February 25 at 11:30 am
Research Talks I
Using Social Work Theories for Data Justice
Jessica K. Camp, PhD, LMSW Clinical and Macro, CAADC, ACTP-E
The paradigmatic framework from which measurements are created, data is collected, and analysis is conducted has a profound impact on human services. As a former social work clinician and community researcher, I want to show how I’m using social work practice theories, like Healing Centered Restorative Engagement (a model of trauma-informed care) , to drive more equitable, inclusive, and socially just approaches to collecting and analyzing quantitative data. This talk will discuss:
1) common pitfalls in measurement creation and data collection that exclude certain individuals and social issues, making them invisible,
2) use one social work practice theory, Healing Centered Restorative Engagement, to demonstrate how social work approaches are valuable to build equitable, inclusive, data that is useful for human service programs,
3) show several concrete examples in data collection and analysis where social work theory improved the quality of data findings.
Adverse Police Experiences Study
Robert Alexander Holt , University of Michigan
Alana Lay-Phillips, University of Michigan
Ashley Akaba, University of Michigan
Leopold Samba, University of Michigan
Fatima Saad, University of Michigan
Roland Alexander Blackwood, Office for Health Equity and Inclusion
Summary:
The actions of the police have continued to be a significant issue affecting individuals and communities, especially individuals of African American descent. Experiences with police have the potential to cause high levels of distrust, anger, mental health issues, and lifelong consequences. Through an anonymous survey distributed to university students, containing a subcommunity of African Americans, an understanding of the types and frequencies of experiences will be further explored to assess the impact that police have in society.
Methods:
Using various communication methods, an anonymous Qualtrics survey was distributed among college students in the United States.
Results:
Of our respondents, 75% were African American, with 12% caucasian. 54.2% have had at least one negative experience with the police and 62.5% were with male officers. 40% of respondents believe they were stopped because of their race. 87% of the people who have had a negative experience were not committing a crime. Of those with negative experiences, about 40% have experienced verbal abuse. 70% reported developing negative feelings after an experience with the police.
Impact:
Adverse Police Experiences (APES) are observed through a variety of different scenarios, but ultimately are traumatic instances that individuals have with police. This trauma often remains with those affected even after the incident. It is crucial to highlight these experiences, as it acknowledges the physical, social, and emotional consequences of adverse police experiences. Authoritative figures possess the potential to oppress and project negative perceptions onto individuals, and such experiences leave lasting impacts within the communities of those most affected by that abuse. Such effects warrant demand for change and reform of police training. Furthermore, we want to promote equity and active positive change for a brighter future while also bringing awareness to the different types of health issues that come from continuous adverse police experiences.
February 25 at 1:30 pm
Research Talks II
A Mediation Analysis of Chemical Exposures and Racial Differences in Telomere Length
Emily Roberts, Department of Biostatistics, University of Michigan
Jonathan Boss, Department of Biostatistics, University of Michigan
Stephen Salerno, Department of Biostatistics, University of Michigan
Dr. Ami Zota, Department of Environmental & Occupational Health, George Washington University
Dr. Bhramar Mukherjee, Department of Biostatistics, University of Michigan & Department of Epidemiology, University of Michigan
Dr. Belinda Needham, Department of Epidemiology, University of Michigan
Given what we know about race differences in diseases of aging and premature mortality, as well as race differences in exposure to socioeconomic disadvantage, psychosocial stress, and other non-genetic risk factors for short leukocyte telomere length (LTL), previous research strongly suggests that US Blacks should have shorter LTL than Whites; yet, this is not the case. Building on recent studies which have shown that exposure to certain environmental toxicants is associated with longer LTL, we examine the hypothesis that Black/White differences in LTL are partially explained by differences in exposure to carcinogens – differences that arise from residential and occupational segregation. To test this hypothesis, we use a study population comprised of 1,169 adults (300 Black, 869 White) from the 1999-2000 and 2001–2002 cycles of the National Health and Nutrition Examination Survey (NHANES). Exposures with more than 50% below their respective detection limits are excluded from the analysis, while non-detects for the remaining 16 exposures are multiply imputed. The primary analytical goal is to estimate the natural direct effect of race on LTL and the natural indirect effect of race on LTL that is mediated through polychlorinated biphenyl (PCB), furan, and dioxin exposure. The natural direct effect and the natural indirect effect estimates can then be used to ascertain how much of the race and LTL association is explained through exposures. We first characterize the single-pollutant mediation effects by constructing survey-weighted mediation models accounting for NHANES’s stratified cluster sampling design and multiple testing. Following pairwise mediation, we then consider several mediation analysis methods which quantify a global mediation effect of all environmental exposures simultaneously, while being cognizant of the highly collinear nature of exposure data. Identifying factors in the causal pathway between race and LTL will lead to a more in-depth understanding of the counterintuitive Black/White discrepancy in LTL.
CanVaxKB: The First Web-Based Cancer Vaccine Knowledge Base and its Data Analysis
Chloe Darancou, College of Literature, Science, and the Arts, University of Michigan
Eliyas Asfaw, School of Public Health, University of Michigan
Cancer is the second leading cause of death in the United States and has become a global public health problem (Stewart, 2004). Cancer is the abnormal or uncontrolled growth of cells in the body caused by mutations of DNA within cells. In recent years, there have been extensive studies experimenting with the concept of cancer vaccines. Cancer vaccines can either be used to prevent or treat cancer through the injection of antigens or cancer cells that can stimulate a beneficial immune response in a patient. The primary goal of this project is to collect, annotate, and analyze various types of cancer vaccines around the world, in order to examine patterns in genomic data, clinical trial assessments, and vaccine preparation. CanVaxKB includes information regarding the genes of the antigens/cells presented in the vaccine while discussing clinical trials associated with each vaccine. CanVaxKB also provides a user-friendly web interface for users to interactively search, compare, and analyze different cancer vaccines. CanVaxKB has more than 600 vaccines with detailed profiles on many arrays of data. Additionally, we analyzed the recurring epitopes of MLANA and ERBB-2 genes, the most frequently used cancer vaccine antigens in CanVaxKB. In our analysis, we also assessed vaccine formulation and clinical trial outcome measurements for vaccines treating melanoma. Furthermore, from this database and compilation of vaccines, we are able to utilize ontology and discover relationships between cancer vaccines to further our understanding and highlight the best approaches to make them.
