation in posting patterns on the Reddit forum SuicideWatch, an online community for individuals who might be at risk of, or who know someone at risk of suicide. Methods We used time series analysis to compare date and time stamps of 90,518 SuicideWatch posts from 1st December 2008 to 31st August 2015 to (i) 6,616,431 posts on the most commonly subscribed general subreddit, AskReddit and (ii) 66,934 of these AskReddit posts, which were posted by the SuicideWatch authors. Results Mondays showed the highest proportion of posts on SuicideWatch. Clear diurnal variation was observed, with a peak in the early morning (2:00–5:00 h), and a subsequent decrease to a trough in late morning/early afternoon (11:00–14:00 h). Conversely, the highest volume of posts in the control data was between 20:00–23:00 h. Conclusions Posts on SuicideWatch occurred most frequently on Mondays: the day most associated with suicide risk. The early morning peak in SuicideWatch posts precedes the time of day during which suicide attempts and deaths most commonly occur. Further research of these weekly and diurnal rhythms should help target populations with support and suicide prevention interventions when needed most.

Background Duration of untreated psychosis (DUP) is an important clinical construct in the field of mental health, as longer DUP can be associated with worse intervention outcomes. DUP estimation requires knowledge about when psychosis symptoms first started (symptom onset), and when psychosis treatment was initiated. Electronic health records (EHRs) represent a useful resource for retrospective clinical studies on DUP, but the core information underlying this construct is most likely to lie in free text, meaning it is not readily available for clinical research. Natural Language Processing (NLP) is a means to addressing this problem by automatically extracting relevant information in a structured form. As a first step, it is important to identify appropriate documents, i.e., those that are likely to include the information of interest. Next, temporal information extraction methods are needed to identify time references for early psychosis symptoms. This NLP challenge requires solving three different tasks: time expression extraction, symptom extraction, and temporal “linking”. In this study, we focus on the first step, using two relevant EHR datasets. Results We applied a rule-based NLP system for time expression extraction that we had previously adapted to a corpus of mental health EHRs from patients with a diagnosis of schizophrenia (first referrals). We extended this work by applying this NLP system to a larger set of documents and patients, to identify additional texts that would be relevant for our long-term goal, and developed a new corpus from a subset of these new texts (early intervention services). Furthermore, we added normalized value annotations (“2011–05”) to the annotated time expressions (“May 2011”) in both corpora. The finalized corpora were used for further NLP development and evaluation, with promising results (normalization accuracy 71–86%). To highlight the specificities of our annotation task, we also applied the final adapted NLP system to a different temporally annotated clinical corpus. Conclusions Developing domain-specific methods is crucial to address complex NLP tasks such as symptom onset extraction and retrospective calculation of duration of a preclinical syndrome. To the best of our knowledge, this is the first clinical text resource annotated for temporal entities in the mental health domain.

Summary We describe the incidence of suicidality (2007–2017) in people with depression treated by secondary mental healthcare services at South London and Maudsley NHS Trust (n = 26 412). We estimated yearly incidence of ‘suicidal ideation’ and ‘high risk of suicide’ from structured and free-text fields of the Clinical Record Interactive Search system. The incidence of suicidal ideation increased from 0.6 (2007) to 1 cases (2017) per 1000 population. The incidence of high risk of suicide, based on risk forms, varied between 0.06 and 0.50 cases per 1000 adult population (2008–2017). Electronic health records provide the opportunity to examine suicidality on a large scale, but the impact of service-related changes in the use of structured risk assessment should be considered.

Background Suicide is a serious public health issue, accounting for 1.4% of all deaths worldwide. Current risk assessment tools are reported as performing little better than chance in predicting suicide. New methods for studying dynamic features in electronic health records (EHRs) are being increasingly explored. One avenue of research involves using sentiment analysis to examine clinicians’ subjective judgments when reporting on patients. Several recent studies have used general-purpose sentiment analysis tools to automatically identify negative and positive words within EHRs to test correlations between sentiment extracted from the texts and specific medical outcomes (eg, risk of suicide or in-hospital mortality). However, little attention has been paid to analyzing the specific words identified by general-purpose sentiment lexicons when applied to EHR corpora. Objective This study aims to quantitatively and qualitatively evaluate the coverage of six general-purpose sentiment lexicons against a corpus of EHR texts to ascertain the extent to which such lexical resources are fit for use in suicide risk assessment. Methods The data for this study were a corpus of 198,451 EHR texts made up of two subcorpora drawn from a 1:4 case-control study comparing clinical notes written over the period leading up to a suicide attempt (cases, n=2913) with those not preceding such an attempt (controls, n=14,727). We calculated word frequency distributions within each subcorpus to identify representative keywords for both the case and control subcorpora. We quantified the relative coverage of the 6 lexicons with respect to this list of representative keywords in terms of weighted precision, recall, and F score. Results The six lexicons achieved reasonable precision (0.53-0.68) but very low recall (0.04-0.36). Many of the most representative keywords in the suicide-related (case) subcorpus were not identified by any of the lexicons. The sentiment-bearing status of these keywords for this use case is thus doubtful. Conclusions Our findings indicate that these 6 sentiment lexicons are not optimal for use in suicide risk assessment. We propose a set of guidelines for the creation of more suitable lexical resources for distinguishing suicide-related from non–suicide-related EHR texts.

In the global effort to prevent death by suicide, many academic medical institutions are implementing natural language processing (NLP) approaches to detect suicidality from unstructured clinical text in electronic health records (EHRs), with the hope of targeting timely, preventative interventions to individuals most at risk of suicide. Despite the international need, the development of these NLP approaches in EHRs has been largely local and not shared across healthcare systems. In this study, we developed a process to share NLP approaches that were individually developed at King’s College London (KCL), UK and Weill Cornell Medicine (WCM), US - two academic medical centers based in different countries with vastly different healthcare systems. After a successful technical porting of the NLP approaches, our quantitative evaluation determined that independently developed NLP approaches can detect suicidality at another healthcare organization with a different EHR system, clinical documentation processes, and culture, yet do not achieve the same level of success as at the institution where the NLP algorithm was developed (KCL approach: F1-score 0.85 vs. 0.68, WCM approach: F1-score 0.87 vs. 0.72). Shared use of these NLP approaches is a critical step forward towards improving data-driven algorithms for early suicide risk identification and timely prevention.

