Slide 1: AI for COVID-19 Response
BOOZ ALLEN HAMILTON’S FEDERAL HEALTH AI EVENT SERIES AI FOR COVID-19 RESPONSE Federal Health Virtual Forum MAY 7, 2020
Slide 2: Welcome & Introductions
WELCOME & INTRODUCTIONS // DR . LAU R EN NEA L + DR . KEV IN V IG ILA NT E // D i r ec t or O f Hea l t h AI & C hi ef Medi c a l O f f i c er , B o o z Al l en Ha m i l t o n
Slide 3: Agenda
AGENDA + WELCOME & INTRODUCTIONS Dr. Kevin Vigilante, Booz Allen Hamilton + DIGITAL EPIDEMIOLOGY AND THE COVID-19 PANDEMIC Dr. John Brownstein, Boston Children's Hospital & Harvard Medical School + NLM DIGITAL DATA MANAGEMENT AND STANDARDS TO SUPPORT COVID-19 RESEARCH Dr. Kim Pruitt, National Librar
Slide 4: How to Maximize Your Time Today
HOW TO MAXIMIZE YOUR TIME TODAY SCREEN VIEW + In the bottom right corner of the video, you will find four options to show the presentation and speaker. You can also switch to full screen mode. Take a moment to choose what view is right for you. QUESTIONS + Following each presentation we will have a
Slide 5: NLM Digital Data Management and Standards to Support COVID-19 Research
NLM DIGITAL DATA MANAGEMENT AND STANDARDS TO SUPPORT COVID-19 RESEARCH // DR . KIM P R U IT T // B r a nc h C hi ef , Inf o r m a t i o n E ng i neer i ng B r a nc h, N a t i o na l L i b r a r y o f Medi c i ne (N L M)
Slide 6: NLM digital data management and standards to support COVID-19 research
NLM digital data management and standards to support COVID-19 research Dr. Kim D. Pruitt Branch Chief, Information Engineering Branch U.S. National Library of Medicine National Center for Biotechnology Information
Slide 7: PubMed Central support of CORD-19 & Kaggle dataset challenge
PubMed Central support of Covid-19 Open Research Dataset (CORD-19) & Kaggle dataset challenge >59,000 free scholarly articles about the novel coronavirus for use by the research community. Goal is to develop new text and data mining techniques that can help the science community answer high- priorit
Slide 8: COVID-19 Initiative CORD-19
COVID-19 Initiative Publishers and societies https://www.ncbi.nlm.nih.gov/pmc/about/covid-19/ Subject: Publications on COVID-19 and coronaviruses, more broadly. National Library of Medicine's PubMed Central License: Must allow for research re-use and CORD-19 secondary analysis. https://www.kaggle.co
Slide 9: SPHERES Consortium
SPHERES CONSORTIUM SARS-CoV-2 Sequencing for Public Health Emergency Response, Epidemiology, and Surveillance May 1, 2020 https://www.cdc.gov/coronavirus/2019-ncov/cases-updates/spheres.html Acknowledgement: Ilene Mizrachi & GenBank team, Rodney Brister & Virus team, Chris O’Sullivan & SRA team
Slide 10: SPHERES
U.S. & International Sequencing Laboratories SPHERES Submit: https://submit.ncbi.nlm.nih.gov/sarscov2/ Subject: SARS-CoV-2 assembled genomes and raw sequence data National Library of Medicine’s GenBank & SRA Tools: genome annotation, rapid Virus Hub: https://go.usa.gov/xdbuc submissions, BLAST, k- m
Slide 11: Data Management
Data Management Support Expand Data Submitters Communication Enable Data Enhance Discovery Workflows Support Enhance Collaborations Tools U.S. National Library of Medicine COVID-19 data is available from: Credit: NationalNational Institute of Allergy and Center for Biotechnology Information PubMed -
Slide 12: COVID-19 Twitter ML Analysis and the COVID Workbench
COVID-19 TWITTER ML ANALYSIS AND THE COVID WORKBENCH // M S . CAT H ER INE O R DU N + DR . EDWA R D R A F F + M R . A LEX H A NS O N // S eni o r As s o c i a t es , B o o z Al l en Ha m i l t o n // DR . SA NJAY P U R U S H OT H A M // U MB C , C o - Aut ho r
Slide 13: What can we learn about COVID-19 on Twitter?
