Richard Dobson

Yuezhou Zhang, Amos A Folarin, Judith Dineley et al.

Language use has been shown to correlate with depression, but large-scale validation is needed. Traditional methods like clinic studies are expensive. So, natural language processing has been employed on social media to predict depression, but limitations remain-lack of validated labels, biased user samples, and no context. Our study identified 29 topics in 3919 smartphone-collected speech recordings from 265 participants using the Whisper tool and BERTopic model. Six topics with a median PHQ-8 greater than or equal to 10 were regarded as risk topics for depression: No Expectations, Sleep, Mental Therapy, Haircut, Studying, and Coursework. To elucidate the topic emergence and associations with depression, we compared behavioral (from wearables) and linguistic characteristics across identified topics. The correlation between topic shifts and changes in depression severity over time was also investigated, indicating the importance of longitudinally monitoring language use. We also tested the BERTopic model on a similar smaller dataset (356 speech recordings from 57 participants), obtaining some consistent results. In summary, our findings demonstrate specific speech topics may indicate depression severity. The presented data-driven workflow provides a practical approach to collecting and analyzing large-scale speech data from real-world settings for digital health research.

Zulqarnain Rashid, Amos A Folarin, Yatharth Ranjan et al.

Digital Biomarkers and remote patient monitoring can provide valuable and timely insights into how a patient is coping with their condition (disease progression, treatment response, etc.), complementing treatment in traditional healthcare settings.Smartphones with embedded and connected sensors have immense potential for improving healthcare through various apps and mHealth (mobile health) platforms. This capability could enable the development of reliable digital biomarkers from long-term longitudinal data collected remotely from patients. We built an open-source platform, RADAR-base, to support large-scale data collection in remote monitoring studies. RADAR-base is a modern remote data collection platform built around Confluent's Apache Kafka, to support scalability, extensibility, security, privacy and quality of data. It provides support for study design and set-up, active (eg PROMs) and passive (eg. phone sensors, wearable devices and IoT) remote data collection capabilities with feature generation (eg. behavioural, environmental and physiological markers). The backend enables secure data transmission, and scalable solutions for data storage, management and data access. The platform has successfully collected longitudinal data for various cohorts in a number of disease areas including Multiple Sclerosis, Depression, Epilepsy, ADHD, Alzheimer, Autism and Lung diseases. Digital biomarkers developed through collected data are providing useful insights into different diseases. RADAR-base provides a modern open-source, community-driven solution for remote monitoring, data collection, and digital phenotyping of physical and mental health diseases. Clinicians can use digital biomarkers to augment their decision making for the prevention, personalisation and early intervention of disease.

Zeljko Kraljevic, Anthony Shek, Joshua Au Yeung et al.

Protecting patient privacy in healthcare records is a top priority, and redaction is a commonly used method for obscuring directly identifiable information in text. Rule-based methods have been widely used, but their precision is often low causing over-redaction of text and frequently not being adaptable enough for non-standardised or unconventional structures of personal health information. Deep learning techniques have emerged as a promising solution, but implementing them in real-world environments poses challenges due to the differences in patient record structure and language across different departments, hospitals, and countries. In this study, we present AnonCAT, a transformer-based model and a blueprint on how deidentification models can be deployed in real-world healthcare. AnonCAT was trained through a process involving manually annotated redactions of real-world documents from three UK hospitals with different electronic health record systems and 3116 documents. The model achieved high performance in all three hospitals with a Recall of 0.99, 0.99 and 0.96. Our findings demonstrate the potential of deep learning techniques for improving the efficiency and accuracy of redaction in global healthcare data and highlight the importance of building workflows which not just use these models but are also able to continually fine-tune and audit the performance of these algorithms to ensure continuing effectiveness in real-world settings. This approach provides a blueprint for the real-world use of de-identifying algorithms through fine-tuning and localisation, the code together with tutorials is available on GitHub (https://github.com/CogStack/MedCAT).

Zeljko Kraljevic, Dan Bean, Anthony Shek et al.

Background: Electronic Health Records hold detailed longitudinal information about each patient's health status and general clinical history, a large portion of which is stored within the unstructured text. Existing approaches focus mostly on structured data and a subset of single-domain outcomes. We explore how temporal modelling of patients from free text and structured data, using deep generative transformers can be used to forecast a wide range of future disorders, substances, procedures or findings. Methods: We present Foresight, a novel transformer-based pipeline that uses named entity recognition and linking tools to convert document text into structured, coded concepts, followed by providing probabilistic forecasts for future medical events such as disorders, substances, procedures and findings. We processed the entire free-text portion from three different hospital datasets totalling 811336 patients covering both physical and mental health. Findings: On tests in two UK hospitals (King's College Hospital, South London and Maudsley) and the US MIMIC-III dataset precision@10 0.68, 0.76 and 0.88 was achieved for forecasting the next disorder in a patient timeline, while precision@10 of 0.80, 0.81 and 0.91 was achieved for forecasting the next biomedical concept. Foresight was also validated on 34 synthetic patient timelines by five clinicians and achieved relevancy of 97% for the top forecasted candidate disorder. As a generative model, it can forecast follow-on biomedical concepts for as many steps as required. Interpretation: Foresight is a general-purpose model for biomedical concept modelling that can be used for real-world risk forecasting, virtual trials and clinical research to study the progression of disorders, simulate interventions and counterfactuals, and educational purposes.