Receiving timely and appropriate treatment is crucial for better health outcomes, and research on the contribution of specific variables is essential. In the mental health domain, an important research variable is the date of psychosis symptom onset, as longer delays in treatment are associated with worse intervention outcomes. The growing adoption of electronic health records (EHRs) within mental health services provides an invaluable opportunity to study this problem at scale retrospectively. However, disease onset information is often only available in open text fields, requiring natural language processing (NLP) techniques for automated analyses. Since this variable can be documented at different points during a patient’s care, NLP methods that model clinical and temporal associations are needed. We address the identification of psychosis onset by: 1) manually annotating a corpus of mental health EHRs with disease onset mentions, 2) modelling the underlying NLP problem as a paragraph classification approach, and 3) combining multiple onset paragraphs at the patient level to generate a ranked list of likely disease onset dates. For 22/31 test patients (71%) the correct onset date was found among the top-3 NLP predictions. The proposed approach was also applied at scale, allowing an onset date to be estimated for 2483 patients.

Suicide is a serious public health issue worldwide, yet current clinical methods for assessing a person's risk of taking their own life remain unreliable and new methods for assessing suicide risk are being explored. The widespread adoption of electronic health records (EHRs) has opened up new possibilities for epidemiological studies of suicide and related behaviour amongst those receiving healthcare. These types of records capture valuable information entered by healthcare practitioners at the point of care. However, much recent work has relied heavily on the structured data of EHRs, whilst much of the important information about a patient's care pathway is recorded in the unstructured text of clinical notes. Accessing and structuring text data for use in clinical research, and particularly for suicide and self-harm research, is a significant challenge that is increasingly being addressed using methods from the fields of natural language processing (NLP) and machine learning (ML). In this review, we provide an overview of the range of suicide-related studies that have been carried out using the Clinical Records Interactive Search (CRIS): a database for epidemiological and clinical research that contains de-identified EHRs from the South London and Maudsley NHS Foundation Trust. We highlight the variety of clinical research questions, cohorts and techniques that have been explored for suicide and related behaviour research using CRIS, including the development of NLP and ML approaches. We demonstrate how EHR data provides comprehensive material to study prevalence of suicide and self-harm in clinical populations. Structured data alone is insufficient and NLP methods are needed to more accurately identify relevant information from EHR data. We also show how the text in clinical notes provide signals for ML approaches to suicide risk assessment. We envision increased progress in the decades to come, particularly in externally validating findings across multiple sites and countries, both in terms of clinical evidence and in terms of NLP and machine learning method transferability.

Background Mobile health apps are increasingly available and used in a clinical context to monitor young people’s mood and mental health. Despite the benefits of accessibility and cost-effectiveness, consumer engagement remains a hurdle for uptake and continued use. Hundreds of mood-monitoring apps are publicly available to young people on app stores; however, few studies have examined consumer perspectives. App store reviews held on Google and Apple platforms provide a large, rich source of naturally generated, publicly available user reviews. Although commercial developers use these data to modify and improve their apps, to date, there has been very little in-depth evaluation of app store user reviews within scientific research, and our current understanding of what makes apps engaging and valuable to young people is limited. Objective This study aims to gain a better understanding of what app users consider useful to encourage frequent and prolonged use of mood-monitoring apps appropriate for young people. Methods A systematic approach was applied to the selection of apps and reviews. We identified mood-monitoring apps (n=53) by a combination of automated application programming interface (API) methods. We only included apps appropriate for young people based on app store age categories (apps available to those younger than 18 years). We subsequently downloaded all available user reviews via API data scraping methods and selected a representative subsample of reviews (n=1803) for manual qualitative content analysis. Results The qualitative content analysis revealed 8 main themes: accessibility (34%), flexibility (21%), recording and representation of mood (18%), user requests (17%), reflecting on mood (16%), technical features (16%), design (13%), and health promotion (11%). A total of 6 minor themes were also identified: notification and reminders; recommendation; privacy, security, and transparency; developer; adverts; and social/community. Conclusions Users value mood-monitoring apps that can be personalized to their needs, have a simple and intuitive design, and allow accurate representation and review of complex and fluctuating moods. App store reviews are a valuable repository of user engagement feedback and provide a wealth of information about what users value in an app and what user needs are not being met. Users perceive mood-monitoring apps positively, but over 20% of reviews identified the need for improvement.

Background How neighbourhood characteristics affect the physical safety of people with mental illness is unclear. Aims To examine neighbourhood effects on physical victimisation towards people using mental health services. Method We developed and evaluated a machine-learning-derived free-text-based natural language processing (NLP) algorithm to ascertain clinical text referring to physical victimisation. This was applied to records on all patients attending National Health Service mental health services in Southeast London. Sociodemographic and clinical data, and diagnostic information on use of acute hospital care (from Hospital Episode Statistics, linked to Clinical Record Interactive Search), were collected in this group, defined as ‘cases’ and concurrently sampled controls. Multilevel logistic regression models estimated associations (odds ratios, ORs) between neighbourhood-level fragmentation, crime, income deprivation, and population density and physical victimisation. Results Based on a human-rated gold standard, the NLP algorithm had a positive predictive value of 0.92 and sensitivity of 0.98 for (clinically recorded) physical victimisation. A 1 s.d. increase in neighbourhood crime was accompanied by a 7% increase in odds of physical victimisation in women and an 13% increase in men (adjusted OR (aOR) for women: 1.07, 95% CI 1.01–1.14, aOR for men: 1.13, 95% CI 1.06–1.21, P for gender interaction, 0.218). Although small, adjusted associations for neighbourhood fragmentation appeared greater in magnitude for women (aOR = 1.05, 95% CI 1.01–1.11) than men, where this association was not statistically significant (aOR = 1.00, 95% CI 0.95–1.04, P for gender interaction, 0.096). Neighbourhood income deprivation was associated with victimisation in men and women with similar magnitudes of association. Conclusions Neighbourhood factors influencing safety, as well as individual characteristics including gender, may be relevant to understanding pathways to physical victimisation towards people with mental illness.

