Maintaining the uninterrupted flow of care for patients with complex medical needs hinges upon effective transitions of care (TOC). Routine work within healthcare organizations could face unforeseen pressure due to the COVID-19 pandemic, potentially taxing the TOC system. To understand how TOC structures have evolved, this study will assess them both before and during the COVID-19 pandemic, and quantify those changes using network analysis. We analyzed a trauma registry repository at Vanderbilt University Medical Center (VUMC) documenting care transitions of 5674 inpatients (2699 pre-COVID-19 and 2975 intra-COVID-19) who were admitted between January 2019 and May 2021. The structures of TOC and their transformations were gauged using network metrics, encompassing assortativity, homophily, and the small-world characteristic. Prior to and during the COVID-19 era, TOC structures exhibited disassortative tendencies, homophilic patterns, and small-world characteristics; the pandemic's impact on these three structural aspects was minimal. Transferring patients between highly-connected triage centers and less-connected receivers is facilitated by the disassortative structure of the TOC; the homophily characteristic reveals that similar patient types are often served by linked care units; and the small-world nature of the network means that transfer routes to highly collaborative units are generally short.Over 78 million individuals are projected to experience dementia by 2030, emphasizing the critical need for early detection in at-risk patients with mild cognitive impairment (MCI) and personalized clinical evaluation steps to diagnose potentially reversible causes. Utilizing real-world electronic health records within the observational medical outcomes partnership (OMOP) data model, we develop machine learning models to forecast Mild Cognitive Impairment (MCI) up to one year prior to its recorded diagnosis. Models for predicting MCI onset, including logistic regression, random forest, and XGBoost, were trained and evaluated on more than 531,000 patient visits. The random forest model displayed an ROC-AUC of 0.68207. The clinical elements most impactful in forecasting MCI, as noted by clinicians, are ascertained. Applying comparable association mining techniques, we derive a data-driven roster of frequently ordered clinical procedures during MCI case investigations, a resource that can be used to create clinical guidelines and pre-defined order sets.In the US, primary open-angle glaucoma (POAG) is the top cause of blindness, specifically targeting African Americans and Hispanics. Fundus image analysis using deep learning in POAG detection frequently yields results that are similar to, or even superior to, the standard of clinical diagnosis. Clinical diagnoses, susceptible to human bias, can unfortunately be mirrored and magnified in widely deployed deep learning models, consequently affecting their efficacy. Prejudices can result in (1) overlooking a condition's presence, thus raising the probability of delayed or inadequate treatment, and (2) wrongly identifying a condition, which might contribute to increased stress, fear, diminished well-being, and needless/costly intervention. In this research, deep learning's deployment for POAG detection, using the Ocular Hypertension Treatment Study (OHTS) dataset from 22 centers in 16 US states, was assessed for its susceptibility to underdiagnosis and overdiagnosis. The conclusions drawn from our research highlight a possible flaw in the popular deep learning model's application to underserved populations, leading to either an underestimation or an overestimation of their conditions. A pronounced instance of underdiagnosis is observed in the group of females under 60 years of age; conversely, the older Black population (60 years and older) is frequently overdiagnosed. Deep learning methodologies in ophthalmology, when employed using traditional techniques, may lead to biased diagnoses, potentially delaying appropriate treatments and exacerbating burdens on underserved populations, consequently presenting significant ethical challenges.The United States' drinking water contains per- and polyfluoroalkyl substances (PFAS), man-made compounds with demonstrated human toxicity, which are now a known contaminant. Our electronic health records were expanded to include geospatial information, which enabled classification of PFAS exposure for our patients in New Jersey. Three popular PFAS exposure classification methods, as seen in the literature, were tested, resulting in the identification of differing boundary types: public water supplier service areas, municipal boundaries, and ZIP code delineations. We investigated the point where all three borders met. To assess potential bias, we analyzed the previously identified associations between PFAS exposure and diseases, namely hypertension, thyroid disease, and parathyroid disease. The classification method for PFAS exposure proved influential in determining the statistical significance and the effect size of the associations. A higher proportion of PFAS