WHAT CAN WE LEARN ABOUT COVID-19 ON TWITTER? @realDonaldTrump In New 700,000 Chinese arrived Thank you for your work. How to prevent or to cure from #covid Zealand, there has been a in NY after Trump's Its a changing picture, we #corona : Daily dose: 4 x 2000mg decrease in the number of travel ban f
Slide 14: To date, few analyses on conversations about COVID-19 in Twitter
TO DATE, FEW ANALYSES ON CONVERSATIONS ABOUT COVID-19 IN TWITTER Keyword Question 1: What high-level trends can be inferred Analysis from COVID-19 tweets? No retweets Topic Question 2: Are there any events that lead to spikes in 5.5 M Tweets Modeling COVID-19 Twitter activity? Uniform Manifold Appro
Slide 15: Data collection consisted of 23.8M tweets
DATA COLLECTION CONSISTED OF 23.8 M TWEETS Booz Allen Hamilton, Inc. | BoozAllen.com/HealthAI
Slide 16: Keyword analysis reveals high level themes
KEYWORD ANALYSIS REVEALS HIGH LEVEL THEMES Booz Allen Hamilton, Inc. | BoozAllen.com/HealthAI
Slide 17: Highly attuned to government broadcasts
HIGHLY ATTUNED TO GOVERNMENT BROADCASTS Booz Allen Hamilton, Inc. | BoozAllen.com/HealthAI
Slide 18: High dimensional visualization
HIGH DIMENSIONAL VISUALIZATION UMAP represents high dimensional data such as text in a low-dimensional 2D space to improve understanding of distinct groups of tweets.
Slide 19: Tweeting is the mechanism of spreading information on Twitter
TWEETING IS THE MECHANISM OF SPREADING INFORMATION ON TWITTER COVID-19, 2020 - MEDIAN TIME 2.78 HOURS H7N9, 2013 - MEDIAN TIME 3.8 HOURS
Slide 20: Density of COVID-19 tweeters over time
DENSITY OF COVID-19 TWEETERS OVER TIME QUITE RAPIDLY AT 19 SECONDS, ALREADY USERS BEING RETWEETED BY 24 HOURS, MULTIPLE DENSE CONNECTIONS BY 1 WEEK, ONE MAJOR ACCOUNT BEING RETWEETED
Slide 21: Key Takeaways
KEY TAKEAWAYS 1. Conversations on Twitter during March 24–April 8, 2020 focused more on healthcare concerns such as lack of masks, ventilators, healthcare workers, and highly attuned following of U.S. Government broadcasts. 2. Machine learning methods not typically applied to Twitter can be used to
Slide 22: State map county map CHIME tweets
Stotemap County ap chime Twsete CHIME “ ive tet ret by county Projected dy adicons Positive Test Results 1 wo 10+
Slide 23: State map county map CHIME tweets
Weeovia woredonch ‘tateMap County Map chime Tweets Backto State Map Georgia | © click county on the map for more details.
