Athletes with Type 1 Diabetes (T1D) often experience unpredictable glucose fluctuations during exercise, yet current technologies rely on static, population-level exercise presets. SportRx-T1D is a simulation-based framework designed to evaluate whether personalized modeling can improve exercise safety. Using retrospective CGM, insulin, carbohydrate, and heart-rate–derived intensity data, an individualized sequence model forecasts glucose 15–90 minutes ahead and produces calibrated hypoglycemia-risk estimates. The FDA-validated UVA/Padova simulator is then used to test guideline-consistent insulin or carbohydrate adjustments informed by these risk predictions. Findings illustrate how personalized forecasting and virtual experimentation can reduce projected hypoglycemia and support more adaptive exercise decision-support tools.

Nursing care facilities in the United States provide essential post-acute and long-term care to clinically complex populations. These facilities are evaluated through metrics for payment, utilization, and quality, making them critical for data-driven performance and risk assessment. This study develops an integrated intelligence framework. The framework analyzes facility-level risk, utilization intensity, and quality variation using Medicare provider data from the Centers for Medicare & Medicaid Services (CMS).CMS public datasets, accessed via BigQuery, were used to construct a facility-level analytic dataset incorporating nursing facility and home health agency data. The key measures, which include episode counts, beneficiary volume, Medicare payments, outlier share, and service intensity, were first transformed into normalized features. Geographic information was derived from ZIP codes to enable spatial comparisons. The analytical framework combined principal component analysis (PCA), K-means clustering, autoencoder-based anomaly detection, and supervised machine learning models (logistic regression, random forest, and XGBoost), with SHAP used for model interpretability. This integrated approach identified distinct facility clusters, uncovered patterns of utilization intensity, and detected anomalous providers with potentially elevated operational risk. Machine learning models demonstrated the ability to differentiate high- and low-risk facility profiles, while SHAP analysis (a method to interpret feature impacts) highlighted key drivers of variation in utilization and payments. The framework provided a multidimensional understanding of facility behavior beyond traditional descriptive metrics. This study demonstrates that combining public CMS data with advanced machine learning and unsupervised methods can enhance healthcare intelligence and oversight. By identifying hidden facility segments and flagging outlier behavior, the proposed framework enables more targeted quality monitoring and informs policy decision-making. As a result, it offers a scalable, data-driven approach for improving transparency and performance evaluation in nursing care systems.

Leaving treatment against medical advice (AMA) among pregnant and postpartum women with substance use disorder (SUD) is influenced by various factors, including housing and other social determinants of health (age, insurance, SUDs and mental state). Housing instability can be a significant barrier for pregnant and postpartum women with SUD seeking treatment. Lack of stable housing can lead to difficulties in accessing care, completing treatment programs, and maintaining recovery.

This study aims to explore the feasibility and effectiveness of utilizing natural language processing to extract housing information from clinical notes, validate model performance. The dataset is coming from a clinical chart for patients in the Rainbow/Mending Rainbow program at Elam Mental Health Center (EMHC) at Meharry.