Thomas Searle, Zina Ibrahim, James Teo et al.

Brief Hospital Course (BHC) summaries are succinct summaries of an entire hospital encounter, embedded within discharge summaries, written by senior clinicians responsible for the overall care of a patient. Methods to automatically produce summaries from inpatient documentation would be invaluable in reducing clinician manual burden of summarising documents under high time-pressure to admit and discharge patients. Automatically producing these summaries from the inpatient course, is a complex, multi-document summarisation task, as source notes are written from various perspectives (e.g. nursing, doctor, radiology), during the course of the hospitalisation. We demonstrate a range of methods for BHC summarisation demonstrating the performance of deep learning summarisation models across extractive and abstractive summarisation scenarios. We also test a novel ensemble extractive and abstractive summarisation model that incorporates a medical concept ontology (SNOMED) as a clinical guidance signal and shows superior performance in 2 real-world clinical data sets.

Yuezhou Zhang, Callum Stewart, Yatharth Ranjan et al.

Digital phenotyping offers a novel and cost-efficient approach for managing depression and anxiety. Previous studies, often limited to small-to-medium or specific populations, may lack generalizability. We conducted a cross-sectional analysis of data from 10,129 participants recruited from a UK-based general population between June 2020 and August 2022. Participants shared wearable (Fitbit) data and self-reported questionnaires on depression (PHQ-8), anxiety (GAD-7), and mood via a study app. We first examined the correlations between PHQ-8/GAD-7 scores and wearable-derived features, demographics, health data, and mood assessments. Subsequently, unsupervised clustering was used to identify behavioural patterns associated with depression or anxiety. Finally, we employed separate XGBoost models to predict depression and anxiety and compared the results using different subsets of features. We observed significant associations between the severity of depression and anxiety with several factors, including mood, age, gender, BMI, sleep patterns, physical activity, and heart rate. Clustering analysis revealed that participants simultaneously exhibiting lower physical activity levels and higher heart rates reported more severe symptoms. Prediction models incorporating all types of variables achieved the best performance ($R^2$=0.41, MAE=3.42 for depression; $R^2$=0.31, MAE=3.50 for anxiety) compared to those using subsets of variables. This study identified potential indicators for depression and anxiety, highlighting the utility of digital phenotyping and machine learning technologies for rapid screening of mental disorders in general populations. These findings provide robust real-world insights for future healthcare applications.

Anastasiia Tokareva, Judith Dineley, Zoe Firth et al.

Background: Captured between clinical appointments using mobile devices, spoken language has potential for objective, more regular assessment of symptom severity and earlier detection of relapse in major depressive disorder. However, research to date has largely been in non-clinical cross-sectional samples of written language using complex machine learning (ML) approaches with limited interpretability. Methods: We describe an initial exploratory analysis of longitudinal speech data and PHQ-8 assessments from 5,836 recordings of 586 participants in the UK, Netherlands, and Spain, collected in the RADAR-MDD study. We sought to identify interpretable lexical features associated with MDD symptom severity with linear mixed-effects modelling. Interpretable features and high-dimensional vector embeddings were also used to test the prediction performance of four regressor ML models. Results: In English data, MDD symptom severity was associated with 7 features including lexical diversity measures and absolutist language. In Dutch, associations were observed with words per sentence and positive word frequency; no associations were observed in recordings collected in Spain. The predictive power of lexical features and vector embeddings was near chance level across all languages. Limitations: Smaller samples in non-English speech and methodological choices, such as the elicitation prompt, may have also limited the effect sizes observable. A lack of NLP tools in languages other than English restricted our feature choice. Conclusion: To understand the value of lexical markers in clinical research and practice, further research is needed in larger samples across several languages using improved protocols, and ML models that account for within- and between-individual variations in language.

Shubham Agarwal, Thomas Searle, Mart Ratas et al.

Electronic Health Records are large repositories of valuable clinical data, with a significant portion stored in unstructured text format. This textual data includes clinical events (e.g., disorders, symptoms, findings, medications and procedures) in context that if extracted accurately at scale can unlock valuable downstream applications such as disease prediction. Using an existing Named Entity Recognition and Linking methodology, MedCAT, these identified concepts need to be further classified (contextualised) for their relevance to the patient, and their temporal and negated status for example, to be useful downstream. This study performs a comparative analysis of various natural language models for medical text classification. Extensive experimentation reveals the effectiveness of transformer-based language models, particularly BERT. When combined with class imbalance mitigation techniques, BERT outperforms Bi-LSTM models by up to 28% and the baseline BERT model by up to 16% for recall of the minority classes. The method has been implemented as part of CogStack/MedCAT framework and made available to the community for further research.

Kawsar Noor, Katherine Smith, Julia Bennett et al.

