ABSTRACT IMPACT: Understanding the longitudinal glucose changes following SARS-CoV-2 infection can inform point-of-care guidelines and elucidate the viral hypothesis of diabetes mellitus pathogenesis. OBJECTIVES/GOALS: Hyperglycemia has emerged as an important manifestation of SARS-CoV-2 infection in both diabetic and non-diabetic patients. Whether clinically-detectable glycemic changes persist following SARS-CoV-2 infection remain to be elucidated. This work aims to characterize temporal patterns in glucose dysregulation following SARS-CoV-2 infection. METHODS/STUDY POPULATION: Electronic health records of patients with a diagnosis of COVID-19, positive laboratory test for SARS-CoV-2, and negative history of Diabetes Mellitus prior to infection were extracted from the TriNetX database. 7,502 patients with at least one blood glucose value 2 years to 2 weeks before, 2 weeks before to 2 weeks after, and 2 weeks after to 1 year after COVID-19 diagnosis were used for analysis. Temporal patterns are characterized by training state-of-the-art clustering algorithms, including fuzzy short time-series clustering, k-means for longitudinal data, and spectral clustering. Clustering performance is evaluated using internal evaluation metrics of the Silhouette coefficient, Calinski-Harabasz score, and Davies Bouldin index. RESULTS/ANTICIPATED RESULTS: Based on the success of prior clustering methods with random blood glucose measurements, we anticipate that the proposed time-series clustering algorithms will appropriately characterize temporal patterns of glycemic dysregulation. The best performing algorithm based on interval evaluation metrics will be selected for further analysis. Associations between blood glucose values and cluster membership will be evaluated using Kruskal-Wallis one-way ANOVA and effect size will be calculated using unbiased Cohen’s d. Clinical phenotypes for each cluster will be characterized in terms of current diagnoses, prior medication use, pertinent laboratory tests, and vital signs. DISCUSSION/SIGNIFICANCE OF FINDINGS: A clearer understanding of the longitudinal glucose changes following SARS-CoV-2 infection can elucidate clinically-detectable patterns of glycemic dysregulation, identify sub-phenotypes of patients who are more susceptive to glycemic dysregulation, and inform appropriate point-of-care guidelines.

ABSTRACT IMPACT: Measuring and analyzing qualitative and quantitative traits using phenomics approaches will yield previously unrecognized heart failure subphenotypes and has the potential to improve our knowledge of heart failure pathophysiology, identify novel biomarkers of disease, and guide the development of targeted therapeutics for heart failure. OBJECTIVES/GOALS: Current classification schemes fail to capture the broader pathophysiologic heterogeneity in heart failure. Phenomics offers a newer unbiased approach to identify subtypes of complex disease syndromes, like heart failure. The goal of this research is to use data-driven associations to redefine the classification of the heart failure syndrome. METHODS/STUDY POPULATION: We will identify < 10 subphenotypes of patients with heart failure using unsupervised machine learning approaches for dense multidimensional quantitative (i.e. demographics, comorbid conditions, physiologic measurements, clinical laboratory, imaging, and medication variables; disease diagnosis, procedure, and billing codes) and qualitative data extracted from an integrated health system electronic health record. The heart failure subphenotypes we identify from the integrated health system electronic health record will be replicated in other heart failure population datasets using unsupervised learning approaches. We will explore the potential to establish associations between identified subphenotypes and clinical outcomes (e.g. all-cause mortality, cardiovascular mortality). RESULTS/ANTICIPATED RESULTS: We expect to identify < 10 mutually exclusive phenogroups of patients with heart failure that have differential risk profiles and clinical trajectories. DISCUSSION/SIGNIFICANCE OF FINDINGS: We will attempt to derive and validate a data-driven unbiased approach to the categorization of novel phenogroups in heart failure. This has the potential to improve our knowledge of heart failure pathophysiology, identify novel biomarkers of disease, and guide the development of targeted therapeutics for heart failure.

