Electronic health records and patient portals are increasingly utilized to enhance research recruitment efficiency, yet response patterns across patient groups remain unclear. We examined 10 studies at Emory Healthcare that used these tools to identify and recruit 24,000 patients over 1 year. Response rates were lower among males and Black individuals, though study interest was higher among respondents. Interest was also greater among those with frequent healthcare interactions and lower comorbidity. In a large academic health system, portal-based recruitment offered a streamlined approach to research recruitment and patient engagement, with minor variations across patient characteristics warranting continued study.

Objectives/Goals: Kentucky (KY) is a high priority ending the HIV epidemic state, with high rates of new HIV diagnoses tied to injection drug use. The overall goal of this pilot is to launch sentinel surveillance of bloodborne infections and drug compounds among people who inject drugs (PWID) to monitor trends in near-real time and inform rapid community response. Methods/Study Population: In collaboration with the Clark County, KY, syringe service program (SSP), the pilot study involves two 1-month waves of data collection: enrolling eligible SSP participants and conducting anonymous behavioral surveys, collection of participants’ syringes, laboratory testing of syringes to detect HIV and hepatitis C (HCV), drug residue testing through National Institute of Standards and Technology, and statistical modeling approaches to produce outputs of bloodborne infection and drug detection. Syringes are tested from each enrolled individual for: 1) HIV antibody; 2) HCV antibody; 3) HIV and HCV PCR; 4) HIV antigen; and 5) drug residue. Collaboration with community and PWID stakeholders will identify optimal messaging for reporting results. Results/Anticipated Results: The first wave community-facing pilot was conducted in September–October 2024. 29 survey responses were obtained; median age of the sample is 42 years, 55.2% are gender female; 37.9% reported unstable housing in the past week. Primary drugs of injection reported via survey in the prior month were methamphetamine (62.1%), heroin (13.8%), fentanyl (13.8%), buprenorphine (10.3%), meth and fentanyl in combination (3.4%). PWID reported returning 900 used syringes and a median of 15 per participant visit. At most recent testing, 69.0% reported a positive HCV test; 0% reported a positive HIV test. Some level of drug checking with fentanyl test strips in past month was reported by 51.7%. Initially, 20 syringes were tested for drug compounds; results are pending. HIV and HCV detection testing will be completed by early 2025. Discussion/Significance of Impact: Early results document proof of concept for our sentinel surveillance study; all individuals screened were willing to participate in surveys and syringe collection. New methods to identify risk for disease outbreaks and emerging drugs can inform rapid allocation of prevention resources at a community level, especially where testing is infrequent.

Objectives/Goals: Retention in care is vital for people living with HIV. We used human-centered design (HCD) to engage a community-based research panel over a 5-year period, allowing us to incorporate their insights on research guidance and interpretation of findings to investigate correlates of HIV care outcomes. Methods/Study Population: We recruited a diverse panel of individuals who were living with HIV, HIV clinicians, and/or providing non-clinical HIV services in Marion County, Indiana. We conducted biannual sessions using a variety of HCD tools and activities to engage participants. Each session took about three hours, and panelists were compensated for their participation. Due to the COVID-19 pandemic, sessions were initially held virtually. Sessions were designed for project discussion and to facilitate exploration of concerns and challenges facing receipt of HIV services. Our HCD approach put participants in the center of discussion and empowered them to externalize ideas and collaborate meaningfully with our team. Results/Anticipated Results: Since project inception, 48 individuals have joined the panel. Thirty-five are actively engaged, participating in one or more of six sessions conducted to date. We have learned much from the panel. One example is that a residential move might be a risk or protective factor for retention in care and the amount of time one had lived with HIV is a crucial factor. Panel insights have helped guide and prioritize analyses, aided in identification of data missing from our ecosystem, helped interpret results, provided feedback on future interventions, led to a quality improvement project with the local health department, and led to a presentation at a local health equity conference. Discussion/Significance of Impact: Community engagement is essential to impactful and sustainable research. HCD was a successful approach to engage our panel to inform interventions more relevant to the community. We anticipate these methods will be important for others conducting community-engaged research.

