2020 ICTS Symposium Poster Abstracts

Informatics Tools
Poster #Poster Details
1 Abass, Babatunde
Washington University in St. Louis
Using REDCap for data collection in resource-poor settings
Babatunde AA; Abente B; Damulira C; Ssentumbwe V; Ozge SB; Ssewamala FM; Kimberly J

  • Introduction: In many resource-poor settings, medical records needed for research are paper-based, making data collection challenging. Research Electronic Data Capture (REDCap), a HIPPA-compliant secure web application designed to capture research data on and offline, has been used by researchers in over 800,000 projects in >130 countries. However, user methods and experience in resource-poor settings are not well documented. Our objective is to describe our experience and data quality challenges observed using REDCap in Suubi4Cancer, a study that aims to identify confirmed and suspected cancer cases in a cohort of >3000 HIV+ youth seen at 39 clinics in Southern Uganda.
  • Methods: Data for this study is extracted from medical records of patients at different health facilities. A REDCap database was designed to capture data from Uganda Ministry of Health forms including demographics, HIV care enrollment, anti-retroviral therapy use, diagnoses at clinic visits, and referrals for suspected cancers. The database was created and stored on a server at Washington University in St. Louis. All field data were entered by data collectors using the REDCap mobile application that was installed on iPads. The data collectors were trained to use the offline REDCap mobile application and after every field data collection day, data was synced to the server from the mobile application when internet became available.
  • Results: Data collection started on 8/12/2019. As of 12/02/2019, 3,449 medical files were collected using REDCap mobile apps and synced to the database. Data collection per record is estimated at a median of 15 minutes (range of 7 -119 minutes) for <=180 responses (depending on skip patterns). The only technical challenge encountered was a sync error when the database was modified but the project was not updated on the mobile application prior to syncing stored data. This was corrected by removing the database modification, syncing data, adding the modification, and updating the mobile app software on the iPads. With respect to data quality, dates were often missing (e.g. birthdate was missing for 42% of cases).
  • Conclusions: REDCap is an effective data collection tool for use in resource-poor settings that often use paper-based records and lack internet access. However, local installation of the REDCap database can be an issue due to server requirements and IT technical knowledge. In addition, missing data, especially for dates, can be a challenge. This may be a general concern for studies involving collection of data from individuals in communities where birthdate tracking is not part of the culture.
2 Antes, Alison PhD
Washington University in St. Louis
The public is lukewarm about the use of artificial intelligence in healthcare according to a survey of U.S. residents
Antes AL; Burrous S; Sisk BA; DuBois JM

  • Introduction: To realize the promise and mitigate potential pitfalls of artificial intelligence (AI) in healthcare, the perspectives of patients must be included in the development and implementation of AI technologies. We developed a new scenario-based measure to investigate public views of AI in healthcare. We examined potential concerns, such as privacy and the patient-doctor relationship, and potential benefits, such as improving access and quality of care, related to AI in healthcare. We explored factors like trust in the healthcare system that might explain views.
  • Methods: U.S. residents (N = 936) were recruited via MTurk. They indicated their level of openness on a 5-point scale to six AI healthcare technologies described in brief scenarios. Following each scenario, participants were presented with additional information reflecting potential benefits and concerns. They indicated whether this information changed their view of the technology positively or negatively on a 7-point bipolar scale. We also measured healthcare system trust, trust in technology, healthcare utilization and satisfaction, conservatism, and demographics.
  • Results: Findings indicate that members of the public are moderately open (M = 3.06, SD = .89) to AI in healthcare. Ratings of the concerns (M = 5.34, SD = .82) indicated that when presented with concerns, participants viewed the technologies somewhat more negatively. Ratings of the benefits (M = 5.49, SD = .75) suggested that when presented benefits, participants felt somewhat more positive. Additional findings show that views of the healthcare system, past healthcare utilization, health status, and gender are associated with views. In particular, trust in technology and in the healthcare system appear to be important.
  • Conclusions: While the public is slightly supportive of AI technologies in healthcare, perceived benefits are endorsed only slightly more strongly than perceived concerns. AI developers, policy makers, and physicians will need to engage the public to advance trust and acceptance of AI technologies. This study suggests important future research, such as examining views among patients with diverse health conditions.
3 Barbour, Dennis MD, PhD
Washington University in St. Louis
Differential inference enables rapid diagnoses with little data
Barbour DL; Larsen TJ; Malkomes G

  • Introduction: Patient workups are traditionally structured to proceed open-loop, with complete data collection before a clinical decision can be rendered. This structure poorly accommodates expensive or time-consuming measures. Here we define the principles of active differential inference, which optimizes data collection to evaluate a specific hypothesis for a specific patient. Clinical decisions can then be made as soon as sufficient data are accumulated to rule in or out possibilities. This project piloted the feasibility of active differential inference for diagnosing hearing loss.
  • Methods: A large database of audiograms (i.e., the standard hearing test) was used to construct a canonical generative model for each of 7 diagnoses of hearing loss. These models were used as references against which to evaluate the estimation quality of a new simulated patient model matching one of the 7 diagnoses. The estimated patient models were formulated as Gaussian processes. All 49 combinations of the 7 reference and 7 new models were assessed in 10 independent experiments each.
  • Results: Even though the models of hearing were complex, fewer than 10 actively acquired samples were needed on average to determine whether or not a simulated current patient matched a particular reference diagnosis at a Bayes factor of 100.
  • Conclusions: Active differential inference has been used to rapidly diagnose simulated patients with hearing loss. Of greater significance, this new form of inference reveals that arbitrarily complex models can be readily compared against one another in real time as data are collected. Clinicians may eventually be able to use tools such as this to direct effort in working up patients toward the most productive diagnostic procedures.
4 Baskett, William
University of Missouri—Columbia
Identifying complex interactions which affect outcomes using proof tree contrast mining
Baskett W; Shyu C

  • Introduction: Contrast pattern mining attempts to discover patterns of features which occur at a significantly different rate between user defined groups. Understanding the underlying differences between patients with different medical outcomes is essential for developing better treatments. We propose a new pattern mining algorithm which is capable of uncovering complex relationships between biomarkers and phenotypes. Patterns uncovered by our algorithm resemble proof trees as they can contain complex structured logical interactions which are easily interpreted. Identifying these interactions may help explain the underlying biological mechanisms and help better understand medical comorbidities. Our objective is to extract patterns of biomarkers which are strongly associated with the development and non-development of specific conditions using both EHR and genetic data.
  • Methods: We have developed a unique contrast mining method capable of extracting patterns of complex logical interactions from large medical datasets in the form of proof trees. This method iteratively builds more and more complex patterns by finding and then combining smaller patterns which differ in frequency between the two defined cases. We applied this method to both anonymized EHR data and to exome data from the Simons SPARK autism dataset.
  • Results: When applied to EHR data, our method was successful at locating patterns of features strongly associated with the development and non-development of specific conditions. Our method was also successful in locating a large number of patterns associated with both the development and non-development of autism as well as patterns which differentiated specific autistic phenotypes.
  • Conclusions: Our method produces a human-readable explanation of how underlying factors are likely interacting to produce the observed result. In healthcare, this kind of analysis can be used to better understand how different medical conditions, medications, and environmental factors interact to influence medical outcomes. Understanding these allows for better treatment and prevention.
5 Baumann, Ana PhD
Washington University in St. Louis
Translation and validation of the evidence-based practice assessment survey to Brazilian Portuguese: Challenges and lessons learned
Baumann AA; Vazquez AA; Carothers B; Coelho L; Juras M; Kohlsdorf M; Lima A; Macchioni AC; Ribeiro M

  • Introduction: Incorporating multiple stakeholder perspectives is important for the successful adoption and ultimately sustainment of any intervention. This study takes place in the preparation phase to implement an evidence-based parent intervention in Brazil. We translated the Evidence-Based Practice Attitude Scale (EBPAS), a survey that asks for professionals’ feelings about using evidence-based interventions, to Portuguese. This study aimed to investigate the psychometric properties of the Brazilian version of the EBPAS. Understanding practitioners’ attitudes can give insights about their readiness to implement EBPs in their services, and contribute to the tailoring of better implementation strategies that strengthen EBP uptake.
  • Methods: The survey was translated from English to Portuguese, and back-translated, and cognitive interviews were done to examine the readability of the items. Participants were recruited via snowball sampling. The survey was administered to the participants using Qualtrics, an online survey tool. Confirmatory factor analysis (CFA) was used to determine if the scale retained its original structure after translation and administration to a Brazilian audience.
  • Results: The four-factor structure estimates 37 parameters for the CFA, meaning the suggested minimum n=370. Current sample size with no missing data is n=323. We included an additional 66 participants who were missing 3 or fewer variables. Missing data were imputed using Random Forest. A baseline model loading all survey items onto one factor was performed. The four-factor model with items loaded to Requirements, Appeal, Openness, and Divergence subscales was tested against the baseline. The baseline model had a poor fit, Chisq(90)=897.36, p<.001, Chisq/df = 9.97, CFI=.913. The four-factor model had a good fit, Chisq(84)=223.94, p<.001, Chisq/df = 2.67, CFI=.985. This difference was significant, ChisqDif(6)=324.9, p<.001.
  • Conclusions: The factor structure of the translated scale is reasonably similar to the original. Participant recruitment will continue to increase the sample size. Challenges involved not being able to pay participants due to ethical guidelines in country, and length of the overall survey.
6 Haithcoat, Timothy PhD Candidate
University of Missouri—Columbia
Bringing health context to the forefront: Geospatial Analytical Research Knowledge-base (GeoSPARK)
Haithcoat TL; Shyu C

