Monitoring adverse drug events (ADEs) is critical for pharmacovigilance and patient safety. However, identifying ADEs remains challenging, as suspected or confirmed side effects are often documented solely in the unstructured text of electronic health records (EHRs). Manually reviewing clinical notes to detect ADEs is labor-intensive and time-consuming, highlighting the need for automated methods capable of analyzing and extracting ADE-related information from clinical documentation. In this short communication, we describe our ongoing research on fine-tuning and evaluating a large language model (LLM) for the detection of ADEs in clinical notes. Preliminary descriptive result of this study indicates that ADEs are poorly documented in discharge notes, with less than 15% explicitly linking ADEs to specific drugs, which highlights the need for improved reporting practices.

Background: Medications are essential for health care but can cause adverse drug events (ADEs), which are harmful and sometimes fatal. Detecting ADEs is a challenging task because they are often not documented in the structured data of electronic health records (EHRs). There is a need for automatically extracting ADE-related information from clinical notes, as manual review is labor-intensive and time-consuming. Objective: This study aims to fine-tune the pretrained clinical language model, Swedish Deidentified Clinical Bidirectional Encoder Representations from Transformers (SweDeClin-BERT), for medical named entity recognition (NER) and relation extraction (RE) tasks, and to implement an integrated NER-RE approach to more effectively identify ADEs in clinical notes from clinical units in Sweden. The performance of this approach is compared with our previous machine learning method, which used conditional random fields (CRFs) and random forest (RF). Methods: A subset of clinical notes from the Stockholm EPR (Electronic Patient Record) Corpus, dated 2009‐2010, containing suspected ADEs based on International Classification of Diseases, 10th Revision ( ICD-10 ) codes in the A.1 and A.2 categories was randomly sampled. These notes were annotated by a physician with ADE-related entities and relations following the ADE annotation guidelines. We fine-tuned the SweDeClin-BERT model for the NER and RE tasks and implemented an integrated NER-RE pipeline to extract entities and relationships from clinical notes. The models were evaluated using 395 clinical notes from clinical units in Sweden. The NER-RE pipeline was then applied to classify the clinical notes as containing or not containing ADEs. In addition, we conducted an error analysis to better understand the model’s behavior and to identify potential areas for improvement. Results: In total, 62% of notes contained an explicit description of an ADE, indicating that an ADE-related ICD-10 code alone does not ensure detailed event documentation. The fine-tuned SweDeClin-BERT model achieved an F ₁ -score of 0.845 for NER and 0.81 for RE task, outperforming the baseline models (CRFs for NER and random forests for RE). In particular, the RE task showed a 53% improvement in macro-average F ₁ -score compared to the baseline. The integrated NER-RE pipeline achieved an overall F ₁ -score of 0.81. Conclusions: Using a domain-specific language model like SweDeClin-BERT for detecting ADEs in clinical notes demonstrates improved classification performance (0.77 in strict and 0.81 in relaxed mode) compared to conventional machine learning models like CRFs and RF. The proposed fine-tuned ADE model requires further refinement and evaluation on annotated clinical notes from another hospital to evaluate the model’s generalizability. In addition, the annotation guidelines should be revised, as there is an overlap of words between the Finding and Disorder entity categories, which were not consistently distinguished by the annotators. Furthermore, future work should address the handling of compound words and split entities to better capture context in the Swedish language.