ABSTRACT IMPACT: This is the first study to use QUINT analyses to examine heterogeneity of treatment effect for group medical visits among individuals with type 2 diabetes. QUINT is a data driven method that assumes no a priori assumptions regarding effect moderators - an important step in the path towards personalized medicine. OBJECTIVES/GOALS: To examine heterogeneity of treatment effect (HTE) in Jump Start, a trial that compared the effectiveness of group medical visits (GMVs) focused on medication management only versus the addition of intensive weight management (WM) on glycemic control for patients with type 2 diabetes and body mass index >=27kg/m^2. METHODS/STUDY POPULATION: Jump Start patients (n=263) were randomized to a GMV-based medication management plus low carbohydrate diet-focused WM program (WM/GMV; n = 127) or GMV-based medication management only (GMV; n = 136) for diabetes control. We used QUalitative INteraction Trees (QUINT), a tree-based clustering method, to determine if there were subgroups of patients who derived greater benefit from either WM/GMV or GMV. Subgroup predictors included 32 baseline demographic, clinical, and psychosocial factors. Outcome was hemoglobin A1c (HbA1c). We conducted internal validation via bootstrap resampling to estimate bias in the range of mean outcome differences among arms. RESULTS/ANTICIPATED RESULTS: QUINT analyses indicated that for patients who had not previously attempted weight loss, WM/GMV resulted in better glycemic control than GMV alone (mean difference in HbA1c improvement = 1.48%). For patients who had previously attempted weight loss and had lower cholesterol and blood urea nitrogen levels, GMV alone was better than WM/GMV (mean difference in HbA1c improvement = 1.51%). Internal validation resulted in moderate corrections in the mean HbA1c differences between arms; however, differences remained in the clinically significant range. DISCUSSION/SIGNIFICANCE OF FINDINGS: Among patients with diabetes and BMI>=27kg/m^2, a low-carbohydrate, weight loss focus may better improve HbA1c in those who have never attempted weight loss. A medication management focus may be better in those who have attempted weight loss and have lower cholesterol and blood urea nitrogen.

ABSTRACT IMPACT: Our findings could potentially identify CVD at-risk persons living with HIV who might benefit from aggressive risk-reduction. OBJECTIVES/GOALS: PWH have higher rates of CVD than the general population yet CVD risk prediction models rely on traditional risk factors and fail to capture the heterogeneity of CVD risk in PWH. Here we identify protein biomarkers that are able to discriminate between CVD cases and controls in PWH, and we assess their added benefit beyond traditional risk factors. METHODS/STUDY POPULATION: We analyzed 459 baseline protein expression levels from five OLINK panels in a matched CVD (MI, coronary revascularization, stroke, CVD death) case-control study with 390 PWH from INSIGHT trials (131 cases, 259 controls). We formed 200 datasets via bootstrap. For each bootstrap set, a two-component partial least squares discriminant model (PLSDA) was fit. The importance of each variable in the discrimination of cases and controls in the PLSDA projection was assessed by the variable importance in projection (VIP) score. Proteins with average VIP scores > 1 were used in penalized logistic regression models with elastic net penalty, and proteins were ranked based on the number of times the protein had a nonzero coefficient. Proteins in the top 25th percentile were considered to have high discrimination. RESULTS/ANTICIPATED RESULTS: Participants had mean age 47 years, 13% were females, 4.9% had CVD at baseline and 69% were on ART at baseline. Eight proteins including the hepatocyte growth factor and interleukin-6 were identified as able to distinguish between CVD cases and controls within PWH. A protein score (PS) of the top-ranked proteins was developed using the bootstrap (for weights) and the entire data. The PS was found to be predictive of CVD independent of established CVD and HIV factors (Odds ratio: 2.17 CI: 1.58-2.99). A model with the PS and traditional risk factors had a 5.9% improvement in AUC over the baseline model (AUC=0.731 vs 0.69), which is an increase in model predictive power of 18%. Individuals with a PS above the median score were 3.1 (CI: 1.83- 5.41) times more likely to develop CVD than those with a protein score below the median score. DISCUSSION/SIGNIFICANCE OF FINDINGS: A protein score developed improved discrimination of PWH with CVD and those without, and helped identify PWH with high risk for developing CVD. If validated, this score and/or the individual proteins could be used in addition with established factors to identify CVD at-risk individuals who might benefit from aggressive risk-reduction.

