OBJECTIVES/GOALS: We seek to develop an automated deep learning-based method for segmentation and volumetric quantification of the fetal brain on T2-weighted fetal MRIs. We will evaluate the performance of the algorithm by comparing it to gold standard manual segmentations. The method will be used to create a normative sample of brain volumes across gestational ages. METHODS/STUDY POPULATION: We will adapt a U-Net convolutional neural network architecture for fetal brain MRIs using 3D volumes. After re-sampling 2D fetal brain acquisitions to 3mm3 3D volumes using linear interpolation, the network will be trained to perform automated brain segmentation on 40 randomly-sampled, normal fetal brain MRI scans of singleton pregnancies. Training will be performed in 3 acquisition planes (axial, coronal, sagittal). Performance will be evaluated on 10 test MRIs (in 3 acquisition planes, 30 total test samples) using Dice scores, compared to radiologists’ manual segmentations. The algorithm’s performance on measuring total brain volume will also be evaluated. RESULTS/ANTICIPATED RESULTS: Based on the success of prior U-net architectures for volumetric segmentation tasks in medical imaging (e.g. Duong et al., 2019), we anticipate that the convolutional neural network will accurately provide segmentations and associated volumetry of fetal brains in fractions of a second. We anticipate median Dice scores greater than 0.8 across our test sample. Once validated, the method will retrospectively generate a normative database of over 1500 fetal brain volumes across gestational ages (18 weeks to 30 weeks) collected at our institution. DISCUSSION/SIGNIFICANCE OF IMPACT: Quantitative estimates of brain volume, and deviations from normative data, would be a major advancement in objective clinical assessments of fetal MRI. Such data can currently only be obtained through laborious manual segmentations; automated deep learning methods have the potential to reduce the time and cost of this process.

OBJECTIVES/GOALS: Autism spectrum disorder (ASD) is a developmental disorder with a high financial and personal burden. Individuals with ASD experience significant comorbid medical conditions. We identified conditions that appear prior to ASD diagnosis in order to potentially improve early screening practices. METHODS/STUDY POPULATION: We used electronic health record data from an anonymized database at Vanderbilt University Medical Center for individuals with ASD and matched controls to analyze comorbid conditions prior to an ASD diagnosis. Data were censored to include only individuals who first appeared in the databank prior to two years of age (ntotal = 1551, ncontrols = 976, nASD = 575). Comorbidities (~1800 conditions) were compared between the ASD and matched control group using a novel tool (pyPheWAS) to examine presence, count, and duration of comorbidities that occurred between 0-2 years old and before ASD diagnosis. RESULTS/ANTICIPATED RESULTS: Convulsions (p = 0.000404, ß = 0.807), constipation (p = 0.000789, ß = 0.894), and strabismus (p = 0.00243, ß = 1.155) were the most significant comorbid conditions prior to age 2 in individuals who would later be diagnosed with ASD. The group with ASD also had more visits associated with convulsions (p = 0.00511, ß = 0.195), diseases of the esophagus (p = 0.0117, ß = 1.675), and allergic reactions to food (p = 0.0119, ß = 0.540) prior to their diagnosis. The ASD group was also seen for a longer duration regarding convulsions (p = 0.000273, ß = 0.695), constipation (p = 0.00157, ß = 0.712), and malaise and fatigue (p = 0.00188, ß = 0.903) before ASD diagnosis. DISCUSSION/SIGNIFICANCE OF IMPACT: Precise comorbid condition profiles in early childhood may help uncover biomarkers leading to better prediction of a future ASD diagnosis. Medical conditions that precede the onset of measurable behavioral symptoms may enhance early screening, treatment, and intervention in ASD.

OBJECTIVES/GOALS: Although referrals for specialty consultations are a core clinical process, they are prone to coordination and communication breakdowns that have led to adverse clinical outcomes. This project’s objective is to improve timely documentation, transmission, access, and quality of consultation notes across healthcare systems. METHODS/STUDY POPULATION: There are two specific aims for this project. In Aim 1, we will characterize clinical workflows and information flow during cross-institutional referrals. In Aim 2, we will develop and test a prototype leveraging electronic health information exchange (HIE) to increase closing the loop for cross-institutional referrals and improve the quality of consultation notes. To accomplish these aims, we will use human factors methods, including data analytics, medical-record reviews, semi-structured interviews of consultants, rapid prototyping, and usability evaluations. RESULTS/ANTICIPATED RESULTS: Results will inform the design and integration of clinician-facing technologies into clinical workflows to close the referral loop and improve diagnostic processes. Aim 1 will provide quantitative evidence about the quality of cross-institutional referrals, inform the eventual implementation of our prototype, and identify user interface features required for successful electronic health information exchange. Based on the results from Aim 1, reports and visual representations will be generated to illustrate information flows and clinical workflows. This will prioritize design efforts for the intervention’s prototype. Aim 2 will translate clinicians’ requirements into prototype features and assess clinicians’ experience with the prototype. DISCUSSION/SIGNIFICANCE OF IMPACT: The use and usefulness of HIE has been limited due to usability and implementation issues. Cross-institutional referrals are complex and dependent on HIE due to EHRs’ lack of interoperability. This project will provide evidence-based recommendations for the use of Fast Healthcare Interoperability Resources (FHIR) to improve HIE during referrals.