Slide 24: Focusing Assessment and Understanding of Human Pathogen Impacts: Integrated COVID-19 Data Analytics
FOCUSING ASSESSMENT AND UNDERSTANDING OF HUMAN PATHOGEN IMPACTS: INTEGRATED COVID-19 DATA ANALYTICS // CA PT KIM B ER LY ELENBERG // D ep a r t m ent of D ef ense, O f f i c e of t he S ec r et a r y of D ef ense (P ol i c y ), C O V ID- 1 9 Ta s k F o r c e ( C V TF )
Slide 25: Focusing Assessment and Understanding of Human Pathogen Impacts
Focusing Assessment and Understanding of Human Pathogen Impacts: Integrated COVID-19 Data Analytics 7 MAY CAPT Kimberly Elenberg, DNP
Slide 26: Decision Support Enablers to Achieve DOD End-State
Decision Support Enablers to Achieve DOD End-State COVID-19 DOD Decision-Support End-State Overall goal of CVTF Data Analysis line of effort is to ensure military readiness is maintained in the near and far-term as the full effects of the COVID-19 pandemic are realized How can COVID-19 Data Analytic
Slide 27: Interagency COVID-19 Analytic Network
Interagency COVID-19 Analytic Network DoD/Interagency Analytic Community of Interest (COI) REAL-TIME RELIABLE ANALYTICS REPORTING Civilian And DoD Partnerships Commercial Data and Data Sources Sources Virus MED Staff Equipment MTF Capacity Testing Transmission Availability
Slide 28: History: Lessons from the Influenza Pandemic 1918-1919
History- Lessons from the U.S. Military and the Influenza Pandemic 1918-1919 Informing Modeling Efforts Today CVTF Tasks: Provide coordinated and integrated COVID-19 disease modeling insights. Scenario-based COVID-19 modeling includes disease impacts on people, facilities and supplies by locality an
Slide 29: Challenge/Solution
Challenge/Solution Multiple Data Collection and Analysis Efforts Complicate Situational Awareness § CVTF is Integrating DoD-wide Analytic and Modeling Approaches o Executive COP utilizing the ADVANA Platform o Employ analytics from multiple stakeholders (CAPE, DTRA, DHA, Services, Interagency partne
Slide 30: Types of Data/Analysis?
Types of Data/Analysis? § COVID-19 Cases and Testing Data Data Analytics and o DoD Modeling Informs DoD/US o HHS COVID-19+ Future COVID-19 o Open Source Impact COP § DoD Medical System Data However- it is very o Bed Data important for decision Military makers to know how o Personnel Data Healthcare
Slide 31: Outside-the-Gate – Current and Future Hotspots
“Outside-the-Gate” – Current and Future Hotspots DoD Installation Military Treatment Facility Area Selected (Ft. Hood + Neighbors) Identify hotspots of COVID-19 transmission relative to DoD installations Advana
Slide 32: Inside the gate – MTFs and COVID-19 Projections
“Inside the gate”– MTFs and COVID-19 Projections Example Scenario COVID-19 Hosp. COVID-19 ICU Cases Total Beds Ventilation Required Scenario analysis available by MTF using real time bed-supply data from DHA Advana
Slide 33: What's Next?
What’s Next? Future efforts include the modeling and simulation of different injects and mitigation measures taken by the DoD, USG, states, and international partners Objectives/goals: § CVTF will coordinate to assess impact of COVID-19 resource allocations on planning (such as Time Phase Force Depl
Slide 34: What's Next?
What’s Next? § Additional waves; Influenza and other influenza like illnesses § Atlantic Hurricane Season starts in June and forecasters are predicting a higher-than-average number of storms this season § Other?