  • Introduction: A consistent finding across population health literature is that location matters. Geographic context is an integral component of population health research. It is paramount to understand the nature of the ‘environment’ in which individuals are located in order to understand the ways that physical, biological, environmental, infrastructural, economic, social, and cultural factors can affect personal health outcomes, disease risk and exposure, and access to healthcare. The collection, integration, and use of varied data are foundational to health research. However, there is no “one-stop shop” available for practitioners and researchers to efficiently perform query, extraction, and spatial analytics across all the data publicly available. To strategically transform health research, a robust integrated data platform is needed.
  • Methods: This research used a geospatial informatics approach to build an integrated contextual framework. This was achieved through the design, compilation, and assembly of the Geospatial Analytical Research Knowledgebase (GeoSPARK). The big table represents 318 million locations across the lower 48 states, each with a myriad of attributes compiled from public data sources across multiple scales, geographies, and times. It applies advanced complex queries across this multi-resolution locational information to form the data framework.
  • Results: The design and approach of GeoSPARK provides the ability to identify and potentially mitigate health disparities. It can provide decision makers with a new tool to evaluate policy implications and identify areas affected. It provides health researchers an integrated data repository that can be searched to enable stronger research designs, spatial interaction models, or simply contextualize a population’s characteristics for better modeling.
  • Conclusions: The GeoSPARK will enable more complex health research and broaden geospatial data use and analytics, all while enabling more cost-effective research. The project’s success is an anticipated change in health researchers’ use of geospatial information and access to big data analytic tools.
7 Liu, Danlu MS
University of Missouri—Columbia
Exploratory data mining for subgroup cohort discoveries and prioritization
Liu D; Baskett W; Shyu CR

  • Introduction: Finding small homogeneous subgroup cohorts in large heterogeneous populations is a critical process for hypothesis development in biomedical research. Concurrent computational approaches are still lacking in robust answers to the question “what hypotheses are likely to be novel and to produce clinically relevant results with well thought-out study designs?” The goal of exploratory data mining for cohort discovery is to provide a robust data-driven framework to tailor potential interventions for precision health automatically. The aims of this research include: (1) crawl a large number of candidate subgroups efficiently, (2) discover significant contrasts between the subgroups and (3) prioritize the candidate subgroup pairs for clinical trials or future studies.
  • Methods: Our floating and path expansion approach utilizes a less greedy and more computational feasible floating selection process to select potential subgroups based on their qualities, and the expansion process controls the ratio of the number of subgroups for later floating iterations. Then, we use a distributed computing framework to identify contrast patterns of features which differentiate groups by exploring patterns that have an imbalanced prevalence between the groups. Finally, we prioritize the candidate subgroup pairs by J-value, an index for evaluating the aggregated contributions of the extracted contrast patterns within each pair of subgroups based on the number of contrast patterns and the significance of those patterns.
  • Results: We used the Simons Foundation Autism Research Initiative (SFARI) Simon’s Simplex Collection (SSC) for autism cohort discoveries. By performing the deep exploratory data mining method with a 20% expanding factor, we discovered 142 contrast subgroups. From all discovered genes or gene combinations in the top 20 subgroup cohorts, 11.57% of 415 relevant genes are in AutDB, nearly 20.72% were identified through the PubMed search, and the remaining genes were considered novel.
  • Conclusions: This research has the potential to enable targeted treatments to improve outcomes, reduce costs, and minimize morbidity associated with misdirected interventions.
8 Lyons, Patrick MD
Washington University in St. Louis
How reliably can we grade ICU-ward transfer notes? Validating the physician documentation quality instrument in assessment of ICU-ward transfer notes
Lyons PG; Rojas JC; Baty J; Santhosh L

  • Introduction: Patient transfers from the ICU to the wards are high-risk, and poor physician communication quality during these transfers is associated with negative patient outcomes. Currently, no tools exist to assess ICU-ward physician transfer note quality. We aimed to test the validity and reliability of the 9-item Physician Documentation Quality Instrument (PDQI-9; validated for assessing progress notes and discharge summary quality) for ICU-ward transfer note quality.
  • Methods: We de-identified 12 ICU-ward transfer notes from the Medical ICUs at Zuckerberg San Francisco General, University of Chicago, and Barnes-Jewish Hospital for evaluation by physicians using the PDQI-9. “Gold-standard” evaluators (n = 9) were Program Directors, Associate Program Directors, or patient safety experts; “test” evaluators were Chief Residents or Fellows (n = 8). Measures were (1) discriminant validity (total scores on best/worst notes according to a general utility question: “Without additional information, could you use this note to manage this patient if called for help?”); (2) criterion-related validity for the total score against the general utility question; (3) internal consistency and reliability; and (4) interrater reliability (intraclass correlation coefficient [ICC]).
  • Results: Total PDQI scores ranged from 8 to 35 (of a possible 35), with medians of 18 and 29 for the lowest- and highest-rated notes. Discriminant validity was present (p < 0.001), criterion-related validity was fair, and internal consistency was good (mean 0.84; range 0.69-0.94). However, interrater reliability was poor among both gold-standard (ICC 0.09, 95% CI -0.01 – 0.37) and test (ICC 0.12, 0.01 – 0.46) reviewers.
  • Conclusions: The PDQI-9 had fair validity and good internal consistency. Interrater reliability was poor within a small sample of notes. However, low interrater reliability raises questions about the PDQI-9 and physician preferences for handoff note content. Further evaluation of the PDQI-9 within a larger corpus of transfer notes could determine whether this tool has sufficient reliability to be used.
10 Lyons, Patrick MD
Washington University in St. Louis
Design thinking to create user-centered ICU-ward handoffs at three academic hospitals
Lyons PG; Rojas JC; Garcia B; Thomashow M; Santhosh L

  • Introduction: Patient transfers from the ICU to the wards are high-risk, and physician communication quality during these transfers is associated with patient outcomes. However, no best practices have been described for standard content or format in ICU-ward transfer notes. We aimed to use Design Thinking – an iterative, collaborative process for user-focused solutions – to develop a structured ICU-ward transfer communication tool for clinicians.
  • Methods: We conducted structured focus groups using Design Thinking methodology at three Internal Medicine residency programs in 2019. These voluntary focus groups included 4-10 2nd- or 3rd-year residents. With participants’ consent, we recorded and anonymously transcribed group discussions. We performed qualitative inquiry on transcripts for coding and thematic content analysis. Coding was performed independently by two coders using both a theory-driven (deductive) and data-driven (inductive) approach; disparate coding was reconciled via in-depth discussion between coders with a 3rd author serving as tiebreaker.
  • Results: Focus groups identified three main themes around limitations of current ICU-ward handoff practices: Patient Safety, Source of Clinician Burnout, and Lack of Standardization. From these limitations, participants identified specific attributes of an ideal handoff tool (Table 1), which coalesced around 3 main goals: (1) notes should include a succinct hospital course and prioritized problem list (“Quantity and Quality of Handoff Information”); (2) the tool should integrate into the electronic health record (EHR) in a customizable and updatable manner (“EHR Integration”); and (3) the tool must have standardized elements to reduce errors of omission (“Standardization of Components”).
  • Conclusions: Across 3 sites, Design Thinking focus groups described key limitations to current ICU-ward handoff processes and identified standard content perceived as necessary for effective and safe ICU-ward clinician handoff communication. Future work will involve the generation and iterative testing of a new user-designed tool to improve physician handoff communication and patient outcomes for patients leaving the ICU.
11 Lyons, Patrick MD
Washington University in St. Louis
Comparison of deep learning, machine learning, and penalized logistic regression for predicting clinical deterioration in oncology inpatients
Lyons PG; Li D; McEvoy CA; Westervelt P; Gage BF; Lu C; Kollef MH

  • Introduction: Oncology patients require intensive care unit (ICU) admission or die on the wards in over 9% of hospitalizations (twice the rate of general inpatients). Applying an accurate electronic early warning system (EWS) to these patients may overcome limitations to EWS predictive values; unfortunately, existing EWS are inaccurate in oncology patients. We aimed to develop an accurate oncology EWS by comparing three approaches to predicting deterioration using electronic health record data: traditional statistical regression, machine learning, and deep learning.
  • Methods: In this retrospective cohort study, we included all oncology ward hospitalizations at Barnes-Jewish Hospital from 1/1/14 to 6/30/17. The primary outcome was clinical deterioration (composite of ward death and ICU transfer). In a random 70% sample of admissions, we developed 3 models to predict clinical deterioration within a discrete-time framework of 6-hour windows: (a) elastic net (EN) logistic regression; (b) a gradient-boosted machine (GBM, “ensembles” of tree-based models); and (c) a deep neural network (DNN). Predictors included patient characteristics, vital signs, lab values, medications (including chemotherapy), and trends. We repeated 10-fold cross-validation 10 times. In the independent 30% (validation cohort), we compared model discrimination to each other and the Modified Early Warning Score (MEWS; a common EWS) using the area under the receiver-operating-characteristic curve (AUC).
  • Results: We evaluated 21,219 admissions from 9,058 patients. Clinical deterioration occurred in 1,965 ward stays: 1,425 ICU transfers and 540 deaths. The GBM model had the highest discrimination (validation AUC 0.91, 95% CI 0.89-0.93), followed by the DNN (0.81, 95% CI 0.80-0.83), and the EN (0.79, 95% CI 0.77-0.80). All were more accurate than the MEWS (AUC 0.65, 95% CI 0.64-0.67, p < 0.001).
  • Conclusions: To our knowledge, these are the first EWS models developed and validated specifically for oncology wards. All were more accurate than the MEWS, and GBM had the highest discrimination of the new models. These models require prospective evaluation and testing of associated interventions to determine if real-time use improves outcomes.
12 Riordan, Raven
Washington University in St. Louis
An exploratory study on the impact of mobile health technology to support the recovery of individuals with opioid use disorder (OUD)
Riordan R; Wilson B; Fentem A; Kasson E; Min C; Garcia M; Goodman M; Kaiser N; Li X; Cavazos-Rehg P