Funder: EPSRC Healtex Feasibility Funding (grant EP/N027280/1): "Towards Shareable Data in Clinical Natural Language Processing: Generating Synthetic Electronic Health Records", Funder: National Institute for Health Research (NIHR) Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and King’s College London, Funder: National Institute for Health Research Post Doctoral Fellowship award (grant number PDF-2017-10-029), Funder: Health Data Research UK, Funder: Swedish Research Council(2015-00359)/the Marie Sklodowska Curie Actions, A serious obstacle to the development of Natural Language Processing (NLP) methods in the clinical domain is the accessibility of textual data. The mental health domain is particularly challenging, partly because clinical documentation relies heavily on free text that is difficult to de-identify completely. This problem could be tackled by using artificial medical data. In this work, we present an approach to generate artificial clinical documents. We apply this approach to discharge summaries from a large mental healthcare provider and discharge summaries from an intensive care unit. We perform an extensive intrinsic evaluation where we (1) apply several measures of text preservation; (2) measure how much the model memorises training data; and (3) estimate clinical validity of the generated text based on a human evaluation task. Furthermore, we perform an extrinsic evaluation by studying the impact of using artificial text in a downstream NLP text classification task. We found that using this artificial data as training data can lead to classification results that are comparable to the original results. Additionally, using only a small amount of information from the original data to condition the generation of the artificial data is successful, which holds promise for reducing the risk of these artificial data retaining rare information from the original data. This is an important finding for our long-term goal of being able to generate artificial clinical data that can be released to the wider research community and accelerate advances in developing computational methods that use healthcare data.

Objective Pre-hospital emergency medical services use clinical decision support systems (CDSS) to triage calls. Call handlers often supplement this by making free text notes covering key incident information. We investigate whether machine learning approaches using features from such free text notes can improve prediction of unconscious patients who require conveyance. Materials and methods We analysed a subset of all London Ambulance Service calls that were triaged through the Medical Priority Dispatch System (MPDS) as involving an unconscious or fainting patient in 2018. We use and compare two machine learning algorithms: random forest (RF) and gradient boosting machine (GBM). For each incident, we predict whether the patient will be conveyed to a hospital emergency department or equivalent using as features 1) the MPDS code, 2) the free text notes and 3) the two together. We evaluate model performance using the area under the curve (AUC) metric. Given the imbalance of outcomes (patient conveyed 71 %, not conveyed 29 %), we also consider sensitivity and specificity. Results Using only the MPDS code resulted in an AUC of 0.57. Using the text notes gave an improved AUC score of 0.63 and combining the two gave an AUC score of 0.64 (scores were similar for RF and GBM). GBM models scored better on sensitivity (0.93 vs 0.62 for RF in the combined model), but specificity was lower (0.17 vs. 0.56 for RF in the combined model). Conclusions Using information contained in the free text notes made by call handlers in combination with MPDS improves prediction of unconscious and fainting patients requiring conveyance to a hospital emergency department (or equivalent) when compared with machine learning models using MPDS codes only. This suggests there is some useful information in unstructured data captured by emergency call handlers that complements MPDS codes. Quantifying this gain can help inform emergency medical service policy when evaluating the decision to expand or augment existing CDSS.

While Dementia with Lewy Bodies (DLB) is the second most common type of neurodegenerative dementia following Alzheimer’s Disease (AD), it is difficult to distinguish from AD. We propose a method for DLB detection by using mental health record (MHR) documents from a (3-month) period before a patient has been diagnosed with DLB or AD. Our objective is to develop a model that could be clinically useful to differentiate between DLB and AD across datasets from different healthcare institutions. We cast this as a classification task using Convolutional Neural Network (CNN), an efficient neural model for text classification. We experiment with different representation models, and explore the features that contribute to model performances. In addition, we apply temperature scaling, a simple but efficient model calibration method, to produce more reliable predictions. We believe the proposed method has important potential for clinical applications using routine healthcare records, and for generalising to other relevant clinical record datasets. To the best of our knowledge, this is the first attempt to distinguish DLB from AD using mental health records, and to improve the reliability of DLB predictions.

Extracting and modeling temporal information in clinical text is an important element for developing timelines and disease trajectories. Time information in written text varies in preciseness and explicitness, posing challenges for NLP approaches that aim to accurately anchor temporal information on a timeline. Relative and incomplete time expressions (RI-Timexes) are expressions that require additional information for their temporal anchor to be resolved, but few studies have addressed this challenge specifically. In this study, we aimed to reproduce and verify a classification approach for identifying anchor dates and relations in clinical text, and propose a novel relation classification approach for this task.

This article describes the development and evaluation of a set of knowledge patterns that provide guidelines and implications of design for developers of mental health portals. The knowledge patterns were based on three foundations: (1) knowledge integration of language technology approaches; (2) experiments with language technology applications and (3) user studies of portal interaction. A mixed-methods approach was employed for the evaluation of the knowledge patterns: formative workshops with knowledge pattern experts and summative surveys with experts in specific domains. The formative evaluation improved the cohesion of the patterns. The results of the summative evaluation showed that the problems discussed in the patterns were relevant for the domain, and that the knowledge embedded was useful to solve them. Ten patterns out of thirteen achieved an average score above 4.0, which is a positive result that leads us to conclude that they can be used as guidelines for developing health portals.

To automatically analyse complex trajectory information enclosed in clinical text (e.g. timing of symptoms, duration of treatment), it is important to understand the related temporal aspects, anchoring each event on an absolute point in time. In the clinical domain, few temporally annotated corpora are currently available. Moreover, underlying annotation schemas - which mainly rely on the TimeML standard - are not necessarily easily applicable for applications such as patient timeline reconstruction. In this work, we investigated how temporal information is documented in clinical text by annotating a corpus of medical reports with time expressions (TIMEXes), based on TimeML. The developed corpus is available to the NLP community. Starting from our annotations, we analysed the suitability of the TimeML TIMEX schema for capturing timeline information, identifying challenges and possible solutions. As a result, we propose a novel annotation schema that could be useful for timeline reconstruction: CALendar EXpression (CALEX).