A clinical and translational scientist (CTS) often seeks to increase their knowledge of statistical topics to effectively conduct biomedical research studies. A common method for obtaining this knowledge is through existing online educational materials that are suggested by a biostatistical collaborator or identified by the CTS. However, the volume of available educational materials on diverse statistical topics makes the task of identifying high-quality educational resources at an appropriate level challenging and time consuming for CTSs and collaborative biostatisticians. In response to these challenges, the Biostats4You website was created, where existing online educational materials for a variety of statistical topics are vetted to identify those most appropriate for CTSs. In this manuscript, we describe the resource review process, provide information about statistical topics and resources currently available, and make recommendations for how CTSs and collaborative biostatisticians can utilize the Biostats4You website to improve training, mentoring, and collaborative research practices.

While clinical research intends to improve health outcomes for all, access to research participation is often limited and inequitable. Geographic proximity is a recognized barrier, thus, systemic infrastructure solutions through federal programs including General Clinical Research Centers and Clinical and Translational Science Awards have sought to improve accessibility. Even with such support, academic medical centers often have limited clinical research-dedicated space apart from shared exam rooms in difficult-to-navigate hospitals or clinics. In 2019, the Duke University School of Medicine looked beyond its medical center campus to identify free-standing sites within Durham communities for participant study visits. Catalyzed by the COVID-19 pandemic, Duke Research at Pickett, a 22 000-square-foot building with a laboratory, 30 exam rooms, and on-site parking, opened in October 2020 to support vaccine and treatment trials. Upon the lifting of many COVID-19 restrictions, and in partnership with the Research Equity and Diversity Initiative (READI) Community Advisory Council, the building was transformed to encourage community gatherings, education, and training programs. To date, Duke Research at Pickett has hosted 2692 participants in 78 research trials and 14 community-engaged activities.

Objectives/Goals: To create, train, and evaluate the FAST-PACE (Promoting Academic and Community Engagement) Toolkit that catalyzes academic-community translation science teams during a public health emergency. The toolkit is a road map based on the Research Readiness and Partnership Protocol (R2P2), which was developed from the Flint Water Crisis. Methods/Study Population: A literature review was conducted by the Michigan Institute for Clinical & Health Research Community Engagement (MICHR CE) program and the Community-Based Organization Partners (CBOP), to identify important and common elements in disaster response protocols with a set of key interviews (n = 31) to glean perspectives from community leaders. Key findings were extracted and reviewed to generate guidelines and recommendations for the R2P2 protocol. The co-developed FAST-PACE Toolkit launched its expansion statewide to address emergencies and health disparities of communities in crisis. The iterative process consisted of community report-outs, gathering input from stakeholders, via discussion, and evaluation surveys. The feedback was used to develop, enhance, and tailor the toolkit and training content. Results/Anticipated Results: Data from training (n = 8) of the critical elements of the FAST-PACE Toolkit, which provides guidance for academic and community team members that includes 1) assessing community assets and needs; 2) engaging in clear and bidirectional communication; 3) facilitating transparency and equitable partnering; 4) identifying health equity and justice issues; and 5) conducting the evaluation of research. The training will be disseminated in-person and virtually across the state of Michigan resulting in participants sharing community-identified health issues and social determinants of health to assist MICHR CE to suggest resources to address health impacts. Discussion/Significance of Impact: The FAST-PACE Toolkit borne from the flint water crisis and confounded by other crises used CEnR principles to create a translation science roadmap. It equips communities and collaborating academic institutions across the state to respond to public health crises and fosters equitable translation science partnerships built on respect and trust.

Objectives/Goals: The goal of this proposal is to better understand how informing African Americans of their genetic risk affects their behavior as part of a cardiovascular disease (CVD) risk reduction intervention. Aim 1: To determine the effect of genetic risk knowledge on CVD health behavior. Aim 2: To determine the effect of genetic risk knowledge on secondary variables. Methods/Study Population: Method: Fifty participants from the Baton Rouge metropolitan area will be recruited. Participants must be African American adults over the age of 18. Potential participants will be recruited using community-based efforts that have been successful in recruiting this population specifically. Participants will be randomized into one of two groups. Genetically Unblinded Group (GU) will be “genetically unblinded” after baseline orientation. Genetically Blinded Group (GB) will be “genetically blinded” until the end of the study. This study design ensures that we can measure the impact of knowledge of genetic risk on participant behavior. Results/Anticipated Results: Baseline participants’ characteristics (body mass index, blood glucose, and cholesterol) will be summarized by intervention group, with counts and percentages for categorical variables and means and 95% confidence intervals for continuous variables. Primary Outcomes: Attendance in intervention sessions will be counted across groups. Effect on genetic risk knowledge will be determined via comparing the difference between the increased healthy lifestyle behaviors at endpoint between Genetically Unblinded (Cases) and Genetically Blinded Groups (Control). Secondary and Tertiary Outcomes: Mean change in secondary outcomes in the GU group will be compared against the mean change in the GB group. Participant’s survey responses and changes in physical measurement from baseline to endpoint will be observed. Discussion/Significance of Impact: This study empowers African Americans in Baton Rouge by providing genetic risk knowledge for cardiovascular disease. By addressing social determinants of health, it promotes behavior change, improves health outcomes, and fosters trust, potentially reducing health disparities and advancing health equity.