exposure was found among Black/African-American patients in multiple cohorts, thus drawing attention to the intricate relationship between knowledge discovery and environmental justice.Patients receiving chemotherapy frequently encounter cancer-related physical impairments and consequent functional decline. While exercise is demonstrably helpful in addressing these symptoms, cancer patients' access to exercise-based rehabilitation is often constrained. Telemedicine-based rehabilitation programs have demonstrated successful outcomes in reducing impediments to treatment access. Designing patient-centered telerehabilitation interfaces and functions hinges on a strong comprehension of how cancer patients view this approach. This study investigated patient viewpoints and experiences with a mobile cancer telerehabilitation system, employing a step-by-step walkthrough for data collection. Semi-structured qualitative interviews with 29 chemotherapy-receiving cancer patients were conducted after the demonstration. Deductive thematic analysis was applied to the interviews, revealing patterns and themes. Glucagon receptor Patients appreciated the telerehabilitation system's convenience and straightforward operation, contributing to its widespread approval. The system presented minimal impediments to adoption for patients who reported low technology literacy. From the thematic analysis results, a thorough comprehension of the needs and preferences of patients concerning the telerehabilitation system's interface and functionality was gleaned. These insightful observations will shape the future creation and deployment of a patient-centered cancer telerehabilitation program.Homelessness in America touched over half a million individuals nightly in 2021, a stark statistic contrasted with the fact that only about 50% of these individuals sought refuge in shelters. We detail the creation of H4H, a large-scale repository of homeless resources, including emergency shelters and housing opportunities. Advanced natural language processing techniques were applied to extract detailed information crucial to the homeless population, including their admission procedures, available services, duration of stays, and eligibility requirements. The method we employ for information extraction is predicated on posing and answering questions. Using a dataset of 2055 question-answer pairs for training and assessment, a two-step classification and question-answering Roberta model with prompting demonstrated the highest performance, resulting in a macro-average F1 score of 7583. The annotated entries of H4H, a public benchmark dataset, are freely available.In response to the escalating stroke death rate observed over the past eight years, clinicians are actively seeking data-supported decision-making tools. Deep-learning models, fine-tuned with electronic health records (EHRs) containing detailed data, have recently shown superior ability in predicting health outcomes. Yet, the utilization of electronic health record-based deep learning models for forecasting outcomes in hemorrhagic stroke patients has not been extensively explored or researched. Employing an ensemble deep learning methodology, this paper aims to forecast early mortality in ICU patients with hemorrhagic stroke. The ensemble model's accuracy, at 83%, surpassed the fusion model and all baseline models, including logistic regression, decision trees, random forests, and XGBoost. Furthermore, SHAP values were employed to interpret the ensemble model and pinpoint crucial features for prediction. This paper, in a significant way, follows the MINIMAR (Minimum Information for Medical Artificial Intelligence Reporting) criterion, representing a substantial step towards establishing trust between artificial intelligence and medical personnel.The fifth iteration of the U.S. national objectives, Healthy People 2030, focuses centrally on health literacy as a key area. People struggling with health literacy frequently face obstacles in understanding medical information, adhering to post-visit instructions, and correctly using medications, leading to poorer health results and significant health inequalities. Employing natural language processing techniques, this study aims to improve health literacy in patient education materials through the automated translation of illiterate languages in each sentence. Patient education materials were harvested from MedlinePlus.gov and three other online health information websites. Drugs.com offers details on medications and their uses. In the vast online landscape, Mayoclinic.org and Reddit.com hold considerable importance. Neural machine translation (NMT) models, representing the pinnacle of current technology, were trained on a silver standard dataset and assessed on a gold standard testing dataset. The experimental data supports a conclusion that BiLSTM NMT models achieved better outcomes compared to the models based on BERT. To assess the performance of NMT models on health-illiterate languages, we compared the frequency of health-illiterate language elements within the sentences.