References:
Stewart, S. L., King, J. B., Thompson, T. D., Friedman, C., & Wingo, P. A. (2004). Cancer mortality surveillance—United States, 1990–2000. MMWR Surveill summ, 53(3), 1-108.
Detecting Domestic Violence in Medical Records
Susan Parker, Doctoral Student, School of Public Health, University of Michigan
Annually, 7 million women in the United States report experiencing domestic violence, and 15% report assaults requiring emergency medical attention. However, less than 5% of these events are recorded in corresponding electronic health records. Underreporting substantially limits research seeking to estimate domestic violence in health settings or assess efforts to reduce its incidence. This study uses machine learning methods to determine which records are related to domestic violence but are not recorded in medical records. Using corrected health records with review of injuries as training data, machine learning models are fitted on data with accurate health record coding. Models fitted on corrected training data are applied to uncorrected records to reduce bias from substantial under-reporting in existing medical records. The machine learning approach is evaluated using natural language processing to assess the accuracy, precision and recall of the approach.
EyeSnap: Machine Learning and Global Partnerships to Prevent Childhood Eye Cancer
Alec Bernard, University of Michigan Medical School
Sahal Saleh, University of Michigan Medical School
Tochukwu Ndukwe, University of Michigan Medical School
Shang Zhou Xia, College of Engineering, Dept of CSE, University of Michigan
Josh Meyer, College of Engineering, Dept of CSE, University of Michigan
Elliot Soloway, Arthur F. Thurnau Professor, College of Engineering, Dept of CSE, University of Michigan
Christine Nelson, MD, Bartley R Frueh, M. D. and Frueh Family Collegiate Professor of Eye Plastics and Orbital Surgery, Professor of Ophthalmology and Visual Sciences and Professor of Surgery, Medical School
Hakan Demirci, MD Richard N and Marilyn K Witham Professor of Ophthalmology and Visual Sciences and Associate Professor of Ophthalmology and Visual Sciences, Medical School Shang Zhou Xia, College of Engineering, Dept of CSE, University of Michigan
Retinoblastoma is a childhood eye cancer with vastly different outcomes in different parts of the world. When diagnosis is missed, this disease is fatal. Leukocoria or “white pupil” is the most common presenting sign of retinoblastoma and occurs as a result of abnormal reflection of light from the eye. This sign can also present in other childhood ocular diseases. Early detection of leukocoria results in earlier, more effective treatment, ultimately saving a child’s vision and, potentially, life.
The current shortage of trained ophthalmology providers in resource-poor settings speaks to a need for inexpensive, accessible screening. A University of Michigan research team is pioneering the development of a system that employs inexpensive smartphone cameras and free, open-source machine learning to detect this childhood cancer at a time when intervention can save life and vision. UM partnered with a team of ophthalmologists and pediatricians in Addis Ababa, Ethiopia. This partnership allowed the collection of data from a population highly likely to benefit from the software application. A nation-wide meeting was held to train and solicit feedback from additional ophthalmologists and broaden the partnership. So far, the dataset that was gathered in Ethiopia contains almost 10,000 images of eyes taken by smartphone-based cameras.
The current dataset is being used to fine-tune an open-source, deep learning model on personal computers. The model has achieved a ~95% accuracy rate. The partnership between stakeholders in University of Michigan Medical School, College of Engineering, and St Paul’s Millennium Medical College in Ethiopia allowed the development of an effective system using machine learning to provide an accurate and timely diagnosis of an important cause of childhood eye and life loss globally. The use of similar, widely-available technologies – low-cost phone-based cameras, open-source ML software, personal-computer computation – has the potential of wide applicability in human health and disease.
February 26 at 11:30 am
Research Talks III
Understanding Student COVID-19 Response from Reddit Data: A Dynamic Topic Modeling Approach
Jiaxi (Jacy) Li, Dual MS, MPSM, School of Environment and Sustainability, University of Michigan
Chendi Zhao, MS Student, MPSM, Institute for Social Research, University of Michigan
The COVID-19 pandemic has spread worldwide, impacting every aspect of daily life. In an emergency event of such kind that persists, it becomes increasingly important to have reliable, up-to-date emergency management assessments. The ability to accurately capture public needs and attitudes are critical to the success of emergency response plans.
The traditional methods, such as surveys, have several pitfalls to understanding these issues. Most surveys introduce hypothetical bias as subjects’ responses are measured under hypothetical scenarios. Besides, as response rates fall, costs rise, and the desire for timely data increases, social media are becoming increasingly popular as a source of data for social science research. Reddit, the self-proclaimed “”front page of the Internet”” with its unique feature of subreddits, opens up potentials to study smaller geographies and rarer subpopulations.
This study focuses on college students’ responses to COVID-19 emergency management and the variability across time and groups in Michigan. Using Reddit data, we collect organic submissions in text formats from corresponding subreddits of the ten largest colleges and universities in Michigan for the period of March 1st, 2020, to October 31st, 2020. Taking a topic modeling approach, we derive COVID-19 related topics of our interest. We further divide reference time by academic schedule (Spring Semester, Summer Vacation, and Fall Semester) and implement a dynamic topic model to capture the information changing over time and variability among universities. Our study presents the student group’s social media landscape and shed light on the implications of different locations and school emergency management plans on student responses.
