Natural Language Research Articles

Review of Convolutional Neural Network

Over the past decade, the domain of deep learning has emerged as a swiftly expanding area of interest among researchers globally. It offers substantial benefits over traditional shallow networks, particularly in the realms of feature extraction and model fitting. Deep learning excels at uncovering intricate, distributed features from raw data, which exhibit robust generalization capabilities. It has triumphantly addressed challenges that were once deemed intractable within the field of artificial intelligence. With the exponential growth in the volume of training data and a marked increase in computational power, deep learning has achieved remarkable strides and found applications across a spectrum of domains, including but not limited to object detection, computer vision, natural language processing, speech recognition, and semantic parsing, thereby accelerating the evolution of artificial intelligence. Deep learning encompasses a series of non-linear transformations organized hierarchically, with deep neural networks representing the predominant form of contemporary deep learning methodologies. Inspired by the neural connections found in the animal visual cortex, these networks are characterized by local connectivity, shared weights, and pooling operations. Such attributes not only simplify the network model and curtail the number of trainable parameters but also engender invariance to shifts, distortions, and scaling, thereby endowing the model with enhanced robustness and fault tolerance. This makes the training and optimization of the network structure more manageable. This paper commences with a retrospective on the evolution of convolutional neural networks (CNNs). Subsequently, it delineates the architectures of neuron models and multilayer perceptron. It proceeds to dissect the CNN architecture, which typically encompasses a sequence of convolutional and pooling layers, culminating in fully connected layers, each serving distinct functions. The paper also elaborates on various enhanced algorithms for CNNs, such as the “Network in Network” and “Spatial Transformer Networks,” while also introducing supervised and unsupervised learning methodologies pertinent to CNNs and highlighting several prevalent open-source tools. Further, the paper scrutinizes the application of CNNs in various fields, including image classification, facial recognition, audio retrieval, electrocardiogram analysis, and object detection. It posits the amalgamation of CNNs with recurrent neural networks as a potential alternative for training datasets. The research culminates in the design of CNN structures with varying parameters and depths, supported by experimental analysis that elucidates the interplay among these parameters and the ramifications of their configuration. Ultimately, the paper synthesizes the merits and lingering challenges associated with CNNs and their practical applications.

Hallucination or vision? Establishing reliability and validity of AI-based qualitative analysis of the unified model of activism in the Scottish independence Twitter debate

PurposeThe present study explores both the validation of the unified model of activism and the methodological reliability of the LlamaParsing approach to natural language processing. Theoretically, it applies the unified model of activism within the context of the Scottish independence movement, evaluating its effectiveness in social media environment.Design/methodology/approachMethodologically, it addresses the reliability and validity challenges associated with AI analyses, particularly the issue of AI hallucinations—instances where AI generates seemingly accurate but incorrect information. By employing the LlamaParsing approach and then comparing and contrasting it with a quantitative content coding process, the study demonstrates how context-specific instructions can enhance the accuracy of AI analyses.FindingsThe findings indicate this approach not only tests and extends the unified model of activism but also offers a robust methodological framework for using NLP and RAG in qualitative research. This dual focus underscores the potential of AI to provide systematic and theoretically valuable insights while highlighting the importance of mitigating its limitations.Originality/valueThis study represents a cutting-edge approach to qualitative data analysis, theory development, and theory testing in communication using a tool that was developed in 2024.

Integrating Optimal Deep Learning with Natural Language Processing for Arabic Spam and Ham Tweets Recognition

Integrating Optimal Deep Learning with Natural Language Processing for Arabic Spam and Ham Tweets Recognition

A New Natural Language Processing-Inspired Methodology (Detection, Initial Characterization, and Semantic Characterization) to Investigate Temporal Shifts (Drifts) in Health Care Data: Quantitative Study.

