Bringing Healthcare Analytics to the Point-of-Care at Mayo Clinic

Mayo Clinic is bringing real-time clinical decision making to the point of care with a vertical application they built using the cloud, Elasticsearch, and D3 widgets. Hear how this application allows physicians to find similar patients and explore what-if scenarios using outcome and intervention parameters.

Transcript

1. 1 Peter Li, PhD February 19, 2016 @precbioinf Bringing Healthcare Analytics to Point-of-Care
2. 2 B
   ioinformatics, LLC
   P
   recision
   Bio •  Background Systems Architect Genome Scientist Bioinformatician Clinical Informatician •  Founder: Precision Bioinformatics, li.peter@precisionbioinformatics.com Peter Li, PhD
3. 3 B
   ioinformatics, LLC
   P
   recision
   Outline Introduction to Medicine Cloud Deployment @ Point of Care Elasticsearch Tools Healthcare Analytics User Interface
4. 4 B
   ioinformatics, LLC
   P
   recision
   Mayo Clinic •  Largest integrated nonprofit medical group practice •  No. 1 on the List of "Best Hospitals in US" 1 •  4,000 physicians and scientists •  50,000 allied health staff •  > 1 million unique patients per year •  Leader in electronic medical record (EMR) applications 12014–2015 U.S. News & World Report
5. 5 B
   ioinformatics, LLC
   P
   recision
   Enhanced Analytics for Surgical Excellence (EASE) •  Objective: Improve surgical outcomes by managing peri-operative care Clinical : comfort, complication, recovery, … Operational : performance, compliance, stress, … Financial : length of stay, reoperation, readmission, … •  Sponsor: Dr. David Larson, Chief of Colorectal Surgery, Mayo Clinic Making data meaningful, accessible, and actionable
6. 6 B
   ioinformatics, LLC
   P
   recision
   Past •  Background Medical History Family History Social History … Data at Point-of-Care •  Current Condition Treatment Plan Diagnoses Medications Lifestyle … Present •  Outcomes Return to Health Status Quo Complication … Future Medical Knowledge + Clinical Experience
7. 7 B
   ioinformatics, LLC
   P
   recision
   Medical Knowledge : Principles of Medicine Training, Fellowships, and CMEs Conferences and Conventions Books and Journals
8. 8 B
   ioinformatics, LLC
   P
   recision
   Clinical Experience : Patient-Specific Patterns Office Visits Inpatient Care Procedures Gained one patient at a time !
9. 9 B
   ioinformatics, LLC
   P
   recision
   How Can We Expand Clinical Experience •  Improve recall of past patients Eliminate subjective bias of recall •  Overcome the “one-at-a-time” limitation Integrate colleagues’ experiences And experiences from other institutions
10. 10 From “Backoffice” Analytics … Notes Claims Struct. Data Limitations: Mining over the “whole” population Not specific to any given patient After care has been performed Notices Alerts Messages Mining Warehouse And Cubes 9   Metrics Models EMR Clinical Activities
11. 11 To Analytics @ Point-of-Care Outcomes Mining 9   Metrics Models Realtime Analytics Informed Decisions @ Point of Care Interactive Visualization Search Engine Notes Claims Struct. Data EMR Clinical Activities
12. 12 B
    ioinformatics, LLC
    P
    recision
    Introduction to Medicine Cloud Deployment @ Point of Care Elasticsearch Tools Healthcare Analytics User Interface
13. 13 B
    ioinformatics, LLC
    P
    recision