ABSTRACT IMPACT: The described framework will enable other sites with a well-defined apparatus for enabling the secondary analysis of EHR data for research through education, team science, and resource consolidation. OBJECTIVES/GOALS: EHR’s potential to improve healthcare outcomes extends far beyond the clinic. This vast repository of clinical insights has dramatic potential for biomedical research. To enhance accessibility for busy clinicians and underserved populations, we describe a framework for interfacing with EHR locally and through national network participation. METHODS/STUDY POPULATION: The Institutional Development Award (IDeA) program, which began in 1993, broadens NIH funding’s geographic distribution for biomedical research. Included in this is the IDeA Networks for Clinical and Translational Research, which focuses on enhancing clinical and translational science across a network of IDeA-states with traditionally underserved communities and rural providers. A prior survey of the needs and capabilities of IDeA-CTR centers identified the need for improved research support. Based on our annual member survey we developed a process for supporting distributed research projects across the GP-CTR. NIH also recently made a funding announcement for the IDeA-CTR community identifying EHR research as a major priority in responding to the COVID-19 pandemic. RESULTS/ANTICIPATED RESULTS: Results from site interviews and member surveys show a clear need for dedicated resources to navigate the process of EHR-derived research. Most described a different set of requirements for increasing accessibility to EHR for research and a strong desire to participate in research networks. Local investigators cited a lack of tools, educational materials, and accessibility. Initial efforts demonstrate strong research questions but limited technical, statistical, and terminological capabilities to succeed. In response, a pipeline for team science and promotion of projects from local phenotypes to national studies. We created a facilitator training program to expand the number of facilitators (n=22), quarterly training for investigators (n=104), and ongoing efforts to advance COVID-19 research. DISCUSSION/SIGNIFICANCE OF FINDINGS: As evidenced in the expanding number of EHR-based research networks there is a need for a system to promote project development and best practices. The proposed model promotes education, resource sharing, and team formation to advance clinical questions from the idea stage toward national research network participation.

Novel endometrial cancer (EC) early-detection approaches may reduce racial disparities in mortality. We conducted six community-based focus groups with White and Black women (N = 57 participants) in February-March 2020 to explore acceptability of a home-based tampon sampling approach for EC. Participants also completed a survey. Data were analyzed using qualitative content analysis. Awareness of EC and risk factors was low. Acceptability regarding home sampling was high, but participants expressed concerns about instruction complexity and potential risks. Black women reported lower comfort with tampons. Increasing EC awareness, self-efficacy, and familiarization with tampons would advance prospects for at-home sample collection for EC testing.

ABSTRACT IMPACT: Laying the groundwork for better predictive algorithms to inform clinical decisions and planning. OBJECTIVES/GOALS: Frailty scores predict poor patient outcomes. Validated against highly relevant outcomes, such scores can be used to inform clinical and resource utilization decisions. We generated and validated an electronic Frailty Index (EFI) from real-world EHR data using the Rockwood deficit-accumulation framework to predict patient safety events. METHODS/STUDY POPULATION: To assure that the research approach reflected perspectives of multiple stakeholders, our multidisciplinary group included an implementation scientist, a geriatrician, an internist, and an informatician. From our large academic health center, we accessed EHR data for 14,844 patients randomly sampled from the data warehouse underlying our ACT/SHRINE node. The per-visit EFI scores were calculated using EHR codes in a rolling 2-year time window. EFI was used as the predictor variable in the analytic design. The primary outcomes were preventable patient-safety events derived from ICD-10 codes including hospital-acquired infections, non-operative hospital-acquired trauma, and cardiac complications. Cox proportional hazard models were used to estimate risk for each outcome. RESULTS/ANTICIPATED RESULTS: We found statistically significant associations of EFI with clinically meaningful outcomes from EHR data. For most outcomes, we found significant correlation with EFI and c-statistics indicating good calibration of the models. The EFI was a strong predictor of clinically relevant outcomes without relying on any data other than diagnoses, vital signs, and laboratory results from the EHR. In contrast to previous studies, we treated EFI as a time-varying predictor with multiple follow-ups per patient, which is more realistic than relying on one static time-point. We used a representative sample of the adult patient population rather than limiting it to older individuals and found EFI to be a useful metric even at relatively young ages. DISCUSSION/SIGNIFICANCE OF FINDINGS: The EFI predicted safety events in adult patients using only routine, structured EHR data and can offer a low-effort, scalable method of risk assessment, valuable to clinical decisions. The capability to harness EHR data and rapidly generate clinical knowledge can be transformative for complex care and contributes to Learning Health Systems.