Proper analysis and interpretation of health care data can significantly improve patient outcomes by enhancing services and revealing the impacts of new technologies and treatments. Understanding the substantial impact of temporal shifts in these data is crucial. For example, COVID-19 vaccination initially lowered the mean age of at-risk patients and later changed the characteristics of those who died. This highlights the importance of understanding these shifts for assessing factors that affect patient outcomes. This study aims to propose detection, initial characterization, and semantic characterization (DIS), a new methodology for analyzingchanges in health outcomes and variables over time while discovering contextual changes for outcomes in large volumes of data. The DIS methodology involves 3 steps: detection, initial characterization, and semantic characterization. Detection uses metrics such as Jensen-Shannon divergence to identify significant data drifts. Initial characterization offers a global analysis of changes in data distribution and predictive feature significance over time. Semantic characterization uses natural language processing-inspired techniques to understand the local context of these changes, helping identify factors driving changes in patient outcomes. By integrating the outcomes from these 3 steps, our results can identify specific factors (eg, interventions and modifications in health care practices) that drive changes in patient outcomes. DIS was applied to the Brazilian COVID-19 Registry and the Medical Information Mart for Intensive Care, version IV (MIMIC-IV) data sets. Our approach allowed us to (1) identify drifts effectively, especially using metrics such as the Jensen-Shannon divergence, and (2) uncover reasons for the decline in overall mortality in both the COVID-19 and MIMIC-IV data sets, as well as changes in the cooccurrence between different diseases and this particular outcome. Factors such as vaccination during the COVID-19 pandemic and reduced iatrogenic events and cancer-related deaths in MIMIC-IV were highlighted. The methodology also pinpointed shifts in patient demographics and disease patterns, providing insights into the evolving health care landscape during the study period. We developed a novel methodology combining machine learningand natural language processing techniques to detect, characterize, and understand temporal shifts in health care data. This understanding can enhance predictive algorithms, improve patient outcomes, and optimize health care resource allocation, ultimatelyimproving the effectiveness of machine learning predictive algorithms applied to health care data. Our methodology can be applied to a variety of scenarios beyond those discussed in this paper.

Flexible Natural Language-Based Image Data Downlink Prioritization for Nanosatellites

Nanosatellites increasingly produce more data than can be downlinked within a reasonable time due to their limited bandwidth and power. Therefore, an on-board system is required to prioritize scientifically significant data for downlinking, as described by scientists. This paper determines whether natural language processing can be used to prioritize remote sensing images on CubeSats with more flexibility compared to existing methods. Two approaches implementing the same conceptual prioritization pipeline are compared. The first uses YOLOv8 and Llama2 to extract image features and compare them with text descriptions via cosine similarity. The second approach employs CLIP, fine-tuned on remote sensing data, to achieve the same. Both approaches are evaluated on real nanosatellite hardware, the VERTECS Camera Control Board. The CLIP approach, particularly the ResNet50-based model, shows the best performance in prioritizing and sequencing remote sensing images. This paper demonstrates that on-orbit prioritization using natural language descriptions is viable and allows for more flexibility than existing methods.

A deep learning model for syncope recognition in the emergency department

Abstract Introduction Syncope is a common presenting symptom in the emergency departments (EDs) and poses challenges in its recognition and differentiation from alternative causes of transient loss of consciousness. Natural Language Processing (NLP) is an artificial intelligence technique that enables the analysis of natural language, such as clinical notes and medical history within electronic medical records (EMRs). The application of NLP into the medical field is surging but its use for the recognition of syncope is still underexplored. Purpose We aimed to develop a NLP-based deep learning model for syncope recognition in the ED. Methods Two models, based on the Italian and Multilingual Bidirectional Encoder Representations from Transformers (BERT) language model, were developed and evaluated using consecutive EMRs from Humanitas Research Hospital ED. The first "triage" model was only based on textual data contained in the "triage" section of the EMR. The second "anamnesis" model also considered textual data contained in the "medical history" section. Performance metrics including area under the curve (AUC), Matthew’s correlation coefficient (MCC) and Brier’s score (BS) were used to compare the discriminative capabilities of different BERT models. Results The models were developed and tested on 15,098 and 15,222 EMRs, respectively. The Multilanguage BERT discriminated patients with syncope with an AUC of 0.85 in the triage model and of 0.92 in the anamnesis model. The Italian BERT identified patients with syncope with a MCC of 0.50 and a BS of 0.02 in the triage model and of 0.64 and 0.01, respectively, in the anamnesis model. Conclusions We propose two NLP-based models that identify syncope patients with a high discriminative capability. The presented algorithm may provide a starting point for the management of syncope patients using machine learning techniques.ROC and calibration curves of modelsFramework for model implementation