Slide 35: Analytics/Modeling Pearls
Analytics/Modeling Pearls § An existing, established common platform for analysis and data sharing could increase efficiency and help mitigate morbidity and mortality; requires, computing power, hardware, architecture, software, personnel to clean and ingest data, and a vehicle that can rapidly scal
Slide 36: Agent-Based Modeling for Resource Planning
AGENT-BASED MODELING FOR RESOURCE PLANNING // DR . A M IR M O KH TA R I + DR . JEF F R EY KR IS EM A N // S eni o r As s o c i a t es , B o o z Al l en Ha m i l t o n
Slide 37: Main Objectives
MAIN OBJECTIVES Quantify the COVID-19 burden of disease and SIMULATE respective interventions using model-based DISEASE BURDEN simulations grounded in epidemiology. EVALUATE IMPACT Estimate the impact of COVID-19 spread on ON THE the healthcare system (e.g., bed and HEALTHCARE ventilator utilization
Slide 38: Combining Agent-Based Modeling with System Dynamics
COMBINING AGENT-BASED MODELING WITH SYSTEM DYNAMICS * Work in-progress COHORT-SPECIFIC NO. OF CHRONIC CONDITIONS SYNTHETIC POPULATION DATA AT COUNTY LEVEL Booz Allen Hamilton, Inc. | BoozAllen.com/HealthAI
Slide 39: Ability to calibrate model parameters using observed cases
ABILITY TO CALIBRATE MODEL PARAMETERS USING OBSERVED CASES OF ILLNESS AND DEATH List of Selected Parameters Used in the Model Calibration: + Under-reporting + Transmission rate for mild, severe, and critical infections + Hospital-acquired transmission + Social distancing + Percent critical infection
Slide 40: County-level predictions for disease dynamics
COUNTY-LEVEL PREDICTIONS FOR DISEASE DYNAMICS Booz Allen Hamilton, Inc. | BoozAllen.com/HealthAI
Slide 41: County-level predictions for hospital utilization
COUNTY-LEVEL PREDICTIONS FOR HOSPITAL UTILIZATION Booz Allen Hamilton, Inc. | BoozAllen.com/HealthAI
Slide 42: Ability to simulate wide-spread testing strategies
ABILITY TO SIMULATE WIDE-SPREAD TESTING STRATEGIES IN ADDITION TO SOCIAL DISTANCING (A) (C) End of social End of social distancing distancing • A: End of social distancing in June • B: End of social distancing in June + 10k/day testing • C: End of social distancing in June + 20k/day testing (B) (D)
Slide 43: Next Steps for Model Development
NEXT STEPS FOR MODEL DEVELOPMENT + Including county-level data regarding number of chronic conditions within each age cohorts and evaluating the impacts on death rates due to critical infections + Addition of “real-time” telemetry data for improved model predictions (e.g., population movements, soci
Slide 44: COVID-19 Simulator
COVID-19 SIMULATOR ‘A Multi-Method Approach for Modeling Infectious Disease Dynamics in Communities OVERVIEW Overview The COVID-19 pandemic is confronting every level of the U.S. economy with an unprecedented challenge, and the goverment must mount a sustained, ambitious economic response Tasting mo
Slide 45: AI & COVID-19
AI & COVID-19 // DR . T H O M A S O S BO R NE // C hi ef Medi c a l Inf o r m a t i c s O f f i c er , D i r ec t or , N a t i o na l C ent er f o r C o l l a b o r a t i ve Hea l t hc a r e Inno va t i o n (N C C HI), V A P a l o Al t o Hea l t h C a r e S y st em
Slide 46: AI & COVID-19
Ai & COVID-19 Thomas Osborne, MD • Director, National Center for Collaborative Healthcare Innovation • Chief Medical Informatics Officer (CMIO), VA Palo Alto
Slide 47: EHR insights, physical distancing, early detection sensors
EHR insights Empower physical distancing Early detection sensors
Slide 48: VEO Veterans Signals: COVID-19 Customer Experience Data
VEO VETERANS SIGNALS: COVID-19 CUSTOMER EXPERIENCE DATA // M R . A A RO N P ET ERS // L ea d As s o c i a t e, B o o z Al l en Ha m i l t o n