After a patient consultation, a clinician determines the steps in the management of the patient. A clinician may for example request to see the patient again or refer them to a specialist. Whilst most clinicians will record their intent as "next steps" in the patient's clinical notes, in some cases the clinician may forget to indicate their intent as an order or request, e.g. failure to place the follow-up order. This consequently results in patients becoming lost-to-follow up and may in some cases lead to adverse consequences. In this paper we train a machine learning model to detect a clinician's intent to follow up with a patient from the patient's clinical notes. Annotators systematically identified 22 possible types of clinical intent and annotated 3000 Bariatric clinical notes. The annotation process revealed a class imbalance in the labeled data and we found that there was only sufficient labeled data to train 11 out of the 22 intents. We used the data to train a BERT based multilabel classification model and reported the following average accuracy metrics for all intents: macro-precision: 0.91, macro-recall: 0.90, macro-f1: 0.90.

Linglong Qian, Tao Wang, Jun Wang et al.

We present a comprehensive analysis of deep learning approaches for Electronic Health Record (EHR) time-series imputation, examining how architectural and framework biases combine to influence model performance. Our investigation reveals varying capabilities of deep imputers in capturing complex spatiotemporal dependencies within EHRs, and that model effectiveness depends on how its combined biases align with medical time-series characteristics. Our experimental evaluation challenges common assumptions about model complexity, demonstrating that larger models do not necessarily improve performance. Rather, carefully designed architectures can better capture the complex patterns inherent in clinical data. The study highlights the need for imputation approaches that prioritise clinically meaningful data reconstruction over statistical accuracy. Our experiments show imputation performance variations of up to 20\% based on preprocessing and implementation choices, emphasising the need for standardised benchmarking methodologies. Finally, we identify critical gaps between current deep imputation methods and medical requirements, highlighting the importance of integrating clinical insights to achieve more reliable imputation approaches for healthcare applications.

Zeljko Kraljevic, Joshua Au Yeung, Daniel Bean et al.

Foresight 2 (FS2) is a large language model fine-tuned on hospital data for modelling patient timelines (GitHub 'removed for anon'). It can understand patients' clinical notes and predict SNOMED codes for a wide range of biomedical use cases, including diagnosis suggestions, risk forecasting, and procedure and medication recommendations. FS2 is trained on the free text portion of the MIMIC-III dataset, firstly through extracting biomedical concepts and then creating contextualised patient timelines, upon which the model is then fine-tuned. The results show significant improvement over the previous state-of-the-art for the next new biomedical concept prediction (P/R - 0.73/0.66 vs 0.52/0.32) and a similar improvement specifically for the next new disorder prediction (P/R - 0.69/0.62 vs 0.46/0.25). Finally, on the task of risk forecast, we compare our model to GPT-4-turbo (and a range of open-source biomedical LLMs) and show that FS2 performs significantly better on such tasks (P@5 - 0.90 vs 0.65). This highlights the need to incorporate hospital data into LLMs and shows that small models outperform much larger ones when fine-tuned on high-quality, specialised data.

Linglong Qian, Joseph Arul Raj, Hugh Logan Ellis et al.

We introduce the Conditional Self-Attention Imputation (CSAI) model, a novel recurrent neural network architecture designed to address the challenges of complex missing data patterns in multivariate time series derived from hospital electronic health records (EHRs). CSAI extends state-of-the-art neural network-based imputation by introducing key modifications specific to EHR data: a) attention-based hidden state initialisation to capture both long- and short-range temporal dependencies prevalent in EHRs, b) domain-informed temporal decay to mimic clinical data recording patterns, and c) a non-uniform masking strategy that models non-random missingness by calibrating weights according to both temporal and cross-sectional data characteristics. Comprehensive evaluation across four EHR benchmark datasets demonstrates CSAI's effectiveness compared to state-of-the-art architectures in data restoration and downstream tasks. CSAI is integrated into PyPOTS, an open-source Python toolbox designed for machine learning tasks on partially observed time series. This work significantly advances the state of neural network imputation applied to EHRs by more closely aligning algorithmic imputation with clinical realities.

Zeljko Kraljevic, Thomas Searle, Anthony Shek et al.

Electronic health records (EHR) contain large volumes of unstructured text, requiring the application of Information Extraction (IE) technologies to enable clinical analysis. We present the open-source Medical Concept Annotation Toolkit (MedCAT) that provides: a) a novel self-supervised machine learning algorithm for extracting concepts using any concept vocabulary including UMLS/SNOMED-CT; b) a feature-rich annotation interface for customising and training IE models; and c) integrations to the broader CogStack ecosystem for vendor-agnostic health system deployment. We show improved performance in extracting UMLS concepts from open datasets (F1:0.448-0.738 vs 0.429-0.650). Further real-world validation demonstrates SNOMED-CT extraction at 3 large London hospitals with self-supervised training over ~8.8B words from ~17M clinical records and further fine-tuning with ~6K clinician annotated examples. We show strong transferability (F1 > 0.94) between hospitals, datasets, and concept types indicating cross-domain EHR-agnostic utility for accelerated clinical and research use cases.

Thomas Searle, Zina Ibrahim, Richard JB Dobson

Clinical coding is currently a labour-intensive, error-prone, but critical administrative process whereby hospital patient episodes are manually assigned codes by qualified staff from large, standardised taxonomic hierarchies of codes. Automating clinical coding has a long history in NLP research and has recently seen novel developments setting new state of the art results. A popular dataset used in this task is MIMIC-III, a large intensive care database that includes clinical free text notes and associated codes. We argue for the reconsideration of the validity MIMIC-III's assigned codes that are often treated as gold-standard, especially when MIMIC-III has not undergone secondary validation. This work presents an open-source, reproducible experimental methodology for assessing the validity of codes derived from EHR discharge summaries. We exemplify the methodology with MIMIC-III discharge summaries and show the most frequently assigned codes in MIMIC-III are under-coded up to 35%.