OBJECTIVES/GOALS: Utilizing clinical electronic health record (eHR) data pulled en masse from data warehouses provides unique challenges when applying it to retrospective studies. Use of this data in conjunction with metabolomic and genomic results to predict response to lisinopril or ondansetron has been completed. METHODS/STUDY POPULATION: Study population consists of >2000 subjects recruited from the Emergency Medicine Specimen Bank at University of Colorado Denver (UCD). All patients presenting to the emergency department are approached to participate which significantly increases demographic diversity of our study populations. Clinical data is pulled from Health Data Compass (data warehouse at UCD that collects all electronic health record (EHR) data to be able to deliver de-identified). Effectiveness of lisinopril and ondansetron were investigated using metabolomic data collected via ultra-high performance liquid chromatography mass spectrometry and genomic data from Illumina chip technology to find relevant correlations. RESULTS/ANTICIPATED RESULTS: Obtaining retrospective clinical data from data warehouses comes with significant challenges to be addressed. Verifying all clinical variables from patient EHRs is a crucial step that requires extensive quality control steps. As well, ensuring data validity, appropriateness of data points pulled as relate to the study criteria and identifying alternate EHR data points is needed. Chart review is a critical step necessary to surmount these challenges. Additionally, use of retrospective EHR data often necessitates the development of novel definitions of clinical effectiveness that can be abstracted from the EHR– such as how to determine decrease in nausea without a visual analogue scale. DISCUSSION/SIGNIFICANCE OF IMPACT: Utilizing data warehouses to deliver EHR data provides a valuable tool for completing retrospective precision medicine projects. The validation of definitions for clinical outcomes identifiable retrospectively are necessary and provide novel guidance for future studies.

OBJECTIVES/GOALS: Currently physicians are able to order CT Chest/Abd/Pelvis images as STAT or Routine. STAT images denote an emergency and are done immediately. We aim to determine the percentage of CT images that are inappropriately ordered as STAT, determine physician image ordering habits, and develop targeted interventions to encourage appropriate STAT image ordering. METHODS/STUDY POPULATION: A fishbone diagram helped reveal possible causes of inappropriate STAT image ordering. Based on the fishbone diagram, a survey was created to assess CT image ordering habits amongst radiology and internal medicine residents and attending physicians. All CT Chest/Abd/Pelvis images ordered over a 3 month period of time (July-Oct 2017) was obtained. The dataset included whether the image was ordered Stat vs Routine, time of image order, physician name and location, and reason for the imaging study.The STAT images were evaluated based on the explanation provided in the CT image order. Currently 2 radiology residents, 2 internal medicine residents, and 2 internal medicine hospitalists are evaluating all STAT CT images to determine appropriateness and how long they are willing to wait for the image to result in a read. RESULTS/ANTICIPATED RESULTS: Analysis of all CT Chest/Abd/Pelvis imaging orders revealed that 51% (1710/3345) of them were ordered as STAT. The preliminary analysis of 227 STAT images showed that 6% were inappropriate. We anticipate results of our survey to show differences in how long a STAT vs Routine image orders should take amongst Radiology and Internal Medicine clinicians. We also anticipate our survey to show differences in factors that warrant STAT imaging amongst the different medical fields. We anticipate that the clinician manual evaluation of all STAT CT image will reveal a large percentage of imaging orders to be inappropriate. All STAT imaging that were flagged as inappropriate will be characterized by the department who ordered the image and the reason provided for the imaging to assess for common themes. DISCUSSION/SIGNIFICANCE OF IMPACT: STAT images are the new routine with more images ordered STAT than Routine. Inappropriate STAT imaging results in truly urgent patients not getting the medical care they need. Many images ordered stat could potentially be switched to routine. By evaluating why clinicians are ordering STAT CT image inappropriately, we will be able to develop targeted interventions to decrease inappropriate STAT CT imaging.