Slide 49: The Veterans Experience Office measures veteran customer experience
THE VETERANS EXPERIENCE OFFICE WORKS WITH PROGRAM OFFICES AROUND THE VA TO MEASURE THE VETERAN CUSTOMER EXPERIENCE Booz Allen Hamilton, Inc. | BoozAllen.com/HealthAI
Slide 50: Veterans Signals is a web-enabled customer feedback tool
VETERANS SIGNALS IS A WEB-ENABLED CUSTOMER FEEDBACK TOOL TO IDENTIFY OPPORTUNITIES IN SERVICE RECOVERY AND SYSTEMS IMPROVEMENT + VEO has deployed over 34 surveys across 14 lines of business within the VA + Veterans are sent an email link shortly after their interaction with the specific service (e.g
Slide 51: Machine learning facilitates classification of veteran-reported experiences
MACHINE LEARNING FACILITATES THE CLASSIFICATION OF VETERAN-REPORTED EXPERIENCES, INCLUDING QUALITY OF CARE ANALYSIS OF VETERAN-REPORTED EXPERIENCE DATA IS OFTEN LIMITED TO STRUCTURED SURVEY FIELDS Self-assessments and survey responses provide insight into the experiences of individuals in the health
Slide 52: NLP and ML risk index for identifying concerns about CX
AN NLP AND MACHINE LEARNING RISK INDEX PROVIDES A PROACTIVE APPROACH FOR IDENTIFYING AND RESPONDING TO CONCERNS ABOUT CX A COMBINATION OF SUPERVISED AND UNSUPERVISED TOPIC MODELING IDENTIFIES EMERGING CONCERNS A machine learning approach developed for the Veterans Health Administration (VHA) identif
Slide 53: VEO rapidly deployed new topic models related to COVID-19
AS COVID-19 EMERGED, VEO WORKED TO RAPIDLY DEPLOY NEW TOPIC MODELS RELATED TO COVID-19 FEEDBACK VA STAKEHOLDERS AND SENIOR LEADERSHIP CAN NOW VIEW COVID-19 FEEDBACK IN REAL TIME When Veterans leave comments about COVID-19, they leave specific When Veterans leave comments about COVID-19, they concern
Slide 54: What's Next?
WHAT’S NEXT? VEO WILL CONTINUE TO USE AI TO MONITOR FOR NEW SUBTOPICS RELATED TO COVID-19 AND REPORT ON VETERAN SENTIMENT
Slide 55: Artificial Intelligence: Addressing a Global Pandemic
ARTIFICIAL INTELLIGENCE: ADDRESSING A GLOBAL PANDEMIC // DR . S U SA N G R EG U R ICK // Asso c i a t e D i r ec t o r f o r D a t a S c i enc e a nd D i r ec t o r o f t he O f f i c e o f D a t a S c i enc e S t r a t eg y (O D S S ), N a t i o na l Ins t i t ut es o f Hea l t h (N IH)
Slide 56: Artificial Intelligence: Addressing a Global Pandemic
Artificial Intelligence: Addressing a Global Pandemic Susan Gregurick, Ph.D. Associate Director for Data Science Office of Data Science Strategy May 7th, 2020
Slide 57: AI in Biomedicine: Opportunities
AI in Biomedicine: Opportunities EMRs/EHRs Interpret genomic sequence Extract medical information from data to understand impact of text in EMRs/EHRs mutations on protein function Monitor sleep and vitals to send Determine which calls to child Read medical images information about welfare systems wa
Slide 58: AI Investments Reach Across NIH
AI Investments Reach Across NIH TOTAL NIH SPENDING IN FY2018 (~$283 MILLION) TR AA AG OD AI NS AR NR LM CA HL HD DA GM DC DK EY EB AA AG AI AR CA DA DC DK EB EY GM HD HL LM NR NS OD TR
Slide 59: Artificial intelligence predicts heart disease risk from CT scans
Artificial intelligence predicts heart disease risk from CT scans A research team led by Drs. Perry J. Pickhardt of the University of Wisconsin and Ronald Summers from the NIH Clinical Center has been developed and applied AI to estimate disease risk from CT scans. The AI programs measured calcifica
Slide 60: NIH NVIDIA COVID CT-AI Classification