  • Introduction: Approximately 16 million citizens worldwide suffer from opioid use disorder (OUD). The opioid epidemic has resulted in widespread negative health outcomes among vulnerable populations, specifically in Missouri. While use of medication assisted treatment (MAT) within a comprehensive treatment plan is the current standard of care, individuals with OUD may struggle to remain motivated and encouraged during their recovery journey. To combat this, we have combined evidence-based content with innovative technology to develop the uMAT-R mobile application. uMAT-R is a theory-based digital therapeutic tool which aims to educate and motivate individuals with OUD throughout their comprehensive treatment plan, including their initiation and continued use of MAT. Digital therapeutic tools (e.g., mobile applications) have been shown to be accessible and low-cost interventions, and this study extends the use of such tools to counter MAT misconceptions and to support OUD recovery. We pilot tested the efficacy of our uMAT-R app using data collected from individuals who were recruited from opioid treatment facilities in Missouri. Specifically, we examined the feasibility, usability, and efficacy of the uMAT-R mobile app in supporting individuals with OUD throughout their recovery.
  • Methods: 119 individuals at opioid treatment facilities have been enrolled into the study thus far. All participants received access to the uMAT-R mobile application. Participants were asked to complete weekly online survey assessments regarding their experience using uMAT-R. After accessing uMAT-R for at least a month, participants were asked to complete a qualitative interview to garner feedback on potential areas for app improvement and engagement.
  • Results: 86% of participants with access to uMAT-R found it either very useful or useful to stay motivated to work on their recovery while 15% reported that the app was not applicable to their recovery. When asked if the app helped participants do what they could to keep themselves healthy, 80% reported the app was very useful or useful, 3% reported not very useful, and 17% reported that the app was not applicable to keeping themselves healthy. Participants were then asked to choose which mobile app features were found to be helpful during their recovery. Results indicated that the educational courses and in-app messaging with a coach were reported to be the most helpful features (59% and 56%, respectively). About one-third of participants reported that the tip of the day, goal setting, and event reminders (appointment scheduling or medication reminders) were helpful; 16% reported that the list of resources were helpful.
  • Conclusions: Our study suggests that digital therapeutics could motivate and educate individuals during their recovery from OUD. Findings following the use of uMAT-R, even in its prototype version, do signal significant public health implications for helping to support individuals with OUD with MAT initiation and adherence and throughout their recovery process. Additionally, participants provided suggestions for supplemental features to be added to our prototype app, such as a networking option within the app and a self-monitoring component. The incorporation of this participant feedback during mobile app development could significantly increase engagement and satisfaction with the tool, promoting further improvements in health and mental health outcomes for this population.
13 Tomov, Dimitre MSee
Washington University in St. Louis
Standardization of organization and encapsulation of processing for imaging data
Tomov DN; Metcalf NV; Cooley SA; Strain JF; Boerwinkle AH; Bowen CL; Ances BM

  • Introduction: The complexity of imaging data management and organization increases as they grow and are being processed. Researchers would like and should be able to focus on the pursuit of answers to the scientific questions at hand. Recently the Brain Imaging Data Structure (BIDS) was adopted as a data organization standard by the Ances Laboratory. Python classes were developed to automate the fine-grained download from XNAT (CNDA) by using the pyXNAT API and to convert to BIDS via the HeuDiConv container. Subsequently a containerized pipeline (Qu|Nex) was employed to quickly produce Myelin surface maps for HIV participants and controls. The BIDS standard and containerization of pipelines provide robust, encapsulated, compact and reproducible setup for imaging data preprocessing, processing to volume and surface maps and the subsequent statistical analyses.
  • Methods: pyXNAT API was used to fine-grain download imaging data from the CNDA instance of the XNAT database. They were automatically converted to BIDS by the HeuDiConv software in its containerized version. A containerized processing pipeline, Qu|Nex based on the Human Connectome Project pipeline was utilized to produce Myelin maps from the T1w and T2w scans of a 424 HIV infected and control participants. The ciftify and fmriPrep containerized pipelines and a non-containerized pipeline were installed on the same hardware.
  • Results: The uniform language approach through the use of Python allowed the fine-grained download imaging modalities and conversion to BIDS to be done in a single pass. The HeuDiConv container converted the dicom files to NIfTI and the associated in strict compliance with the BIDS specification. None of the tested pipelines produced output strictly following the BIDS-derivatives specification.
  • Conclusions: Availability of fine-grained download from XNAT and conversion to strict BIDS-raw in a seamless manner. Standardized data are convenient to share with collaborators and pipeline developers. Containerized pipelines isolate their own dependencies, don’t interfere with software on the host, coexist on the same host, can easily be sent to or received from collaborators.
14 Zhuang, Yan
University of Missouri—Columbia
Applying blockchain technology for health information exchange and persistent monitoring for clinical trials
Zhuang Y; Sheets L; Shae ZY; Shyu CR

  • Introduction: Health Information Exchange (HIE) is a process of sharing patients’ healthcare records among healthcare providers and patients. Timely HIE facilitates the meaningful use of health records such as decision supports and improving patients’ outcomes. However, there are multiple challenges that impede data collaboration across institutions such as data security, patient privacy, and time consumption. Data sharing issues also exist in the current clinical trial system. Data inconsistency in clinical trials results in the infamous “imprecision medicine” problem. Data falsification and human entry errors could happen between trial sites and trial sponsors or between trial sponsors and the FDA. Blockchain is a distributed ledger technology, considered to be a potential solution to the current HIE issues, ensuring data security, immutability, and user’s privacy. With the “Smart Contract” which is a programmable self-executing computer protocol added into the blockchain system, all the HIE transactions can be validated automatically.
  • Methods: We have proposed and implemented a private blockchain system to provide a feasible simulation for potential solutions to monitor clinical trials across different census regions persistently. Various levels of data access privileges have been designed to utilize a suite of customized Smart Contract settings. These settings emulate the workflow protocols for all the participants. We have also conducted a large-scale simulation of HIE across different healthcare facilities.
  • Results: We have simulated 1,553,635 HIE transactions in four months by running the scripts continuously. All the transactions have been successfully validated and exchanged the data in an average of 23.8 seconds. We have also intentionally falsified the transactions by tampering with the data, those transactions are detected and discarded automatically.
  • Conclusions: Our proof-of-concept work provides the informatics community with a prototype system that goes beyond planning and high-level discussions of blockchain in healthcare. We have shown that blockchain technology is able to facilitate an automatic validation process for HIE and clinical trials without third-party involvement.
Clinical Focus
Poster #Poster Details
15 Chandrasekaran, Vinay
Washington University in St. Louis
Imaging pipeline to identify functional mutations in genes of clinical significance
Chandrasekaran VD; Bramley JC; Kremitzki C; Waligorski J; Xu EX; Yenkin AL; Liebskind MJ; Buchser WB

  • Introduction: Although gene editing technologies for clinical applications are emerging, the question of which specific mutations cause disease phenotypes remains largely unknown – especially for diseases with polygenic phenotypes (e.g. glaucoma). To identify causative mutations in disease genes the Functional Imaging for Variant Elucidation (FIVE) lab uses an image analysis pipeline to identify and isolate cells with functional perturbations as caused by mutations.
  • Methods: To identify cells with functional mutations, a population of cells is perturbed with CRISPR-Cas9 to induce a range of mutations in a gene of interest. The pool of cells are then stained, imaged, and the perturbed cells are identified for isolation. Software performs feature engineering by analyzing imageable phenotypes and classifying cells as wild-type or mutated. Simpler phenotypes such as fluorescence intensity can be identified using straightforward algorithms, but complex phenotypes like mitochondrial properties are identified with machine learning algorithms. Cells of interest are isolated from the perturbed population using raft-based technology. To test pipeline efficacy, isogenic clones of mutated cells were mixed with isogenic wild-type cells. The genotypes of both cell groups were known. The FIVE imaging pipeline was used to unmix the cells, they were single-cell genotyped, and the unique sequence was used to determine sensitivity and specificity of the selected populations.
  • Results: The discriminatory power of the pipeline was measured by generating ROC curves from sensitivity / specificity measurements. The area under the curve (AUC) measures how well the mutated cells can be distinguished. In cases where the phenotype is defined by fluorescence (like unmixing cells that express GFP and RFP), the pipeline showed strong discriminatory power between populations with AUC of > 0.98. The pipeline has also been tested on more complicated phenotypes like mitochondrial dimensions, nuclear morphology, and cytoskeleton structure. AUC vary between 0.6 and 0.9!
  • Conclusions: The FIVE imaging pipeline can be used to identify and isolate cells with functional mutations.
16 Chininis, Jeffrey MEng
Washington University in St. Louis
Remote telemonitoring of patient prehabilitation activity predicts surgical outcomes
Chininis JA; Williams GA; Li D; Dai R; Lu C; Hammill CW