COVID-19 Response Modeling using MI Symptoms App Data
Christopher Crowe, MS, Population and Health Sciences, School of Public Health
Partnership: The MI Symptoms application is the result of a state-university partnership between the State of Michigan and the University of Michigan. It was created in consultation with public health experts from the University of Michigan and the Michigan Department of Health and Human Services for the State of Michigan. MI Symptoms is an application for Michigan residents and employers to report symptoms of COVID-19. The goal of collecting symptom data is to enable early detection of COVID-19 illness and allow immediate response to potential outbreaks.
Research and Methods: We have used the data collected from the MI Symptoms app and applied a lagged correlation time series analysis to analyze its usefulness as a tool for early detection of outbreaks.
Main Results: Results of the analysis show that MI Symptoms data produces a lead time of up to 9 days over MDHHS case data. Results were statistically significant with a correlation of over 0.8 and held true at both the state and MERC region level. This finding shows that MI Symptoms is a potentially useful tool to help slow the spread of the SARS-CoV-2 virus in Michigan communities.
Impact: The impact of these findings shows that MI symptoms can be a useful tool for public health professionals to prevent large outbreaks. Employers may take the lead time to isolate and contact trace based on reported symptoms to avoid lost productivity. However, the largest impact of the findings will be the lives saved by using the early warning given by the application to slow the spread of the virus. Michigan has been hard-hit by the COVID-19 pandemic, with over 13,300 deaths reported as of January 3, 2021. The predictive abilities found by analyzing MI Symptoms survey data could be used to save potentially thousands of Michiganders from this tragic fate.
February 26 at 1:30 pm
Research Talks IV
Improving Researcher Access to Data through the Michigan Education Data Center
Nicole Wagner Lam, University of Michigan, Education Policy Initiative (EPI)
Jasmina Camo, University of Michigan, Education Policy Initiative (EPI)
Kyle Kwaiser, University of Michigan, Education Policy Initiative (EPI)
Rod Bernosky, State of Michigan, Center for Educational Performance and Information (CEPI)
Johan Mosquera, University of Michigan, Michigan Education Data Center (at EPI)
Jonathan Hartman, University of Michigan, Michigan Education Data Center (at EPI)
The State of Michigan has a long history of offering students and families educational opportunities unparalleled in the Midwest. With high quality public universities and local communities supporting K-12 school systems aimed at preparing students for postsecondary and beyond, it is unfortunate that educational inequalities, achievement, and attainment gaps have grown markedly over time. Education researchers have worked for over a decade to understand the policies contributing to widening educational inequalities. Despite the need for analysis, researchers have experienced substantial barriers to accessing education data. Therefore, a partnership between the State of Michigan, UM, and MSU was formalized to accelerate research efforts. Within the Michigan Education Research Institute (MERI), UM acts as the main data steward and operates the Michigan Education Data Center (MEDC). The benefits associated with MEDC have been to introduce a) improvements to Michigan’s education data, b) a transparent process for researchers, c) a sophisticated method to merge education records to other data sets, and to d) align researcher curiosities with MDE’s research needs, e) expand the set of researchers solving education programs, f) capture substantial resource efficiencies, g) a streamline the process by which research findings make their way to MDE/inform education policy and practice. Examples of questions answered with MERI/MEDC data include:
• Does waiting a year to start kindergarten lead to better academic outcomes?
• How does summer youth employment impact graduation and college enrollment?
• What career are Michigan’s students pursuing?
An emerging concern for MEDC are gaps and the potential presence of error within administrative data, which may introduce biases into research outcomes. In this presentation, we will briefly introduce MEDC, after which we will turn our attention to existing literature on biases introduced when using administrative data. Finally, we will propose a framework for identifying and communicating this area of emerging concern to researchers.
Evolving Data Dashboards to Engage Key Partners
Cathy Hearn, Program Manager, Center for Education Design, Evaluation, and Research
Jennifer Nulty, Evaluation Specialist, Center for Education Design, Evaluation, and Research
Data dashboards are a valuable tool for communicating responsive, relevant updates to program partners and participants. At the 2020 Data for the Public Good Symposium, the Center for Education Design, Evaluation, and Research (CEDER) demonstrated how we created a customizable data dashboard to track Bosch Eco+STEM Teacher (BEST) Grant applications. The BEST Teacher Grant Program is a partnership between CEDER and the Bosch Community Fund that advances sustainability and STEM education in ways that inspire, excite, and engage students.
This year, we will be shifting our focus to outline how we responded to outreach partner needs through improvements to the dashboard. This presentation will cover the strategies we used to address evolving partner needs and improve dashboard accessibility. These strategies included
– Recalibrating dashboard metrics and visuals to address inter-district differences,
– Developing a more user-friendly communication strategy for sharing the dashboard, and
– Leveraging technology to collect real-time feedback on the usefulness of the dashboard.
Then, we will share feedback from our partners about the ways in which they used this year’s dashboard and outline our plans for continued improvement of this tool. Participants will walk away with strategies for communicating program data in ways that engage and respond to partner needs.