Zeljko Kraljevic, Daniel Bean, Aurelie Mascio et al.

Biomedical documents such as Electronic Health Records (EHRs) contain a large amount of information in an unstructured format. The data in EHRs is a hugely valuable resource documenting clinical narratives and decisions, but whilst the text can be easily understood by human doctors it is challenging to use in research and clinical applications. To uncover the potential of biomedical documents we need to extract and structure the information they contain. The task at hand is Named Entity Recognition and Linking (NER+L). The number of entities, ambiguity of words, overlapping and nesting make the biomedical area significantly more difficult than many others. To overcome these difficulties, we have developed the Medical Concept Annotation Tool (MedCAT), an open-source unsupervised approach to NER+L. MedCAT uses unsupervised machine learning to disambiguate entities. It was validated on MIMIC-III (a freely accessible critical care database) and MedMentions (Biomedical papers annotated with mentions from the Unified Medical Language System). In case of NER+L, the comparison with existing tools shows that MedCAT improves the previous best with only unsupervised learning (F1=0.848 vs 0.691 for disease detection; F1=0.710 vs. 0.222 for general concept detection). A qualitative analysis of the vector embeddings learnt by MedCAT shows that it captures latent medical knowledge available in EHRs (MIMIC-III). Unsupervised learning can improve the performance of large scale entity extraction, but it has some limitations when working with only a couple of entities and a small dataset. In that case options are supervised learning or active learning, both of which are supported in MedCAT via the MedCATtrainer extension. Our approach can detect and link millions of different biomedical concepts with state-of-the-art performance, whilst being lightweight, fast and easy to use.

Linglong Qian, Zina Ibrahim, Richard Dobson

Missingness is ubiquitous in multivariate time series and poses an obstacle to reliable downstream analysis. Although recurrent network imputation achieved the SOTA, existing models do not scale to deep architectures that can potentially alleviate issues arising in complex data. Moreover, imputation carries the risk of biased estimations of the ground truth. Yet, confidence in the imputed values is always unmeasured or computed post hoc from model output. We propose DEep Attention Recurrent Imputation (DEARI), which jointly estimates missing values and their associated uncertainty in heterogeneous multivariate time series. By jointly representing feature-wise correlations and temporal dynamics, we adopt a self attention mechanism, along with an effective residual component, to achieve a deep recurrent neural network with good imputation performance and stable convergence. We also leverage self-supervised metric learning to boost performance by optimizing sample similarity. Finally, we transform DEARI into a Bayesian neural network through a novel Bayesian marginalization strategy to produce stochastic DEARI, which outperforms its deterministic equivalent. Experiments show that DEARI surpasses the SOTA in diverse imputation tasks using real-world datasets, namely air quality control, healthcare and traffic.

Joshua Au Yeung, Jacopo Dalmasso, Luca Foschini et al.

Background: Emerging reports of "AI psychosis" are on the rise, where user-LLM interactions may exacerbate or induce psychosis or adverse psychological symptoms. Whilst the sycophantic and agreeable nature of LLMs can be beneficial, it becomes a vector for harm by reinforcing delusional beliefs in vulnerable users. Methods: Psychosis-bench is a novel benchmark designed to systematically evaluate the psychogenicity of LLMs comprises 16 structured, 12-turn conversational scenarios simulating the progression of delusional themes(Erotic Delusions, Grandiose/Messianic Delusions, Referential Delusions) and potential harms. We evaluated eight prominent LLMs for Delusion Confirmation (DCS), Harm Enablement (HES), and Safety Intervention(SIS) across explicit and implicit conversational contexts. Findings: Across 1,536 simulated conversation turns, all LLMs demonstrated psychogenic potential, showing a strong tendency to perpetuate rather than challenge delusions (mean DCS of 0.91 $\pm$0.88). Models frequently enabled harmful user requests (mean HES of 0.69 $\pm$0.84) and offered safety interventions in only roughly a third of applicable turns (mean SIS of 0.37 $\pm$0.48). 51 / 128 (39.8%) of scenarios had no safety interventions offered. Performance was significantly worse in implicit scenarios, models were more likely to confirm delusions and enable harm while offering fewer interventions (p < .001). A strong correlation was found between DCS and HES (rs = .77). Model performance varied widely, indicating that safety is not an emergent property of scale alone. Conclusion: This study establishes LLM psychogenicity as a quantifiable risk and underscores the urgent need for re-thinking how we train LLMs. We frame this issue not merely as a technical challenge but as a public health imperative requiring collaboration between developers, policymakers, and healthcare professionals.

Yuezhou Zhang, Amos A. Folarin, Callum Stewart et al.