Objectives/Goals: This study evaluates the role of visual machine learning algorithms (VMLA) in automating a predictive model of central sarcopenia in geriatric trauma patients based on the psoas:lumbar vertebral index (PLVI) and trauma-specific frailty index (TSFI). Methods/Study Population: 150 trauma patients seen at Jon Michael Moore Trauma Center within J.W Ruby Memorial Hospital in rural West Virginia were included in this investigation across the life spectrum. The VMLA was trained on their standard of care trauma panoramic CT scans. Five expert reviewers segmented bilateral psoas muscles and the L4 vertebrae of each CT image at one slice inferior to the posterior elements of the L4 vertebrae. The data were read into a U-net convoluted neural network as ground truth. Labels were preprocessed to focus on the regions of interest and standardized into four classes: right psoas, left psoas, L4 vertebrae, and background. Performance was evaluated using accuracy, Dice coefficient, and F1 score. Results/Anticipated Results: Between our expert reviewer segmentations, we had significant inter-reader reliability with a Kappa greater than 0.8 and a mean standard deviation of the PLVI of 0.10mm^2. Preliminary VMLA testing on a subset of 70 patients yielded a validation accuracy of 88.5%, a Dice coefficient of 0.86, and an F1 score of 0.87 after 20 epochs. There was a moderate interclass correlation between PLVI and TSFI even though the TSFI lacks sensitivity. In fact, the PLVI is a more accurate predictor of frailty in trauma patients based on various outcome measures such as corrected length of stay. Our ongoing efforts are centered around improving the VMLA. Discussion/Significance of Impact: Our VMLA outperforms the current clinical standard, TSFI. Integration of our VMLA into the clinical workflow has the potential to revolutionize geriatric trauma care by providing rapid, accurate, identification of high-risk frail patients.

The consent process for research studies can be burdensome for potential participants due to complex information and lengthy consent forms. This pragmatic study aimed to improve the consent experience and evaluate its impact on participant decision making, study knowledge, and satisfaction with the In Our DNA SC program, a population-based genomic screening initiative. We compared two consent procedures: standard consent (SC) involving a PDF document and enhanced consent (EC) incorporating a pictograph and true or false questions. Decision-making control, study knowledge, satisfaction, and time to consent were assessed. We analyzed data for 109 individuals who completed the SC and 96 who completed the EC. Results indicated strong decision-making control and high levels of knowledge and satisfaction in both groups. While no significant differences were found between the two groups, the EC experience took longer for participants to complete. Future modifications include incorporating video modules and launching a Spanish version of the consent experience. Overall, this study contributes to the growing literature on consent improvements and highlights the need to assess salient components and explore participant preferences for receiving consent information.

Objectives/Goals: In recent years, there has been growing interest in the development of air pollution prediction models, particularly in low- and middle-income countries that are disproportionately impacted by the effects of air pollution. Recent methodological advancements, particularly in machine learning, provide novel opportunities for modeling efforts. Methods/Study Population: We estimate daily ground-level fine particulate matter (PM2.5) concentrations in the Mexico City Metropolitan Area at 1-km2 grids from 2005 to 2023 using a multistage approach. Spatial and temporal predictor variables include data from the moderate resolution imaging spectroradiometer (MODIS), Copernicus Atmosphere Monitoring Service (CAMS), and additional meteorological and land use variables. We employed machine-learning-based approaches (random forest and gradient boosting algorithms) to downscale satellite measurements and incorporate local sources, then utilized a generalized additive model (GAM) to geographically weight predictions from the initial models. Model performance was evaluated using 10-fold cross-validation. Results/Anticipated Results: On average, the random forest, gradient boosting, and GAM models explained 75, 82, and 83% of variations measured in PM2.5 concentrations. PM2.5 levels were generally higher in densely populated urban centers and lower in suburban and rural areas. Important predictors of ground-level PM2.5 included wind (both u and v components), 2-meter mean air temperature, elevation, and the normalized difference vegetation index (NDVI). Discussion/Significance of Impact: Using novel machine learning-based approaches, we developed robust models with fine-scale spatial (1-km2) and temporal (daily) variations of PM2.5 in Mexico City from 2005 to 2023. The predicted PM2.5 concentrations can further advance public health research on air pollution in Mexico City and beyond.