Chetah: Fast and Intelligent Search Engine for UN and NGO reports
Shivika K Bisen, Computer Science and Engineering, LSA, University of Michigan Data4Good, University of Michigan School of Information
“Start using the data to make better decisions about the world we want to see”- Jake Porway In the era of data, there are many tools that are able to retrieve the relevant information. Google has been dominant as a search engine and Semantic Scholar is advancing in the use of AI in search. Such tools have been transformational. However, the non-profit sector is under-represented as there is often bias towards the commercial sector. UN and NGOs release information that is often unstructured; this makes search challenging. Chetah, aims to solve these problems. It implements natural language processing for intelligent search of UN and NGO reports and summarises reports for fast digest of information. This research is conducted under supervision of Professor Edward Happ at Data4Good center in University of Michigan Ann Arbor. Our partners are Microsoft and NetHope. Chetah retrieves relevant reports and summarizes each report with a deep learning algorithm, BERT. Another innovation is that it is built to classify reports based on the eleven UN Clusters. Knowing in which cluster a user’s query lies helps in engagement with authorities, industry experts and speeds up the resolution of issues. The relevancy ranking approach is BM25 and the results have been proven better than the Google for Non-profit sector, with an F1-score of 0.78. Chetah is impactful in following ways. It has equity in UN and NGO reports and is not biased towards UN reports only, unlike Google. Also, it is designed to help NGO program managers, policy makers and philanthropic foundations to design programs and grant funds. Finally, Chetah retrieves evidence-based program reports and not just annual reports, unlike Google. This tool can lead to better answers for nonprofit work and eventually help solve the crucial real problems that NGO and foundations are facing.
Data & Evaluation in Substance Use Disorder: Evaluating Quick Response Teams
Timothy NeCamp, PhD
David Mackey, BS
Quick Response Teams (QRT) are pre-arrest diversion programs that involve interdisciplinary overdose follow-up and engagement with survivors to link individuals to treatment options immediately following an overdose. Families Against Narcotics (FAN) had a pilot QRT program in place with a police department in Sterling Heights, MI. They did not have a process for determining the outcomes of their visits or a way to measure the success of the program. They needed quantitative data to support the positive feedback they received from survivors and law enforcement in the field.
Data Bloom developed and implemented a comprehensive data strategy which allows FAN to see the real time impact the Comeback program is having. This included facilitating goal alignment for FAN, defining the collection and organization process, creating a form to track visit information, training team members in the process, and creating a dashboard to convey results and outcomes in real-time. This project has helped FAN get a very detailed understanding of their QRT visits including demographic information on who they were visiting and the outcomes of each visit. This understanding will help FAN understand the impact of their program (for both internal use and for acquiring future funding) and has helped them improve their QRT visits and outreach. Since we started this project, the Comeback program has expanded to (6) additional police departments throughout Michigan. We’ve tracked (344) QRT visits and now have 20-30 occurring weekly. (176) survivors have found treatment through the Comeback program with more being added every day.
In this presentation, we will overview the entire data pipeline: strategizing, collection, analysis, and reporting, and present initial results from the program. We will talk about the impact of our work on FAN and individuals struggling with substance use disorder. We will also mention future goals of the project.
2:40 PM – Teaching Python for Data Science in a Girls Who Code Club
Data for Good Organizations
Audrey Drotos, Neuroscience Graduate Program
Sarah Haynes, Department of Pathology
Vy Nguyen, Department of Computational Medicine and Bioinformatics
Kelly Sovacool, Department of Computational Medicine and Bioinformatics
The goal of Girls Who Code (GWC) at the University of Michigan Department of Computational Medicine and Bioinformatics is to provide a collaborative and supportive environment for high school women of all skills levels and backgrounds interested in learning how to code. To do this, we developed a computational data analysis curriculum to teach area high school women basic programming in Python at a weekly coding club. At each club meeting, students learn new skills through a live-coded lesson in Python and practice coding in small groups. Later in the year, students work in teams to apply their data analysis skills to a real-world data set and present their findings to their peers. An additional focus of GWC is to increase the visibility and representation of women in computer science-related fields by hosting a guest speaker at every club meeting and targeting enrollment to student populations which are underrepresented in STEM. GWC has graduated 93 students since 2017. Alumni self-report improvements in critical thinking/problem solving abilities, understanding of data science, and programming skills. GWC uses a variety of free and open-source technologies, including Jupyter notebooks for live coding and GitHub for collaborative development of our curriculum. These tools have proven key for allowing the club to successfully transition online during the COVID-19 pandemic. Overall, GWC provides underrepresented groups in STEM with data analysis skills and exposure to careers in data science.
Contributed Talks (prerecorded)
Pesticide Exposure Levels in the US and their Bioactivity
Public Health & Medicine
Chanese A. Forté
Justin Colacino
Pesticides are heavily used in agriculture globally and are potentially harmful to human health. Using 2 large publicly available datasets, this study aims to 1) quantify differences in pesticide exposure levels of the US general population in comparison to farmworkers, and 2) assess biological activity of these pesticides at human relevant doses using ToxCast high throughput screening toxicity data. The National Health and Nutrition Examination Survey (NHANES) is a cross-sectional study representative of the US population with oversampling weights for minoritized populations. NHANES was used to quantify pesticide exposure among US farmworkers and the general population who responded to NHANES between 1999 and 2016. Overall, there are 65,893 the NHANES survey with occupation data available with at least one pesticide biomarker present.. Biomarker measurements were available for 28 different pesticides, which also have been assessed for dose-dependent toxicity using high throughput assays in ToxCast. Of the 28 pesticides, o,p’-DDT had the lowest average biomarker concentration at 0.03µM among US farmworkers, whereas p,p’-DDE had the highest exposure level at 0.5µM. In ToxCast, the AC50 represents the concentration at 50% of the maximum biological activity. The average AC50 for p,p’-DDE is 1.3µM suggesting that activation levels are higher than that of the US population exposure level range. Ongoing work integrates pesticide biomarker concentrations across different occupational and demographic groups with biologically active concentrations based on high throughput toxicity data. By comparing population exposure data to toxicological assay data, our goal is to create an overarching view of how pesticides may be affecting the body at a human population level.