Continuous monitoring of behavior and physiology via wearable devices offers a novel, objective method for the early detection of worsening depression and anxiety. In this study, we present an explainable anomaly detection framework that identifies clinically meaningful increases in symptom severity using consumer-grade wearable data. Leveraging data from 2,023 participants with defined healthy baselines, our LSTM autoencoder model learned normal health patterns of sleep duration, step count, and resting heart rate. Anomalies were flagged when self-reported depression or anxiety scores increased by >=5 points (a threshold considered clinically significant). The model achieved an adjusted F1-score of 0.80 (precision = 0.73, recall = 0.88) in detecting 393 symptom-worsening episodes across 341 participants, with higher performance observed for episodes involving concurrent depression and anxiety escalation (F1 = 0.84) and for more pronounced symptom changes (>=10-point increases, F1 = 0.85). Model interpretability was supported by SHAP-based analysis, which identified resting heart rate as the most influential feature in 71.4 percentage of detected anomalies, followed by physical activity and sleep. Together, our findings highlight the potential of explainable anomaly detection to enable personalized, scalable, and proactive mental health monitoring in real-world settings.

Julia Ive, Olatomiwa Olukoya, Jonathan P. Funnell et al.

Introduction: Timely care in a specialised neuro-intensive therapy unit (ITU) reduces mortality and hospital stays, with planned admissions being safer than unplanned ones. However, post-operative care decisions remain subjective. This study used artificial intelligence (AI), specifically natural language processing (NLP) to analyse electronic health records (EHRs) and predict ITU admissions for elective surgery patients. Methods: This study analysed the EHRs of elective neurosurgery patients from University College London Hospital (UCLH) using NLP. Patients were categorised into planned high dependency unit (HDU) or ITU admission; unplanned HDU or ITU admission; or ward / overnight recovery (ONR). The Medical Concept Annotation Tool (MedCAT) was used to identify SNOMED-CT concepts within the clinical notes. We then explored the utility of these identified concepts for a range of AI algorithms trained to predict ITU admission. Results: The CogStack-MedCAT NLP model, initially trained on hospital-wide EHRs, underwent two refinements: first with data from patients with Normal Pressure Hydrocephalus (NPH) and then with data from Vestibular Schwannoma (VS) patients, achieving a concept detection F1-score of 0.93. This refined model was then used to extract concepts from EHR notes of 2,268 eligible neurosurgical patients. We integrated the extracted concepts into AI models, including a decision tree model and a neural time-series model. Using the simpler decision tree model, we achieved a recall of 0.87 (CI 0.82 - 0.91) for ITU admissions, reducing the proportion of unplanned ITU cases missed by human experts from 36% to 4%. Conclusion: The NLP model, refined for accuracy, has proven its efficiency in extracting relevant concepts, providing a reliable basis for predictive AI models to use in clinically valid applications.

Shubham Agarwal, Vlad Dinu, Thomas Searle et al.

This study introduces RelCAT (Relation Concept Annotation Toolkit), an interactive tool, library, and workflow designed to classify relations between entities extracted from clinical narratives. Building upon the CogStack MedCAT framework, RelCAT addresses the challenge of capturing complete clinical relations dispersed within text. The toolkit implements state-of-the-art machine learning models such as BERT and Llama along with proven evaluation and training methods. We demonstrate a dataset annotation tool (built within MedCATTrainer), model training, and evaluate our methodology on both openly available gold-standard and real-world UK National Health Service (NHS) hospital clinical datasets. We perform extensive experimentation and a comparative analysis of the various publicly available models with varied approaches selected for model fine-tuning. Finally, we achieve macro F1-scores of 0.977 on the gold-standard n2c2, surpassing the previous state-of-the-art performance, and achieve performance of >=0.93 F1 on our NHS gathered datasets.

Kawsar Noor, Lukasz Roguski, Alex Handy et al.

As more healthcare organisations transition to using electronic health record (EHR) systems it is important for these organisations to maximise the secondary use of their data to support service improvement and clinical research. These organisations will find it challenging to have systems which can mine information from the unstructured data fields in the record (clinical notes, letters etc) and more practically have such systems interact with all of the hospitals data systems (legacy and current). To tackle this problem at University College London Hospitals, we have deployed an enhanced version of the CogStack platform; an information retrieval platform with natural language processing capabilities which we have configured to process the hospital's existing and legacy records. The platform has improved data ingestion capabilities as well as better tools for natural language processing. To date we have processed over 18 million records and the insights produced from CogStack have informed a number of clinical research use cases at the hospitals.

Zeljko Kraljevic, Anthony Shek, Daniel Bean et al.

The data available in Electronic Health Records (EHRs) provides the opportunity to transform care, and the best way to provide better care for one patient is through learning from the data available on all other patients. Temporal modelling of a patient's medical history, which takes into account the sequence of past events, can be used to predict future events such as a diagnosis of a new disorder or complication of a previous or existing disorder. While most prediction approaches use mostly the structured data in EHRs or a subset of single-domain predictions and outcomes, we present MedGPT a novel transformer-based pipeline that uses Named Entity Recognition and Linking tools (i.e. MedCAT) to structure and organize the free text portion of EHRs and anticipate a range of future medical events (initially disorders). Since a large portion of EHR data is in text form, such an approach benefits from a granular and detailed view of a patient while introducing modest additional noise. MedGPT effectively deals with the noise and the added granularity, and achieves a precision of 0.344, 0.552 and 0.640 (vs LSTM 0.329, 0.538 and 0.633) when predicting the top 1, 3 and 5 candidate future disorders on real world hospital data from King's College Hospital, London, UK (\textasciitilde600k patients). We also show that our model captures medical knowledge by testing it on an experimental medical multiple choice question answering task, and by examining the attentional focus of the model using gradient-based saliency methods.