OBJECTIVES/GOALS: Histopathological descriptions of kidney tissue provide more information about kidney disease severity and prognosis than serum measures, yet most patients with chronic kidney disease do not undergo kidney biopsy. We aim to develop a method to determine the degree of renal injury in patients with diabetes and hypertension without the need for biopsy. METHODS/STUDY POPULATION: Clinical data and renal tissue samples were collected from 864 patients undergoing tumor-associated nephrectomy in seven medical centers in the United States. Exclusion criteria included age < 18, presence of pyelonephritis or non-diabetic or hypertensive renal disease or incomplete clinical or histopathologic data. 19 histologic parameters were scored in a blinded manner by one renal pathologist. We examined the relationship between and functional variables (such as estimated glomerular filtration rate (eGFR)). Polynomial regression analysis was performed to model histopathologic variables and important clinical parameters such as eGFR RESULTS/ANTICIPATED RESULTS: 607 samples met inclusion criteria and were stratified as: control (no history of diabetes or hypertension, n = 160), hypertension alone (n = 224) and both diabetes and hypertension (n = 223). Interstitial fibrosis (IF) and glomerulosclerosis (GS) had the strongest correlations with eGFR. Regression analysis was used to model histopathologic score for a given eGFR. We found that diabetes and hypertension modified the relationship between tubulointerstitial fibrosis and eGFR. For example, while hypertensive patients without diabetes had 33% IF at an eGFR of 30 ml/min/1.73m2 (r2 = 0.64, p<0.01), hypertensive patients with diabetes had 32% IF at an eGFR of 30 ml/min/1.73m2 (r2 = .43, p<0.01) and control patients had approximately 23% IF at an eGFR of 30 ml/min/1.73m2 (r2 = 0.22, p<0.01). DISCUSSION/SIGNIFICANCE OF IMPACT: Here, we describe the relationship between renal structural changes and renal function and show that hypertension significantly modifies the relationship at a given eGFR. These data can be used to reasonably predict renal structural changes given clinical information without the need for renal biopsy and may provide prognostic value.

OBJECTIVES/GOALS: Flavorings differ between brands and tobacco products, potentially altering the sensory perceptions. This study aimed to examine discrepancies in flavor preference across various non-cigarette tobacco products among a national representative sample of US adult regular tobacco users. METHODS/STUDY POPULATION: Data from the Population Assessment of Tobacco and Health (PATH) Study Wave 3 (W3) were used. Weighted prevalence of flavor preference for various tobacco products, including electronic nicotine delivery systems (ENDS), traditional cigars, cigarillos/filtered cigars, hookah and snus/smokeless, was presented for 9,037 adult current and new former users of multiple flavored tobacco products. Within-subject flavor discrepancies were assessed using generalized estimating equations (GEE) models considering the complex sampling design of the PATH study. RESULTS/ANTICIPATED RESULTS: Most regular users of a flavored tobacco products reported using one flavor category per product. Fruit flavors, followed by tobacco, were the most common flavor categories among ENDS (32% and 25%, respectively) and hookah users (44% and 36%, respectively). Tobacco flavor was the most common among regular users of traditional cigars (80%), cigarillos/filtered cigars (55%), and smokeless tobacco (79%). Polytobacco users of ENDS and traditional cigars had the largest discrepancy, where about 68-76%% used different flavor categories when switching products. Conversely, polytobacco users of traditional cigars and cigarillos/filtered cigars had the lowest discrepancy (23-25%). DISCUSSION/SIGNIFICANCE OF IMPACT: Many consumers of multiple tobacco products had different flavor preferences when switching between products. In the event of a partial or full flavor ban for ENDS, these findings raise questions about consumer loyalty to a particular tobacco product or a particular flavor category. Conflict of Interest Description: MLG serves as a paid consultant for Johnson & Johnson and has received research grant from Pfizer, manufacturers of smoking cessation medications. The other authors have no conflicts to declare. CONFLICT OF INTEREST DESCRIPTION: MLG serves as a paid consultant for Johnson & Johnson and has received research grant from Pfizer, manufacturers of smoking cessation medications. The other authors have no conflicts to declare.