ArgoSSM: A Bayesian state-space framework for predicting the location of missing temperature sensors in the Southern Ocean
Climate Change and Sustainability
Derek Hansen, Department of Statistics
Drew Yarger, Department of Statistics
The Argo project deploys a fleet of sensors that collect information such as the temperature and salinity at varying depths of the ocean. These sensors are attached to floats that drift with the ocean currents. In the Southern Ocean, these floats occasionally end up under ice, and their location can no longer be tracked via GPS.
We introduce a novel framework, called ArgoSSM, to predict the location of these floats while they are under ice-cover. ArgoSSM is a fully probabilistic Bayesian state-space model which provides both point estimates and uncertainty in the missing location measurements. Moreover, it can incorporate additional information like potential vorticity in the predicted locations. We compare our approach to existing approaches in the oceanographic literature, such as linear interpolation and interpolation in potential vorticity coordinates.
By providing a posterior distribution of potential paths the floats could have taken under ice, our modelled predictions and uncertainty can improve downstream tasks like temperature estimation in the Southern Ocean, allowing for better scientific understanding and monitoring.
Prediction of repurposed drugs for treating lung injury in COVID-19
Public Health & Medicine
Bing He, Department of Computational Medicine and Bioinformatics, Medical School, University of Michigan.
Lana Garmire, Department of Computational Medicine and Bioinformatics, Medical School, University of Michigan.
Background: Coronavirus disease (COVID-19) is an infectious disease discovered in 2019 and currently in outbreak across the world. Lung injury with severe respiratory failure is the leading cause of death in COVID-19, caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). However, there still lacks efficient treatment for COVID-19 induced lung injury and acute respiratory failure.
Methods: Inhibition of angiotensin-converting enzyme 2 (ACE2) caused by the spike protein of SARS-CoV-2 is the most plausible mechanism of lung injury in COVID-19. We performed drug repositioning analysis to identify drug candidates that reverse gene expression pattern in L1000 lung cell line HCC515 treated with ACE2 inhibitor. We confirmed these drug candidates by similar bioinformatics analysis using lung tissues from patients deceased from COVID-19. We further investigated deregulated genes and pathways related to lung injury, as well as the gene-pathway-drug candidate relationships.
Results: We propose two candidate drugs, COL-3 (a chemically modified tetracycline) and CGP-60474 (a cyclin-dependent kinase inhibitor), for treating lung injuries in COVID-19. Further bioinformatics analysis shows that 12 significantly enriched pathways (P-value <0.05) overlap between HCC515 cells treated with ACE2 inhibitor and human COVID-19 patient lung tissues. These include signaling pathways known to be associated with lung injury such as TNF signaling, MAPK signaling and chemokine signaling pathways. All 12 pathways are targeted in COL-3 treated HCC515 cells, in which genes such as RHOA, RAC2, FAS, CDC42 have reduced expression. CGP-60474 shares 11 of 12 pathways with COL-3 and common target genes such as RHOA. It also uniquely targets other genes related to lung injury, such as CALR and MMP14.
Conclusions: This study shows that ACE2 inhibition is likely part of the mechanisms leading to lung injury in COVID-19, and that compounds such as COL-3 and CGP-60474 have potential as repurposed drugs for its treatment.
Increasing Access to Specialized Dermatology Care: a Retrospective Study Investigating Clinical Operation and Impact of a University-Affiliated Free Clinic
Ongoing Community-University Partnership
Thomas Hester, B.S. Department of Dermatology
Reinie Thomas, B.A. Department of Dermatology
Jean Cederna, M.D. Hope Clinic
Ann Marie Peterson, B.S. Hope Clinic
Julie Brown, R.N. Hope Clinic
Timothy M. Johnson, M.D. Department of Dermatology
Kelly B. Cha, M.D., Ph.D. Department of Dermatology
Partnership: Hope@UMHS is a partnership between University of Michigan Health System (UMHS) and Hope Clinic (HC) which provides free specialty consultations to Southeast Michigan’s uninsured. Patients referred to Hope@UMHS Dermatology from HC primary care providers are scheduled for a quarterly Saturday morning clinic at UMHS outpatient dermatology facility. Clinics are staffed by volunteer UMHS dermatology employees. The clinic operates in our typical dermatology setting, and patients receive on-site biopsies and procedures. Patients may also be referred on the same day or future clinic dates to other subspecialty services involved in the Hope@UMHS collaboration, including plastic surgery and otolaryngology, who provide care in close proximity to the dermatology clinic.
Methods: A retrospective chart review was completed for patients referred to UMHS Dermatology clinic as part of the HOPE@UMHS collaboration from April 2012 through February 2020.
Results: Two hundred forty-six of 294 referred patients were evaluated in 30 clinic sessions over 8 years, staffed by 92 unique volunteers. Forty-eight percent of patients identified as non-White and 33.5% spoke a primary language other than English. Patients most commonly presented with atopic dermatitis (10.5%), seborrheic dermatitis (7.9%), and actinic keratosis (7.4%). The majority of patients (68.2%) were prescribed at least one new medication. There were 102 procedures performed. Patients most commonly received care consistent with New Patient Level 3 evaluation & management.
Impact: Nine skin cancers, including one melanoma, were diagnosed and treated. Eighty-seven percent of patients received conclusive evaluation and treatment at the time of their consultation.
