4:00 PM – 4:45 PM | Merten Hall, Room 1204
Merten Hall, Room 1204
Centering Malleable Factors in Black Women’s Mental Health: How Psychological Armoring and Giving More Connect to Trauma Symptoms
Shane Stori (College of Humanities and Social Sciences)
Prior research centering Black women’s mental health has established a link between gendered racism and psychological distress, with perceived low social support and disengagement coping exacerbating adverse mental health outcomes, but these constructs and relationships among them require elaboration to inform culturally relevant care. The current study aimed to extend the literature by examining the mediating role of psychological armoring (a culturally tailored frame for Acceptance and Commitment Therapy’s psychological inflexibility model) as a coping response in the connection between gendered racism and trauma symptoms from discrimination among a community sample of 187 Black American women. We introduced the concept of disparity of social support roles (e.g., giving more support than receiving) to assess the impact of this factor on perceived low social support and evaluated the moderating role of satisfaction with balance of social support roles in the gendered racism to disengagement coping link. Results revealed that psychological armoring partially mediated the relationship between gendered racism and trauma symptoms from discrimination. While there was no moderating support for disparity of social support roles, giving more support than receiving predicted low social support satisfaction, and decreased satisfaction with balance of social support roles was related to higher psychological armoring, psychological distress, and trauma symptoms from discrimination. Our results expand upon empirical research that connects gendered racism to adverse psychological outcomes and lends support to psychological armoring and disparity of social support roles as malleable mechanisms that can be targeted for wellness promoting interventions.
Solar Flare Prediction using Deep Learning Models
Joao Pereira (College of Science)
Solar flares (and accompanying coronal mass ejections) are known to be causes and drivers of severe space weather near Earth and pose significant risks to technological systems on Earth. Despite significant advancements in AI and machine learning, solar flare prediction still needs to improve due to the elusive nature of their underlying physical mechanisms and the limitations of current predictive tools. To improve solar flare predictability, we compare a novel spatiotemporal explanation supervision (STES) framework for prediction to other machine learning techniques, specifically support vector machines (SVM) and Random Forests (RF), that have been used on a benchmark multivariate time series dataset, known as the Space-Weather ANalytics for Solar Flares (SWAN-SF). By comparing the results between STES and traditional machine learning techniques on a benchmark dataset, we adequately measure the effectiveness of this new image-based prediction technique compared to parameter-based prediction, finding that it reaches prediction scores in agreement with the traditional methods, providing new insights for improving solar flare predictability rates.
Lessons Learned from Developing the Support Group “Parents Helping Parents”: Building Strong Families
Maribel Tohara Nakamatsu (College of Education and Human Development), Rachael Goodman
This research presents insights gained from developing the support group Parents Helping Parents: Building Strong Families (Padres Ayudando a Padres: Construyendo Families Fuertes), designed for Central American immigrant mothers in Alexandria, Virginia. The initiative aimed to enhance parenting skills through interactive activities such as role-playing, group discussions, and case study presentations. Throughout the process, several challenges and successes emerged. Obstacles included logistical constraints, cultural differences in parenting approaches, and initial hesitancy among participants to engage. However, these were balanced by the strengths of community support, the mothers’ willingness to learn, and the effectiveness of hands-on learning strategies. The experience provided valuable lessons on fostering trust, adapting strategies to participants’ needs, and maintaining engagement. These takeaways will be instrumental for those working to support immigrant families or in communities facing similar challenges. By reflecting on these experiences, this research highlights best practices for developing culturally responsive support programs that empower parents and strengthen families.
To what extent do Farsi/Dari-language online webpages for antenatal care comply with WHO guidelines among pregnant women in Afghanistan?
Narges Ghafary (College of Humanities and Social Sciences)
Access to reliable antenatal care information is critical for maternal health, particularly in Afghanistan, where political instability, economic hardship, and restrictive policies severely limit women’s access to healthcare services. With the Taliban’s restrictions on women’s mobility and the shortage of female healthcare providers, many Afghan women turn to online sources for prenatal care guidance. However, the accuracy and comprehensiveness of these resources remain largely unexamined. This study investigates the extent to which Farsi/Dari-language online webpages providing antenatal care information comply with World Health Organization (WHO) guidelines.
Guided by the research question To what extent do Farsi/Dari-language online webpages for antenatal care comply with WHO guidelines for pregnant women in Afghanistan? this study employs a media content analysis of 16 systematically selected webpages. A structured codebook assesses ten prenatal care variables, including dietary recommendations, supplement intake, physical activity, and danger signs during pregnancy. Findings reveal substantial gaps in online health information. While 80% of webpages mention healthy diets, only 13.33% provide comprehensive guidance. Additionally, 62.5% fail to address potential pregnancy complications, and only 20% specify the recommended number of antenatal visits. Critical topics such as salt intake, nausea management, and iron supplementation are often inadequately covered or omitted entirely. These findings indicate that online prenatal care resources in Farsi/Dari often fail to align with WHO guidelines. Evidence-based digital interventions are needed to enhance maternal health literacy and access to accurate care information.