The development of Natural Language Processing (NLP) solutions for information extraction from electronic health records (EHRs) has grown in recent years, as most clinically relevant information in EHRs is documented only in free text. One of the core tasks for any NLP system is to extract clinically relevant concepts such as symptoms. This information can then be used for more complex problems such as determining symptom onset, which requires temporal information. In the mental health domain, comprehensive vocabularies for specific disorders are scarce, and rarely contain keywords that reflect real-world terminology use. We explore the use of embedding techniques to automatically generate lexical variants of psychosis symptoms into vocabularies, that can be used in complex downstream NLP tasks. We study the impact of the underlying text material on generating useful lexical entries, experimenting with different corpora and with unigram/bigram models. We also propose a method to automatically compute thresholds for choosing the most relevant terms. Our main contribution is a systematic study of unsupervised vocabulary generation using different corpora for an understudied clinical use-case. Resulting lexicons are publicly available.

Risk assessment of suicidal behavior is a time-consuming but notoriously inaccurate activity formental health services globally. In the last 50 years a large number of tools have been designed for suicide risk assessment, and tested in a wide variety of populations, but studies show that these tools suffer fromlow positive predictive values.More recently, advances in research fields such as machine learning and natural language processing applied on large datasets have shown promising results for health care, and may enable an important shift in advancing precision medicine. In this conceptual review, we discuss established risk assessment tools and examples of novel data-driven approaches that have been used for identification of suicidal behavior and risk. We provide a perspective on the strengths and weaknesses of these applications to mental health-related data, and suggest research directions to enable improvement in clinical practice, This manuscript was written as a result of a workshop that was held at the Institute of Psychiatry, Psychology and Neuroscience, King’s College London, financially supported by the European Science Foundation (ESF) Research Networking Programme Evaluating Information Access Systems: http://elias-network.eu/. SV is supported by the Swedish Research Council (2015-00359) and the Marie Skłodowska Curie Actions, Cofund, Project INCA 600398. EB-G is partially supported by grants from Instituto de Salud Carlos III (ISCIII PI13/02200; PI16/01852), Delegación del Gobierno para el Plan Nacional de Drogas (20151073); American Foundation for Suicide Prevention (AFSP) (LSRG- 1-005-16). NW is supported by the UCLH NIHR Biomedical Research Centre. DN is supported by The Alan Turing Institute under the EPSRC grant EP/N510129/1, with an Alan Turing Institute Fellowship (TU/A/000006). RP has received support from a Medical Research Council (MRC) Health Data Research UK Fellowship (MR/S003118/1) and a Starter Grant for Clinical Lecturers (SGL015/1020) supported by the Academy of Medical Sciences, The Wellcome Trust, MRC, British Heart Foundation, Arthritis Research UK, the Royal College of Physicians and Diabetes UK. DL is supported by the UK Medical Research Council under grant MR/N028244/2 and the King’s Centre for Military Health Research. JD is supported by a Medical Research Council (MRC) Clinical Research Training Fellowship (MR/L017105/1). RD is funded by a Clinician Scientist Fellowship (research project e-HOST-IT) from the Health Foundation in partnership with the Academy of Medical Sciences

A major obstacle to the development of Natural Language Processing (NLP) methods in the biomedical domain is data accessibility. This problem can be addressed by generating medical data artificially. Most previous studies have focused on the generation of short clinical text, and evaluation of the data utility has been limited. We propose a generic methodology to guide the generation of clinical text with key phrases. We use the artificial data as additional training data in two key biomedical NLP tasks: text classification and temporal relation extraction. We show that artificially generated training data used in conjunction with real training data can lead to performance boosts for data-greedy neural network algorithms. We also demonstrate the usefulness of the generated data for NLP setups where it fully replaces real training data., BioNLP 2019

Background Electronic health care records (EHRs) are a rich source of health-related information, with potential for secondary research use. In the United Kingdom, there is no national marker for identifying those who have previously served in the Armed Forces, making analysis of the health and well-being of veterans using EHRs difficult. Objective This study aimed to develop a tool to identify veterans from free-text clinical documents recorded in a psychiatric EHR database. Methods Veterans were manually identified using the South London and Maudsley (SLaM) Biomedical Research Centre Clinical Record Interactive Search—a database holding secondary mental health care electronic records for the SLaM National Health Service Foundation Trust. An iterative approach was taken; first, a structured query language (SQL) method was developed, which was then refined using natural language processing and machine learning to create the Military Service Identification Tool (MSIT) to identify if a patient was a civilian or veteran. Performance, defined as correct classification of veterans compared with incorrect classification, was measured using positive predictive value, negative predictive value, sensitivity, F1 score, and accuracy (otherwise termed Youden Index). Results A gold standard dataset of 6672 free-text clinical documents was manually annotated by human coders. Of these documents, 66.00% (4470/6672) were then used to train the SQL and MSIT approaches and 34.00% (2202/6672) were used for testing the approaches. To develop the MSIT, an iterative 2-stage approach was undertaken. In the first stage, an SQL method was developed to identify veterans using a keyword rule–based approach. This approach obtained an accuracy of 0.93 in correctly predicting civilians and veterans, a positive predictive value of 0.81, a sensitivity of 0.75, and a negative predictive value of 0.95. This method informed the second stage, which was the development of the MSIT using machine learning, which, when tested, obtained an accuracy of 0.97, a positive predictive value of 0.90, a sensitivity of 0.91, and a negative predictive value of 0.98. Conclusions The MSIT has the potential to be used in identifying veterans in the United Kingdom from free-text clinical documents, providing new and unique insights into the health and well-being of this population and their use of mental health care services.