OBJECTIVES/GOALS: Inflammation following traumatic injury to the spinal cord persists long after the primary insult and is known to increase complication rates and prolong recovery time. We investigated the effects of Erythropoietin (EPO) in combination with Glucocorticoids on the levels of erythrocyte sedimentation rate (ESR), an overall measure of inflammation. METHODS/STUDY POPULATION: Electronic medical records from approximately 38 million patients in 27 Healthcare Organizations were analyzed using the TriNetX Analytics platform. Patients with spinal cord injuries (SCI) were defined with the ICD-10 code, G95 and two unique cohorts were defined for patients treated with injectable EPO in combination with injectable Glucocorticoids within 6 months of SCI or only injectable Glucocorticoids with no injectable EPO. ESR rates were queried from patient cohorts to evaluate the potential effects of the two treatment pathways on the ESR. Most recent lab results within 6 months before initiating treatment and 1-year post-treatment were defined as “before” and “after” treatment, respectively. Changes in ESR lab results were evaluated using unpaired t-test with Welch’s Correction. RESULTS/ANTICIPATED RESULTS: A total of 14,370 patients satisfied the inclusion criteria. 89 patients were treated with injectable EPO in combination with Glucocorticoids within 6 months of SCI. The ESR lab results were available for 33 patients before treatment with a mean of 63±33 mm/h. The ESR lab results were available for 22 patients after treatment with a mean of 51.7±34.1 mm/h. 14,281 patients were treated with Glucocorticoids (no injectable EPO) within 6 months of SCI. The ESR lab results were available for 2,042 patients before treatment with a mean of 29.2±30.5 mm/h. The ESR lab results were available for 2,184 patients after treatment with a mean of 32.6±30 mm/h. Patients treated with combinational therapy showed a reduction in ESR of 11.3 mm/h, while those treated with only Glucocorticoids showed an increase in ESR of 3.4 mm/h. DISCUSSION/SIGNIFICANCE OF IMPACT: The present results demonstrated that combinational therapy with injectable, EPO and glucocorticoids exhibited a significant reduction in ESR level. The study suggests that EPO and glucocorticoid might have a synergistic effect on reducing the inflammation following SCI. This approach might help reduce the therapeutic dose of glucocorticoids. Conflict of Interest Description: The authors declare that they have no competing interests. CONFLICT OF INTEREST DESCRIPTION: The authors declare that they have no competing interests.

OBJECTIVES/GOALS: The Dixon up-and-down method (U/D), original developed for testing explosives, is especially common in anesthesia research studies. The objective of this research is to compare the performance of the U/D method for obtaining and analyzing sensitivity data with that of the Bayesian Optimal Interval (BOIN) method. METHODS/STUDY POPULATION: A simulation study will compare the performance of the U/D method with the BOIN design. The two study designs offer alternative decision-making algorithms with respect to the dose at which the next experimental unit is treated. These alternative decisions may impact the precision of point estimates of the mean and standard deviation of the effective dose to elicit a response. Transition probability matrices are developed, and maximum likelihood estimates of the unknown parameters assessed for accuracy. For simulation, the effective dose is assumed to be randomly distributed with a known mean and standard deviation. Fixed dose levels are defined, and decisions for what level the next experimental unit should be treated at are defined by the Dixon up-and-down method and the BOIN design. For the U/D method, the stimulus is increased by one level in the absence of a response or decreased if a response occurs from an initial stimulus. A target toxicity probability of 0.50 is used to define the dose escalation or de-escalation rules for the application of the BOIN design. RESULTS/ANTICIPATED RESULTS: A feature of both methods is that the consecutive observations are concentrated about the mean value of the effective dose. However, the BOIN design tends to be more concentrated between these two dose levels. In the presence of severe adverse events, the BOIN design can choose to eliminate doses that are too toxic whereas the U/D design cannot eliminate any dose levels. Transition probability matrices are defined and parameters for the distribution of the effective dose are estimated using maximum likelihood estimation. Mean squared errors for the estimated mean and standard deviations compare the two study designs. DISCUSSION/SIGNIFICANCE OF IMPACT: The BOIN design offers an alternative method for decision-making compared with the U/D method. The BOIN design tends to concentrate dose levels about the mean more than the U/D. This may provide better estimates of the mean and standard deviation of the effective dose for eliciting a response in some circumstances.

OBJECTIVES/GOALS: Harnessing the EHR to support clinical research is critical for the study of rare diseases such as interstitial lung disease (ILD). However no studies have compared EHR and research-quality data in the ILD population. Our objectives were to (1) identify ILD patients and extract clinical data from an EHR system and (2) assess the performance of ILD data capture. METHODS/STUDY POPULATION: Case validated algorithms were implemented to identify patients from the University of California San Francisco EHR and extract key ILD clinical information including, demographic variables, process measures and patient outcomes. Key clinical information were defined based on consensus statements and ILD clinical trials. A subset of ILD patients, had variables recorded in both the EHR and a separate ILD longitudinal research database. The completeness of EHR data capture and level of agreement were compared between three data collection methods: (1) data manually and systematically collected for an ILD research database (gold standard), (2) data automatically extracted from structured fields in the EHR, and (3) data extracted from unstructured data sources. RESULTS/ANTICIPATED RESULTS: We identified 5857 ILD patients in the EHR, of which 2100 patients had data available in the both the EHR and research database. Baseline demographic variables, co-morbidities, use of diagnostic testing, pharmacotherapy were accurately extracted from structured fields. Outcome measures, including lung physiology, radiographic patterns, pathology results, and health related quality of life (HRQoL) were unevenly extracted from structured fields alone. With the exception of HRQoL, these measures were accurately captured in unstructured EHR sources. Notably, certain metrics were better defined in the EHR, including health care resource utilization metrics, acute exacerbations, medication side effects, supplemental oxygen use and specialty care referrals (rheumatology, lung transplant, palliative care, etc). DISCUSSION/SIGNIFICANCE OF IMPACT: A large real-world ILD cohort can be algorithmically extracted from the EHR along with key clinical variables with accuracy comparable to protocol-driven research databases. Rigorous assessment of the types of disease-specific variables that are present in EHR-derived data will inform future interventions to improve the fidelity, accessibility and use of the EHR in clinical research.