  • Introduction: Pancreatectomy is a specifically morbid and complex operation with post-operative complications ranging from 40-60%. Novel innovative approaches are urgently needed to address the morbidity associated with pancreatectomies. We have implemented a clinical trial leveraging machine learning technologies via off-the-shelf remote telemonitoring devices to predict poor surgical outcomes.
  • Methods: From March 2019 to August 2019 patients undergoing pancreatectomy were provided a remote telemonitoring device to be worn before surgery, during their hospital stay, and 30 days post-discharge. Postoperative complications were prospectively tracked. Remote telemonitoring devices collected 32 activity metric features that were applied to machine learning models and traditional statistical analyses to predict complications.
  • Results: 21 patients underwent pancreatectomy during the study period. The median age of participants was 61 (IQR 57–71) years with a majority being female (71%). Patients wore the remote telemonitoring device for a median of 17 (IQR 14–17) days before surgery, walked a median of 2.5 (IQR 2–3) miles per day, and were active a median of 36 (IQR 16–48) minutes per day. The median post-operative length of stay was 6 (IQR 4–11) days with 4 (19%) of the patients being readmitted. Post-operative morbidity included 13 (62%) patients experiencing a complication with 3 (14%) being severe. Support Vector Machine Learning algorithms outperformed all other analyses showing high accuracy (75%) and specificity (86%) in this preliminary training data to predict the onset of a surgical complication.
  • Conclusions: This preliminary analysis demonstrates that remote patient activity metrics can be effectively tracked. Using machine learning analysis we have efficaciously demonstrated the ability of this novel technology to predict surgical outcomes. Future refinement of our machine learning models can provide healthcare teams evidence based decisions to (1) determine preoperatively which patients are fit for surgery and (2) identify complications early in their course alerting healthcare providers to intervene.
17 Gupta, Aditi PhD
Washington University in St. Louis
Informatics-based sub-phenotyping of children with neurofibromatosis type 1 using electronic medical records reveals new clinical associations
Gupta A; Morris SM; Kim S; Foraker R; Gutmann DH; Payne PRO

  • Introduction: Neurofibromatosis type 1 (NF1) is one of the most common neurogenetic disorders, occurring in 1 of every 3,000 births, and affecting all races, ethnic groups, and both sexes. While establishing the diagnosis of NF1 is usually straightforward, it is currently not possible to accurately predict which medical problems a person with NF1 will experience during his/her lifetime. The study employs statistical and machine learning techniques using patient-level clinical features to sub-phenotype patients with NF1.
  • Methods: Statistical and informatics-based approaches were used to perform a longitudinal analysis of clinical features associated with NF1, followed by stratification of these features across demographic characteristics (sex, race, age) using data collected from an existing clinical database of individuals with NF1 in combination with data curated from the electronic health record (EHR). Using this information, we developed predictive models for identifying patients with optic pathway gliomas (OPG) and attention deficit hyperactivity disorder (ADHD).
  • Results: Data for 798 individuals were available in the NF1 Clinical Registry, and 3,451 individuals in EHR. Longitudinal analysis revealed new associations, including sexually dimorphic differences in several clinical features, such as ADHD and malignant peripheral nerve sheath tumors (MPNSTs). Prediction models performed well in our analyses, where their performance increased with the addition of clinical features. The ADHD models predicted the clinical diagnosis with 78% accuracy, while OPG had best performance of 88%.
  • Conclusions: The use of statistical and machine learning based techniques that combine features extracted from clinical databases and electronic health records can identify novel prognostic clinical markers, which can be used to develop predictive models for improving the management of children and adults with NF1.
18 Ingaiza, Lucy
Washington University in St. Louis
A qualitative analysis on adapting lessons from HIV systems of care to hypertension prevention and treatment
Ingaiza LM; McKay VG; Davila-Roman V; Mutabazi V; Baumann A; Brown A; Hooley C; Twagirumukiza M; Proctor E; Mutimura E

  • Introduction: In the country of Rwanda, pilot data suggest that 36% of adults have hypertension and 33% are unaware of it. Furthermore, there is inadequate information on infrastructure and capacity to engage in translational research for hypertension control in Rwanda.Often cited as an example of good health system responses to the HIV epidemic, our pilot study aimed to compare the systems of HIV and hypertension care and delivery for Rwandese patients in hopes of adopting a similar model to hypertension prevention and treatment.
  • Methods: We conducted a comparative case study of HIV and hypertension systems of care. Focus groups and observational data were collected in August of 2019 during the 3rd annual Noncommunicable Disease Symposium hosted in Kigali, Rwanda. Participants were recruited at the start of the symposium and included a national representation of Rwandan policy makers, academics, clinicians, practitioners and stakeholders (N=35). Questions centered around the current state of hypertension and HIV care in Rwanda, and the role evidence has in dissemination and implementation of these two epidemics. We used the Social Ecological and Health Belief models to develop a code-book and guide analysis.
  • Results: Participants presented barriers and facilitators to hypertension and HIV care at multiple levels. For example, in comparison to HIV, awareness among individuals was low for hypertension. At the institution and policy levels, the participants discussed a lack of research and priority setting as barriers for hypertension prevention and treatment. Lastly, participants presented recommendations on how approaches to HIV care could be applied to improve hypertension care.
  • Conclusions: Factors at multiple socio-ecological levels are related to how hypertension care is delivered and patient outcomes. Understanding the historical success of HIV care may provide direction for informing approaches to hypertension prevention and treatment.
19 Leary, Emily PhD
University of Missouri—Columbia
Translational informatics to improve clinical management of patients with hypertension
Leary EL; Manring ND; Crotty SM; Crockett EE; Li J; Emter CA

  • Introduction: Hypertension is a complicated expression of various factors which include heart function as well as blood flow and pressures. It is often difficult to identify the root cause for high blood pressure in a specific patient. For these patients, careful clinical monitoring of cardiac function is critical for their healthcare and these measurements, peripheral resistance and aortic stiffness, cannot be obtained in current clinical practice. In this study, a more comprehensive computation algorithm was developed to express hemodynamic profiles, which will characterize peripheral resistance and aortic stiffness, allowing individualized management for patients with hypertension.
  • Methods: We refined the ability to characterize the hemodynamics of the cardiovascular system using a fluid mechanics version of the Windkessel model and an average value of the aortic blood flow. Pre-clinical swine data were collected to develop characterizations for peripheral resistance and aortic stiffness measures. A human clinical data repository was created which contained echocardiogram results and longitudinal health data from patients (hypertensive and non-hypertensive) seen in our health system over a period of three years. This extensive collection of pre-clinical data and patient profiles from the human clinical data repository were used to validate our hemodynamic model.
  • Results: Results showed an expected correlation between vascular resistance and capacitance, with hypertensive patients demonstrating an increase in resistance and a decrease in capacitance, compared to healthy controls. This indicates that our model describes what would be biologically expected from non-hypertensive vs hypertensive patients.
  • Conclusions: Our algorithm can describe cardiovascular measures which have been shown in the literature to be exceptional predictors for adverse cardiovascular events, and which are currently unavailable in current clinical practice. Using translational informatics, we created bench-to-bedside outputs to better describe patient cardiovascular function using features easily obtained from clinical testing procedures, i.e. echocardiogram. This process improves patient monitoring and clinical documentation, decreases costs to patients and will facilitate individualized patient care.
20 Lenard, Emily MSW
St. Louis College of Pharmacy
New methods for clinical trials recruitment: Do EPIC and Facebook improve recruitment of older adults?
Lenard E; Stephens ME; Pennock S; Lenze EJ