ABSTRACT IMPACT: This review will encourage further development of novel uses of REDCap for the benefit of the research community. OBJECTIVES/GOALS: REDCap is a clinical research data collection platform that is primarily used as intended. However, little is known about its more novel uses, specifically in clinical decision support in patient care and in clinical research management. Thus, the purpose of this review is to examine peer reviewed literature identifying and describing such novel uses. METHODS/STUDY POPULATION: A systematic search was conducted in both PubMed and Google Scholar using the equation ((REDCap) OR ('Research Electronic Data Capture')) AND ((Clinical Trial Management) OR (Clinical Research)).’ Articles were screened by title, then abstract, and then were reviewed in full if they met inclusion criteria. Articles were included if they had potential relevance to the topic of REDCap or if they mentioned activities related to fields of clinical and translational science including operational support in areas such as clinical research management. Articles were excluded if they focused on common clinical research activities relating to data collection software such as survey administration, database building or data collection for clinical trials, registries, and cohort studies. RESULTS/ANTICIPATED RESULTS: The initial search yielded 390 results, of which 40 underwent an abstract review; only 8 of these underwent full text review. Of these, 5 discussed uses of REDCap in the context of operational support in clinical research management; 3 were related to clinical decision support in patient care. For the 5 articles focused on operational support in clinical research management, topics include e-consenting procedures, collection and storage of protected health information (PHI), patient recruitment and tracking stakeholder engagement. The 3 articles about clinical decision support discuss REDCap tools for generating risk predictions for post-surgical clinical outcomes, generating recommendations and STI test orders, and increasing efficiency in hand-offs to enhance care of surgical oncology patients. DISCUSSION/SIGNIFICANCE OF FINDINGS: Considering that only a small percentage of peer reviewed research reports out on novel uses of REDCap, there is a need for the REDCap consortium to do further work to fulfill its mission to adopt, innovate, and suggest novel uses of REDCap, thus expanding the understanding of its functionalities and therefore its utility in the research community.

ABSTRACT IMPACT: A machine learning approach using electronic health records can combine descriptive, population-level factors of pressure injury outcomes. OBJECTIVES/GOALS: Pressure injuries cause 60,000 deaths and cost $26 billion annually in the US, but prevention is laborious. We used clinical data to develop a machine learning algorithm for predicting pressure injury risk and prescribe the timing of intervention to help clinicians balance competing priorities. METHODS/STUDY POPULATION: We obtained 94,745 electronic health records with 7,000 predictors to calibrate a predictive algorithm of pressure injury risk. Machine learning was used to mine features predicting changes in pressure injury risk; random forests outperformed neural networks, boosting and bagging in feature selection. These features were fit to multilevel ordered logistic regression to create an algorithm that generated empirical Bayes estimates informing a decision-rule for follow-up based on individual risk trajectories over time. We used cross-validation to verify predictive validity, and constrained optimization to select a best-fit algorithm that reduced the time required to trigger patient follow-up. RESULTS/ANTICIPATED RESULTS: The algorithm significantly improved prediction of pressure injury risk (p<0.001) with an area under the ROC curve of 0.60 compared to the Braden Scale, a traditional clinician instrument of pressure injury risk. At a specificity of 0.50, the model achieved a sensitivity of 0.63 within 2.5 patient-days. Machine learning identified categorical increases in risk when patients were prescribed vasopressors (OR=16.4, p<0.001), beta-blockers (OR=4.8, p<0.001), erythropoietin stimulating agents (OR=3.0, p<0.001), or were ordered a urinalysis screen (OR=9.1, p<0.001), lipid panel (OR=5.7, p<0.001) or pre-albumin panel (OR=2.0, p<0.001). DISCUSSION/SIGNIFICANCE OF FINDINGS: This algorithm could help hospitals conserve resources within a critical period of patient vulnerability for pressure injury not reimbursed by Medicare. Savings generated by this approach could justify investment in machine learning to develop electronic warning systems for many iatrogenic injuries.

Homo sapiens is currently living in serious disharmony with the rest of the natural world. For our species to survive, and for our well-being, we must gather knowledge from multiple perspectives and actively engage in studies of planetary health. The enormous diversity of species, one of the most striking aspects of life on our planet, provides a source of solutions that have been developed through evolution by natural selection by animals living in extreme environments. The food system is central to finding solutions; our current global eating patterns have a negative impact on human health, driven climate change and loss of biodiversity. We propose that the use of solutions derived from nature, an approach termed biomimetics, could mitigate the effects of a changing climate on planetary health as well as human health. For example, activation of the transcription factor Nrf2 may play a role in protecting animals living in extreme environments, or animals exposed to heat stress, pollution and pesticides. In order to meet these challenges, we call for the creation of novel interdisciplinary planetary health research teams.