Objectives/Goals: Idiopathic inflammatory myopathies (IIMs) are autoimmune diseases influenced by genetic and environmental factors. This study aims to explore infection patterns preceding IIM onset by applying temporal data mining and machine learning to deidentified patient records and corroborate results from molecular analysis. Methods/Study Population: The dataset used in this work was extracted from TriNetX with a focus on patients who have IIM. Risks for developing the outcomes were assessed using case–control cohorts. For each participant, information was extracted about diagnosis code, date of infection, and study visit in which the infection was reported. This data were then temporally encoded and used to generate sequence files for each of the outcomes. Unsupervised temporal machine learning was then preformed on these files to detect frequent subsequences of infections. Python library scikit-learn was used to perform the unsupervised machine learning with k-means clustering. Results/Anticipated Results: The results of this study identify infections associated with the onset of IIM by analyzing temporal infection patterns. Frequent sequences of infections uncovered, with specific patterns linked to different cohorts, offer insights into the etiology of IIM. Common and cohort-specific infection sequences will help validate existing research and provide new avenues for exploring the disease mechanisms. The findings will highlight significant infection patterns, which will inform our understanding of IIM onset across various patient populations. Discussion/Significance of Impact: The results will provide key insights into pre-symptomatic infection sequences related to IIM onset, enhancing understanding of its etiology and pathogenesis. These findings may aid in developing more precise screening methods for early detection and confirm previous results from analyzing immune signatures of infections in IIM.

Objectives/Goals: The objective of this study is to explore strategies for AI-physician collaboration in diagnosing acute respiratory distress syndrome (ARDS) using chest X-rays. By comparing the diagnostic accuracy of different AI deployment methods, the study aims to identify optimal strategies that leverage both AI and physician expertise to improve outcomes. Methods/Study Population: The study analyzed 414 frontal chest X-rays from 115 patients hospitalized between August 15 and October 2, 2017, at the University of Michigan. Each X-ray was reviewed by six physicians for ARDS presence and diagnostic confidence. We developed a deep learning AI model for detecting ARDS and explored the strengths, weaknesses, and blind spots of both physicians and AI systems to inform optimal system deployment. We then investigated several AI-physician collaboration strategies, including: 1) AI-aided physician: physicians interpret chest X-rays first and defer to the AI model if uncertain, 2) physician-aided AI: the AI model interprets chest X-rays first and defers to a physician if uncertain, and 3) AI model and physician interpreting chest X-rays separately and then averaging their interpretations. Results/Anticipated Results: While the AI model (84.7% accuracy) had higher accuracy than physicians (80.8%), we found evidence that AI and physician expertise are complementary. When physicians lacked confidence in a chest X-ray’s interpretation, the AI model had higher accuracy. Conversely, in cases of AI uncertainty, physicians were more accurate. The AI excelled with easier cases, while physicians were better with difficult cases, defined as those where at least two physicians disagreed with the majority label. Collaboration strategies tested include AI-aided physician (82.4%), physician-aided AI (86.9%), and averaging interpretations (86%). The physician-aided AI approach had the highest accuracy, could off-load the human expert workload on the reading of up to 79% chest X-rays, allowing physicians to focus on challenging cases. Discussion/Significance of Impact: This study shows AI and physicians complement each other in ARDS diagnosis, improving accuracy when combined. A physician-aided AI strategy, where the AI defers to physicians when uncertain, proved most effective. Implementing AI-physician collaborations in clinical settings could enhance ARDS care, especially in low-resource environments.