Lp(a) testing and LDL-C levels among australian ASCVD patients in 2022: a cross-sectional study from the advara heartcare network (PREVAIL)

Abstract Background The proportion of Australians with atherosclerotic cardiovascular disease (ASCVD) who meet secondary prevention low-density lipoprotein cholesterol (LDL-C) targets is uncertain, particularly among those undergoing lipoprotein(a) (Lp(a)) testing. This study describes the characteristics and achievement of LDL-C targets in ASCVD patients, both with and without Lp(a) tests. Methods A cross-sectional study was performed utilising secondary data from the Advara HeartCare cardiology network electronic medical records (EMR). Natural language processing (NLP) was used to extract data from clinical letters between 2011 and 2022 for patients attended during 2022. The 2022 baseline report captures clinical characteristics (i.e,. comorbidities, medications, lab values) in those with and without Lp(a) tests. Results In 2022, a total of 55,427 ASCVD patients were identified, predominantly male (65.1%) with a mean age of 71.2 (12.3). 61% had dyslipidaemia, 58.6% had hypertension, and 21.4% had diabetes. 72.7% were on statins, 19.4% on ezetimibe and 5.6% on PCSK9i. LDL-C information was available for 60.2% (33,381) of all ASCVD patients, with a mean LDL-C of 2.0 mmol/dL. 47.6% and 21.4% had LDL-C levels <1.8 mmol/dL and <1.4 mmol/dL respectively. Lp(a) test results were available for 2.6% (n=1,460) of ASCVD patients, with 87.7% having LDL-C information available, and a mean LDL-C of 1.9 mmol/dL. Among this group, 54.5% and 25.1% had LDL-C levels <1.8 mmol/dL and <1.4 mmol/dL respectively. Conclusions The study highlights suboptimal achievement of LDL-C targets among both ASCVD patients and those tested for Lp(a), underscoring the need to improve secondary prevention strategies and Lp(a) measurement in the ASCVD population.

Monitoring event data extracted from online news for outbreak detection

Abstract Introduction Digital disease surveillance (DDS) detects public health events from internet-based data e.g., online news. Event features depicting epidemiological and social characteristics of health events can be extracted from news using the natural language process techniques. However, few studies have leveraged the event features to support anomaly detection in DDS. We aimed to understand the distribution of the event features and explore anomaly detection using the frequency of these features. Methods We collected event data from COVID-19-related news collected from October 1 to December 31, 2021, sourced from BioCaster, an infectious-disease-focused DDS system. The predefined event features in BioCaster include disease, pathogen, location and 14 binary features, such as if an event was caused by an unclassified virus. We described the distribution of the features and detected changes in the frequency of event features using a Bayesian online change point detection. We compared the change points with the number of new cases and of genomic samples collected. Results We included 170,168 news articles reporting COVID-19 in 155 countries. The event feature indicating that an event was caused by an unclassified virus was identified as positive among 3831 (2.25%) news and 12.91% of news had positive value for the feature indicating cases who had travelled across international borders. The change points detected from these two features were temporally correlated to the emergence of the Omicron variant in corresponding countries, which was more significant in countries with at least 300 news articles. Conversely, event features irrelevant to this case study, e.g., if the cases were military workers, were identified as negative in all news and no change points were detected. Conclusions Our study highlights the potential of monitoring the frequency of event features extracted from online news for anomaly detection in DDS, which relies on sufficient news coverage. Key messages • Monitoring the event features extracted from online news provide is useful approach for automatic anomaly detection in digital disease surveillance. • Increasing media coverage is fundamental for improving the early detection in a digital disease surveillance system.