ABSTRACT IMPACT: E-values can help quantify the amount of unmeasured confounded necessary to fully explain away a relationship between treatment and outcomes in observational data. OBJECTIVES/GOALS: Older patients with HL have worse outcomes than younger patients, which may reflect treatment choice (e.g., fewer chemotherapy cycles). We studied the relationship between treatment intensity and 3-year overall survival (OS) in SEER-Medicare. We calculated an E-value to quantify the unmeasured confounding needed to explain away any relationship. METHODS/STUDY POPULATION: This retrospective cohort study of SEER-Medicare data from 1999-2016 included 1131 patients diagnosed with advanced stage HL at age ≥65 years. Treatment was categorized as: (1) full chemotherapy regimens (‘full regimen’, n=689); (2) partial chemotherapy regimen (‘partial regimen’, n=175); (3) single chemotherapy agent or radiotherapy (‘single agent/RT’, n=102), or (4) no treatment (n=165). A multivariable Cox regression model estimated the relationship between treatment and 3-year OS, adjusting for disease and patient factors. An E-value was computed to quantify the minimum strength of association that an unmeasured confounder would need to have with both the treatment and OS to completely explain away a significant association between treatment and OS based on the multivariable model. RESULTS/ANTICIPATED RESULTS: Results from the multivariable model found higher hazards of death for partial regimens (HR=1.81, 95% CI=1.43, 2.29), single agent/RT (HR=1.74, 95% CI=1.30, 2.34), or no treatment (HR=1.98, 95% CI=1.56, 2.552) compared to full regimens. We calculated an E-value for single agent/RT because it has the smallest HR of the treatment levels. The observed HR of 1.74 could be explained away by an unmeasured confounder that was associated with both treatment and OS with a HR of 2.29, above and beyond the measured confounders; the 95% CI could be moved to include the null by an unmeasured confounder that was associated with both the treatment and OS with a HR of 1.69. Of the measured confounders, B symptoms had the strongest relationship with treatment (HR=2.08) and OS (HR=1.38), which was below the E-value. DISCUSSION/SIGNIFICANCE OF FINDINGS: Patients with advanced stage HL who did not receive full chemotherapy regimens had worse 3-year OS, even after adjusting for potential confounders related to the patient and disease. The E-value analysis made explicit the amount of unmeasured confounding necessary to fully explain away the relationship between treatment and OS.

ABSTRACT IMPACT: This project will aid in the optimization of treatment for those with heart failure with a reduced ejection fraction in order to both maximize health benefits and minimize financial burdens. OBJECTIVES/GOALS: To evaluate the accuracy and clinical applicability of a novel web-based application programming interface in the optimization of care for patients being treated for heart failure with reduced ejection fraction (HFrEF). The purpose of this validation is to ensure the translatability of this algorithm to a clinical setting using real-world data. METHODS/STUDY POPULATION: This study is a retrospective analysis of a previously created algorithm designed to optimize therapy for patients currently diagnosed with HFrEF. Patients that are seen for HFrEF treatment at Michigan Medicine are enrolled in a heart failure registry and were included in this study. Exceptions include those with heart transplants, LVAD, and those undergoing treatment with chronic inotropes (milrinone/dobutamine). Clinically relevant information (demographics, vital statistics, labs, and medications including dose and frequency) was taken from their respective electronic health record (EHR) and this data was used as the input for the algorithm. The therapy recommendations were collected and manually compared to the 2017 ACC/AHA/HFSA guidelines to verify the accuracy of the algorithm outputs. RESULTS/ANTICIPATED RESULTS: Data is currently being collected and analyzed. At first glance, our algorithm has been successful at detecting patients that are good candidates for therapy optimization. Based on inputs given, the treatment recommendations have been appropriate when compared to the most up-to-date HF treatment guidelines. The algorithm has also correctly identified levels of urgency for therapeutic recommendations. Finally, we have also shown the algorithm to have effectiveness for identifying areas of inappropriately adjusted therapy. Preliminary results have led to changes to the functionality of the algorithm, including how medications are retrieved from the EHR’s and how medication doses are identified. Previous iterations created discrepancies in dosing and the algorithm has since been adjusted. DISCUSSION/SIGNIFICANCE OF FINDINGS: By verifying its validity, our algorithm can accurately flag patients with HFrEF that are eligible for therapy optimization and give providers the opportunity to make appropriate changes. Given the high health and financial burdens of HFrEF, our algorithm has the ability to provide significant morbidity, mortality, and financial benefits.