Investigating Cross-Domain Binary Relation Classification in Biomedical Natural Language Processing.

This paper addresses the challenge of binary relation classification in biomedical Natural Language Processing (NLP), focusing on diverse domains including gene-disease associations, compound protein interactions, and social determinants of health (SDOH). We evaluate different approaches, including fine-tuning Bidirectional Encoder Representations from Transformers (BERT) models and generative Large Language Models (LLMs), and examine their performance in zero and few-shot settings. We also introduce a novel dataset of biomedical text annotated with social and clinical entities to facilitate research into relation classification. Our results underscore the continued complexity of this task for both humans and models. BERT-based models trained on domain-specific data excelled in certain domains and achieved comparable performance and generalization power to generative LLMs in others. Despite these encouraging results, these models are still far from achieving human-level performance. We also highlight the significance of high-quality training data and domain-specific fine-tuning on the performance of all the considered models.

Pragmatic De-Identification of Cross-Domain Unstructured Documents: A Utility-Preserving Approach with Relation Extraction Filtering.

The volume of information, and in particular personal information, generated each day is increasing at a staggering rate. The ability to leverage such information depends greatly on being able to satisfy the many compliance and privacy regulations that are appearing all over the world. We present READI, a utility preserving framework for the unstructured document de-identification. READI leverages Named Entity Recognition and Relation Extraction technology to improve the quality of the entity detection, thus improving the overall quality of the data de-identification process. In this proof of concept study, we evaluate the proposed approach on two different datasets and compare with the existing state-of-the-art approaches. We show that Relation Extraction-based Approach for De-Identification (READI) notably reduces the number of false positives and improves the utility of the de-identified text.

Serum Albumin and Post-Stroke Outcomes: Analysis of UK Regional Registry Data, Systematic Review, and Meta-Analysis.

Hypoalbuminemia associates with poor acute ischemic stroke (AIS) outcomes. We hypothesised a non-linear relationship and aimed to systematically assess this association using prospective stroke data from the Norfolk and Norwich Stroke and TIA Register. Consecutive AIS patients aged ≥40 years admitted December 2003-December 2016 were included. Outcomes: In-hospital mortality, poor discharge, functional outcome (modified Rankin score 3-6), prolonged length of stay (PLoS) > 4 days, and long-term mortality. Restricted cubic spline regressions investigated the albumin-outcome relationship. We updated a systematic review (PubMed, Scopus, and Embase databases, January 2020-June 2023) and undertook a meta-analysis. A total of 9979 patients were included; mean age (standard deviation) = 78.3 (11.2) years; mean serum albumin 36.69 g/L (5.38). Compared to the cohort median, albumin < 37 g/L associated with up to two-fold higher long-term mortality (HRmax; 95% CI = 2.01; 1.61-2.49) and in-hospital mortality (RRmax; 95% CI = 1.48; 1.21-1.80). Albumin > 44 g/L associated with up to 12% higher long-term mortality (HRmax1.12; 1.06-1.19). Nine studies met our inclusion criteria totalling 23,597 patients. Low albumin associated with increased risk of long-term mortality (two studies; relative risk 1.57 (95% CI 1.11-2.22; I2 = 81.28)), as did low-normal albumin (RR 1.10 (95% CI 1.01-1.20; I2 = 0.00)). Strong evidence indicates increased long-term mortality in AIS patients with low or low-normal albumin on admission.

Automated Feature Selection from Medical Literature.

We propose an automated approach to rank the most salient variables related to a certain clinical phenomenon from scientific literature. Our solution is an automated approach to improve the efficiency of the collection of different health-related measures from a population, and to accelerate the discovery of novel associations and dependencies between health-related concepts.

Automatic Mapping of Terminology Items with Transformers.

Biomedical ontologies are a key component in many systems for the analysis of textual clinical data. They are employed to organize information about a certain domain relying on a hierarchy of different classes. Each class maps a concept to items in a terminology developed by domain experts. These mappings are then leveraged to organize the information extracted by Natural Language Processing (NLP) models to build knowledge graphs for inferences. The creation of these associations, however, requires extensive manual review. In this paper, we present an automated approach and repeatable framework to learn a mapping between ontology classes and terminology terms derived from vocabularies in the Unified Medical Language System (UMLS) metathesaurus. According to our evaluation, the proposed system achieves a performance close to humans and provides a substantial improvement over existing systems developed by the National Library of Medicine to assist researchers through this process.