Enhancing Public Health: training an AI chatbot for complex tasks support

Abstract Background The evolution of Large Language Models (LLMs) such as GPT-4, Gemini, and Copilot has transformed the field of Natural Language Processing, demonstrating human-level capabilities and potential applications in Public Health. Here, we report on the training of a chatbot model based on GPT-4 designed to assist operators with relevant and complex public health tasks. Methods A scoping review of existing studies on the application of LLMs in Public Health settings was conducted to understand the landscape and identify gaps in their current implementations. We then proceeded to analyse the necessary skills and competencies required for public health professionals to adequately instruct the model on required tasks. Finally, we selected the Italian regulations and guidelines with the greatest relevance in the public health field to provide the knowledge base to the model. Results The review revealed no relevant documents on the use of LLMs in areas related to Public Health. Skills and competencies were retrieved from the Guidelines of the Italian Society of Hygiene for Public Health operators and used to instruct the model on how to address public health tasks. As for the knowledge base, we retrieved 25 documents regulating specific areas of the public health field, such as vaccinations, hygiene and safety at work, infection prevention, and school hygiene. Such documents were used to train the model, after formatting them appropriately. Conclusions Despite challenges with non-standardized and technically complex documentation, preliminary testing of our model has shown its versatility and utility in supporting daily public health operations. A validation phase is under definition and will involve a DELPHI process, in which products of the custom GPT model will be evaluated by a panel of public health operators to evaluate its performance in real-world settings, ensuring compliance with national health regulations. Key messages • Training a GPT-4-based chatbot to support complex tasks in Public Health, tailored by relevant national guidelines. • Our custom model, informed by public health guidelines, showed potential in preliminary tests.

Leveraging ChatGPT for systematic reviews: how accurate is it for screening by title and abstract?

Abstract Background Systematic reviews are at the top of the evidence pyramid, but the whole process can be labor-intensive, especially in the screening by title and abstract of retrieved records. Large language models (LLMs), such as ChatGPT, are proficient in natural language processing tasks and offer potential for accelerating this phase. The aim of this study is to assess ChatGPT’s performance in screening records for an already published systematic review (doi: 10.1080/21645515.2023.2300848). Methods 1601 records with title and abstract were evaluated with ChatGPT 4, using two different prompts to instruct the model: one asking the model to classify each record as included or excluded (Prompt A), and one asking the model to rate each record from 1 to 5 based on inclusion confidence (Prompt B), to compute the model’s performance metrics. Results The review included 64 records after title and abstract screening, of which 18 after full text screening. Using records included by title and abstract as a reference, prompt B with a rating cut-off of 3 provided 82% sensitivity, 88% specificity and 99% negative predictive value (NPV). With articles included by full text as a reference, 100% sensitivity, 86% specificity, and 100% NPV. An 85% workload saving was reached. Prompts A showed higher workload savings (∼93%), high NPVs (∼100%) and specificity (∼96%), but lower sensitivity (62-72%). Conclusions Prompts with a rating cut-off of 3 achieved a better performance, and still relevant workload savings compared to binary classification prompts. These findings can inform prompt-engineering strategies to instruct LLMs and reach both high sensitivity and workload savings. Key messages • Systematic reviews provide evidence synthesis to guide policy and decision-making in healthcare and public health. Such process can be time-consuming, and need for timely evidence is often stringent. • LLMs can help reduce workload and speed up the title and abstract screening phase of a systematic review, which is considered a major bottleneck in the review process.

Automatic Negation Detection for Semantic Analysis in Arabic Hotel Reviews Through Lexical and Structural Features: A Supervised Classification

One significant challenge in sentiment analysis is the presence of negation, which reverses the meanings of sentences, transformingpositive statements into negative ones and impacting the sentiment conveyed in the text. This issue is particularly pronounced in Arabic, a language known for its complex morphology. Detecting negation is crucial for enhancing sentiment analysis performance and various natural language processing applications. This paper presents an approach for automatically detecting negation in user-generated Arabic hotel reviews through lexical and structural features. It comprises several stages: data collection, text pre-processing, feature extraction, supervised learning classification, and evaluation. The study employed multiple supervised classification techniques, including naïve Bayes, random forest, logistic regression, support vector machines, and deep learning, to analyse lexical and structural features extracted from the dataset. The results of the experiments yielded promising outcomes, demonstrating the feasibility of the approach for practical applications. The classifiers exhibited highly comparable performance in identifying negation, with only marginal deviations in their performance metrics. Notably, the deep learning classifier consistently emerged as the top performer, achieving an exceptionally high overall accuracy rate of 99.24 percent, surpassing established benchmarks in Arabic text processing and underscoring its potential for practical applications. These findings hold significant implications for advancing Arabic text processing, particularly in sentiment analysis and related NLP tasks. The high accuracy of 99.24 percent achieved by the deep learning classifier highlights its robustness in accurately detecting negation, a critical challenge in sentiment analysis. This classifier performance demonstrates the potential to be integrated into real-world applications, such as automated review systems and opinion mining tools, where accurate sentiment interpretation is essential.