Thomas Searle, Zina Ibrahim, James Teo et al.

The current mode of use of Electronic Health Record (EHR) elicits text redundancy. Clinicians often populate new documents by duplicating existing notes, then updating accordingly. Data duplication can lead to a propagation of errors, inconsistencies and misreporting of care. Therefore, quantifying information redundancy can play an essential role in evaluating innovations that operate on clinical narratives. This work is a quantitative examination of information redundancy in EHR notes. We present and evaluate two strategies to measure redundancy: an information-theoretic approach and a lexicosyntactic and semantic model. We evaluate the measures by training large Transformer-based language models using clinical text from a large openly available US-based ICU dataset and a large multi-site UK based Trust. By comparing the information-theoretic content of the trained models with open-domain language models, the language models trained using clinical text have shown ~1.5x to ~3x less efficient than open-domain corpora. Manual evaluation shows a high correlation with lexicosyntactic and semantic redundancy, with averages ~43 to ~65%.

Yuezhou Zhang, Amos A Folarin, Shaoxiong Sun et al.

The Bluetooth sensor embedded in mobile phones provides an unobtrusive, continuous, and cost-efficient means to capture individuals' proximity information, such as the nearby Bluetooth devices count (NBDC). The continuous NBDC data can partially reflect individuals' behaviors and status, such as social connections and interactions, working status, mobility, and social isolation and loneliness, which were found to be significantly associated with depression by previous survey-based studies. This paper aims to explore the NBDC data's value in predicting depressive symptom severity as measured via the 8-item Patient Health Questionnaire (PHQ-8). The data used in this paper included 2,886 bi-weekly PHQ-8 records collected from 316 participants recruited from three study sites in the Netherlands, Spain, and the UK as part of the EU RADAR-CNS study. From the NBDC data two weeks prior to each PHQ-8 score, we extracted 49 Bluetooth features, including statistical features and nonlinear features for measuring periodicity and regularity of individuals' life rhythms. Linear mixed-effect models were used to explore associations between Bluetooth features and the PHQ-8 score. We then applied hierarchical Bayesian linear regression models to predict the PHQ-8 score from the extracted Bluetooth features. A number of significant associations were found between Bluetooth features and depressive symptom severity. Compared with commonly used machine learning models, the proposed hierarchical Bayesian linear regression model achieved the best prediction metrics, R2= 0.526, and root mean squared error (RMSE) of 3.891. Bluetooth features can explain an extra 18.8% of the variance in the PHQ-8 score relative to the baseline model without Bluetooth features (R2=0.338, RMSE = 4.547).

Shuo Liu, Jing Han, Estela Laporta Puyal et al.

This study investigates the potential of deep learning methods to identify individuals with suspected COVID-19 infection using remotely collected heart-rate data. The study utilises data from the ongoing EU IMI RADAR-CNS research project that is investigating the feasibility of wearable devices and smart phones to monitor individuals with multiple sclerosis (MS), depression or epilepsy. Aspart of the project protocol, heart-rate data was collected from participants using a Fitbit wristband. The presence of COVID-19 in the cohort in this work was either confirmed through a positive swab test, or inferred through the self-reporting of a combination of symptoms including fever, respiratory symptoms, loss of smell or taste, tiredness and gastrointestinal symptoms. Experimental results indicate that our proposed contrastive convolutional auto-encoder (contrastive CAE), i. e., a combined architecture of an auto-encoder and contrastive loss, outperforms a conventional convolutional neural network (CNN), as well as a convolutional auto-encoder (CAE) without using contrastive loss. Our final contrastive CAE achieves 95.3% unweighted average recall, 86.4% precision, anF1 measure of 88.2%, a sensitivity of 100% and a specificity of 90.6% on a testset of 19 participants with MS who reported symptoms of COVID-19. Each of these participants was paired with a participant with MS with no COVID-19 symptoms.

Judith Dineley, Grace Lavelle, Daniel Leightley et al.

The ease of in-the-wild speech recording using smartphones has sparked considerable interest in the combined application of speech, remote measurement technology (RMT) and advanced analytics as a research and healthcare tool. For this to be realised, the acceptability of remote speech collection to the user must be established, in addition to feasibility from an analytical perspective. To understand the acceptance, facilitators, and barriers of smartphone-based speech recording, we invited 384 individuals with major depressive disorder (MDD) from the Remote Assessment of Disease and Relapse - Central Nervous System (RADAR-CNS) research programme in Spain and the UK to complete a survey on their experiences recording their speech. In this analysis, we demonstrate that study participants were more comfortable completing a scripted speech task than a free speech task. For both speech tasks, we found depression severity and country to be significant predictors of comfort. Not seeing smartphone notifications of the scheduled speech tasks, low mood and forgetfulness were the most commonly reported obstacles to providing speech recordings.

Zina M Ibrahim, Daniel Bean, Thomas Searle et al.