This paper describes a continuing initiative of the International Society for Disease Surveillance designed to bring together public health practitioners and analytics solution developers from both academia and industry. Funded by the Defense Threat Reduction Agency, a series of consultancies have been conducted on a range of topics of pressing concern to public health (e.g. developing methods to enhance prediction of asthma exacerbation, developing tools for asyndromic surveillance from chief complaints). The topic of this final consultancy, conducted at the University of Utah in January 2017, is focused on defining a roadmap for the development of algorithms, tools, and datasets for improving the capabilities of text processing algorithms to identify negated terms (i.e. negation detection) in free-text chief complaints and triage reports.

International audience; Background:Natural language processing applied to clinical text or aimed at a clinical outcome has been thriving inrecent years. This paper offers the first broad overview of clinical Natural Language Processing (NLP) for languagesother than English. Recent studies are summarized to offer insights and outline opportunities in this area.Main Body:We envision three groups of intended readers: (1) NLP researchers leveraging experience gained in otherlanguages, (2) NLP researchers faced with establishing clinical text processing in a language other than English, and(3) clinical informatics researchers and practitioners looking for resources in their languages in order to apply NLPtechniques and tools to clinical practice and/or investigation. We review work in clinical NLP in languages other thanEnglish. We classify these studies into three groups: (i) studies describing the development of new NLP systems orcomponents de novo, (ii) studies describing the adaptation of NLP architectures developed for English to anotherlanguage, and (iii) studies focusing on a particular clinical application.Conclusion:We show the advantages and drawbacks of each method, and highlight the appropriate applicationcontext. Finally, we identify major challenges and opportunities that will affect the impact of NLP on clinical practiceand public health studies in a context that encompasses English as well as other languages.

For psychiatric disorders such as schizophrenia, longer durations of untreated psychosis are associated with worse intervention outcomes. Data included in electronic health records (EHRs) can be useful for retrospective clinical studies, but much of this is stored as unstructured text which cannot be directly used in computation. Natural Language Processing (NLP) methods can be used to extract this data, in order to identify symptoms and treatments from mental health records, and temporally anchor the first emergence of these. We are developing an EHR corpus annotated with time expressions, clinical entities and their relations, to be used for NLP development. In this study, we focus on the first step, identifying time expressions in EHRs for patients with schizophrenia. We developed a gold standard corpus, compared this corpus to other related corpora in terms of content and time expression prevalence, and adapted two NLP systems for extracting time expressions. To the best of our knowledge, this is the first resource annotated for temporal entities in the mental health domain.

The number of people affected by mental illness is on the increase and with it the burden on health and social care use, as well as the loss of both productivity and quality-adjusted life-years. Natural language processing of electronic health records is increasingly used to study mental health conditions and risk behaviours on a large scale. However, narrative notes written by clinicians do not capture first-hand the patients' own experiences, and only record cross-sectional, professional impressions at the point of care. Social media platforms have become a source of 'in the moment' daily exchange, with topics including well-being and mental health. In this study, we analysed posts from the social media platform Reddit and developed classifiers to recognise and classify posts related to mental illness according to 11 disorder themes. Using a neural network and deep learning approach, we could automatically recognise mental illness-related posts in our balenced dataset with an accuracy of 91.08% and select the correct theme with a weighted average accuracy of 71.37%. We believe that these results are a first step in developing methods to characterise large amounts of user-generated content that could support content curation and targeted interventions.

Research into suicide prevention has been hampered by methodological limitations such as low sample size and recall bias. Recently, Natural Language Processing (NLP) strategies have been used with Electronic Health Records to increase information extraction from free text notes as well as structured fields concerning suicidality and this allows access to much larger cohorts than previously possible. This paper presents two novel NLP approaches – a rule-based approach to classify the presence of suicide ideation and a hybrid machine learning and rule-based approach to identify suicide attempts in a psychiatric clinical database. Good performance of the two classifiers in the evaluation study suggest they can be used to accurately detect mentions of suicide ideation and attempt within free-text documents in this psychiatric database. The novelty of the two approaches lies in the malleability of each classifier if a need to refine performance, or meet alternate classification requirements arises. The algorithms can also be adapted to fit infrastructures of other clinical datasets given sufficient clinical recording practice knowledge, without dependency on medical codes or additional data extraction of known risk factors to predict suicidal behaviour.

The number of people affected by mental illness is on the increase and with it the burden on health and social care use, as well as the loss of both productivity and quality-adjusted life-years. Natural language processing of electronic health records is increasingly used to study mental health conditions and risk behaviours on a large scale. However, narrative notes written by clinicians do not capture first-hand the patients’ own experiences, and only record cross-sectional, professional impressions at the point of care. Social media platforms have become a source of ‘in the moment’ daily exchange, with topics including well-being and mental health. In this study, we analysed posts from the social media platform Reddit and developed classifiers to recognise and classify posts related to mental illness according to 11 disorder themes. Using a neural network and deep learning approach, we could automatically recognise mental illness-related posts in our balenced dataset with an accuracy of 91.08% and select the correct theme with a weighted average accuracy of 71.37%. We believe that these results are a first step in developing methods to characterise large amounts of user-generated content that could support content curation and targeted interventions.

Background The ShARe/CLEF eHealth challenge lab aims to stimulate development of natural language processing and information retrieval technologies to aid patients in understanding their clinical reports. In clinical text, acronyms and abbreviations, also referenced as short forms, can be difficult for patients to understand. For one of three shared tasks in 2013 (Task 2), we generated a reference standard of clinical short forms normalized to the Unified Medical Language System. This reference standard can be used to improve patient understanding by linking to web sources with lay descriptions of annotated short forms or by substituting short forms with a more simplified, lay term. Methods In this study, we evaluate 1) accuracy of participating systems’ normalizing short forms compared to a majority sense baseline approach, 2) performance of participants’ systems for short forms with variable majority sense distributions, and 3) report the accuracy of participating systems’ normalizing shared normalized concepts between the test set and the Consumer Health Vocabulary, a vocabulary of lay medical terms. Results The best systems submitted by the five participating teams performed with accuracies ranging from 43 to 72 %. A majority sense baseline approach achieved the second best performance. The performance of participating systems for normalizing short forms with two or more senses with low ambiguity (majority sense greater than 80 %) ranged from 52 to 78 % accuracy, with two or more senses with moderate ambiguity (majority sense between 50 and 80 %) ranged from 23 to 57 % accuracy, and with two or more senses with high ambiguity (majority sense less than 50 %) ranged from 2 to 45 % accuracy. With respect to the ShARe test set, 69 % of short form annotations contained common concept unique identifiers with the Consumer Health Vocabulary. For these 2594 possible annotations, the performance of participating systems ranged from 50 to 75 % accuracy. Conclusion Short form normalization continues to be a challenging problem. Short form normalization systems perform with moderate to reasonable accuracies. The Consumer Health Vocabulary could enrich its knowledge base with missed concept unique identifiers from the ShARe test set to further support patient understanding of unfamiliar medical terms.

