
Introduction: Medication administration is a high-risk clinical task requiring accurate interaction with digital systems such as electronic medical records (EMRs) and barcode medication administration. Simulation-based training is used to prepare learners for these workflows; however, how users visually process training content remains underexplored. This study evaluates visual attention patterns during EMR-based nursing simulation training using areas of interest (AOI) based eye-tracking. Methods : A total of 37 participants (15 clinicians, 22 non-clinicians) completed a six-part interactive EMR training video using SimEMR©. Eye-tracking data were collected using Sticky by Tobii to capture fixation count, return fixations, fixation duration, and AOI coverage. Training performance was measured using post-video multiple-choice responses and analyzed using independent-samples t-tests and variance tests. Results : Non-clinicians demonstrated significantly higher fixation counts (~19 vs ~12, p < 0.001) and more return fixations (~52 vs ~37, p < 0.001) compared to clinicians. Variability in revisit behavior was lower among clinicians (p = 0.03). Total viewing time (p = 0.230) and AOI coverage (~91% for both groups, p = 0.865) were similar. Training performance did not differ between groups (~88.9% vs ~87.6%, p = 0.798). Conclusion : Clinicians and non-clinicians achieved comparable training outcomes despite differences in visual attention patterns. Non-clinicians exhibited higher visual effort and reprocessing behavior, while clinicians demonstrated more efficient and consistent attention strategies. AOI-based metrics provide a direct method to evaluate how users process training content and can inform the design of more effective nursing simulation training systems.
Mental health conditions are widespread, yet scalable assessment remains limited, as clinical interviews are time-intensive and self-report screeners are episodic and prone to bias, motivating passive, data-driven approaches. This study investigates whether short-term physical activity patterns captured by consumer wearables can classify self-rated mental health in a large, diverse cohort. Using the All of Us Research Program’s Registered Tier Dataset v7, we aligned Fitbit activity data with self-rated mental health, binarized into positive (good/very good/excellent) and negative (poor/fair) classes. Addressing class imbalance (88.5% positive, 11.5% negative) issue through undersampling resulted in a final cohort of 2,092 participants (837 negative, 1,255 positive). In the cohort, each participant was represented by eleven seven-day physical activity features together with age. Five models, logistic regression (LR), support vector machine (SVM), decision tree (DT), random forest (RF), and XGBoost, were trained and evaluated using AUC, accuracy, precision, recall, and F1. Result indicated that XGBoost performed best (AUC = 0.77, F1 = 0.774, recall = 0.883), followed by RF, while LR and SVM showed similar performance (AUC ≈ 0.75) and DT performed worst (AUC = 0.63). These outcomes suggest that short-term wearable activity data can inform mental health classification, with ensemble methods performing best. In practice, such systems could enable low-burden, continuous monitoring to support early identification of at-risk individuals, but should be designed as decision-support tools that complement, but not replace, clinical assessment, with careful attention to interpretability, user trust, and appropriate handling of uncertainty.
Hospital readmission risk is typically assessed at or near discharge, limiting opportunities to address modifiable risk factors such as social determinants of health (SDOH). The Arrival Readmission Report (ARR), a proof-of-concept, AI-assisted, HIE-enabled tool was designed to facilitate readmission risk assessment at emergency department admission using an interpretable, structured format. The report incorporates 30-day readmission predictions from pre-admission variables, and SHapley Additive exPlanations (SHAP) to identify individual readmission risk drivers. Results are presented in structured modules, including a high-level risk snapshot, cohort comparisons, and patient-specific explanations. Preliminary formative evaluation highlighted strengths in interpretability and structure and opportunities for improving contextualization and decision support for earlier intervention.
Alcohol Use Disorder (AUD) is a multifactorial condition with severe individual and societal impacts. Ex- tending our 2024 study, this work examines lifestyle, background, and family history determinants of AUD using an expanded dataset from the All of Us Research Program. The updated analysis includes approxi- mately 2.5 times more participants than the prior study, enabling improved statistical power and evaluation of result stability over time. Using interpretable machine learning models and statistical analyses, we identified annual income, residential stability, recreational drug use, sex/gender, marital status, education, and family history as key contributors to AUD risk. Annual income remained the most influential predictor across both datasets, while other feature rankings showed modest shifts. Family history factors continued to demonstrate non-linear effects, with close relatives’ AUD status remaining influential despite differences between statisti- cal association and predictive importance. In predicting AUD versus non-AUD status, Random forest models achieved the highest classification accuracy (81%), consistent with 2024 results but with improved precision for identifying AUD cases. Overall, the findings confirm the robustness of previously identified AUD de- terminants and underscore the need for coordinated, multi-level prevention strategies addressing behavioral, familial, and structural factors contributing to AUD.
This study examines how trust and privacy perceptions vary across hospital scenarios with different levels of Human Robot Interaction (HRI) using a Lean UX research approach. Sixteen participants (24.81 ± 2.23 years, twelve females, four males) engaged with five storyboard-based scenarios: a) appointment support, b) wayfinding, c) visitor assistance, d) emergency evacuation, and e) autonomous logistics, using visual artifacts and modality tokens. Semi-structured interviews captured user expectations, trust, and privacy concerns. Results show that trust is scenario-dependent and strongly influenced by perceived competence, while privacy concerns are most salient in information-sensitive contexts. A clear pattern emerged: users prefer non-verbal communication for navigation and low-interaction tasks, with verbal communication reserved for staff handoffs. Robots that approach users proactively with multimodal cues elicit higher engagement. Participants consistently rejected audible disclosure of sensitive information, favoring device-based or visual communication. The Lean UX approach enabled rapid, low-cost exploration of user expectations, providing actionable, scenario-specific design insights for trustworthy and privacy-aware robot integration in healthcare settings.