OBJECTIVES/GOALS: Identify the interactions between nasopharyngeal bacterial pathogens, commensals, and patient clinical characteristics in relation to the development of viral upper respiratory tract infections (URI) and acute otitis media (AOM) in infants. METHODS/STUDY POPULATION: The subjects were part of a prospective, longitudinal study (2008–2014) of infants to evaluate the prevalence and risks for the development of URI and AOM. Healthy infants (n = 362) were enrolled from near birth and followed to the first episode of AOM up to 12 months of age. Nasopharyngeal specimens and clinical traits were collected at monthly intervals between 1-6 months, month 9, and during viral URI episodes. Subjects were closely followed for AOM development. 16S rRNA sequencing was performed on the nasopharyngeal swabs to identify their bacterial composition. Multidimensional (2, 3, and 4 dimensional) co-presence, co-exclusion, and one-way relation patterns were identified between the microbiome compositions, health status, and other collected clinical traits. RESULTS/ANTICIPATED RESULTS: We analyzed 971 specimens collected monthly and during URI and AOM episodes from 139 infants. Of the 139 enrolled subjects, 96% had 2 or more healthy samples, 77% contracted URI/AOM during the study period, and 60% had at least 1 healthy sample before URI/AOM onset. Otopathogens (Moraxella, Haemophilus, and Streptococcus), Staphylococcus, and Pseudomonas were the most common pathogenic genera. Corynebacterium, Dolosigranulum, and Acinetobacter were 3 most abundant commensal bacterial genera. Samples from infants with AOM in the first year had a significantly higher relative abundance of Haemophilus, Enterobacter, and Yersinia, and lower relative abundance of Corynebacterium, and Pseudomonas compared to samples from infants who did not develop AOM. DISCUSSION/SIGNIFICANCE OF IMPACT: Identification of complex multidimensional interaction patterns within microbial communities and environmental factors is vital to understanding disease onset risk and prevention. Prophylactic microbiome and environmental factor modulation between enterotypes could be used to reduce URI/AOM onset in infants.

OBJECTIVES/GOALS: Metabolic syndrome (MetSyn) is a risk for World Trade Center-Lung Injury (WTC-LI; defined as developing FEV1<lower limit of normal [LLN]). Metabolic health is a modifiable disease risk factor. We propose to characterize how time-dependent covariates of MetSyn are longitudinally associated with WTC-LI. METHODS/STUDY POPULATION: WTC-particulate exposed firefighters, consented, with pre-9/11 FEV1 LLN (N = 5,746). Data assessed from last pre-9/11 till August 1, 2017. Longitudinal MetSyn characteristics were assessed using 3 models: i. A linear mixed effect model to assess the effect size of longitudinal MetSyn and its components on longitudinal FEV1% predicted as an outcome; ii. a time-dependent Cox regression to assess the associations of MetSyn to time of onset of WTC-LI; iii. a novel, partially linear single index regression model with repeatedly measured MetSyn to assess their joint effects and delineate their relative contribution on the longitudinal lung function in the WTC-FDNY cohort. RESULTS/ANTICIPATED RESULTS: In Model I, BMI 30 kg/m2 had the largest effect size compared to ever-smoking, with −2.524 (95%CI: −2.708,−2.340) compared to −1.681(−2.325,−1.038) respectively. Having MetSyn, defined as 3/5 risk factors, had an effect size of −2.319(−2.526,−2.112). In Model II, hazards of triglycerides 150mg/dL were highest at 1.497(1.336, 1.677), followed by BMI 30 kg/m2 at 1.406(1.256, 1.575), and HDL<40mg/dL 1.355(1.176-1.561), compared to ever-smoking (1.201, p = 0.002). Having high exposure to PM by being present in the morning of 9/11 was a significant covariate only in Model II investigating HDL<40mg/dL or triglycerides 150mg/dL. Model III The proposed methods will be applied to our cohort study. DISCUSSION/SIGNIFICANCE OF IMPACT: MetSyn is both a predictor and concurrent marker of WTC-LI. The single index model can not only reduce dimensionality of the covariates, but also provides efficient estimates of the joint MetSyn effects, allowing linear or nonlinear effects. Future studies will investigate dietary intervention as a potential disease-modifying factor. CONFLICT OF INTEREST DESCRIPTION: NA, nothing to disclose.