  • Introduction: Recruiting participants for clinical trials is a well known challenge. Failure to recruit adequate and diverse samples reduces the ability to draw meaningful conclusions from research data. More novel recruitment methods, including EPIC electronic health records-based and social media-based techniques, could improve both the overall recruitment and the diversity of participants in clinical trials. While electronic health records-based recruitment has been shown to be a viable option for recruiting individuals with a specific diagnosis, this study seeks to determine feasibility for recruiting patients with depression, a common diagnosis that is inconsistently documented in patient charts. This study examines novel recruitment methods, including EPIC electronic health records-based and social media-based techniques, in a clinical trial of older adults with depression.
  • Methods: In the first 2 years of recruitment, Optimum relied on referrals from primary care and mental health providers. These recruitment methods were not sufficient to meet the study’s recruitment goals. The research team created a Facebook advertising campaign to increase referrals. The research team also began to utilize informatics tools, namely the Epic Research Alert and MyChart messaging in order to reach more patients efficiently and cost-effectively. As well, the team used traditional newspaper advertising.
  • Results: Traditional physician referrals accounted for the largest number of referrals and patients randomized, followed by Facebook advertising. Print advertising and posters generated the smallest number of referrals but had the highest proportion of patients who were randomized. The Epic Research alert was implemented later in the trial (May 2018) and the proportion of referrals who were randomized was comparable to traditional physician referrals. Epic MyChart messaging was recently implemented (June 2019) and has led to more referrals than print advertising and posters, but has a lower proportion of referrals randomized compared with other methods.
  • Conclusions: Informatics tools for research recruitment (Epic Research Alert and MyChart messaging) are promising strategies for clinical trial recruitment. These tools are well-accepted by providers and patients, while being convenient and cost-effective for researchers. These tools are most effective for recruiting participants who are already receiving care in-system, and for studies with eligibility criteria that are consistently documented in the medical record. Informatics tools may need to be supplemented with other strategies in order to meet recruitment goals. Facebook and print advertising is effective in reaching people who are not already receiving care in-system. These methods are also helpful for studies of symptoms or conditions that can be self-identified but may not be consistently documented in medical charts. This study adds to our knowledge because it is applicable to mental disorders such as depression and other studies of symptoms or conditions that may not be consistently documented in medical charts.
21 Li, Yu
University of Missouri—Columbia
Deciphering visual diagnosis processes for evidence-based medical image interpretation
Li Y; Shyu CR

  • Introduction: Visual reasoning processes are critical in medical image interpretation such as radiology and pathology. But they are mostly subconscious and rely on experience in such high demanding tasks, which makes it difficult to explain and educate the expertise. Moreover, there can be diagnostic discrepancies between specialists. It is urgently needed to computationally capture and model the reasoning processes for evidence-based medical image interpretation. Using eye tracking, we can reveal the real-time visual attention and locate important regions. However, most analysis methods focus on quantitatively differentiating visual behaviors but rarely produce interpretable results which is necessary for visual reasoning understanding and radiologist/pathologist education. Without extracted and organized eye movement sequences, it is difficult for specialists to explain the reasoning processes behind diagnosis.
  • Methods: Based on the observation that the process of medical image interpretation normally consists of multiple tasks, we propose the concepts of common and contrast visual reasoning patterns. With adapted frequent and contrast pattern mining algorithms, our method can highlight the common and unique visual activities that fulfill a task for certain groups of viewers. The patterns can be directly mapped back to the original medical image for specialists to further examine the visual features and explain the reasoning processes. We also design spatial and temporal distances to quantitatively measure the reasoning processes and help specialists to standardize evidence-based medical image interpretation.
  • Results: We conducted experiments with expert and novice radiographers, and the patterns are reviewed and explained by a senior expert. The results show several unique reasoning processes shared by experts while rarely exhibited by novices, which highlight the knowledge gap in high-level tasks with subtle visual cues.
  • Conclusions: The proposed method has shown its potential for explaining and educating medical image interpretation. The explainable results are essential to help specialists to understand and standardize the interpretation processes which are mostly subconscious and rarely articulated.
22 Raju, Murugesan PhD
University of Missouri—Columbia
Pattern discovery and contrast data mining for glaucoma risk assessments
Raju M; Liu D; Shanmugam KP; Shyu CR

  • Introduction: Glaucoma is the second leading cause of irreversible blindness across the world, about 70 million people have glaucoma and about 4.4 million people are blind due to undiagnosed glaucoma by optic nerve damage worldwide. To address this problem, we applied a subgroup contrast set mining for glaucoma risk assessment. Contrast mining has been successfully applied in health care data analytics and demonstrated in recent work from our lab using a large volume of EHR (electronic health records) data analysis. The main goal of this method is to identify patterns within a subset of a dataset that shows an interesting behavior concerning selective attributes associated with one group but not other groups. Certainly, the investigation of potential risk factors associated with the development of glaucoma could provide a preventive measure to control the diseases.
  • Methods: Cerner Health Facts HER database was used for this study. We used international classification of diseases diagnostic codes for retrieving glaucoma related cases from 2001 to 2015 with the inclusion and exclusion criteria of this study. The EHR data was sliced at one-year intervals and used as input data for contrast mining. A deep exploratory mining process was applied that identifies small homogenous sub-data from a large, diverse and heterogeneous dataset by using floating and Path Expansion, contrast pattern mining and prioritization of subgroups based on J-value.
  • Results: The outcome of the analysis shows high contrast patterns with hypertension, alcohol use and African American race with glaucoma condition compared to non-glaucoma condition (p -value < 0.05) The detailed results will be presented during the poster session.
  • Conclusions: The above results demonstrate that alcohol use and hypertension are significant risk factors for the development of Glaucoma conditions. However, the risk associated with hypertension along with alcohol use is three-fold higher among the African American race compared to other races. This method can be applied to many other diseases for risk assessments.
23 Said, Abdullah
Washington University in St. Louis
PROMISing information – Cognitive function and stigma in patients with single suture craniosynostosis
Said AM; Skolnick G; Niadoo SD; Smyth M; Patel K

  • Introduction: Children with craniofacial anomalies are often stigmatized and at greater risk for psychosocial disturbances and cognitive impairment than their peers. Studies have shown that children with craniosynostosis have lower appearance ratings and modest differences in neurodevelopment and intelligence compared to healthy children. To date, there is no reliable, convenient, and validated method to measure psychosocial parameters in children with nonsyndromic single suture craniosynostosis (SSC). The purpose of our study is to use the Patient Reported Outcomes Measurement Information System (PROMIS) to evaluate the perception of stigma and cognitive function in nonsyndromic SSC patients.
  • Methods: Stigma and cognitive function were measured in 42 consecutive patients, 5 years and older, presenting to clinic with repaired nonsyndromic SSC using the National Institute of Health’s PROMIS questionnaire with transformed scores normalized for the general population (mean ± standard deviation, 50 ± 10). Questionnaires were administered as part of clinical care via Research Electronic Data Capture (REDCap) from July 2018 to May 2019. Scores were automatically transferred to electronic medical records. Parents were asked to fill out the Parent Proxy Cognitive Function Questionnaire if the child was under 8 years old (n=21). The child’s responses were entered for the stigma questionnaire regardless of age. Computerized-adaptive testing (CAT) was utilized to reduce survey burden and improve sensitivity. As a result, the number of questions answered by each participant varied.
  • Results: Forty-two patients (50 % male, 39 white patients, 2 African-American patients, and 1 Asian patient) were treated for nonsyndromic SSC and participated in this study (18 sagittal, 12 unicoronal, 11 metopic, and 1 frontosphenoidal). The average age at follow-up was 8.4 years (range 5 -18 years). Our cohort had equivalent cognitive function scores (mean 51.8±10.2, 95% Cl [48.7-55.0], p < 0.001) when compared to children without SSC. Children with repaired SSC perceived less stigma (41.7±6.8, 95% Cl [39.6-43.8], p < 0.001) compared to healthy children. There was a significant negative correlation between the two scales (Spearman’s rho= -.629, p < 0.001).
  • Conclusions: Based on patient-reported outcomes, children with repaired SSC have equivalent cognitive function and feel less stigmatized than healthy controls. Lower perceptions of stigma were associated with higher cognitive function scores. These results do not align with previously reported in-depth assessments in patients with repaired SSC which found small but significant deficits in intelligence and small but significant elevations in perceived stigma. Future research is needed to explain the discrepancy. Nevertheless, PROMIS questionnaires via CAT offer a convenient, validated method of measuring psychosocial parameters in children with SSC which might otherwise be difficult to obtain during standard follow-up visits.
24 Stoneking, Faith MA
Washington University in St. Louis
Implementation of ELEVATE (Electronic Health Record-Enabled Evidence-Based Smoking Cessation Treatment) increases smoking cessation treatment and abstinence rates among cancer patients
Stoneking FS; Ramsey AT; Smock N; Chen J; Chiu AC; Bierut LJ; Chen LS

  • Introduction: Quitting smoking after a cancer diagnosis positively modifies responses to treatment and lowers the risk of developing secondary cancers, but is often unaddressed in cancer care due to time and resource constraints. This study provides data on the effectiveness for a low-burden point of care treatment model that offers a low-burden approach to delivering evidence-based treatment to cancer patients who smoke, developed through the Cancer Center Cessation Initiative (C3I) Cancer Moonshot program.
  • Methods: Electronic Health Record-Enabled Evidence-Based Smoking Cessation Treatment (ELEVATE), a point-of-care smoking cessation strategy, was implemented in June of 2018. One pre- and two post-implementation six-month blocks were used to assess treatment and abstinence rates of the smoking cessation program across two departments, one with the intervention and one control.
  • Results: One pre- and two post-implementation six-month blocks for Medical Oncology (MO) were analyzed and compared to Surgical Oncology (SO). Treatment: The percentage of MO patients treated for smoking increased from 1.6% to 27.5% (Ns=808; 2,466, z=25.86, p<0.0001) to 42.8% (Ns=2,466; 2,433, z=11.36, p<0.0001). The percentage of MO patients treated for smoking is significantly greater than that of SO post-implementation block one (27.5% vs 11.8%; Ns=2,466; 575, z=9.70, p<0.00001) and block two (42.8% vs 6.6%; Ns=2,433; 499, z=24.18, p<0.00001). Abstinence: The percentage of MO patients quit from smoking increased from 12.0% to 17.1% (Ns=808; 2,466, z=3.72, p=0.0002) and percentage of MO patients quit from smoking post-implementation is significantly greater than that of SO (17.1% vs 9.9%; Ns=2,466; 575, z=4.94, p<0.00001). Quit rate data for the second post-module block is still being collected, but we anticipate quit rates within MO will be higher than SO.
  • Conclusions: Implementation of ELEVATE leads to greater treatment and cessation rates among cancer patients. The effects of the module are sustainable, as indicated by the post-intervention data. Implementation of the module could allow for increased treatment and quit rates in multiple populations and those with lower access to care.
26 Tetteh, Emmanuel MD
Washington University in St. Louis
Persistence vs de-implementation of HIV interventions: A mixed-method analysis of the reasons why
Tetteh EK; Mckay VR; Combs TB; Reid M