NIH NVIDIA COVID CT-AI Classification Preprocessing Segmentation Lung Segmentation Mask CT images have been Conversion to nifti with 1x1x1 resampling AH-Net Architecture used in Asia to detect COVID-19 virus in Apply Mask Image Classification patients dicoms nifti Classification Likely COVID 3D-Dens
Slide 61: Examples of Use Cases of AI in Pharmaceutical Arena
Examples of Use Cases of AI in Pharmaceutical Arena • Precision drug discovery: use of deep learning, super computers • Time saved: from years to days • Identifying targets in drug design • Predicting drug response requires data on: • Molecular profiles • Genomic aberrations • Point mutation • Delet
Slide 62: Research in Developing and Using AI
Research in Developing and Using AI AI Applied to Drug Design R01GM138340, David Koes, University of R01GM126189, Guowei Wei, Michigan State Pittsburgh University Goal: Improved protein-ligand scoring by Goal: Develop deep learning methods for applying deep learning and image recognition prediction
Slide 63: AI Approaches to Predicting Drug Response: Points for Consideration
AI Approaches to Predicting Drug Response: Points for Consideration • Which datasets should be selected for training: o To detect putative association between molecular features and drug response o Requires large volume of data o Needs to pre-process data to reduce noise o Needs to correctly label d
Slide 64: NIAMS Requesting Input to Advance AI-ready Data Generation
NIAMS Requesting Input to Advance AI-ready Data Generation and Scalable Computational Approaches The National Institute of Arthritis and Musculoskeletal and Skin Diseases (NIAMS) is seeking public comments on the needs and opportunities to develop, advance, and implement scalable computational appro
Slide 65: COVID-19 open-access data and computational resources
• COVID-19 open-access data and computational resources are being provided by federal agencies, including NIH, public consortia, and private entities • The COVID-19 HPC Consortium comes together to provide computing capabilities from some of the most powerful and advanced computers in the world • NI
Slide 66: COVID-19 Digital Pathology Repository (DPR)
COVID-19 Digital Pathology Repository (DPR) Curated digital whole slide image reference dataset for COVID and related pathogens (MERS, SARS-1, H1N1) for pathologists world-wide • Leverages existing NCI and NIH cloud infrastructure Based on HALO Link, an ODSS-supported NCI cloud project • Utilizes th
Slide 67: Office of Data Science Strategy
Office of Data Science Strategy www.datascience.nih.gov A modernized, integrated, FAIR biomedical data ecosystem @NIHDataScience /NIH.DataScience datascience@nih.gov
Slide 68: Open Questions
OPEN QUESTIONS // DR . LAU R EN NEA L // D i r ec t or O f Hea l t h AI, B o o z Al l en Ha m i l t o n
Slide 69: Closing Remarks
CLOSING REMARKS // DR . LAU R EN NEA L // D i r ec t or O f Hea l t h AI, B o oz Al l en Ha m i l t o n
Slide 70: Additional Resources
ADDITIONAL RESOURCES + Request a Demo or Meeting: ai@bah.com + Booz Allen’s Health Analytics and Artificial Intelligence Insights: boozallen.com/HealthAI + Perspective on AI bias in healthcare: https://www.boozallen.com/c/insight/blog/ai-bias-in-healthcare.html + Article describing why data manageme
Slide 71: About Booz Allen Acknowledgments
ABOUT BOOZ ALLEN ACKNOWLEDGMENTS For more than 100 years, business, SPEAKERS government, and military leaders DR. JOHN BROWNSTEIN have turned to Booz Allen Hamilton DR. KIM PRUITT to solve their most complex problems. They trust us to bring together the MS. CATHERINE ORDUN, DR. EDWARD RAFF, AND MR.
Slide 72: Thank You
THANK YOU // LEA R N M O R E // B o o zAl l en.c o m /Hea l t hAI // R EQ U EST A M EET ING // a i @ b a h.c om