OBJECTIVES/GOALS: Access to biostatistics expertise is essential for a successful clinical and translational research program. However, demand for statistical support at academic research centers can strain the capacity of biostatistics units. Our objective was to efficiently increase access to statistical expertise. METHODS/STUDY POPULATION: In cooperation with the Cancer Center Biostatistics Shared Resource, we replaced an informal 1-hour drop-in consultation program with structured office hours to provide statistical support to clinical and translational researchers at the University of California, Davis Medical Center. We doubled office hours to 2 hours per week and established six 20-minute appointments. Two Ph.D. level statisticians staff office hours. Researchers schedule appointments through Acuity Scheduling, a free on-line resource. Availability of the service is advertised monthly by sending an informational flyer to various university listservs. RESULTS/ANTICIPATED RESULTS: Prior to implementing the program in 2014, we averaged 91 office hour consults per year. Subsequently, consultations jumped to 171 in 2014 and have averaged 150 per year since then. Office hours attract students, residents, staff and faculty from a wide range of disciplines including the Schools of Medicine, Nursing, Veterinary Medicine and basic science departments. Project types span the clinical and translational spectrum covering lab, animal, clinical and population-level studies. Most consults related to data analysis and interpretation (57%) followed by sample size calculations/study design (29%) and response to reviewers (4%), with general statistical advice as the remainder. DISCUSSION/SIGNIFICANCE OF IMPACT: With 6 micro-consults per week, we can meet with many investigators and triage their statistical support needs. This program has proved very popular and was highly rated in a recent user survey, with several investigators noting that the consults facilitated successful publications and proposals.

OBJECTIVES/GOALS: We modeled risk of reintubation within 48 hours of cardiac surgery using variables available in the electronic health record (EHR). This model will guide recruitment for a prospective, pragmatic clinical trial entirely embedded within the EHR among those at high risk of reintubation. METHODS/STUDY POPULATION: All adult patients admitted to the cardiac intensive care unit following cardiac surgery involving thoracotomy or sternotomy were eligible for inclusion. Data were obtained from operational and analytical databases integrated into the Epic EHR, as well as institutional and departmental-derived data warehouses, using structured query language. Variables were screened for inclusion in the model based on clinical relevance, availability in the EHR as structured data, and likelihood of timely documentation during routine clinical care, in the hopes of obtaining a maximally-pragmatic model. RESULTS/ANTICIPATED RESULTS: A total of 2325 patients met inclusion criteria between November 2, 2017 and November 2, 2019. Of these patients, 68.4% were male. Median age was 63.0. The primary outcome of reintubation occurred in 112/2325 (4.8%) of patients within 48 hours and 177/2325 (7.6%) at any point in the subsequent hospital encounter. Univariate screening and iterative model development revealed numerous strong candidate predictors (ANOVA plot, figure 1), resulting in a model with acceptable calibration (calibration plot, figure 2), c = 0.666. DISCUSSION/SIGNIFICANCE OF IMPACT: Reintubation is common after cardiac surgery. Risk factors are available in the EHR. We are integrating this model into the EHR to support real-time risk estimation and to recruit and randomize high-risk patients into a clinical trial comparing post-extubation high flow nasal cannula with usual care. CONFLICT OF INTEREST DESCRIPTION: REF has received grant funding and consulting fees from Medtronic for research on inpatient monitoring.

OBJECTIVES/GOALS: To describe principles, best practices, and techniques recommended to instill deep understanding of the application and interpretation of statistical techniques and statistical inference among translational scientists and trainees, that best support the concepts of scientific Rigor, Reproducibility and Reporting. METHODS/STUDY POPULATION: Each of the six New York City Area Biostatistics, Epidemiology and Research Design (BERD) resources have strong educational programs, novel curricular components, and creative strategies, implemented by award winning educators. To capitalize on shared knowledge, innovation, and resources, the six teams formed the New York City Area BERD Collaborative (NYC-ABC) comprised of BERD resources from Mt. Sinai, Cornell, Einstein, Columbia, Rockefeller, and NYU. The collaborative suggests principles, concepts, tools and approaches to support the concepts of scientific Rigor, Reproducibility and Reporting in translational science. RESULTS/ANTICIPATED RESULTS: Principles:

• Value of team science approach and including biostatisticians early and often.

• Carefully designing experiments to reduce bias and increase precision.

• Trainees’ focus is often on “statistical significance” and the p-value. Consequences of data dredging/p-hacking, and the impact of sample size and other factors on statistical significance.

• Emphasizing the effect size and answering the scientific hypothesis when reporting results.

• Statistical code used to produce results should be well annotated and raw data posted online to enhance reproducibility.

• Incorporate effective multiple modalities (i.e. didactic, demonstrative, hands on workshops, applications, and tools).