Objectives/Goals: This study aimed to enhance clinical trial data management through large language model information retrieval and generation techniques within the clinical trial reporting workflow. We focused on improving compliance with reporting, reducing human labor, and promoting standardized reporting structure and data quality oversight. Methods/Study Population: We used approved study protocols from UC Davis IRB-approved investigator-initiated studies compared to the same studies reported to ClinicalTrials.gov. Our baseline data extraction system employs commercial large language models (LLMs) and retrieval augmented generation (RAG) to isolate data sources within the secure extraction environment. We stratified protocol documents into easy, complex, and random categories based on study focus, document complexity, the extent of amendments or modifications, and completion metrics from ClinicalTrials.gov. We developed a pilot web-based architecture to capture variations in categorization, labeling, and reporting style and compared generated extraction data. We primarily focused on qualitative evaluation through a review of expert staff. Results/Anticipated Results: Our results revealed significant variations in reporting quality, with dependencies stemming from multiple authors and stages throughout the clinical trial protocol lifecycle. Based on these variations, we used prompt engineering to improve the pilot application’s output compliance with the protocol registration and results system (PRS) structured data format for various study types. We piloted the assisted workflow with prospective studies by partnering with study investigators and the clinical trial office staff to assist in review and clinical trial reporting creation. Initial studies reported by our system were approved and released to the public by PRS staff. We are refining content generation and workflows to different components of studies and evaluating their use in quality and training areas. Discussion/Significance of Impact: Our system fosters collaboration, efficient review, and compliance with clinical trial reporting standards. It supports the promise of AI-driven assistance in clinical trial management, design, and reporting. We focus on the multiple stakeholders, expertise, and data flows in the organizational management of clinical and translational science.

Clinical research professionals (CRPs) are integral to the academic medical center workforce, research operations, and daily clinical research tasks; however, due to inconsistent training, there is a shortage of qualified CRPs. The Joint Task Force for Clinical Trial Competency created a competency framework for CRPs, which has demonstrated positive results from various institutions, but training programs have been limited in standardization, replicability, and dissemination. To improve this, we designed the University of Texas Southwestern (UTSW) Medical Center Clinical Research Foundations (CRF) training program, which is a competency-based online self-paced CRP training curriculum hosted via the Collaborative Institutional Training Initiative (CITI) portal. We examined feasibility, acceptability, and uptake of the UTSW CRF training on an institutional scale and were pleased to find this curriculum is not only feasible but has high levels of acceptability. Furthermore, faculty, clinicians, and trainees voluntarily completed this training program indicating utility across diverse groups. The UTSW CRF combines the existing CITI training modules with UTSW-created material, providing an optimal balance between generalized clinical research education and institutionally tailored content. We believe the UTSW CRF curriculum could serve as a plug-and-play foundational model for other research centers to tailor according to their audience and institutional needs.

Objectives/Goals: Magnetic resonance imaging (MRI) reports are stored as unstructured text in the electronic health record (EHR), rendering the data inaccessible. Large language models (LLM) are a new tool for analyzing and generating unstructured text. We aimed to evaluate how well an LLM extracts data from MRI reports compared to manually abstracted data. Methods/Study Population: The University of California, San Francisco has deployed a HIPAA-compliant internal LLM tool utilizing GPT-4 technology and approved for PHI use. We developed a detailed prompt instructing the LLM to extract data elements from prostate MRI reports and to output the results in a structured, computer-readable format. A data pipeline was built using the OpenAI Application Programming Interface (API) to automatically extract distinct data elements from the MRI report that are important in prostate cancer care. Each prompt was executed five times and data were compared with the modal responses to determine variability of responses. Accuracy was also assessed. Results/Anticipated Results: Across 424 prostate MRI reports, GPT-4 response accuracy was consistently above 95% for most parameters. Individual field accuracies were 98.3% (96.3–99.3%) for PSA density, 97.4% (95.4–98.7%) for extracapsular extension, 98.1% (96.3–99.2%) for TNM Stage, had an overall median of 98.1% (96.3–99.2%), a mean of 97.2% (95.2–98.3%), and a range of 99.8% (98.7–100.0%) to 87.7% (84.2–90.7%). Response variability over five repeated runs ranged from 0.14% to 3.61%, differed based on the data element extracted (p Discussion/Significance of Impact: GPT-4 was highly accurate in extracting data points from prostate cancer MRI reports with low upfront programming requirements. This represents an effective tool to expedite medical data extraction for clinical and research use cases.