  • Introduction: Low-value HIV interventions continue to persist amidst growing evidence against their effectiveness. Organizations providing these interventions have continued to do so for of a number of reasons. We conducted a mixed-method study with organizations providing HIV prevention services to understand reasons influencing organizations to continue or end interventions.
  • Methods: Organizations were recruited from the CDC’s website gettested.org and were eligible to participate if the organization had provided at least one of 37 HIV prevention interventions identified as inefficient by the CDC services, in metropolitan areas with the highest HIV incidence. One staff member with intervention oversight was asked about intervention implementation and the decision to de-implement or continue the intervention. 877 organizations were recruited with a response rate of 66% (n= 578). Twenty-four percent (n=213) of organizations met the eligibility criteria, and 188 organizations from 19 cities participated in the survey. We conducted a mixed-method analysis of closed- and open-ended questions using qualitative results to elaborate on quantitative results.
  • Results: Organizations reported 359 instances implementing low-value interventions. Of interventions implemented, only 46% had been de-implemented (n=167). Funding and client demand for services were the primary driving factors when deciding to either continue or end an intervention. Our qualitative results elaborated on quantitative results suggesting programs were continued if client demand for services that they liked or found useful was sustained or if staff perceived interventions as improving client behavior and health outcomes. Scientific evidence was a rarely reported reason for continuing or de-implementing.
  • Conclusions: The decision to continue or end interventions is influenced by a number of factors regardless of scientific evidence. Facilitating a smooth de-implementation process of low-value interventions should consider reasons reported by practitioners.
27 Walsh, Ryan MSOT
Washington University in St. Louis
Acceptance of passive digital biomarker monitoring via mobile health among stroke survivors in the St. Louis area
Walsh RJ; Santos H; Lee Y; Lau SCL; Baum CM; Wong AWK

  • Introduction: Stroke survivors face barriers to participation and activity engagement after discharge from institutional care. Passively monitoring digital biomarkers of participation and activity engagement (e.g., distance traveled or activity levels) via global positioning systems (GPS) or motion sensors in mobile health (mHealth) may enhance community care. However, we know little about characteristics of stroke survivors that predict acceptance of passive digital biomarker monitoring. Therefore, our aim is to describe characteristics of stroke survivors that predict acceptance of passive digital biomarker monitoring via mHealth.
  • Methods: We completed a phone survey of 135 St. Louis area volunteers from a stroke registry. We analyzed two logistic regression models to determine which characteristics of stroke survivors are more likely to predict acceptance of motion sensors or GPS for passive digital biomarker monitoring via mHealth. We gathered information about sociodemographic factors as well as use of and access to technology.
  • Results: Sociodemographic factors such as age, race, and education level were not significant factors predicting acceptance of motion sensors and GPS. Factors such as desiring more mHealth services and having greater access to wireless internet in the community were significant factors predicting acceptance of motion sensors and GPS. Acceptance of more mHealth modes of delivery was a significant factor predicting acceptance of motion sensors but not of GPS. Factors within both models accounted for over 40% of variance and correctly classified over 77% of participants as accepting or rejecting either passive digital biomarker monitoring via mHealth.
  • Conclusions: Stroke survivors’ access to and use of technologies predict acceptance of passive digital biomarker monitoring more strongly than sociodemographic factors. Thus, institutional and community stroke service providers may consider better understanding how stroke survivors, regardless of age, race, and education level, access and use technologies in their everyday lives to determine who may accept passive digital biomarker monitoring via mHealth to optimize participation and activity engagement.
28 Wheelock, Muriah PhD
Washington University in St. Louis
Network level analysis tools elucidate functional connectivity network disruption underlying domain specific impairments in attention for children born very preterm
Wheelock MD; Lean RE; Bora S; Austin NC; Melzer TR; Woodward LJ; Eggebrecht AT; Smyser CD

  • Introduction: Despite increased rates of ADHD in very preterm (VPT) children, the neural correlates underlying these attention problems remain poorly understood. We sought to understand the neural correlates of aberrant attention processes using an innovative new analysis technique.
  • Methods: Neuroimaging and attention assessments were analyzed from an existing longitudinal sample of 123 children, including 58 VPT (mean gestational age=28 wks, scan age=12.3 yrs) and 64 full-term (FT; mean gestational age=39 wks, scan age=12.2 yrs) children. Resting state-functional MRI data were acquired on a 3T GE scanner. Neuroimaging data were pre-processed and corrected for motion artifacts (FD<.2mm), with 5 minutes of low-motion data required for a child to be included in analyses. Selective, sustained and executive attention were assessed with the Test of Everyday Attention-Children (TEACh). VPT children obtained poorer sustained and executive attention scores than FT children (p<.05). Associations between attentional performance and functional connectivity (FC) were assessed within and between groups using a novel Network Level Analysis (NLA) software. Briefly, NLA used non-parametric correlations to determine associations between FC and TEACh scores for FT and VPT children. Next, network level associations with attention measures were determined using chi-squared and hypergeometric tests and network significance was determined via randomization (p<.05).
  • Results: NLA revealed that FT, but not VPT children, demonstrated expected brain-attention associations involving frontoparietal (FPN), dorsal attention (DAN), and cingulo opercular (CO) networks. In contrast, VPT children exhibited stronger attention associations with ventral attention and visual networks.
  • Conclusions: These results support the importance of FPN, DAN and CO networks in regulating attention in typically developing children. They also highlight differences in functional brain systems underlying attention and executive function impairments in VPT children. NLA software is an informative tool for determining the biological mechanisms underlying the development of attention and understanding associated attention deficits and clinical outcomes.
29 Lee, Yejin MSOT
Washington University in St. Louis
Understanding patient engagement intervention for persons with chronic diseases: Systematic review and meta-analysis
Lee Y; Kang E; Walsh R; Wong AWK

  • Introduction: Chronic diseases are long-lasting conditions; individuals may need to go through long periods of treatment and have to maintain self-care behaviors. To endure long periods of treatment, it is important to enhance patients’ motivation. Patient engagement is a key concept enhancing motivation of persons with chronic diseases. In order to design the optimal patient engagement intervention (PEI) for persons with chronic diseases, the components of PEIs and their effects should be investigated. Therefore, this review aimed to investigate the components and effects of PEIs for persons with chronic diseases.
  • Methods: Randomized controlled trials (RCTs) were included via searching the following databases: Ovid Medline, Embase, CINAHL, and Cochrane Central Register of Controlled Trials (CENTRAL). Effect sizes were estimated by calculating the standardized mean differences (SMDs) or mean differences (MDs).
  • Results: Of 6114 identified studies, eight RCTs were included. The findings showed that PEI is a multi-dimensional and component approach. The majority of included studies included three dimensions (educational, behavioral, and affective) with behavioral dimension as most. The most commonly applied components for PEI were verbal educations and reinforcement. All included studies showed beneficial effects of PEI on outcomes. The effect size on independence of activities of daily living was statistically significant in favor of PEI (MD [Fixed], g: 6.07 95% CI: 0.77 – 11.38; p = 0.02; I2 = 12%) compared to control conditions. No SMDs were significant on self-efficacy.
  • Conclusions: Our findings proposed that PEI is a multi-dimensional and component intervention. This intervention might have a beneficial effect on independence of daily living compared to usual care in persons with chronic diseases. Our findings could work as a good resource for health professionals to design the optimal PEI for chronic diseases. However, further studies are required to examine individual effects from each dimension and component consisting of PEI.
Cancer Focus
Poster #Poster Details
30 Al-Taie, Zainab Master in Computer Science
University of Missouri—Columbia
Drug repositioning and subgroup discovery for precision medicine implementation
Al-Taie Z; Liu D; Papageorgiou C; Mitchem J; Shyu CR