Examination of Mental Health Indicators Among Arab Americans in the Metro-Detroit Area
Public Health & Medicine
Malak Kabalan, Arab American Health Initiative and Michigan State University
Marwa Saad, Arab American Health Initiative and University of Michigan-Dearborn
Mahdi Saab, University of Michigan-Dearborn
Amneah Chaaban, University of Michigan-Dearborn
Hussein Bazzi, Wayne State University
Wassim El-Sayed, Wayne State University School of Medicine
Zain Jawad, Arab American Health Initiative and Wayne State University School of Medicine
Dr. R. Alexander Blackwood, Arab American Health Initiative, Office for Health Equity and Inclusion and Michigan Medicine Department of Pediatric Infectious Diseases
Introduction: Within the Arab American (AA) community, mental health is stigmatized and rarely discussed. There is scarce literature on mental health indicators relating to the AA population. This study aims to examine the correlation between depression and loneliness and the indications this can have on mental health.
Methodology: Following being granted the exemption status by the University of Michigan-IRB, an anonymous, digital survey conducted through Qualtrics software was distributed via social media to adults (ages 18+) within Metro-Detroit. The De Jong Scale for Emotional and Social Loneliness questionnaire was used as an indicator of the objective and subjective facets of loneliness to give an overall loneliness score. The CES-D Scale is a validated scale used as a measurement of depression. Statistical analysis was conducted using Chi-squared, ANOVA, and two-sample independent T-tests.
Results: Approximately 75% (190/255) of participants were AA and 25% (65/255) of participants were non-AA. Of the participants, 74% (188/255) were female and 26% (67/255) were male. Our findings display that 47% (89/190) of the AA respondents reported feeling lonely and 58% (110/190) were depressed, compared to non-AA, 52% (34/65) of whom were lonely and 68% (44/65) were depressed. When stratified by immigration status, 34% (15/44) of AA immigrants were lonely and 40% (18/44) were depressed, compared to 50% (74/146) and 63% (92/146) of non-immigrant AAs respectively.
Impact: Our data suggest that AAs have a lower rate of depression and have a lower sense of loneliness than their non-Arab counterparts. This study will serve as a resource for professionals to better understand AA patients reporting mental health symptoms. Given the stigma associated with mental health, we hope that our study provides a basis for initiating new studies and interventions for AAs.
Facilitating Data Usage in Small Health Care Centers
Public Health & Medicine
Maya Lapp
As health care documentation transitions to a digital format, health care data is becoming more and more accessible. Often these data can be extremely useful in identifying areas of patient care that need to be improved, as well as determining the success or failure of interventions intending to address these areas. Unfortunately, small health care centers often do not have the resources or personnel to interpret and analyze the data that is collected. Primary Care Health Services in Pittsburgh, PA, is one organization that is facing this sort of problem. We created an R Shiny App to convert quarterly reports on cancer screening rates into a printable report of data visualizations that easily show trends in screening rates over time.
Mining Big Data to Make Sense of Massive Scale Learning
K-12 or Higher Education
Samaa Haniya, Education
Technology innovation is revolutionizing higher education allowing new modes of learning to emerge. One of these emerging modes of learning is the phenomenon of Massive Open Online Courses, known as MOOCs. Despite its popularity in media and research between the supporters and the opponents, little is known about how learning happens in a large scale educational context and how to design learning environments so they are most effective, innovative, and transparent to learners’ needs. Therefore, in this presentation I will investigate the dynamic nature of learning at scale by using a case study of a Coursera MOOC offered by the University of Illinois Urbana-Champaign. Using a combination of educational data mining and qualitative methods, I will trace learners’ behaviors as they engage with the digital environment and examine the relationship between these behaviors and the different factors that motivate and limit learners to learn. Data was collected from clickstream files and archival pre and post course survey of the course. The study provides significant insights on the dynamic nature of learning at scale. Consequently, this could help to personalize educational experiences to reflect on learners’ unique needs.
Global Partnerships for Global Problems: Assessment of the impact of the COVID-19 Pandemic on Medical Education Across Africa
Public Health & Medicine
Bereket Alemayehu, St.Paul’s Hospital Millennium Medical College, Addis Ababa, Ethiopia
Nicole Byl, University of Michigan Medical School, Ann Arbor, MI
Amani Nuredin, St.Paul’s Hospital Millennium Medical College, Addis Ababa, Ethiopia
Dawit Tesfagiorgis, St.Paul’s Hospital Millennium Medical College, Addis Ababa, Ethiopia
Gnendy Indig, University of Michigan Medical School, Ann Arbor, MI
Alec Bernard, University of Michigan Medical School, Ann Arbor, MI
The COVID-19 pandemic has caused unprecedented disruption to education and health systems across the globe. Our research team is composed of members of Ethiopian Medical Students’ Association (EMSA) and students at the University of Michigan Medical School. EMSA is a member of International Federation of Medical Students’ Association (IFMSA) based in Ethiopia and is working towards improving medical education and public health. The team used the EMSA network to reach medical students across Africa. Via this network, an anonymous electronic survey was sent to medical schools across thirty-three African countries. The survey consisted of a 39-item survey composed of Likert scale (LS), dichotomous, and free-response items and assessed various domains of class structure and timing, patient interactions, exam administration, learning environment satisfaction, mental health impacts, and volunteer opportunities/engagement.
694 medical students across 33 African countries completed our survey. 87.8 % (609) of students reported that their medical schools had been initially closed; among those, only 64.2 % (329) had reopened. 4.9% (34) of the students reported that their school continued without closure through the pandemic. Students from the medical schools that changed their approach saw a shift to online classes while physical interaction with patients interaction decreased by 87%. 20.8 % of students also reported that all forms of examination had been dismissed. The number of students who felt they were not in a supportive learning environment grew from 52 to 232.
The majority of students surveyed had their education and learning environment disrupted due to the COVID-19 pandemic. This disruption may have important global public health implications in the future. Furthermore; the development of this international research team fostered a friendly environment for sharing cultures while simultaneously giving a global dimension for the data, which is a necessity for tackling global problems.