• Approach from “the drivers’ seat” perspective, rather than strictly mathematical.

• Endorse flipped classroom approach

DISCUSSION/SIGNIFICANCE OF IMPACT: Like any complex discipline, biostatistical education can be approached from several dimensions, but it remains essential to focus on fundamental goals of science. We remind our trainees that the goal of science is to create knowledge, not to “find significance”. Deep understanding of inferential methods and proper interpretation of results are key. CONFLICT OF INTEREST DESCRIPTION: None.

OBJECTIVES/GOALS:

• A web interface that allows for easy upload of CSV text data to time-based visualizations

• Implementation of change points analysis to identify and display points where event rates increased or decreased

• customizable plots where the user can change point shapes, color, etc.

• customizable and advanced filtering support

• support for plot comparisons and exports

METHODS/STUDY POPULATION: We used the R/Shiny framework to develop a web application for visualization of time stamped data. The Research and Exploratory Analysis Driven Time-data Visualization (READ-TV) application allows for user-friendly mining for longitudinal patterns in data. READ-TV is built specifically for FD analysis, but is easily adaptable to other clinical use cases, as we allow for the use of general metadata on events and cases.The building of a quantitative framework for event analysis starts with the application of homogeneous Poisson processes, which relate the times of occurrence of events in terms of an underlying rate. To understand the changes in this underlying rate, changepoint analysis is used to model the rate as a function of time using piecewise constant approximations. The changepoint analysis allows us to identify the specific periods of time where the rate of FD is increased relative to a baseline or a desired operating range. RESULTS/ANTICIPATED RESULTS: READ-TV application allows for import of time stamped event data from multiple cases. Event and case metadata are supported to facilitate filtering and mining of interesting subsets of data. Stem plots are used for visualization of selected event timelines in chosen cases. This visualization is accompanied with summary of the number and estimates of rates of occurrence of specific event types (e.g. types of FD). Change-point analysis is implemented using the ‘changepoint‘ R library. These analyses allow the users to quickly understand whether the rates of events (FD) is changing across the case timeline and where exactly these changes are occurring. DISCUSSION/SIGNIFICANCE OF IMPACT: We have demonstrated the READ-TV application to the team of the AHRQ-funded Human Factors and Systems Integration in High Technology Surgery (HF-SIgHTS) study. The ability to visualize and perform quantitative analysis of the study data was received with unanimous positive feedback and enthusiasm. We continue READ-TV development focusing on (1) increased user-friendliness using the HF-SIgHTS as our focus group, (2) increased functionality, and (3) use of more general localization terminology to allow for other applications.

OBJECTIVES/GOALS: Diverse medication-based studies require longitudinal drug dose information. EHRs can provide such data, but multiple mentions of a drug in the same clinical note can yield conflicting dose. We aimed to develop statistical methods which address this challenge by predicting the valid dose in the event that conflicting doses are extracted. METHODS/STUDY POPULATION: We extracted dose information for two test drugs, tacrolimus and lamotrigine, from Vanderbilt EHRs using a natural language processing system, medExtractR, which was developed by our team. A random forest classifier was used to estimate the probability of correctness for each extracted dose on the basis of subject longitudinal dosing patterns and extracted EHR note context. Using this feasibility measure and other features such as a summary of subject dosing history, we developed several statistical models to predict the dose on the basis of the extracted doses. The models developed based on supervised methods included a separate random forest regression, a transition model, and a boosting model. We also considered unsupervised methods and developed a Bayesian hierarchical model. RESULTS/ANTICIPATED RESULTS: We compared model-predicted doses to physician-validated doses to evaluate model performance. A random forest regression model outperformed all proposed models. As this model is a supervised model, its utility would depend on availability of validated dose. Our preliminary result from a Bayesian hierarchical model showed that it can be a promising alternative although performing less optimally. The Bayesian hierarchical model would be especially useful when validated dose data are not available, as it was developed in unsupervised modeling framework and hence does not require validated dose that can be difficult and time consuming to obtain. We evaluated the feasibility of each method for automatic implementation in our drug dosing extraction and processing system we have been developing. DISCUSSION/SIGNIFICANCE OF IMPACT: We will incorporate the developed methods as a part of our complete medication extraction system, which will allow to automatically prepare large longitudinal medication dose datasets for researchers. Availability of such data will enable diverse medication-based studies with drastically reduced barriers to data collection.