    Platform-as-a-Service •  Patient data is uniform: all patient has the same data types •  Patient data can scale practice (clinic): ~1 K – 10 K patients department (regional): ~10 K – 1 M patients institutional (multi): >1 M patients •  Candidate for a PaaS cloud solution: Uniform data in a scalable environment Configuration-driven, on-demand deployment •  Straightforward, except for presence of Protected Health Information
14. 14 B
    ioinformatics, LLC
    P
    recision
    Privacy and Security •  Protected Health Information (PHI) IDs Names Addresses Phone Numbers •  HIPAA Requirements Encryption at rest (files) Encryption in flight (network) User authentication (who is on) Activity logging (who saw what) •  Institutional Requirements Passwords, keys, and certificates must be managed locally User identity and authentication must be compatible with existing infrastructure
15. 15 B
    ioinformatics, LLC
    P
    recision
    Architecture of Cloud Deployment Encrypted Data Volum e Azure Firewall Azure Cloud Environm ent Corporate VPN G ateway Azure VPN G ateway Cloud Nodes Cloud-based VM s Cloud Nodes Cloud-based VM s Internet Encrypted Traffic SSL encrypted packets LUKS encrypted blocks Azure Portal Deploym ent Node /s Linux/shell or W indows/powershell Secure RESTful Azure Com m ands and Response Azure Deploym ents Intranet Environm ent Corporate Firewall SSH M ulti-Factor Authentication Config Files Restricted Perm issions SSH for nodes Deployed Assets Firewall Encrypted Data Volum e Deploym ent Data, Keys, Passwords Encryption At Rest Encryption In Flight
16. 16 B
    ioinformatics, LLC
    P
    recision
    Encryption at Rest Unattached Disk Volumes Password OS + Processes Attaching A Disk Detaching A Disk "Physical" ”Logical" Encryption Decryption Engine Block Cipher Original file system structure, encrypted blocks LUKS dm-crypt 64+ random char Sent at startup
17. 17 B
    ioinformatics, LLC
    P
    recision
    Encryption In Flight Deployment Node/s Linux/shell or Windows/powershell https User ES Slave VM Encrypted Data Volume ES Slave VM Encrypted Data Volume ES Master VM Encrypted Data Volume ... ES Master VM Encrypted Data Volume LDAP Nodejs VM Config & Encrypted Certificate Files Config & Enscrypted Certificate Files VPN Tunnel ssh/scp https https ssh configuration •  SHIELD •  But can’t leave any passwords in config file •  Use ${prompt.secret} •  Send keystore password during startup through ssh •  Helper process to manage stdin/out and daemonization
18. 18 B
    ioinformatics, LLC
    P
    recision
    Subnet and Firewalls MasterNode(s) Data Nodes Data Volumes Virtual Subnet 3 Web Nodejs Virtual Subnet 1 •  OS-level firewall •  limit open ports •  virtual subnet •  iptables (or equivalent) Users Virtual Subnet 2
19. 19 B
    ioinformatics, LLC
    P
    recision
    User/Role Authentication Operating System Disk Encryption Network Firewall Vended Apps Project Software Vended Apps IT/System Administrator: 1.  VM 2.  OS 3.  Disk 4.  Network 5.  Logs App/Project Administrator: 1.  Project 2.  Application 3.  Logs •  Separation of IT/System Admin from App/Project Admin roles •  User identity/authentication managed by Intranet LDAP Policies •  LDAP client needs to be in the Intranet (no SHIELD LDAP realm in the cloud) •  User activity via “run as” (untested)
20. 20 B
    ioinformatics, LLC
    P
    recision
    Outline Introduction to Medicine Cloud Deployment @ Point of Care Elasticsearch Tools Healthcare Analytics User Interface
21. 21 B
    ioinformatics, LLC
    P
    recision
    Surgery Med Outcome LabB Prediction Using The Past Patient 0 Surgery LabA Vital Patient 1 Patient 2 Surgery LabB Outcome LabA LabA LabA ??