Subtyping Gastrointestinal Surgical Outcomes from Real World Data: A Comprehensive Analysis of UK Biobank.

Chronic gastrointestinal (GI) conditions, such as inflammatory bowel diseases (IBD), offer a promising opportunity to create classification systems that can enhance the accuracy of predicting the most effective therapies and prognosis for each patient. Here, we present a novel methodology to explore disease subtypes using our open-sourced BiomedSciAI toolkit. Applying methods available in this toolkit on the UK Biobank, including subpopulation-based feature selection and multi-dimensional subset scanning, we aimed to discover unique subgroups from GI surgery cohorts. Of a 12,073-patient cohort, a subgroup of 440 IBD patients was discovered with an increased risk of a subsequent GI surgery (OR: 2.21, 95% CI [1.81-2.69]). We iteratively demonstrate the discovery process using an additional cohort (with a narrower definition of GI surgery). Our results show that the iterative process can refine the subgroup discovery process and generate novel hypotheses to investigate determinants of treatment response.

Predicting 10-year stroke mortality: development and validation of a nomogram.

Predicting long-term stroke mortality is a clinically important and unmet need. We aimed to develop and internally validate a 10-year ischaemic stroke mortality prediction score. In this UK cohort study, 10,366 patients with first-ever ischaemic stroke between January 2003 and December 2016 were followed up for a median (interquartile range) of 5.47 (2.96-9.15) years. A Cox proportional-hazards model was used to predict 10-year post-admission mortality. The predictors associated with 10-year mortality included age, sex, Oxfordshire Community Stroke Project classification, estimated glomerular filtration rate (eGFR), pre-stroke modified Rankin Score, admission haemoglobin, sodium, white blood cell count and comorbidities (atrial fibrillation, coronary heart disease, heart failure, cancer, hypertension, chronic obstructive pulmonary disease, liver disease and peripheral vascular disease). The model was internally validated using bootstrap resampling to assess optimism in discrimination and calibration. A nomogram was created to facilitate application of the score at the point of care. Mean age (SD) was 78.5 ± 10.9 years, 52% female. Most strokes were partial anterior circulation syndromes (38%). 10-year mortality predictors were: total anterior circulation stroke (hazard ratio, 95% confidence intervals) (2.87, 2.62-3.14), eGFR < 15 (1.97, 1.55-2.52), 1-year increment in age (1.04, 1.04-1.05), liver disease (1.50, 1.20-1.87), peripheral vascular disease (1.39, 1.23-1.57), cancers (1.37, 1.27-1.47), heart failure (1.24, 1.15-1.34), 1-point increment in pre-stroke mRS (1.20, 1.17-1.22), atrial fibrillation (1.17, 1.10-1.24), coronary heart disease (1.09, 1.02-1.16), chronic obstructive pulmonary disease (1.13, 1.03-1.25) and hypertension (0.77, 0.72-0.82). Upon internal validation, the optimism-adjusted c-statistic was 0.76 and calibration slope was 0.98. Our 10-year mortality model uses routinely collected point-of-care information. It is the first 10-year mortality score in stroke. While the model was internally validated, further external validation is also warranted.

Encoding Health Records into Pathway Representations for Deep Learning.

There is a growing trend in building deep learning patient representations from health records to obtain a comprehensive view of a patient's data for machine learning tasks. This paper proposes a reproducible approach to generate patient pathways from health records and to transform them into a machine-processable image-like structure useful for deep learning tasks. Based on this approach, we generated over a million pathways from FAIR synthetic health records and used them to train a convolutional neural network. Our initial experiments show the accuracy of the CNN on a prediction task is comparable or better than other autoencoders trained on the same data, while requiring significantly less computational resources for training. We also assess the impact of the size of the training dataset on autoencoders performances. The source code for generating pathways from health records is provided as open source.

Deep EHR Spotlight: a Framework and Mechanism to Highlight Events in Electronic Health Records for Explainable Predictions.