Unpaid Internships: Exploring the Hardships of Social Work Students
Betzy Balladares Oviedo (College of Public Health)
There is little empirical evidence regarding the consequences of unpaid internships, specifically among social work students. Despite unpaid internships bringing professional experience, this practice perpetuates socio-economic disparities and continues a cycle of exploitation. Under the Council of Social Work Education, bachelor’s students must complete a minimum of 400 hours of unpaid internships and master’s students must complete a minimum of 900 hours in a two year program. This study aims to identify and explore specific challenges faced by social work students in their senior year and Masters at George Mason University. Researchers distributed an exploratory survey to social work students through email. The survey included open-ended and closed-ended questions regarding the impact of unpaid internships on their mental health, financial circumstances, and living stability. Responses were systematically organized in Microsoft Excel for data collection and analysis. Researchers conducted a peer review process to examine each other’s work to confirm the fidelity of data collection procedure and to ensure accuracy and reliability. The findings of this study demonstrate that social work students are at risk for mental health problems, emotional distress, and financial instability due to the demands of unpaid internships. This study highlights the intensified challenges encountered by social work students at the Bachelor’s and Master’s level. While this serves as an initial step in identifying the impact of unpaid practicums on students, future research with a representative sample is necessary to fully capture the challenges and experiences of social work students.
Indigenous Languages Spoken in Argentina: A Survey of NLP and Speech Resources
Belu Ticona Oquendo (College of Engineering and Computing)
Argentina has a large yet little-known Indigenous linguistic diversity, encompassing at least 40 different languages. The majority of these languages are at risk of disappearing, resulting in a significant loss of world heritage and cultural knowledge. Currently, unified information on speakers and computational tools is lacking for these languages. In this work, we present a systematization of the Indigenous languages spoken in Argentina, classifying them into seven language families: Mapuche, Tupí-Guaraní, Guaycurú, Quechua, Mataco-Mataguaya, Aymara, and Chon. For each one, we present an estimation of the national Indigenous population size, based on the most recent Argentinian census. We discuss potential reasons why the census questionnaire design may underestimate the actual number of speakers. We also provide a concise survey of computational resources available for these languages, whether or not they were specifically developed for Argentinian varieties.
Machine Learning Models across Cancer Types: Mortality Prediction Analysis
Huan Ju Shih (College of Public Health)
This study evaluates the generalizability and performance of machine learning-based models across two common cancer types (prostate and breast cancer) by testing models trained on one cancer type against test sets from another. The two cancers were selected because of their commonality and presence in different populations (male/female). The analysis was conducted for four different outcomes: one-year and five-year cancer-specific mortality and one-year and five-year all-cause mortality. For each outcome, models trained on prostate cancer were tested on breast cancer datasets, and vice versa, to assess whether predictive patterns hold across different cancer types. Additionally, combined models trained on both prostate and breast cancer were tested to evaluate whether integrating data from both cancer types improves prediction accuracy. These analyses help determine whether models trained on one cancer type can generalize to another and assess the effectiveness of combined training approaches. This cross-application was possible because models did not include any cancer-specific information (i.e., specific biomarkers). Results are reported in terms of model performance (AUC, precision, recall) and visualized using model correlation plots.
Impact of Decision-Making on Cohort Quality in COVID-19 Data Analytics
Atefehsadat Haghighathoseini (College of Public Health)
This study aimed at analyzing decisions affecting cohort quality in COVID-19 data analytics. It highlights the critical role of strategic decisions in cohort selection for secondary data analyses, and their adverse consequences and selection bias. The analysis was done within N3C data Enclave (22 million patients, 8.8 million COVID-19 patients). It started by identifying COVID-19 cases and inpatient hospitalization records – cases from August 1, 2020, to December 31, 2021, to ensure data reliability. A systematic application of four data-related decisions (identifying COVID-19 cases, recognizing inpatient hospitalizations, determining COVID-19-related hospitalizations, and excluding records without specific admission timestamp) resulted in 16 distinct datasets. With additional decisions (provider IDs and location IDs) this increased to 64 different cohorts. Statistical techniques were applied to evaluate the impact of decisions on cohort quality. Visualization methods were used to demonstrate differences in distribution of demographic groups. No decisions applied in data processing were related to patient demographics; yet resulted in different distributions of demographics. The decisions most affecting demographics were Time and Provider ID. Initially, these factors created differences of up to 0.77% for female patients, 1.17% for Black patients, and 5.84% for Hispanic patients. In the enhanced analysis, the impact increased, with differences reaching 2.68% for female patients, 5.15% for Black patients, and 8.21% for Hispanic patients. The study outlined that arbitrary decisions can significantly affect patient distribution and outcomes across demographic categories. This research emphasizes the need for informed, data-driven strategies to capture the complexities of the pandemic’s effects on diverse populations.