Mental Health Records (MHRs) contain free- text documentation about patients’ suicide and suicidality. In this paper, we address the prob- lem of determining whether grammatic vari- ants (inflections) of the word “suicide” are af- firmed or negated. To achieve this, we pop- ulate and annotate a dataset with over 6,000 sentences originating from a large repository of MHRs. The resulting dataset has high Inter- Annotator Agreement ( κ 0.93). Furthermore, we develop and propose a negation detection method that leverages syntactic features of text 1 . Using parse trees, we build a set of ba- sic rules that rely on minimum domain knowl- edge and render the problem as binary clas- sification (affirmed vs. negated). Since the overall goal is to identify patients who are ex- pected to be at high risk of suicide, we focus on the evaluation of positive (affirmed) cases as determined by our classifier. Our negation detection approach yields a recall (sensitivity) value of 94.6% for the positive cases and an overall accuracy value of 91.9%. We believe that our approach can be integrated with other clinical Natural Language Processing tools in order to further advance information extrac- tion capabilities

Background Electronic patient records (EPRs) contain a large amount of information written in free text. This information is considered very valuable for research but is also very sensitive since the free text parts may contain information that could reveal the identity of a patient. Therefore, methods for de-identifying EPRs are needed. The work presented here aims to perform a manual and automatic Protected Health Information (PHI)-annotation trial for EPRs written in Swedish. Methods This study consists of two main parts: the initial creation of a manually PHI-annotated gold standard, and the porting and evaluation of an existing de-identification software written for American English to Swedish in a preliminary automatic de-identification trial. Results are measured with precision, recall and F -measure. Results This study reports fairly high Inter-Annotator Agreement (IAA) results on the manually created gold standard, especially for specific tags such as names. The average IAA over all tags was 0.65 F -measure (0.84 F -measure highest pairwise agreement). For name tags the average IAA was 0.80 F -measure (0.91 F -measure highest pairwise agreement). Porting a de-identification software written for American English to Swedish directly was unfortunately non-trivial, yielding poor results. Conclusion Developing gold standard sets as well as automatic systems for de-identification tasks in Swedish is feasible. However, discussions and definitions on identifiable information is needed, as well as further developments both on the tag sets and the annotation guidelines, in order to get a reliable gold standard. A completely new de-identification software needs to be developed.

Summary Objectives: We present a review of recent advances in clinical Natural Language Processing (NLP), with a focus on semantic analysis and key subtasks that support such analysis. Methods: We conducted a literature review of clinical NLP research from 2008 to 2014, emphasizing recent publications (2012-2014), based on PubMed and ACL proceedings as well as relevant referenced publications from the included papers. Results: Significant articles published within this time-span were included and are discussed from the perspective of semantic analysis. Three key clinical NLP subtasks that enable such analysis were identified: 1) developing more efficient methods for corpus creation (annotation and de-identification), 2) generating building blocks for extracting meaning (morphological, syntactic, and semantic subtasks), and 3) leveraging NLP for clinical utility (NLP applications and infrastructure for clinical use cases). Finally, we provide a reflection upon most recent developments and potential areas of future NLP development and applications. Conclusions: There has been an increase of advances within key NLP subtasks that support semantic analysis. Performance of NLP semantic analysis is, in many cases, close to that of agreement between humans. The creation and release of corpora annotated with complex semantic information models has greatly supported the development of new tools and approaches. Research on non-English languages is continuously growing. NLP methods have sometimes been successfully employed in real-world clinical tasks. However, there is still a gap between the development of advanced resources and their utilization in clinical settings. A plethora of new clinical use cases are emerging due to established health care initiatives and additional patient-generated sources through the extensive use of social media and other devices.

Mining and automating analysis of information from health documents holds great potential for improving health care in many aspects. Health documents include text sources such as medical records, scientific publications, and user-generated texts in e.g. social media. Research in the area of health text mining has grown and matured in recent years. Research on texts written in English is dominating, but in the last ten years research on other languages has evolved, including work on French, Spanish, Swedish, Danish, Finnish, German, Chinese and Japanese health documents. Initially, scientific biomedical articles were the main data source, after which research on clinical notes, blogs and patient forums, etc. has also increased. Moreover, the number of practical applications has increased [1], including tools for de-identification of Protected Health Information (PHI) in clinical text [2], automatic diagnosis code assignment [3], clinical decision support [4], detection of adverse drug events [5], as well as other adverse hospital events such as detection of hospital acquired infections [6], extraction of structured information from pathology reports [7], text simplification of clinical text for patient use [8], as well as pre-processing pipelines and information extraction in tools such as cTAKES [9]. Working with sensitive data such as medical records requires considerable efforts and considerations in terms of ensuring individual privacy, which makes access to such data an obstacle. Through the organization of shared tasks such as i2b2 (https://www.i2b2.org/NLP/, accessed 2 Nov. 2014), ShARe/CLEF eHealth (http://clefehealth2014.dcu.ie/, accessed 2 Nov. 2014) and SemEval (http://alt.qcri.org/semeval2014/task7/, accessed 2 Nov. 2014), English datasets have been made more easily available for research groups, which has been a crucial step towards strengthening method development and ensuring comparability as well as reproducibility. For other languages, very few such resources are currently available. Similarly, a very important resource for performing research in the health text domain is the availability of terminologies and ontologies. SNOMED CT is a global comprehensive clinical terminology available in English, French, Danish, Dutch, Spanish and Swedish. Extending this and similar terminologies to new languages is a necessity. This supplement aims at contributing to the research community in the area of information analysis in the health domain by presenting work describing approaches, solutions and progress in the challenges identified above. Specifically, it is of importance that research on languages other than English is presented, as well as novel data resources. Research in natural language processing and text mining of health documents is increasing and has extended to several new languages such as German, Finnish, Spanish and Basque, as well as to new application areas. We believe that the evolvement and progress in this area will increase the efficiency and quality of health care in the near future.