Background: Maintaining physical activity (PA) and functioning (mobility, balance) is essential for older adults’ well-being and quality of life. However, current methods (functional tests, self-reports) and available techniques (accelerometers, sensors, advanced movement analysis systems) for assessing physical activity and functioning have shown to be less reliable, time- and resource-consuming with limited routine usage in clinical practice. There is a need to simplify the assessment of physical activity and functioning among older adults both in health care and clinical studies. This work presents a study on using Skeleton Avatar Technology (SAT) for this assessment. SAT analyzes human movement videos using artificial intelligence (AI). The study compares handy SAT based on 2D camera technology (2D SAT) with previously studied 3D SAT for assessing physical activity and functioning in older adults. Objective: To explore whether 2D SAT yields accurate results in physical activity and functioning assessment in healthy older adults, statistically compared to the accuracy of 3D SAT. Method: The mobile pose estimation model provided by Tensorflow was used to extract 2D skeletons from the video recordings of functional test movements. Deep neural networks were used to predict the outcomes of functional tests (FT), expert-based movement quality assessment (EA), accelerometer-based assessments (AC), and self-assessments of PA (SA). To compare the accuracy with 3D SAT models, statistical analysis was used to test whether the difference in the predictions between 2D and 3D models is significant or not. Results: Overall, the accuracy of 2D SAT is lower than 3D SAT in predicting FTs and EA. 2D SAT was able to predict AC with 7% Mean Absolute Error (MAE), and self-assessed PA (SA) with 16% MAE. On average MAE was 4% higher for 2D than for 3D SAT. There was no significant difference found between the 2D and the 3D model for AC and for two FTs (30 seconds chair stand test, 30sCST and Timed up and go, TUG). A significant difference was found for the 2D- and 3D-model of another FT (4-stage balance test, 4SBT). Conclusion: Altogether, the results show that handy 2D SAT might be used for assessing physical activity in older adults without a significant loss of accuracy compared to time-consuming standard tests and to bulky 3D SAT-based assessments. However, the accuracy of 2D SAT in assessing physical functioning should be improved. Taken together, this study shows promising results to use 2D SAT for assessing physical activity in healthy older adults in future clinical studies and clinical practice.

Adverse drug events (ADEs) occur frequently and are a common cause of suffering, hospitalizations, or death, and can be caused by harmful combinations of medications. One method used to prevent ADEs is by using clinical decision support systems (CDSSs). Janusmed Risk Profile is a CDSS evaluating the risk for nine common or serious ADEs resulting from combined pharmacodynamic effects. The aim of this study was to examine the prevalence of potential ADEs identified using CDSS algorithms from Janusmed Risk Profile. This retrospective, cross-sectional study covered the population of a Swedish region (n = 246,010 inhabitants in year 2020) using data on all medications dispensed and administered. More than 20% of patients had an increased risk of bleeding, constipation, orthostatism, or renal toxicity based on their medications. The proportion of patients with an increased risk varied from 3.5% to almost 30% across the nine categories of ADEs. A higher age was associated with an increased risk of potential ADEs and there were gender differences. A cluster analysis identified groups of patients with an increased risk for several categories of ADEs. This study shows that combinations of medications that could increase the risk of ADEs are common. Future studies should examine how this correlates with observed ADEs.

Maritime operations are increasingly reliant on sensor data to drive efficiency and enhance decision-making. However, despite rapid advances in large language models, including expanded context windows and stronger generative capabilities, critical industrial settings still require secure, role-constrained access to enterprise data and explicit limitation of model context. Retrieval-Augmented Generation (RAG) remains essential to enforce data minimization, preserve privacy, support verifiability, and meet regulatory obligations by retrieving only permissioned, provenance-tracked slices of information at query time. However, current RAG solutions lack robust validation protocols for numerical accuracy for high-stakes industrial applications. This paper introduces Lighthouse Bot, a novel Agentic RAG system specifically designed to provide natural-language access to complex maritime sensor data, including time-series and relational sensor data. The system addresses a critical need for verifiable autonomous data analysis within the Artificial Intelligence of Things (AIoT) domain, which we explore through a case study on optimizing ferry operations. We present a detailed architecture that integrates a Large Language Model with a specialized database and coding agents to transform natural language into executable tasks, enabling core AIoT capabilities such as generating Python code for time-series analysis, executing complex SQL queries on relational sensor databases, and automating workflows, while keeping sensitive data outside the prompt and ensuring auditable, policy-aligned tool use. To evaluate performance, we designed a test suite of 24 questions with ground-truth answers, categorized by query complexity (simple, moderate, complex) and data interaction type (retrieval, aggregation, analysis). Our results show robust, controlled data access with high factual fidelity: the proprietary Claude 3.7 achieved close to 90% overall factual correctness, while the open-source Qwen 72B achieved 66% overall and 99% on simple retrieval and aggregation queries. These findings underscore the need for a secure limited-context RAG in maritime AIoT and the potential for cost-effective automation of routine exploratory analyses.