The wide adoption of Electronic Health Records (EHR) has resulted in large amounts of clinical data becoming available, which promises to support service delivery and advance clinical and informatics research. Deep learning techniques have demonstrated performance in predictive analytic tasks using EHRs yet they typically lack model result transparency or explainability functionalities and require cumbersome pre-processing tasks. Moreover, EHRs contain heterogeneous and multi-modal data points such as text, numbers and time series which further hinder visualisation and interpretability. This paper proposes a deep learning framework to: 1) encode patient pathways from EHRs into images, 2) highlight important events within pathway images, and 3) enable more complex predictions with additional intelligibility. The proposed method relies on a deep attention mechanism for visualisation of the predictions and allows predicting multiple sequential outcomes.

Social Determinant Trends of COVID-19: An Analysis Using Knowledge Graphs from Published Evidence and Online Trends.

This paper presents the results of a new approach to discover related health and social factors during the COVID-19 pandemic. The approach leverages a knowledge graph of related concepts mined from a corpus of published evidence (PubMed) prior to the pandemic. Population trends from online searches were used to identify social determinants of health (SDoH) concepts that trended high at the outset of the pandemic from a list of SDoH topics from the World Health Organization (WHO). The trending concepts were then mapped to the knowledge graph and a subsequent analysis of the derived insights, spanning two years, was conducted. This paper suggests an approach to derive new related health and social factors that may have either played a role in, or been affected by, the onset of the global COVID-19 pandemic. In particular, our results show how, from a list of SDoH topics, Food Security, Unemployment trended the highest at the start of the pandemic. Further work is needed to continue to ascertain the validity of the derived relations in a population health context and to improve mining insights from published evidence.

Predictors of mortality and disability in stroke-associated pneumonia.

Whilst stroke-associated pneumonia (SAP) is common and associated with poor outcomes, less is known about the determinants of these adverse clinical outcomes in SAP. To identify the factors that influence mortality and morbidity in SAP. Data for patients with SAP (n = 854) were extracted from a regional Hospital Stroke Register in Norfolk, UK (2003-2015). SAP was defined as pneumonia occurring within 7 days of admission by the treating clinicians. Mutlivariable regression models were constructed to assess factors influencing survival and the level of disability at discharge using modified Rankin Scale [mRS]. Mean (SD) age was 83.0 (8.7) years and ischaemic stroke occurred in 727 (85.0%). Mortality was 19.0% at 30 days and 44.0% at 6 months. Stroke severity assessment using National Institutes of Health Stroke Scale was not recorded in the data set although Oxfordshire Community Stroke Project was Classification. In the multivariable analyses, 30-day mortality was independently associated with age (OR 1.04, 95% CI 1.01-1.07, p = 0.01), haemorrhagic stroke (2.27, 1.07-4.78, p = 0.03) and pre-stroke disability (mRS 4-5 v 0-1: 6.45, 3.12-13.35, p < 0.001). 6-month mortality was independently associated with age (< 0.001), pre-stroke disability (p < 0.001) and certain comorbidities, including the following: dementia (6.53, 4.73-9.03, p < 0.001), lung cancer (2.07, 1.14-3.77, p = 0.017) and previous transient ischemic attack (1.94, 1.12-3.36, p = 0.019). Disability defined by mRS at discharge was independently associated with age (1.10, 1.05-1.16, p < 0.001) and plasma C-reactive protein (1.02, 1.01-1.03, p = 0.012). We have identified non-modifiable determinants of poor prognosis in patients with SAP. Further studies are required to identify modifiable factors which may guide areas for intervention to improve the prognosis in SAP in these patients.

Discovering Associations between Social Determinants and Health Outcomes: Merging Knowledge Graphs from Literature and Electronic Health Data.

Social Determinants of Health (SDoH) are an increasingly important part of the broader research and public health efforts in understanding individuals' physical and mental well-being. Despite this, non-clinical factors affecting health are poorly recorded in electronic health databases and techniques to study how SDoH might relate to population outcomes are lacking. This paper proposes an approach to systematically identify and quantify associations between SDoH and health-related outcomes in a specific cohort of people by (1) leveraging published evidence from literature to build a knowledge graph of health and social factor associations and (2) analysing a large dataset of claims and medical records where those associations may be found. This work demonstrates how the proposed approach could be used to generate hypotheses and inform further research on SDoH in a data-driven manner.