The Impact of Stratified Variable Encoding Methods on Prediction Model Performance
Lemba Priscille Ngana (College of Public Health)
Introduction Variable encoding is a critical yet often overlooked step in clinical data pre-processing for traditional statistical analysis and machine learning (ML) models. Researchers often construct variables by summarizing temporal clinical data, such as labs or vitals, using various encoding methods, yet few studies systematically assess their impact on model performance. Despite its significance, encoding choices remain largely subjective, affecting model accuracy and robustness. This study evaluates the influence of encoding techniques on the quality of ML-based models constructed for predicting in-hospital mortality in systolic heart failure patients. Methods MIMIC-IV v3.0 data was used to develop mortality prediction models based on patient demographics and 43 laboratory tests from 1,037 ICU patients within their first 24 hours of admission. Four encoding methods were evaluated: (1) present/not present, (2) average/minimum/maximum, (3) first/last, and (4) a combined approach. Logistic regression, random forest and gradient boosting models were trained and evaluated using standardized performance metrics. Feature importance and stratified analyses identified optimal encoding techniques in sub-populations. Results Encoding methods affect the measured quality of models. Overall, combination encoding performed the best, followed by average/minimum/maximum and first/last, while binary encoding performed the worst. For example, Random Forest achieved AUCs of 0.920 (first/last) and 0.833 (binary). Stratified analyses showed the highest AUCs for the highest average (0.939), highest minimum (0.930) and highest maximum-minimum (0.907) datasets. Conclusions Variable encoding significantly impacts model performance. No single variable encoding method is universally optimal, underscoring the importance of evaluating encoding methods for all ML-based models in clinical applications.
Enhancing Mental Well-Being and Academic Success: Institutional Support for First-Generation Students
Eunjin Han (College of Education and Human Development)
First-generation college students (FGCS) often encounter significant academic and mental health challenges, including financial strain, imposter syndrome, and limited access to institutional support. Unlike their peers, FGCS frequently lack essential resources, such as family guidance and academic insights from their communities, making it more difficult to navigate higher education. This study conducts a systematic literature review to explore effective institutional strategies for supporting FGCS and enhancing their academic success and well-being. Guided by the research questions (1) What institutional resources have been shown to improve academic performance and well-being among FGCS? (2) What gaps exist in current institutional support systems? this study synthesizes findings from peer-reviewed journal articles, policy reports, and empirical studies on higher education support structures. Key themes include mentorship programs, mental health services, financial aid accessibility, and academic advising tailored to student needs. Preliminary findings suggest that comprehensive support systems—particularly those integrating academic and mental health interventions—significantly improve student retention, resilience, and academic success. However, persistent gaps in mental health support and financial accessibility highlight the need for more inclusive institutional strategies. This review underscores the importance of holistic institutional approaches that integrate academic, financial, and psychological support to foster long-term success for FGCS. By analyzing existing research, this study provides evidence-based recommendations for universities to strengthen their support systems and improve student outcomes in higher education.
DNA-based Nanoprobes for Fluorescence K+ Sensing in Neural Systems
Bryce Dunn (College of Engineering and Computing)
This study presents an approach for functional neural imaging and sensing using fluorescent nanosensors designed to detect potassium (K+) flux in living systems. The nanoprobes exhibit a change in fluorescence that is dependent on local K+ concentration to facilitate real-time sensing of ionic gradients crucial for understanding neuronal activity and ion channel dynamics. We demonstrated K+ sensing with our nanoprobes by electrically stimulating murine brain slices to induce ionic flux. This method offers a useful tool for studying ion-dependent processes in real-time, providing new insights into the complex functioning of neural networks.
Power Plant Emission Trend from 1940 to 2020 in the United States
Xiaorong (Sherry) Shan (College of Engineering and Computing), Lucas Henneman
Historically, power plants have been major contributors to poor air quality in the United States and greenhouse gas emission. Power plant emissions have increased and decreased since the early 1900s because of factors such as environmental legislation, electricity demand, and available fuels. The objective of this project is to quantify and characterize the variation in emissions and exposure from power plants in the United States from 1940 to 2020. For periods without direct measurements, we estimate SO₂, NOx, PM, Hg, and CO₂ emissions using emission factors derived from existing US power plants. The emission factors are calculated based on fuel types and facility unit characteristics. We use reduced complexity atmospheric models including the HYSPLIT average dispersion (HyADS) model, to quantify PM2.5 concentrations attributable to power plant emissions. SO₂ emissions, 90% of which originated from coal power plants primarily located in the Midwest and the West, surged by 545% from 1940 to their peak in 1970, followed by a 10% decline from 1970 to 1990. Between 1995 and 2020, SO₂ emissions further plummeted by 93%, with the most significant reductions occurring between 2005 and 2012. Similarly, NOx emissions experienced a dramatic increase of approximately 1117% from 1940 to 1990 but subsequently decreased by 87% from 1995 to 2020, exhibiting a steady decline during the 2005-2012 period. These reductions can be attributed to key regulations aimed at reducing air pollution and costs of fuel. Programs implemented for SO₂ and NOx, utilizing various control technologies and fuel switches significantly decreased emissions.