Arabic Question Generation Using Transformers

After the increased reliance on online education, online assessment became an essential tool for educators to remotely monitor and evaluate students’ understanding in order to assist them properly. However, the laborious process of creating exam questions is a challenge for most teachers. Thus, automated Question Generation aims to assist teachers by generating questions from given data. Limited research has been conducted to tackle this issue in the Arabic Language due to the complexity of the language and the limited amount of available Arabic data. This paper explores different implementations of the transformer models, that demonstrated their superiority in natural language processing. Three approaches were introduced to tackle this problem with Arabic data using an Arabic-based transformer, an English-based transformer, and a multilingual-based transformer. Each of the fine-tuned models was trained using ARCD, XGLUE, DialectBench, and ArabicQA data sets and evaluated on automatic and manual metrics. Two of the proposed models achieve state-of-the-art results on the Arabic question generation task. The English transformer obtained a ROUGE score of 0.59 on XGLUE, while the Arabic transformer model achieves 0.49 on ARCD. Both of these models demonstrate excellent quality of questions through human-conducted evaluations by achieving low WER and high GC, U, and A scores.

Aortic valve replacement and mortality with left ventricular hypertrophy in setting of normal LVEF

Abstract Background Without symptoms, indications for aortic valve replacement (AVR) in severe aortic stenosis (AS) differ for individuals with normal vs. low/decreasing left ventricular ejection fraction (LVEF). Conceptually, the development of left ventricular hypertrophy (LVH) could also indicate ventricular injury and potential benefit with earlier AVR. However, little is known about the comparative effectiveness of AVR in individuals with LVH. As such, unlike low or dropping LVEF, development of LVH is not in itself an indication for AVR in clinical guidelines. Methods Initially, 1,232,492 echocardiography reports from 12 hospitals were identified from 1 January 2000 to 22 November 2022, and previously validated natural language processing modules were then applied to identify aortic gradients, LVEF, and presence of LVH. These reports were linked to mortality data, key comorbidities with a 2-year lookback covariate window, and date of any AVR. We then isolated echocardiograms demonstrating normal flow severe AS, LVEF ≥ 55% and LVH. Chart reviews were conducted to reduce missing data and outcomes were censored on the date of the chart review (October 4, 2023). To assess the association between AVR and mortality, Cox proportional hazards models considering treatment status (ie, AVR or not) were used with the index echocardiogram as time zero. Covariates used in risk-adjustment were chosen a priori based on conditions and laboratory values used in cardiac surgery mortality risk-adjustment, since those variables could plausibly confound the treatment decision. To address immortal time bias, AVR was considered a time-varying covariate. Multiple additional approaches with propensity scores were performed as sensitivity analyses. Results After application of inclusion and exclusion criteria, 4856 unique patients remained for analysis. Of those, 2043 (42.1%) received AVR with 841/2043 (41.2%) receiving surgical AVR and 1202/2043 (58.8%) receiving transcatheter AVR. Patients who received AVR had lower unadjusted mortality than those who did not receive AVR (33.6% vs. 47.1%, p < 0.001). After adjustment, and accounting for time-dependence of AVR, AVR was associated with reduced mortality (HR 0.81, 95% CI 0.73-0.89, p < 0.0001). The sensitivity analyses were all consistent with the primary analysis. Conclusions AVR was associated with reduced mortality for patients with severe AS, preserved LVEF, and LVH. This association persisted even after accounting for time-dependence, which tends to reduce the observed effectiveness of procedures in observational data. With over a million echocardiograms over more than 20 years, this is the largest known analysis of AVR outcomes in individuals with severe AS and LVH. Our findings suggest LVH may be a high-risk phenotype of individuals with AS, similar to dropping LVEF. As such, these results raise the potential of more proactive AVR clinical guidelines in individuals with LVH, even when LVEF is normal.Survival in AVR Vs No AVR

Natural language processing realises the potential of electronic health records: identifying risk factors for disease progression and major adverse limb events in intermittent claudication