Association of chronic kidney disease with outcomes in acute stroke.

Previous studies have found an association between chronic kidney disease and poor outcomes in stroke patients. However, there is a paucity of literature evaluating this association by stroke type. We therefore aimed to explore the association between CKD and stroke outcomes according to type of stroke. The data consisting of 594,681 stroke patients were acquired from Universal Coverage Health Security Insurance Scheme Database in Thailand. Binary logistic regression was used to assess the relationship of CKD and outcomes, which were as follows; in-hospital mortality, long length of stay (>3 days), pneumonia, sepsis, respiratory failure and myocardial infarction. Results: after fully adjusting for covariates, CKD was associated with increased odds of in-hospital mortality in patients with ischemic (OR 1.32; 95% CI = 1.27-1.38), haemorrhagic (OR 1.31; 95% CI = 1.24-1.39), and other undetermined stroke type (OR 1.44; 95% CI = 1.21-1.73). CKD was found to be associated with increased odds of pneumonia, sepsis, respiratory failure and myocardial infarction in ischaemic stroke. While CKD was found to be associated with increase odds of sepsis, respiratory failure, and myocardial infarction, decrease odds of pneumonia was observed in patients with haemorrhagic stroke. In other undetermined stroke type, CKD was found to only be associated with increase odds of sepsis and respiratory failure, while there is no significant association of CKD and increase or decrease odds with pneumonia and myocardial infarction. CKD was associated with poor outcomes in all stroke types. CKD should be considered as part of stroke prognosis as well as identifying at risk patient population for in-hospital complications.

Exploring the Social Drivers of Health During a Pandemic: Leveraging Knowledge Graphs and Population Trends in COVID-19.

Social determinants of health (SDoH) are the factors which lie outside of the traditional health system, such as employment or access to nutritious foods, that influence health outcomes. Some efforts have focused on identifying vulnerable populations during the COVID-19 pandemic, however, both the short- and long-term social impacts of the pandemic on individuals and populations are not well understood. This paper presents a pipeline to discover health outcomes and related social factors based on trending SDoH at population-level using Google Trends. A knowledge graph was built from a corpus of research literature (PubMed) and the social determinants that trended high at the start of the pandemic were examined. This paper reports on related social and health concepts which may be impacted by the COVID-19 outbreak and may be important to monitor as the pandemic evolves. The proposed pipeline should have wider applicability in surfacing related social or clinical characteristics of interest, outbreak surveillance, or to mine relations between social and health concepts that can, in turn, help inform and support citizen-centred services.

Seasonality of stroke: Winter admissions and mortality excess: A Thailand National Stroke population database study.

OBJECTIVES: We examined the existence and potential burden of seasonality of stroke admissions and mortality within a tropical climate using cohort data collected between 1 st November 2003 and 31 st October 2012. PATIENTS AND METHODS: In a prospective cohort of hospitalised stroke patients from the catchment of ∼75 % of the Thai population (n = 569,307; mean SD age = 64(14.5)), incident stroke admissions, in-hospital mortality, prolonged hospitalisations, and stroke related complications by season were determined. Rates of incident stroke admissions by month and season were plotted. Winter excess indexes for study outcomes expressed as a percentage were calculated. Using logistic regression we examined the association between winter admission and in-hospital mortality (non-winter admission as reference) adjusting for age, sex, stroke type, year of admission, and presence of pre-existing comorbidities. RESULTS: We observed a winter excess in mortality during hospitalisation (+10.3 %) and prolonged length of stay (+7.3 %). Respective winter excess indexes for dyslipidaemias, arrhythmias, anaemia, and alcohol related disorders in patients that died during hospitalisation were +1.4 %, +6.2 %, +0.2 %, +1.5 %. In these patients, respective winter excess indexes for post-stroke complications of pneumonia and sepsis were +6.7 % and +3.2 %. In fully adjusted analyses, winter admission (compared to non-winter admission) was associated with increased odds of in-hospital mortality (OR (95 % CI) = 1.023 (1.006-1.040)). CONCLUSIONS: We provide robust evidence for the existence of an excess in winter stroke admissions and subsequent in-hospital deaths within a tropical region.