GET Access: Geriatric Education on Telehealth. An Initiative to Improve Access to Virtual Healthcare for Geriatric Patients
Public Health & Medicine
Cayla M. Pichan
Clare E. Anderson
Emily L. Vogt
Lillian C. Min, MD
Mary C. Blazek, MD, MEHP
University of Michigan Medical School
Introduction: The Covid-19 pandemic placed an unprecedented demand on health systems to rapidly shift ambulatory in-person care to virtual care. Geriatric patients, at increased risk of morbidly and mortality from Covid-19 infection, face more challenges with video visit access compared to younger patients due to discomfort with technology and less access to necessary devices and broadband internet. Thus, it is a priority to enhance remote connections between geriatric patients and their providers.
Objectives: Medical student volunteers created an initiative to improve access to and comfort with video visits for geriatric patients. They provided patients one-to-one remote guidance, identifying and overcoming barriers, with practice sessions to increase comfort. This was achieved by 1) exploring educational opportunities for the virtual delivery of personalized training to older adults, 2) creating instruction materials for volunteers to successfully navigate conversations with patients and caregivers, 3) sharing with the larger health system, and 4) making the program sustainable.
Results: In 10 weeks, organization leaders created a 28-page manual and trained 26 volunteers. Of 95 contacted patients, 80 agreed to participate, and 71 switched from a phone visit to video visit, representing a conversion rate of 88.75% from phone to video visits. During this time, 1942 patients had 2865 visits. Of 59 providers within the clinics studied, 12 providers with the lowest video visit rates received help from GET Access. Averaged over the 10 weeks of the program, visits after participation in the program were associated with video format 43% compared to 19.2% for visits prior to participation (adjusted OR 3.38 [95% CI 2.49, 4.59]) and visit by non-participating providers (32.1%, adjusted OR 1.65 [1.31, 2.08]).
Conclusion: A program dedicated to personalized virtual technological instruction and practice helped geriatric patients transition from phone to video visits, to provide a platform for stronger connection with their providers.
The Cost of Lubricant Eye Medications Dispensed On a Monthly Basis for the Treatment of Chronic Dry Eye Disease
Public Health & Medicine
Eman Mohamed Elsayed M.B.B.Ch. Claims Department, Globemed Egypt, street 15, Maadi, Cairo, Egypt
Nada Galal Eldin Salama M.B.B.Ch. Dip. Internal Medicine, Internal Medicine Department, New Cairo Hospital, First Settlement, New Cairo, Egypt
Background: Dry eye disease is a common condition worldwide and its economic impact is growing by time. The majority of the cost is represented by medical treatment specially lubricants. The aim of this study is to assess expenses of the lubricant eye medications for the treatment of chronic dry eye thus, decrease the financial load.
Methods: A retrospective study was carried out at Royal Care International in Egypt. The data was retrieved from the claims of chronic diseases (1359) from 1st of January 2018 till the 31st of December 2018. Patients who were ≥ 18 years old, diagnosed with chronic dry eye by ophthalmologists, and received eye lubricant medications were included (152 claims).
Results: The cost of lubricant medications was estimated to be 0.064 million US dollars per 1000 patients per annum, 64.41 dollars per patient per annum, and 5.37 US dollars per patient per month. There was a 19.1% increase in the expenses and an increase of 11% in the number of prescriptions from the first quarter to the fourth quarter. One-way sensitivity analysis was carried out by assessing the impact of varying the growth rate of the number of prescriptions, on the growth rate for the cost of medications among quarters. The result shows that the growth rate for the cost of medications is directly influenced by the growth rate of the number of prescriptions. The prevalence is 11.18%.
Conclusion: Based on our results and the prevalence of Dry eye, the expenses are considered a growing burden on the healthcare sector. The financial consequences are considerable on the community and on a personal basis. Other consequences such as the workplace productivity and the subject’s quality of life will likely be influenced. Our results are to alert healthcare policymakers and ophthalmologists to recommend cheaper and effective treatments.
Understanding the Impact of the CoVid-19 Pandemic on Young Adults in the Health Professions Pipeline
Public Health & Medicine
Marina Khreizat, Office for Health Equity and Inclusion (OHEI)
Fatima Saad, Office for Health Equity and Inclusion (OHEI)
Malak Almasnaah, Office for Health Equity and Inclusion (OHEI)
Mustapha Jaber, Office for Health Equity and Inclusion (OHEI)
Dr. R Alexander Blackwood, Office for Health Equity and Inclusion (OHEI)
Introduction: CoVid-19 has greatly impacted students all throughout the health professions pipeline. Students had to abruptly adjust to virtual formats, move out of their living spaces, and cope with the emotional challenges of the pandemic. This study seeks to highlight the academic, personal, and professional impact of the pandemic from the perspective of students.
Methods: A Qualtrics survey was administered to pre-health and health professions students from May through September 2020 using various online platforms such as Instagram and Facebook, student group chats, and by reaching out to student organizations. The survey collected data regarding knowledge of CoVid-19 and the impact it has had on students’ academic and personal lives. Additionally, participants answered a series of 20 questions related to mental health using the Center for Epidemiologic Studies Depression Scale (CES-D).
Results: Of the 224 total participants in the study, over 50% (117/224) received a score of 16 or higher, which indicates depression, according to the CES-D score. 75% (168/224) of participants responded that they were worried about the outbreak and economic impact of COVID-19 and nearly 40% (71/195) reported that someone in their family lost a job, many with no other source of income.
Conclusion: The COVID-19 pandemic continues to affect students throughout the health professions pipeline. While students were provided with resources if they received a CES-D score indicating depression, future studies could assess the long-term impact on mental health to changes associated with the pandemic and the long term implications on students’ education.