Abstract Background/Introduction Peripheral Arterial Disease (PAD) is a common pathology, affecting 4.5% of the UK population. Symptomatic PAD manifests as intermittent claudication (IC). Predictors of IC progression to Chronic-Limb-Threatening Ischaemia (CLTI) and complications including Major Adverse Limb Events (MALE), such as smoking history and diabetes, have been described previously in resource-intensive cohort studies. Novel Natural Language Processing (NLP) approaches transform routinely collected, unstructured health records into datasets for real-world analysis of PAD. Purpose This study leverages NLP based interrogation of electronic health records to efficiently identify risk factors for disease progression and MALE in patients presenting with IC. Methods A retrospective cohort study of patients with PAD at a large, tertiary vascular referral centre identified using the SNOMED terms of IC in the Medcat-NLP-AI toolkit. Demographics, Index of Multiple Deprivation (IMD), and indicators of disease progression, and revascularisation and amputation (MALE) were analysed using Kaplan-Meier survival and Cox Proportional-Hazards analysis. Results 5,027 patients (Mean age 73.7(61.8-85.6), Males 66.1% (n=2,781) were identified. Self-reported population ethnicity was 4.72% Asian, 15.5% Black, 79.7% White. All-cause mortality was 19.5% (n=979) after 10 years of follow-up, with 5.85% of patients progressing to CLTI (n=294). 2.32% of Asian patients, 5.78% of Black patients and 5.04% of White patients progressed to CLTI, respectively. 5.09% of Asian patients, 8.99% of Black patients, and 7.36% of White patients underwent amputation. Multivariate analysis of risk factors for progression to CLTI, demonstrated that diabetes (HR 2.36, 95%CI: 1.22 - 4.48, p = .01) and smoking history (HR 1.76, 95%CI: 1.19 - 2.59, p = .004) independently predicted progression. No risk factors for MALE reached significance. Conclusions Diabetes and smoking behaviour are well-described in the literature as risk factors for progression to CLTI. This study demonstrates the utility of NLP as a resource efficient, validated, methodological tool. Disparities in MALE may exist between patients within a universal healthcare system, although our sample size was insufficient to demonstrate statistical significance. A larger, multicentre analysis is warranted. NLP represents a valuable tool for efficiently identifying and analysing risk factors for PAD progression and holds potential to refine management algorithms.KM curves of CLTI-free survival (DM)

Sentiment classification for insider threat identification using metaheuristic optimized machine learning classifiers

This study examines the formidable and complex challenge of insider threats to organizational security, addressing risks such as ransomware incidents, data breaches, and extortion attempts. The research involves six experiments utilizing email, HTTP, and file content data. To combat insider threats, emerging Natural Language Processing techniques are employed in conjunction with powerful Machine Learning classifiers, specifically XGBoost and AdaBoost. The focus is on recognizing the sentiment and context of malicious actions, which are considered less prone to change compared to commonly tracked metrics like location and time of access. To enhance detection, a term frequency-inverse document frequency-based approach is introduced, providing a more robust, adaptable, and maintainable method. Moreover, the study acknowledges the significant impact of hyperparameter selection on classifier performance and employs various contemporary optimizers, including a modified version of the red fox optimization algorithm. The proposed approach undergoes testing in three simulated scenarios using a public dataset, showcasing commendable outcomes.

Epidemiology, clinical characterization, and outcomes for heart failure with improved ejection fraction