OBJECTIVES/GOALS: In a previous genome-wide association study of European Americans, rs12772424, a variant in the gene encoding transcription factor 7-like 2 (TCF7L2), was shown to have a SNP-BMI interaction association with bipolar disorder (BD). The objective of this study was to replicate this finding in an African American (AA) sample. METHODS/STUDY POPULATION: Using a sample of 659 controls and 323 BD cases from the Genetic Association Information (GAIN) Study of BD, we conducted analyses to assess association between BD and the interaction of BMI with genetic variation in TCF7L2. For this study we identified 4572 single nucleotide polymorphisms (SNPs) in a 1Mb region around the candidate SNP rs1272424. Based on variants identified in the prior analysis of the larger European American dataset we selected SNPs for analysis in the AA data. This allowed for a smaller AA sample, and still maintained adequate power for statistical significance. RESULTS/ANTICIPATED RESULTS: We anticipate observing an effect of TCF7L2 variation on the relationship between BD and BMI in the AA data. We also anticipate that combining results from the European and African American patients may help narrow down potentially functional variants in TCF7L2 that influence the association between BD and BMI. DISCUSSION/SIGNIFICANCE OF IMPACT: Psychiatric genetics studies lack ancestral diversity among participants, decreasing their generalizability, and possibly increasing health disparities, especially for diseases like BD, which is often misdiagnosed and untreated in AAs. This work propels us towards understanding the genetics of BD and obesity in this underrepresented population.

OBJECTIVES/GOALS: Multivariate regression is used for surgical outcomes analyses; but does not allow for evaluation of all variables. Machine learning could be the perfect alternative to address this issue. Our aim was to evaluate whether machine learning is a feasible alternative to evaluate surgical outcomes. METHODS/STUDY POPULATION: H-CUP National Inpatient Sample database was queried for adult patients with colorectal cancer who underwent colorectal resection, while the NSQIP database was queried for adult patients with rectal cancer who underwent proctectomy. A multivariate regression analysis was performed to assess risk factors associated with 30-day complications following those procedures. Subsequently, machine learning techniques of under-sampling and oversampling were applied to the same datasets for the evaluation of risk factors for the same outcome. These techniques were used to achieve a larger population sample size and to detect statistical significance. Results between the two methodologies were compared. RESULTS/ANTICIPATED RESULTS: Multivariate regression revealed that open approach, gender, race, geographic location, number of comorbidities, and type of insurance was associated with increased 30-day mortality in colorectal resection patients. Conversely, the use of machine learning revealed that preoperative weight loss, preexistent chest heart failure, renal failure or perivascular disease were strongly associated with 30-day mortality. For proctectomy patients, multivariate regression found no association between surgical approach and 30-day mortality. However, machine learning revealed gender, hypertension, and reoperation to be strongly associated with 30-day mortality. DISCUSSION/SIGNIFICANCE OF IMPACT: Machine learning enabled multiple combinations that were not possible to examine in a conventional multivariate regression analysis. Machine learning compared to traditional multivariate regression produced significantly different outcomes, highlighting the need for in depth of these methodologies.

OBJECTIVES/GOALS: Wheezing has been shown to be associated with use of cigarettes, and more recently, electronic nicotine delivery systems (ENDS). This study assessed the association of poly use of tobacco products with wheezing among a national representative sample of US adult current tobacco users. METHODS/STUDY POPULATION: Data from the Population Assessment of Tobacco and Health (PATH) Study Wave 3 (W3) were used. Weighted prevalences of self-reported wheezing and related respiratory symptoms for non-users compared to users of cigarettes, ENDS, cigars, and any combination of these products (poly use of tobacco products) were presented for 28,082 adults. The cross-sectional association of tobacco use with self-reported wheezing and other related respiratory symptoms was assessed using weighted multivariable and ordinal logistic regression with consideration of complex sampling design. RESULTS/ANTICIPATED RESULTS: Most adults who reported on wheezing symptoms did not currently use cigarettes, ENDS or cigars (79%), 15% used cigarettes, 3% used a combination of cigarettes, ENDS and cigars, 1% used ENDS, and 1% used cigars. Significantly higher odds of ever had wheezing or whistling in chest at any time in the past was observed among current cigarette (adjusted OR: 2.62, 95%CI: 2.35, 2.91), ENDS (1.49, 95%CI: 1.14, 1.95), and poly users (2.67, 95%CI: 2.26, 3.16) compared to non-users. No differences were seen for cigar use. Polytobacco use was associated with a higher odds of ever wheezing when compared to ENDS (1.61, 95%CI: 1.19, 2.17) and cigar use (2.87, 95%CI: 1.93, 4.26), but not cigarettes. DISCUSSION/SIGNIFICANCE OF IMPACT: Wheezing is associated with the use of cigarettes, ENDS, or any combination of cigarette, ENDS and cigars likely due to the inhalation of noxious chemicals and gases found in the smoke of cigarettes and ENDS that are likely to increase the odds of experiencing wheezing. CONFLICT OF INTEREST DESCRIPTION: MLG serves as a paid consultant for Johnson & Johnson and has received research grant from Pfizer, manufacturers of smoking cessation medications. The other authors have no conflicts to declare.