The 2015 Clinical TempEval Challenge addressed the problem of temporal reasoning in the clinical domain by providing an annotated corpus of pathology and clinical notes related to colon cancer patients. The challenge consisted of six subtasks: TIMEX3 and event span detection, TIMEX3 and event attribute classification, document relation time and narrative container relation classification. Our BluLab team participated in all six subtasks. For the TIMEX3 and event subtasks, we developed a ClearTK support vector machine pipeline using mainly simple lexical features along with information from rule-based systems. For the relation subtasks, we employed a conditional random fields classification approach, with input from a rule-based system for the narrative container relation subtask. Our team ranked first for all TIMEX3 and event subtasks, as well as for the document relation subtask.

This paper reports on the 2nd ShARe/CLEFeHealth evaluation lab which continues our evaluation resource building activities for the medical domain. In this lab we focus on patients’ information needs as opposed to the more common campaign focus of the specialised information needs of physicians and other healthcare workers. The usage scenario of the lab is to ease patients and next-of-kins’ ease in understanding eHealth information, in particular clinical reports. The 1st ShARe/CLEFeHealth evaluation lab was held in 2013. This lab consisted of three tasks. Task 1 focused on named entity recognition and normalization of disorders; Task 2 on normalization of acronyms/abbreviations; and Task 3 on information retrieval to address questions patients may have when reading clinical reports. This year’s lab introduces a new challenge in Task 1 on visual-interactive search and exploration of eHealth data. Its aim is to help patients (or their next-of-kin) in readability issues related to their hospital discharge documents and related information search on the Internet. Task 2 then continues the information extraction work of the 2013 lab, specifically focusing on disorder attribute identification and normalization from clinical text. Finally, this year’s Task 3 further extends the 2013 information retrieval task, by cleaning the 2013 document collection and introducing a new query generation method and multilingual queries. De-identified clinical reports used by the three tasks were from US intensive care and originated from the MIMIC II database. Other text documents for Tasks 1 and 3 were from the Internet and originated from the Khresmoi project. Task 2 annotations originated from the ShARe annotations. For Tasks 1 and 3, new annotations, queries, and relevance assessments were created. 50, 79, and 91 people registered their interest in Tasks 1, 2, and 3, respectively. 24 unique teams participated with 1, 10, and 14 teams in Tasks 1, 2 and 3, respectively. The teams were from Africa, Asia, Canada, Europe, and North America. The Task 1 submission, reviewed by 5 expert peers, related to the task evaluation category of Effective use of interaction and targeted the needs of both expert and novice users. The best system had an Accuracy of 0.868 in Task 2a, an F1-score of 0.576 in Task 2b, and Precision at 10 (P@10) of 0.756 in Task 3. The results demonstrate the substantial community interest and capabilities of these systems in making clinical reports easier to understand for patients. The organisers have made data and tools available for future research and development.

This paper describes part of an ongoing effort to improve the readability of Swedish electronic health records (EHRs). An EHR contains systematic documentation of a single patient’s medical history across time, entered by healthcare professionals with the purpose of enabling safe and informed care. Linguistically, medical records exemplify a highly specialised domain, which can be superficially characterised as having telegraphic sentences involving displaced or missing words, abundant abbreviations, spelling variations including misspellings, and terminology. We report results on lexical simplification of Swedish EHRs, by which we mean detecting the unknown, out-ofdictionary words and trying to resolve them either as compounded known words, abbreviations or misspellings.

One of the most important features of health care is to be able to follow a patient’s progress over time and identify events in a temporal order. We describe initial steps in creating resources for automatic temporal reasoning of Swedish medical text. As a first step, we focus on the identification of temporal expressions by exploiting existing resources and systems available for English. We adapt the HeidelTime system and manually evaluate its performance on a small subset of Swedish intensive care unit documents. On this subset, the adapted version of HeidelTime achieves a precision of 92% and a recall of 66%. We also extract the most frequent temporal expressions from a separate, larger subset, and note that most expressions concern parts of days or specific times. We intend to further develop resources for temporal reasoning of Swedish medical text by creating a gold standard corpus also annotated with events and temporal links, in addition to temporal expressions and their normalised values.

Discharge summaries and other free-text reports in healthcare transfer information between working shifts and geographic locations. Patients are likely to have difficulties in understanding their content, because of their medical jargon, non-standard abbreviations, and ward-specific idioms. This paper reports on an evaluation lab with an aim to support the continuum of care by developing methods and resources that make clinical reports in English easier to understand for patients, and which helps them in finding information related to their condition. This ShARe/CLEFeHealth2013 lab offered student mentoring and shared tasks: identification and normalisation of disorders (1a and 1b) and normalisation of abbreviations and acronyms (2) in clinical reports with respect to terminology standards in healthcare as well as information retrieval (3) to address questions patients may have when reading clinical reports. The focus on patients’ information needs as opposed to the specialised information needs of physicians and other healthcare workers was the main feature of the lab distinguishing it from previous shared tasks. De-identified clinical reports for the three tasks were from US intensive care and originated from the MIMIC II database. Other text documents for Task 3 were from the Internet and originated from the Khresmoi project. Task 1 annotations originated from the ShARe annotations. For Tasks 2 and 3, new annotations, queries, and relevance assessments were created. 64, 56, and 55 people registered their interest in Tasks 1, 2, and 3, respectively. 34 unique teams (3 members per team on average) participated with 22, 17, 5, and 9 teams in Tasks 1a, 1b, 2 and 3, respectively. The teams were from Australia, China, France, India, Ireland, Republic of Korea, Spain, UK, and USA. Some teams developed and used additional annotations, but this strategy contributed to the system performance only in Task 2. The best systems had the F1 score of 0.75 in Task 1a; Accuracies of 0.59 and 0.72 in Tasks 1b and 2; and Precision at 10 of 0.52 in Task 3. The results demonstrate the substantial community interest and capabilities of these systems in making clinical reports easier to understand for patients. The organisers have made data and tools available for future research and development.