Abstract Background Contemporary guidelines propose a new classification of heart failure with improved ejection fraction (HFimpEF). The prevalence of HFimpEF is expected to increase with the implementation of guideline-directed medical therapy (GDMT); however, there are limited data on the epidemiology of and outcomes for this novel clinical entity. Purpose We describe the prevalence, GDMT use and outcomes among patients with HFimpEF within a large, integrated healthcare delivery system. Methods We identified adults ≥18 years with incident HF with reduced EF (HFrEF ≤40%) between January 2013 and December 2018 from a large, integrated healthcare system serving >4.5 million members. HFimpEF was defined as known HFrEF with a follow-up EF measurement >40% within 12 months after incident HFrEF. Among HFimpEF patients, we examined GDMT use at discharge after incident HFrEF, incident HFimpEF, and at 6- and 12-months post-HFimpEF. From 12 months post initial HFrEF diagnosis, we calculated incidence rates of subsequent worsening HF (WHF) events (i.e., HF-related hospitalization, emergency department or urgent outpatient visits adjudicated using validated natural language processing-based algorithms) and all-cause death through December 31, 2019. We compared rates between patients with HFimpEF and those with persistent HFrEF (i.e., having all LVEF measurements within 12 months of incident HFrEF ≤40%) using Cox proportional hazards models adjusted for the EF value at incident HFrEF. Results Among 12,639 patients with incident HFrEF, 4,124 (33%) experienced HFimpEF within 12 months. Among patients with HFimpEF, use of β-blockers (82%) and angiotensin-converting enzyme inhibitor/angiotensin II receptor blockers (86%) was high, while angiotensin receptor-neprilysin inhibitor (1.2%) or mineralocorticoid receptor antagonist (28.2%) use was low. β-blockers, ACEi/ARB, and MRA use declined after incident HFimpEF (Figure, Panel A). Rates of WHF were 19.7 (95% Confidence Interval [CI]: 18.7 to 20.6) per 100 person-years among patients with HFimpEF vs. 40.0 (95% CI: 38.5 to 41.5) among patients with persistent HFrEF (hazard ratio [HR]: 0.55, 95% CI: 0.49 to 0.62; Figure, Panel B). Rates of death were 5.2 (95% CI: 4.7 to 5.7) and 8.4 (95% CI: 7.8 to 9.1) per 100 person-years among HFimpEF and persistent HFrEF, respectively (HR: 0.59, 95% CI: 0.52 to 0.67). Among patients with HFimpEF, those with an absolute EF improvement of ≥20% were less likely to experience WHF (HR: 0.82, 95% CI: 0.67 to 0.99) and death (HR: 0.85, 95% CI: 0.69 to 1.04) compared with those with <20% improvement. Conclusions One in three patients experienced improvement in EF within 12 months of initial HFrEF diagnosis. HFimpEF is associated with a reduced risk of WHF events and death, although these patients remain at significant clinical risk. Further research is necessary to assess the potential impact of sustained adherence to GDMT and therapeutic optimization on HFimpEF patient outcomes.

Future Trends of Open-Source AI in Libraries: Implications for Librarianship and Service Delivery

This paper explores the future trends and implications of open-source artificial intelligence (AI) for libraries, focusing on predicted technological advancements, long-term impacts on library operations, and the evolving role of librarians. Key advancements, such as enhanced natural language processing, intelligent recommendation systems, and advanced data analytics, are expected to significantly improve user experience and operational efficiency. The implications of these technologies include more personalized and responsive service delivery, streamlined operations, and an evolution in the roles and responsibilities of library staff. Librarians will need to develop new skills and advocate for ethical AI use, ensuring that AI applications align with the library’s values of inclusivity and accessibility. Additionally, the paper discusses the challenges of adopting open-source AI, including technological complexity, resource constraints, and data privacy concerns. The paper concludes that embracing open-source AI fosters innovation and collaboration, positioning libraries as vital hubs of knowledge and community engagement in the future.

Developing a multi-center registry of peripheral artery disease using natural language processing: unlocking the power of the electronic health record

Abstract Background Although peripheral artery disease (PAD) is common and associated with a high risk of mortality, contemporary PAD research remains limited due to poor accuracy of diagnosis codes in the electronic health record (EHR). We previously developed and validated a novel natural language processing (NLP) algorithm that can extract ankle brachial index (ABI) and toe-brachial index (TBI) values from clinical documents and identify patients with PAD with high accuracy (positive predictive value 92.4%). Purpose Scale the NLP system for developing a nationwide registry of PAD in the Veterans Health Administration (VHA), the largest integrated health system in the U.S. that provides care to ~9 million Veterans across 130 sites. Describe the baseline characteristics and mortality outcomes among participants and identify predictors of long-term mortality. Methods A total of 1,182,025 documents from vascular and radiology procedures during 2013-2020 were processed by the NLP system. After excluding non-ABI vascular studies, patients with a known diagnosis of PAD prior to 2015, patients younger than 40 years of age, and those who do not regularly follow-up in the VHA, we identified 107,711 patients with newly diagnosed PAD during 2015-2020. Patients were followed longitudinally using the VA’s electronic health record data. The primary outcome was all-cause mortality. We used Cox proportional hazards regression to examine predictors of long-term mortality. Results The mean age was 70.5 years; 97.3% were males, and 18.5% were of Black race. The mean ABI value was 0.78 (SD: 0.26) and the mean TBI value was 0.51 (SD: 0.19). There was a high prevalence of hypertension (83.6%), heart failure (20.8%), diabetes (53.6%), renal failure (22.1%) and chronic obstructive pulmonary disease (32.4%, Table 1). At 1-year, 9.6% of patients had died and this proportion increased to 25.4% at 3 years. Although nearly all patient variables were significantly associated with mortality, older age, severe PAD at diagnosis, diabetes, heart failure, renal failure and COPD were the strongest predictors of long-term mortality. Black race was associated with lower hazard of long-term mortality. Conclusions Using a previously validated novel NLP algorithm that extracts ABI and TBI values from clinical documents, we have successfully developed a registry of newly diagnosed PAD in the VHA. Ongoing studies from our cohort will yield important insights into the association between clinical management of PAD and clinical outcomes, with the goal of identifying opportunities for improving care.