Infective endocarditis is associated with worse outcomes in stroke: A Thailand National Database Study.

BACKGROUND: There is lack of data on the association between infective endocarditis (IE) and outcomes of mortality and complications in stroke. We aimed to compare characteristics and outcomes of stroke patients with and without IE. METHODS: We retrospectively examined the above association using data obtained from an insurance database which covers ~75% of the Thai population. All hospitalised strokes between 8 January 2003 and 31 December 2013 were included in the current study. Characteristics and outcomes were compared between stroke patients with or without IE, and then between two main stroke types. Multiple logistic regression models including propensity score-matched analyses were constructed to assess study outcomes controlling for age, sex, stroke type and comorbidities. RESULTS: A total of 590 115 stroke patients (mean (SD) age = 64.2 ± 13.7 years; ischaemic = 51.7%; haemorrhagic = 32.6%; undetermined = 15.7%) were included, of whom 2129 (0.36%) had stroke associated with IE. After adjustment, we found that IE was significantly associated with the following complications: arrhythmias (adjusted odds ratio (95% CI) 6.94 (6.29-7.66)), sepsis (1.24 (1.01-1.52)), pneumonia (1.34 (1.17-1.53)), respiratory failure (1.43 (1.24-1.66)) and in-hospital mortality (1.29 (1.13-1.47)) (P for all <.001). Patients with haemorrhagic stroke with IE had poorer outcomes for in-hospital mortality and respiratory failure compared with their counterparts with ischaemic stroke. Propensity score-matched analysis showed similar results. CONCLUSIONS: Our results suggest that stroke patients with IE differ from that of the general stroke population and these patients have worse outcomes. Future studies are needed to determine the best treatment strategies for stroke patients with IE.

Discovering New Social Determinants of Health Concepts from Unstructured Data: Framework and Evaluation.

Social determinants of health (SDoH) are the complex set of circumstances in which individuals are born, or with which they live, that impact their health. Integrating SDoH into practice requires that information systems are able to identify SDoH-related concepts from charts and case notes through vocabularies or terminologies. Despite significant standardisation efforts across healthcare domains, SDoH coverage remains sparse in existing terminologies due to the broad spectrum of this domain, ranging from family relations, risk factors, to social programs and benefits, which are not consistently captured across administrative and clinical settings. This paper presents a framework to mine, evaluate and recommend new multidisciplinary concepts that relate to or impact the health and well-being of individuals using a word embedding model trained from a large dynamic corpus of unstructured data. Five key SDoH domains were selected and evaluated by domain experts. The concepts resulting from the trained model were matched against well-established meta-thesaurus UMLS and terminology SNOMED-CT and, overall, a significant proportion of concepts from a set of 10,000 candidates were not found (31% and 28% respectively). The results confirm both the gaps in current terminologies and the feasibility and impact of the methods presented in this paper for the incremental discovery and validation of new SDoH concepts together with domain experts. This sustainable approach facilitates the development and refinement of new and existing terminologies and, in turn, it allows systems such as Natural Language Processing (NLP) annotators to leverage SDoH concepts across integrated care settings.

Impact of Diabetes on Complications, Long Term Mortality and Recurrence in 608,890 Hospitalised Patients with Stroke.