Towards safe and private machine learning methods in healthcare
Public Health & Medicine
Maria Han Veiga
This project focuses on the impact and feasibility of privacy aware techniques in the context of medical AI. In particular, the focus is on the impact on prediction of patient outcomes in peritoneal dialysis. Peritoneal dialysis (PD) is a home-based renal replacement therapy for kidney failure. BRAZPD is a nationwide cohort study collecting data from adult patients on PD from 122 centers in Brazil. Patient demographics, socioeconomic, and laboratory values were collected, and patients were followed from December 2004 to January 2011.
First, we show that of art techniques in machine learning lead to both accurate and interpretable models. With well understood models such as gradient boosting, we are able to outperform previous predictive models employed on this dataset. In addition, through SHAP values we were able to recover significant predictors in line with expert knowledge.
The second purpose of this project was to design of a safe, privacy preserving framework to perform patient outcome inference. For a long time, the ideas of federated learning and homomorphic encryption were mostly academic, however, in the last few years improvements on the theory and software has allowed for these techniques to be incorporated in traditional machine learning methods. In the second part of this project, we access the impact of these techniques on the quality of previously established inference models, in mind to provide an inference engine framework that is both decentralized and encrypted.
The Prevalence of Suicidal Ideation Among Arab Americans in the Metro Detroit Area
Public Health & Medicine
Marwa Saad, Arab American Health Initiative and University of Michigan-Dearborn
Malak Kabalan, Arab American Health Initiative and Michigan State University
Jacob Agemy, University of Michigan-Dearborn
Nadeen Awada, University of Michigan-Dearborn
Wassim El-Sayed, Wayne State University School of Medicine
Zain Jawad, Arab American Health Initiative and Wayne State University School of Medicine
Dr. R. Alexander Blackwood, Arab American Health Initiative and Office for Health Equity and Inclusion and Michigan Medicine Department of Pediatric Infectious Diseases)
Introduction: Within the Arab American (AA) community, suicide and suicidal ideations are highly stigmatized and rarely spoken about publicly. Previous studies have identified suicide rates to be less prevalent among AA than non-ethnic whites. The aim of this study is to determine the prevalence of suicidal ideation among the AA population compared to the non-AA general population and examine the different contributing factors such as acculturation.
Methodology: Following being granted the exemption status by the University of Michigan-IRB, an anonymous, digital survey conducted through Qualtrics software was distributed via social media to adults (ages 18+) within Metro-Detroit. The Suicidal Behavior Questionnaire-Revised was administered to evaluate lifetime suicidal ideation, suicide attempts, and the self-reported likelihood of future suicidal behavior. This survey gives the opportunity to obtain sensitive information from individuals who may have difficulty revealing them in other settings which may not be anonymous. Statistical analysis will be conducted using Chi-squared, Two-Sample Independent T-tests and ANOVA.
Results: Data was collected from October 2019 until August 2020. Approximately 75% (190/255) of participants were AA and 25% (65/255) were non-AA. Of the participants, 74% (188/255) were female while 26% (67/255) were male. Our findings suggest that 28% (54/190) of the AA respondents displayed suicidal ideation compared to 49% (31/65) for non-AA respondents. A comparison between AA immigrants and non-immigrants demonstrated that 25% (11/44) of AA immigrants had suicidal ideation versus the 29% (43/146) prevalence of suicidal ideation among AA non-immigrants.
Impact: Overall, the trends demonstrate that AAs have a lower prevalence of suicidal ideation compared to non-AAs, our study will demonstrate the need for bilingual programs such as suicide prevention trainings, suicide hotlines, and support groups.
Searching for Post-Translationally Modified Proteins in an ALS Patient Cohort
Public Health and Medicine
Kevin L. Yang, Department of Computational Medicine and Bioinformatics
Daniel J. Geiszler, Department of Computational Medicine and Bioinformatics
Sarah E. Haynes, Department of Pathology
Fengchao Yu, Department of Pathology
Daniel A. Polasky, Department of Pathology
Hui-Yin Chang, Department of Pathology
Alexey I. Nesvizhskii, Department of Computational Medicine and Bioinformatics, Department of Pathology
Amyotrophic lateral sclerosis (also known as ALS or Lou Gehrig’s disease) is characterized by the degeneration of motor neurons and affects hundreds of thousands of people across the world. Although treatment is available to improve quality of life, there is no known cure for ALS. To investigate the ALS proteome, the collection of all proteins in a biological sample, the international #ALSMinePTMs challenge was created. Teams from across the globe reanalyzed a publicly-available liquid chromatography-tandem mass spectrometry (LC-MS/MS) dataset of cerebrospinal fluid samples from 29 healthy and 32 ALS patients, with a focus on post-translational modifications (PTMs) of proteins that could serve as disease biomarkers. Using our in-house proteomics software tools, we analyzed the data, and out of over 25 submissions from around the world, our work was selected as the winning submission by the judges. Our ability to perform quick “mass-tolerant searches” identified potential PTMs across samples. We then explored individual PTMs via more sensitive and traditional database searches. Comparing amounts of modified peptides between the healthy and ALS cohorts, we identified significant changes in abundance of 10 dichlorinated, 10 phosphorylated, and 6 glycosylated peptides. Many of these differentially modified peptides were from proteins previously implicated in neurodegenerative diseases, including O-glycosylation of amyloid-beta precursor protein and N-glycosylation of ceruloplasmin. We also found differential phosphorylation of the iron-binding proteins serotransferrin and hemopexin. Finally, we searched the literature to propose possible reasons for the differences in PTMs, including iron dysregulation, handling of oxidative stress, and myeloperoxidase activity. The hosts of the competition are currently compiling the findings of all the teams into a manuscript that will provide deeper insight into the ALS proteome and help guide future research towards PTM biomarker discovery of this serious disease.