The ability to perform accurate prognosis of patients is crucial for proactive clinical decision making, informed resource management and personalised care. Existing outcome prediction models suffer from a low recall of infrequent positive outcomes. We present a highly-scalable and robust machine learning framework to automatically predict adversity represented by mortality and ICU admission from time-series vital signs and laboratory results obtained within the first 24 hours of hospital admission. The stacked platform comprises two components: a) an unsupervised LSTM Autoencoder that learns an optimal representation of the time-series, using it to differentiate the less frequent patterns which conclude with an adverse event from the majority patterns that do not, and b) a gradient boosting model, which relies on the constructed representation to refine prediction, incorporating static features of demographics, admission details and clinical summaries. The model is used to assess a patient's risk of adversity over time and provides visual justifications of its prediction based on the patient's static features and dynamic signals. Results of three case studies for predicting mortality and ICU admission show that the model outperforms all existing outcome prediction models, achieving PR-AUC of 0.891 (95$%$ CI: 0.878 - 0.969) in predicting mortality in ICU and general ward settings and 0.908 (95$%$ CI: 0.870-0.935) in predicting ICU admission.

Thomas Searle, Zina Ibrahim, Richard Dobson

Alzheimer's Dementia (AD) is an incurable, debilitating, and progressive neurodegenerative condition that affects cognitive function. Early diagnosis is important as therapeutics can delay progression and give those diagnosed vital time. Developing models that analyse spontaneous speech could eventually provide an efficient diagnostic modality for earlier diagnosis of AD. The Alzheimer's Dementia Recognition through Spontaneous Speech task offers acoustically pre-processed and balanced datasets for the classification and prediction of AD and associated phenotypes through the modelling of spontaneous speech. We exclusively analyse the supplied textual transcripts of the spontaneous speech dataset, building and comparing performance across numerous models for the classification of AD vs controls and the prediction of Mental Mini State Exam scores. We rigorously train and evaluate Support Vector Machines (SVMs), Gradient Boosting Decision Trees (GBDT), and Conditional Random Fields (CRFs) alongside deep learning Transformer based models. We find our top performing models to be a simple Term Frequency-Inverse Document Frequency (TF-IDF) vectoriser as input into a SVM model and a pre-trained Transformer based model `DistilBERT' when used as an embedding layer into simple linear models. We demonstrate test set scores of 0.81-0.82 across classification metrics and a RMSE of 4.58.

Aurelie Mascio, Zeljko Kraljevic, Daniel Bean et al.

Text classification tasks which aim at harvesting and/or organizing information from electronic health records are pivotal to support clinical and translational research. However these present specific challenges compared to other classification tasks, notably due to the particular nature of the medical lexicon and language used in clinical records. Recent advances in embedding methods have shown promising results for several clinical tasks, yet there is no exhaustive comparison of such approaches with other commonly used word representations and classification models. In this work, we analyse the impact of various word representations, text pre-processing and classification algorithms on the performance of four different text classification tasks. The results show that traditional approaches, when tailored to the specific language and structure of the text inherent to the classification task, can achieve or exceed the performance of more recent ones based on contextual embeddings such as BERT.

Zina Ibrahim, Honghan Wu, Richard Dobson

Many areas of research are characterised by the deluge of large-scale highly-dimensional time-series data. However, using the data available for prediction and decision making is hampered by the current lag in our ability to uncover and quantify true interactions that explain the outcomes.We are interested in areas such as intensive care medicine, which are characterised by i) continuous monitoring of multivariate variables and non-uniform sampling of data streams, ii) the outcomes are generally governed by interactions between a small set of rare events, iii) these interactions are not necessarily definable by specific values (or value ranges) of a given group of variables, but rather, by the deviations of these values from the normal state recorded over time, iv) the need to explain the predictions made by the model. Here, while numerous data mining models have been formulated for outcome prediction, they are unable to explain their predictions. We present a model for uncovering interactions with the highest likelihood of generating the outcomes seen from highly-dimensional time series data. Interactions among variables are represented by a relational graph structure, which relies on qualitative abstractions to overcome non-uniform sampling and to capture the semantics of the interactions corresponding to the changes and deviations from normality of variables of interest over time. Using the assumption that similar templates of small interactions are responsible for the outcomes (as prevalent in the medical domains), we reformulate the discovery task to retrieve the most-likely templates from the data.

Rebecca Bendayan, Honghan Wu, Zeljko Kraljevic et al.

Multimorbidity research in mental health services requires data from physical health conditions which is traditionally limited in mental health care electronic health records. In this study, we aimed to extract data from physical health conditions from clinical notes using SemEHR. Data was extracted from Clinical Record Interactive Search (CRIS) system at South London and Maudsley Biomedical Research Centre (SLaM BRC) and the cohort consisted of all individuals who had received a primary or secondary diagnosis of severe mental illness between 2007 and 2018. Three pairs of annotators annotated 2403 documents with an average Cohen's Kappa of 0.757. Results show that the NLP performance varies across different diseases areas (F1 0.601 - 0.954) suggesting that the language patterns or terminologies of different condition groups entail different technical challenges to the same NLP task.

Ehtesham Iqbal, Risha Govind, Alvin Romero et al.