Background: Patients with diabetes mellitus (DM) have been found to be at an increased risk of suffering a stroke. However, research on the impact of DM on stroke outcomes is limited. Objectives: We aimed to examine the influence of DM on outcomes in ischaemic (IS) and haemorrhagic stroke (HS) patients. Methods: We included 608,890 consecutive stroke patients from the Thailand national insurance registry. In-hospital mortality, sepsis, pneumonia, acute kidney injury (AKI), urinary tract infection (UTI) and cardiovascular events were evaluated using logistic regressions. Long-term analysis was performed on first-stroke patients with a determined pathology (n = 398,663) using Royston-Parmar models. Median follow-ups were 4.21 and 4.78 years for IS and HS, respectively. All analyses were stratified by stroke sub-type. Results: Mean age (SD) was 64.3 (13.7) years, 44.9% were female with 61% IS, 28% HS and 11% undetermined strokes. DM was associated with in-hospital death, pneumonia, sepsis, AKI and cardiovascular events (odds ratios ranging from 1.13-1.78, p < 0.01) in both stroke types. In IS, DM was associated with long-term mortality and recurrence throughout the follow-up: HRmax (99% CI) at t = 4108 days: 1.54 (1.27, 1.86) and HR (99% CI) = 1.27(1.23,1.32), respectively. In HS, HRmax (t = 4108 days) for long-term mortality was 2.10 (1.87, 2.37), significant after day 14 post-discharge. HRmax (t = 455) for long-term recurrence of HS was 1.29 (1.09, 1.53), significant after day 116 post-discharge. Conclusions: Regardless of stroke type, DM was associated with in-hospital death and complications, long-term mortality and stroke recurrence.

Anaemia and incidence of post stroke dementia.

OBJECTIVES: To assess the impact of anaemia on incidence of post-stroke dementia. PATIENTS AND METHODS: We used data from a UK regional stroke register. To be eligible, patient must have survived to discharge and had anaemia by WHO criteria. Dementia status and other prevalent co-morbidities were assessed using ICD-10 codes. Patients were followed till May 2015 (mean follow-up 3.7 years, total person years = 27,769). Hazard Ratio for incident dementia was calculated using Cox-proportional hazards model controlling for potential confounders. Fine and Gray model was additionally constructed using mortality as the competing risk. RESULTS: A total of 7454 stroke patients were included with mean age SD of 75.912.3 years 50.2 % men). Those with anaemia were older, has higher disability and co-morbidity burden prior to stroke. We observed a large amount of variation in the dementia incidence rates over time and that the hazard ratio increased every year. The significant association between anaemia and dementia incidence was lost after controlling for pre-stroke Modified Rankin score (HR1.17(0.97,1.40)). With every 20 g/dL increase in Hb was associated with a significant reduction in the risk of dementia after adjustment for age, sex, stroke factors and disability but lost significance after adjustment for vascular risk factors. Competing risk analyses showed similar results. CONCLUSION: Whilst we found no evidence of anaemia as a risk factor for post-stroke dementia, the findings may be limited by potential under recognition of post stroke dementia.

Does prior antithrombotic therapy influence recurrence and bleeding risk in stroke patients with atrial fibrillation or atrial flutter?

BACKGROUND: Whilst antithrombotic therapy is recommended in people with atrial fibrillation, little is known about the survival benefits of antithrombotic treatment in those with both high ischaemic and bleeding risk scores. We aim to describe the distribution of these risk scores in those with a prior diagnosis of atrial fibrillation who have suffered stroke and to determine the net clinical benefit of antithrombotic treatment. METHODS: We used regional stroke register data in the UK. Patients with a prior diagnosis of atrial fibrillation and ischaemic or haemorrhagic stroke patients were selected and their ischaemic stroke risk score (CHA2DS2-VASc) and bleeding risk score (HEMORR2HAGES) scores retrospectively calculated. Logistic regression and Cox proportional hazards models were constructed to determine the association between antithrombotic therapy prior to stroke and in-hospital and long-term mortality. RESULTS: A total of 1928 stroke patients (mean age 81.3 years (standard deviation 8.5), 56.8% women) with prior atrial fibrillation were included. Of these, 1761 (91.3%) suffered ischaemic stroke. The most common phenotype (64%) was of those with both high CHA2DS2-VASc (≥2) and high HEMORR2HAGES score (≥4). In our fully adjusted model, patients on antithrombotic treatment with both high ischaemic and bleeding risk had a significant reduction in odds of 31% for in-hospital mortality (odds ratio 0.69 (95% confidence interval 0.48-1.00: p = 0.049)) and 17% relative risk reduction for long-term mortality (hazard ratio 0.83 (95% confidence interval 0.71-0.97: p = 0.02)). CONCLUSIONS: Our study suggests that antithrombotic treatment has a prognostic benefit following incident stroke in those with both high ischaemic risk and high bleeding risk. This should be considered when choosing treatment options in this group of patients.