  • Introduction: Colorectal cancer (CRC) is the second leading cancer killer in the US. One critical issue is that patients may have a different reaction to the same drug. Because of this, the drug discovery process needs to be more individualized, using principles of precision medicine (PM). However, de novo drug discovery is a time-consuming and high-cost process with a low success rate. Drug repositioning (DR), or the ability to repurpose existing FDA approved therapeutics for the treatment(s) of additional diseases, takes advantage of existing drug therapies that have been approved for human use. To marry precision medicine with drug repositioning, in this study, we have developed a computational method using novel patient cohort stratification and DR to identify therapeutic options for patients with CRC.
  • Methods: CRC patients’ genotypic and clinical data were downloaded from The Cancer Genome Atlas (TCGA). Patient subgroups were identified using contrast mining and network analysis using a recently developed novel computational method. This analysis was integrated with a drug repositioning database containing genes targeted/affected, drugs, pathways, diseases, and symptoms, among others, to address the heterogeneity of the biological system. Drugs were suggested for repositioning for each subgroup based on their unique biomedical characteristics.
  • Results: For each subgroup, our approach results in a network with heterogeneous biomedical components, including different drugs that are more related to a given subgroup. We found seven subgroups among CRC patients. Most of the top-recommended drugs for each subgroup were currently utilized for therapy in malignancies, including CRC. This indicates the potential of our method to identify drugs currently used in other malignancies with potential utility in patients with CRC that are not currently utilized.
  • Conclusions: Subgroup discovery is an essential step toward applying PM in the healthcare system. Data mining and network analysis-based subgrouping and DR have a promising impact on improving patient care by enabling tailoring of the drugs for a group of patients with similar characteristics that differentiate each subpopulation from the whole disease population.
31 Hummel, Justin Master of Science
University of Missouri—Columbia
Network analysis reveals BTK as a critical link between myeloid and T cells in TLR signaling in colorectal cancer.
Hummel JJ; Shen Y; Innokenteva I; Papageorgiou C; Shyu CR; Mitchem JB

  • Introduction: Colorectal cancer (CRC) is the second-leading cause of cancer mortality in the US today. Recent advances in immunotherapy have only been shown to benefit a limited subgroup of patients with CRC. In other malignancies, activation of Toll-like receptors (TLRs) has been shown to overcome resistance to immunotherapy, such as immune checkpoint inhibition (ICI). In this study, using publicly available data and informatics-based analysis, we identified BTK as a critical link between TLR signaling and T cells in the CRC tumor microenvironment.
  • Methods: Using RNA-seq data from The Cancer Genome Atlas (TCGA) and the Microenvironment Cell Populations (MCP)-Counter, abundance scores were generated for the tumor microenvironment of each patient. A curated TLR gene panel was generated using Reactome™ and GO™. Pearson analysis was used to evaluate each pairwise combination of genes and cell-types. Significance was determined by the correlation coefficient, r ≥ | 0.7 | with a p-value < 0.05. Network analysis was performed using the Girvan–Newman algorithm to establish critical connections across these features.
  • Results: After establishing a 453 gene TLR panel and creating MCP-Counter scores, correlation analysis demonstrated strong correlations between 54 different genes and 7 cell-types. As expected the most genes were associated with ‘monocytic lineage cells’ (30) and ‘myeloid dendritic cells’ (7). Only 5 genes were significantly associated with ‘T cells’. Genes and cell-types that were highly correlated were then further analyzed for network association. From this analysis, BTK was identified as a critical edge acting as the primary link between ‘myeloid cells’ and ‘T cells’.
  • Conclusions: Developing novel strategies for the treatment of CRC is critical and immunotherapy represents an area ripe for advancement. Informatics based analysis combined with publicly available data provides us with an opportunity to shape pre-clinical and translational studies. Using this approach, we have identified BTK as a critical link between myeloid cells and T cells in the tumor microenvironment in CRC. Further studies in our laboratory will focus on confirming these findings for translation into patients.
32 Kovalenko, Mikhail MS
University of Missouri—Columbia
Classification of monocytes and their precursors using pre-trained CNNs
Kovalenko M; Kholod O; Hammer R; Shin D

  • Introduction: The correct identification and distinction of cell populations of monocytic lineage plays a crucial role in diagnosing hematopoietic diseases. However, high similarity among monocytic precursors presents a challenge even for experienced pathologists who use additional non-morphologic diagnostic methods to ensure the avoidance of a diagnostic error. Given the recent advances of machine learning in detection and classification of images, as well as extracting characteristic features that cannot be captured by conventional methods, we investigated the ability of three popular CNNs to match or exceed the most recently published concordance rate among pathologists.
  • Methods: We obtained 136 images of monocytes and monocyte precursors from patients with chronic myelomonocytic leukemia. They required pre-processing and augmentation to ensure best quality and sufficient quantity for training. We utilized pre-trained models of Alexnet, GoogLeNet, and Inception V3 in Matlab with Adam and sgdm optimizers. The networks were trained for a maximum of 100 epochs.
  • Results: Pre-processing of training images had to eliminate artifacts added by the capturing equipment such as outline frames and shading. The networks reached validation accuracy between 58.54% and 63.41%. The models did not seem to take advantage of all morphological features present in the images, particularly N-C ratio and chromatin patterns that are among the key distinguishing features of monocyte classification. Ongoing research is focusing on the optimization of training parameters and additional pre-processing of images to better reflect the morphologic features not captured by convolution filters.
  • Conclusions: The models in their default configuration did not achieve the expected improvement over the published concordance rate of 76.6%. The number of training images was likely insufficient although cell images are easy to augment by rotating and reflecting. Customized models that offer better accuracy would assist pathologists in eliminating diagnostic pitfalls when diagnosing chronic and acute leukemia.
33 Miller, Amanda
University of Missouri—Columbia
Identification of somatic mutations as novel biomarkers for non-small cell lung cancer by whole-exome sequencing
Miller AA; Barbirou M; Manjunath Y; Raju M; Staveley-O’Carroll KF; Li G; Avella DM; Tonellato PJ; Warren W; Kaifi JT

  • Introduction: Non-small cell lung cancer (NSCLC) leads to the majority of deaths attributable to cancer worldwide and is significantly associated with somatic genetic alterations that in most cases are attributed to tobacco exposure. To uncover the mutational landscape responsible for NSCLC, we conducted a whole-exome analysis approach to detect susceptible genetic variants associated with NSCLC by exome sequencing in tumor and peritumoral unaffected healthy lung tissues from patients with different stages of NSCLC.
  • Methods: We performed whole-exome sequencing in 41 NSCLC patients. Thirty-six primary tumor tissue and their peritumoral healthy lung tissue samples and five primary tumor tissue without matched peritumoral healthy lung tissue samples were collected. DNA was extracted, and DNA libraries were prepared for whole-exome sequencing using the Illumina NovaSeq6000 Genome Analyzer. Somatic single nucleotide variants and insertions/deletions were called using MuTect2 from the Genome Analysis Toolkit 3.8-1. Variants that passed MuTect2 filters were annotated using web ANNOVAR, and variants with allele frequency < 0.01 in 1000 Genomes and the Exome Aggregation Consortium were retained.
  • Results: The 41 NSCLC patients had the following distribution of clinical tumor stages: stage I, 25; stage II, 10; stage III, 4; stage IV, 2. After somatic variant calling and filtering, 24,442 unique and 24,615 total somatic variants located within 11,710 genes across the 36 tumor-peritumoral paired and five tumor unpaired samples were identified. When examining only variants annotated as exonic, there were 12,367 unique and 12,461 total somatic variants located within 6,980 genes.
  • Conclusions: The study provides insights into mutational processes, cellular pathways and gene networks associated with NSCLC. Several highly mutated genes identified in our study may be promising druggable targets in cancer therapy including TTN, RYR2, MUC19, TP53, CSMD3 and MUC16.
34 Shen, Yuanyuan MD
University of Missouri—Columbia
Regulation of gene expression by DNA methylation with cytotoxic T lymphocytes evaluation in consensus molecular subtypes of colorectal cancer
Shen Y; Hummel J; Trindade IC; Papageorgiou C; Shyu CR; Mitchem JB

  • Introduction: Low cytotoxic T lymphocyte (CTLs) infiltration and poor immunogenicity in colorectal cancer (CRC) microenvironment is a challenge to treatment with immune checkpoint (ICI) inhibitors. One group of patients, those with microsatellite instability-high (MSI-H), have demonstrated the response to this therapy thought largely to be due to high levels of CTL infiltration. In the recent molecular classification of CRC these patients are in the consensus molecular subtype (CMS) 1 group. Epigenetic modification by DNA methylation plays a critical role in gene expression and resistance to therapy. In this study, we compare gene expression, DNA methylation, and CTL infiltration in CMS1 patients to other subtypes (CMS2-4) to identify targets for improving CRC patient treatment.
  • Methods: The Cancer Genome Atlas (TCGA) RNA-seq (n=511) and DNA methylation (n=316) databases were used to determine gene expression and methylation profiles based on molecular subtypes. CMS1 was used as a reference and compared to other subtypes (CMS2-4) due to high levels of immune cell infiltration. Microenvironment Cell Populations-counter (MCPcounter) was used to determine tumor CTL infiltration. Genes with significantly different expression (p<0.01, LogFC≥|1.5|) and difference of mean methylation β value ≥|0.25| were integrated for Pearson correlation coefficient analysis with CTL MCPcounter score (r>|0.7|).
  • Results: Comparing CMS1 and CMS2 patients, three genes were differentially methylated and correlated with CTL scores (ARHGAP9, TBX21, LAG3) with ARHGAP9 and TBX21 decreased and hypomethylated in CMS2 patients. Comparing CMS1 and CMS3 tumors, five genes (ARHGAP9, TBX21, FMNL1, HLA-DPB1, STX11) were downregulated in CMS3 patients and highly correlated with CTL scores; however, ARHGAP9, FMNL1, HLA-DPB1, STX11 were all hypomethylated and TBX21 was methylated in both CMS1 and CMS3, but in CMS1 had a higher methylation ratio. Finally, when comparing CMS1 and CMS4, TBX21 was the only gene downregulated and hypomethylated in CMS4 patients and highly correlated with CTL scores.
  • Conclusions: We found six genes differentially expressed, differentially methylated, and highly correlated with CTL infiltration when comparing CMS1. Specifically, TBX21 was the only gene highly correlated with CTL scores and differential gene expression and methylation in CMS2-4 when compared to CMS1. Thus, T-bet may be the critical regulator of T cell responses in CRC.
Institutional Resources
Poster#Poster Details
35 Biostatistics, Epidemiology, and Research Design