Addressing the risks of generative AI for the judiciary: The accountability framework(s) under the EU AI Act

The rapid advancements in natural language processing, particularly the development of generative large language models (LLMs), have renewed interest in using artificial intelligence (AI) for judicial decision-making. While these technological breakthroughs present new possibilities for legal automation, they also raise concerns about over-reliance and automation bias. Drawing insights from the COMPAS case, this paper examines the implications of deploying generative LLMs in the judicial domain. It identifies the persistent factors that contributed to an accountability gap when AI systems were previously used for judicial decision-making. To address these risks, the paper analyses the relevant provisions of the EU Artificial Intelligence Act, outlining a comprehensive accountability framework based on the regulation's risk-based approach. The paper concludes that the successful integration of generative LLMs in judicial decision-making requires a holistic approach addressing cognitive biases. By emphasising shared responsibility and the imperative of AI literacy across the AI value chain, the regulatory framework can help mitigate the risks of automation bias and preserve the rule of law.

Close clinical monitoring of patients after TAVI implantation using artificial intelligence with a virtual voice assistant

Abstract Introduction The number of procedures and patients candidates for transcatheter aortic valve implantation (TAVI) has increased considerably due to population ageing and the rise in scientific evidence supporting this technique in patients with increasingly lower surgical risk. Early discharge would allow for a more efficient use of available resources, in addition to improving patients’ functionality and quality of life. Objective The use of a virtual voice assistant based on artificial intelligence, could be an effective and safe alternative to closely monitor patients after TAVI implantation in order to facilitate early identification of complications and a more efficient use of the resources available in Cardiology services. Material and methods Observational, prospective, single-center and consecutive study of the implementation of a technology based on artificial intelligence and natural language processing to monitor all patients undergoing transfemoral TAVI implantation, according to usual clinical practice, in the hemodynamics unit of our hospital in the year 2023. The virtual voice assistant, "Lola", carries out a telephonic follow-up during week 1, week 2, month 1, month 3 and 12. In these calls, it asks a series of questions mainly related to vascular access and with the patient's cardiovascular situation. After finishing the call, all the information collected is uploaded to a web platform where we monitor the collected data and thus take the appropriate measures. Results After 12 months, 274 patients have been included to whom 1,039 calls have been made with a total of 385 hours of autonomous conversation. In 44% of the calls, no alerts were detected and, therefore, did not required further review. From the remaining calls, a total of 926 alerts were collected, the predominant ones being: dizziness (24.9%), contact with the emergency services (11.3%) and alteration of general status (10.4%). The number of alerts decreases as the follow-up progresses, reflecting the need for closer monitoring in the initial stages after implantation. These alerts generated at least one intervention in 57% of the calls, the predominant ones being: medical contact (43.0%), nursing contact (26.3%) and medication adjustment (4.0%). These calls resulted in a total of 15 in-person examinations (5.4% of patients), detecting in 2 of them a 3rd degree atrioventricular block, in 3 a femoral pseudoaneurysm and in 1 a critical stenosis of the common femoral artery. According to a patient survey, 88.9% of patients were satisfied or very satisfied (Customer Satisfaction score 4.68/5). Conclusions The application of artificial intelligence permits strict control of the patient after TAVI implantation, thus facilitating early detection of complications. Patient satisfaction regarding this follow-up system was very high.