An important task in information access methods is distinguishing factual information from speculative or negated information. Fine-grained certainty levels of diagnostic statements in Swedish clinical text are annotated in a corpus from a medical university hospital. The annotation model has two polarities (positive and negative) and three certainty levels. However, there are many e-health scenarios where such fine-grained certainty levels are not practical for information extraction. Instead, more coarse-grained groups are needed. We present three scenarios: adverse event surveillance, decision support alerts and automatic summaries and collapse the fine-grained certainty level classifications into coarser-grained groups. We build automatic classifiers for each scenario and analyze the results quantitatively. Annotation discrepancies are analyzed qualitatively through manual corpus analysis. Our main findings are that it is feasible to use a corpus of fine-grained certainty level annotations to build classifiers for coarser-grained real-world scenarios: 0.89, 0.91 and 0.8 F-score (overall average).

Background Free text is helpful for entering information into electronic health records, but reusing it is a challenge. The need for language technology for processing Finnish and Swedish healthcare text is therefore evident; however, Finnish and Swedish are linguistically very dissimilar. In this paper we present a comparison of characteristics in Finnish and Swedish free-text nursing narratives from intensive care. This creates a framework for characterising and comparing clinical text and lays the groundwork for developing clinical language technologies. Methods Our material included daily nursing narratives from one intensive care unit in Finland and one in Sweden. Inclusion criteria for patients were an inpatient period of least five days and an age of at least 16 years. We performed a comparative analysis as part of a collaborative effort between Finnish- and Swedish-speaking healthcare and language technology professionals that included both qualitative and quantitative aspects. The qualitative analysis addressed the content and structure of three average-sized health records from each country. In the quantitative analysis 514 Finnish and 379 Swedish health records were studied using various language technology tools. Results Although the two languages are not closely related, nursing narratives in Finland and Sweden had many properties in common. Both made use of specialised jargon and their content was very similar. However, many of these characteristics were challenging regarding development of language technology to support producing and using clinical documentation. Conclusions The way Finnish and Swedish intensive care nursing was documented, was not country or language dependent, but shared a common context, principles and structural features and even similar vocabulary elements. Technology solutions are therefore likely to be applicable to a wider range of natural languages, but they need linguistic tailoring. Availability The Finnish and Swedish data can be found at: http://www.dsv.su.se/hexanord/data/.

Our system for the CoNLL 2008 shared task uses a set of individual parsers, a set of stand-alone semantic role labellers, and a joint system for parsing and semantic role labelling, all blended together. The system achieved a macro averaged labelled F1- score of 79.79 (WSJ 80.92, Brown 70.49) for the overall task. The labelled attachment score for syntactic dependencies was 86.63 (WSJ 87.36, Brown 80.77) and the labelled F1-score for semantic dependencies was 72.94 (WSJ 74.47, Brown 60.18). Publisher: Coling 2008 Organizing Committee. Pages: 248 - 252.

Halla Norden is a web site with information regarding mobility between the Nordic countries in five different languages; Swedish, Danish, Norwegian, Icelandic and Finnish. We wanted to create a Nordic cross-language dictionary for the use in a cross-language search engine for Halla Norden. The entire set of texts on the web site was treated as one multilingual parallel corpus. From this we extracted parallel corpora for each language pair. The corpora were very sparse, containing on average less than 80 000 words per language pair. We have used the Uplug word alignment system (Tiedemann 2003a), for the creation of the dictionaries. The results gave on average 213 new dictionary words (frequency > 3) per language pair. The average error rate was 16 percent. Different combinations with Finnish had a higher error rate, 33 percent, whereas the error rate for the remaining language pairs only yielded on average 9 percent errors. The high error rate for Finnish is possibly due to the fact that the Finnish language belongs to a different language family. Although the corpora were very sparse the word alignment results for the combinations of Swedish, Danish, Norwegian and Icelandic were surprisingly good compared to other experiments with larger corpora.

Online social media, such as Reddit, has become an important resource to share personal experiences and communicate with others. Among other personal information, some social media users communicate about mental health problems they are experiencing, with the intention of getting advice, support or empathy from other users. Here, we investigate the language of Reddit posts specific to mental health, to define linguistic characteristics that could be helpful for further applications. The latter include attempting to identify posts that need urgent attention due to their nature, e.g. when someone announces their intentions of ending their life by suicide or harming others. Our results show that there are a variety of linguistic features that are discriminative across mental health user communities and that can be further exploited in subsequent classification tasks. Furthermore, while negative sentiment is almost uniformly expressed across the entire data set, we demonstrate that there are also condition-specific vocabularies used in social media to communicate about particular disorders. Source code and related materials are available from: https: //github.com/gkotsis/ reddit-mental-health.

Mental health problems represent a major public health challenge. Automated analysis of text related to mental health is aimed to help medical decision-making, public health policies and to improve health care. Such analysis may involve text classification. Traditionally, automated classification has been performed mainly using machine learning methods involving costly feature engineering. Recently, the performance of those methods has been dramatically improved by neural methods. However, mainly Convolutional neural networks (CNNs) have been explored. In this paper, we apply a hierarchical Recurrent neural network (RNN) architecture with an attention mechanism on social media data related to mental health. We show that this architecture improves overall classification results as compared to previously reported results on the same data. Benefitting from the attention mechanism, it can also efficiently select text elements crucial for classification decisions, which can also be used for in-depth analysis.