BERD is the Biostatistics, Epidemiology, and Research Design core for the ICTS. We assist ICTS investigators and scholars with many tasks. Primarily we are here to help with data analysis, design of clinical/translational research, protocol preparation, manuscript preparation and analytic techniques. Our BERD members become an integral part of the research team on projects. Our consultations start with a basic assessment of your needs to facilitate more extensive ongoing support.
36 Bernard Becker Medical Library

Becker Medical Library maintains a world-class collection of information resources, accessible anytime, anywhere. Staff offer expertise in research computing, data management and sharing, publishing, author analytics and support, grant application and compliance, science communication, plain language review, information retrieval and evaluation, and information management. Classes on these topics and more are offered regularly, and customized presentations are available for departments, groups, or as individual consultations. Send questions big or small to askbecker@wustl.edu. Find Becker on Twitter, Facebook and Instagram using @BeckerLibrary.
37 Bioethics Research Center

The Bioethics Research Center (BRC) provides ICTS members with education and consultation services. The BRC offers a variety of educational programming including a Responsible Conduct of Research (RCR) course and a national workshop designed for investigators who have had difficulties with research compliance, The Professionalism and Integrity (P.I.) Program. Additionally, BRC faculty serve as course masters for the research ethics courses within the WUSM Master of Science in Clinical Investigation and the Master of Science in Biostatistics programs. BRC faculty are available for consults on matters of research ethics and the design of research on ethical issues in healthcare and research. BRC makes available to researchers a variety of validated tests and survey measures that explore professional decision making, knowledge of research ethics and good clinical practice, values in science, and attitudes toward compliance, genomics, precision medicine, and AI in healthcare. Please contact us at brc@wustl.edu or 314-747-4220.
38 Center for Community Health Partnership & Research

The Center for Community Health Partnership & Research (CCHPR) is a joint entity of the Institute of Clinical and Translational Sciences and Institute for Public Health. With direction from Dr. Angela Brown and Dr. Vetta Sanders Thompson, the mission of CCHPR is to reduce disparities and improve health and wellness in the region by fostering and supporting meaningful engagement between communities and researchers. CCHPR can assist with developing key stakeholder and community engagement activities, from planning through dissemination. Submit a consultation request here.
39 Clinical Research Training Center

The Clinical Research Training Center (CRTC) is the educational and training core of the Institute of Clinical and Translational Sciences (ICTS) that provides clinical and translational research training for undergraduate students, predoctoral students, house-staff, postdoctoral scholars, fellows, staff and junior faculty. The CRTC provides full-circle assistance through pre-award grant preparation, post-award training grant management, and didactic coursework that can lead to a Certificate or Master’s Degree in Clinical Investigation. For more information on how we can assist with your pre-award grant preparation, or any of our degree or training programs, please visit our website.
40 Clinical and Translational Imaging Unit

The CTIU is comprised of three imaging facilities at Mallinckrodt Institute of Radiology (MIR): The Center for Clinical Imaging Research (CCIR), the Radiology Clinical Research Core (RadCore) and the Cyclotron Facility and Nuclear Pharmacy. The CCIR is a research-dedicated biomedical imaging center and the only one of its kind in the U.S. The facility provides the latest in advanced imaging technology, equipment and expertise to support basic and translational inpatient and outpatient clinical research. RadCore provides a wide range of services to support clinical research needs of Washington University investigators, including research coordination, federal and local regulatory submission support, research program development, grant and protocol design, clinical study and financial management. The Cyclotron Facility and Nuclear Pharmacy provides radionuclides and standard and novel radiopharmaceuticals to the research and medical communities. Its staff is experienced in Food and Drug Administration (FDA) Investigational New Drug (IND) submissions for first-in-human translation. All three facilities are under the direction of Pamela K. Woodard, senior vice chair and division director of radiology research facilities at MIR.
41 ICTS Precision Medicine Function

The goal of the Precision Medicine function is to streamline precision medicine research processes across the university. Three separate initiatives supported this effort: (1) outreach to community and faculty members to assess interest and concerns regarding participation in precision medicine research, (2) convening of key stakeholders to develop a standard genomic consent and institutional database to store and integrate electronic health record and genomic data and (3) support and develop resources to facilitate local access to large shared datasets, including UK Biobank. First, our faculty survey revealed that the vast majority agreed that a shared, institutional-level database that holds genomic/genetic data and electronic health record data is needed and that they would contribute their data to this resource. Community focus group members, while eager to participate and benefit from precision medicine and genetic testing, identified concerns about both consent and data privacy. Second, we developed a standardized genomic consent that alleviates the need for investigators to maintain individual consents. Third, a standardized model for genomic consent facilitates data sharing across the institution and will be an essential component of a common institutional database that stores both electronic health records and genomic/genetic data. Future goals include expanding community engagement to understand the extent of data privacy concerns and improving public dissemination of the benefits of participating in precision medicine research.
42 ICTS Research Development Program

The goal of Research Development Program (RDP) is to provide support for ICTS investigators seeking to accelerate and augment the quality and impact of their clinical and translational science research. Through our services, the RDP advances career development and funding success for ICTS researchers.

  • Research Navigation provides personalized consultation to advise ICTS members on available resources (ICTS Cores, educational options and funding).
  • Research Forum Program supports interactive feedback and project development with experts and core/service specialists on grant research design – connecting investigators to collaborators, mentors and resources.
  • Scientific Editing Service works with ICTS Members to strengthen and clarify their extramural grant applications through support provided by the ICTS and the Department of Medicine.
  • NIH Mock Study Section provides bi-annual review of clinical and translational R, K and F series grants. Applications are reviewed and scored by three faculty and other study section members with feedback distributed prior to NIH grant deadlines.
  • WUSTL Grants Library is a centralized resource for grant writings containing stock language, tools and awarded NIH, DoD and foundation grants.
43 Institute for Informatics

Washington University created the Institute for Informatics (I2) to provide an academic and professional home for informatics science and practice across the university and BJC. From consultations to cutting-edge software, the Institute for Informatics (I2) offers an array of data-mining technologies. We provide these services and tools to faculty, researchers, students and professionals so that together we can answer some of the biggest questions in the healthcare and life sciences fields.
44 mHealth Research Core

mHealth is the use of sensors, native or web-based apps, social media, and/or other smartphone technology to measure and improve health research and outcomes. mHealth is widespread and understudied relative to its importance and potential. For the individual investigator, getting started in mHealth research can be challenging on multiple fronts – scientific, technical, strategic, and regulatory. The ICTS leadership recognized these challenges and initiated an mHealth Research Core in fall 2019, toward the goal of supporting researchers planning and conducting mHealth research projects. The ICTS established the mHealth Research Core in partnership with the Institute for Informatics (I2), the HealthCare Innovation Lab, and the Healthy Mind Lab. The mHealth Research Core has three components. Consultations: We conduct consultations with investigators wishing to receive technical, regulatory, scientific, and/or strategic advice. Meetings: We hold regular meetings to learn about new resources and trends in mHealth research; present and discuss grant ideas, aims pages, and grant proposals. Resources: the Core’s website contains scientific, technical, and regulatory resources and navigation for researchers. As of December 2019, we have conducted 16 consults, 2 meetings, and created the website with technical, regulatory, and scientific resources. Our goal is to increase the number and quality of grant applications in mHealth at WU and to navigate investigators through regulatory and other challenges.
45 Trial-CARE Support Team

In a collaborative effort between the Center for Clinical Studies (CCS) and the Institute of Clinical and Translational Sciences (ICTS), the Trial-CARE (Trial Coordination, Acceleration, & Recruitment Enhancement) Center provides specialized expertise and trial implementation support, with a particular focus on enabling the funding and completion of high-impact multicenter clinical trials. The Trial-CARE support team provides free consultations to investigators and research teams to help them plan, initiate, and successfully complete challenging clinical studies. After each initial consultation, faculty members receive formal recommendations from an expert support team on key steps that can be taken to move their project forward efficiently, ranging from guidance on getting funded, rapidly starting a study, and overcoming real-world barriers to trial conduct. Faculty subsequently receive active Trial-CARE assistance in developing study organization plans, addressing regulatory issues, developing budgets, interacting with sponsors, and building well-targeted recruitment plans. In addition, for applicable studies, Trial-CARE offers full trial coordination services in flexible models of partnership with department-based study teams and investigators. The Trial-CARE team is experienced and facile with studies that are funded by either NIH or industry. To date, Trial-CARE has provided initial consultations to over 50 WU investigator teams, of whom many continue to receive ongoing support during trial conduct. This includes several high-impact multicenter trials for which Trial-CARE partners with department-based faculty to deliver full trial coordination that encompasses all aspects of study start-up and implementation. Peer-to-peer guidance and mentorship are also provided to faculty seeking U-award funding from the NIH.