Objective: Mining the data contained within Electronic Health Records (EHRs) can potentially generate a greater understanding of medication effects in the real world, complementing what we know from Randomised control trials (RCTs). We Propose a text mining approach to detect adverse events and medication episodes from the clinical text to enhance our understanding of adverse effects related to Clozapine, the most effective antipsychotic drug for the management of treatment-resistant schizophrenia, but underutilised due to concerns over its side effects. Material and Methods: We used data from de-identified EHRs of three mental health trusts in the UK (>50 million documents, over 500,000 patients, 2835 of which were prescribed Clozapine). We explored the prevalence of 33 adverse effects by age, gender, ethnicity, smoking status and admission type three months before and after the patients started Clozapine treatment. We compared the prevalence of adverse effects with those reported in the Side Effects Resource (SIDER) where possible. Results: Sedation, fatigue, agitation, dizziness, hypersalivation, weight gain, tachycardia, headache, constipation and confusion were amongst the highest recorded Clozapine adverse effect in the three months following the start of treatment. Higher percentages of all adverse effects were found in the first month of Clozapine therapy. Using a significance level of (p< 0.05) out chi-square tests show a significant association between most of the ADRs in smoking status and hospital admissions and some in gender and age groups. Further, the data was combined from three trusts, and chi-square tests were applied to estimate the average effect of ADRs in each monthly interval. Conclusion: A better understanding of how the drug works in the real world can complement clinical trials and precision medicine.

Zina Ibrahim, Honghan Wu, Ahmed Hamoud et al.

Current machine learning models aiming to predict sepsis from Electronic Health Records (EHR) do not account for the heterogeneity of the condition, despite its emerging importance in prognosis and treatment. This work demonstrates the added value of stratifying the types of organ dysfunction observed in patients who develop sepsis in the ICU in improving the ability to recognise patients at risk of sepsis from their EHR data. Using an ICU dataset of 13,728 records, we identify clinically significant sepsis subpopulations with distinct organ dysfunction patterns. Classification experiments using Random Forest, Gradient Boost Trees and Support Vector Machines, aiming to distinguish patients who develop sepsis in the ICU from those who do not, show that features selected using sepsis subpopulations as background knowledge yield a superior performance regardless of the classification model used. Our findings can steer machine learning efforts towards more personalised models for complex conditions including sepsis.

Thomas Searle, Zeljko Kraljevic, Rebecca Bendayan et al.

We present MedCATTrainer an interface for building, improving and customising a given Named Entity Recognition and Linking (NER+L) model for biomedical domain text. NER+L is often used as a first step in deriving value from clinical text. Collecting labelled data for training models is difficult due to the need for specialist domain knowledge. MedCATTrainer offers an interactive web-interface to inspect and improve recognised entities from an underlying NER+L model via active learning. Secondary use of data for clinical research often has task and context specific criteria. MedCATTrainer provides a further interface to define and collect supervised learning training data for researcher specific use cases. Initial results suggest our approach allows for efficient and accurate collection of research use case specific training data.

Honghan Wu, Karen Hodgson, Sue Dyson et al.

Background: Many efforts have been put into the use of automated approaches, such as natural language processing (NLP), to mine or extract data from free-text medical records to construct comprehensive patient profiles for delivering better health-care. Reusing NLP models in new settings, however, remains cumbersome - requiring validation and/or retraining on new data iteratively to achieve convergent results. Objective: The aim of this work is to minimize the effort involved in reusing NLP models on free-text medical records. Methods: We formally define and analyse the model adaptation problem in phenotype-mention identification tasks. We identify "duplicate waste" and "imbalance waste", which collectively impede efficient model reuse. We propose a phenotype embedding based approach to minimize these sources of waste without the need for labelled data from new settings. Results: We conduct experiments on data from a large mental health registry to reuse NLP models in four phenotype-mention identification tasks. The proposed approach can choose the best model for a new task, identifying up to 76% (duplicate waste), i.e. phenotype mentions without the need for validation and model retraining, and with very good performance (93-97% accuracy). It can also provide guidance for validating and retraining the selected model for novel language patterns in new tasks, saving around 80% (imbalance waste), i.e. the effort required in "blind" model-adaptation approaches. Conclusions: Adapting pre-trained NLP models for new tasks can be more efficient and effective if the language pattern landscapes of old settings and new settings can be made explicit and comparable. Our experiments show that the phenotype-mention embedding approach is an effective way to model language patterns for phenotype-mention identification tasks and that its use can guide efficient NLP model reuse.

Spiros Denaxas, Pontus Stenetorp, Sebastian Riedel et al.

Electronic health records (EHR) are increasingly being used for constructing disease risk prediction models. Feature engineering in EHR data however is challenging due to their highly dimensional and heterogeneous nature. Low-dimensional representations of EHR data can potentially mitigate these challenges. In this paper, we use global vectors (GloVe) to learn word embeddings for diagnoses and procedures recorded using 13 million ontology terms across 2.7 million hospitalisations in national UK EHR. We demonstrate the utility of these embeddings by evaluating their performance in identifying patients which are at higher risk of being hospitalised for congestive heart failure. Our findings indicate that embeddings can enable the creation of robust EHR-derived disease risk prediction models and address some the limitations associated with manual clinical feature engineering.