22. 22 B
    ioinformatics, LLC
    P
    recision
    Reference Events Event Event Event Event Date and Time of •  Symptom Onset •  First Diagnosis •  Admission •  Surgery •  Treatment •  Planning •  Complications •  Resolution •  Discharge •  Medications •  Lab Values •  Observables •  Flowsheet Any event can be set as reference event +Z time -Y time -X time
23. 23 B
    ioinformatics, LLC
    P
    recision
    Id source value Name first last BirthDate Gender Race … Id source value Type ClinDate Display Observation value Order … Id source value RefEvent RelTime … Id source value RefDef … Id source value Display EventType EventTime PatientQual … Patient-Event Relationships Patient Event Relative Time Ref Event Ref Defn contains is-a contains generates generates •  Medicine is big-data complex J >10 K Events per patient case Volume, Variety, Velocity, Veracity •  Constant (and relentless) streaming of Events Detect Reference Events Generate RelativeTimes •  à Parent-Child relationship
24. 24 B
    ioinformatics, LLC
    P
    recision
    Percolator for Streaming Data Electronic Medical Record Event index Percolator For RelTime Rel Time creates triggers index Percolator for RefEvent Ref Event Generates RelTimes creates RefEvent Query Web Service stream
25. 25 B
    ioinformatics, LLC
    P
    recision
    Id source value Name first last BirthDate Gender Race … Id source value Type ClinDate Display Observation value Order … Id source value RefEvent RelTime … Reduce Query Complexity Patient Event Relative Time contains contains A B C Patients who: Live in MN Has hemoglobin lab value <= 10 1 day prior to surgery And return those values Elasticsearch Pseudoquery: { query: { <clause A>, child: { Event : { <clause B>, child: { RelativeTime : { <clause C> } } } } }, inner : { Event : { query : { <clause B>, child : { <clause C> } }, inner : { RelativeTime : { query : { <clause C> } } } } } }
26. 26 B
    ioinformatics, LLC
    P
    recision
    
27. 27 B
    ioinformatics, LLC
    P
    recision
    { "query": { "_source": [ "addresses", "names" ], "filter": { "and": [ { "nested": { "path": "addresses", "filter": { "term": { "addresses.state._raw": "MN" } } } }, { "has_child": { "type": "Event", "filter": { "and": [ { "and": [ { "term": { "observation.name.display._raw": "Hemoglobin" } }, { "range": { "observation.value.float": { "lt": 10 } } } ] }, { "has_child": { "type": "RelativeTime", "filter": { "range": { "relativeDate.mid": { "gte": 0, "lt": -86400000} } } } } ] } } } ] },
28. 28 B
    ioinformatics, LLC
    P
    recision
    Introduction to Medicine Cloud Deployment @ Point of Care Elasticsearch Tools Healthcare Analytics User Interface
29. 29 B
    ioinformatics, LLC
    P
    recision
    Patient A had a surgery… Hemoglobin Surgery on May 22, 2015 Where would the value go ? Is patient at risk for bleeding complication (low hemoglobin) ? Reference Event = surgery Events of interest = diagnoses and lab values
30. 30 B
    ioinformatics, LLC
    P
    recision
    Find other patients, align, and overlay data lines … Patient B Patient C Patient D Patient E
31. 31 B
    ioinformatics, LLC
    P
    recision
    Growth Chart (aka Contour Graphs)
32. 32 B
    ioinformatics, LLC
    P
    recision
    Construct Contours •  Create line segments between adjacent data points •  Divide window into vertical slices (>20) •  For each slice, calculate intersections for each data line •  Determine the 5, 10, 25, … percentile Y intercept-value in each vertical slice •  Connect the percentile (5, 10, 25, …) points into smoothed curves
33. 33 B
    ioinformatics, LLC
    P
    recision
    Realtime User Interaction Add Constraints: Select only patient data lines that pass through bounding box Cohort (set of patients) then becomes more like our patient hemoglobin pain score urine output heart rate systolic bp
34. 34 B
    ioinformatics, LLC
    P
    recision
    Multiple @ Point-of-Care Applications •  Clinical Registries: cohort for diseases, treatments •  Retrospective Evaluation: compare treatment, provider outcomes •  Targeted Outcome: likelihood of success, necessary actions •  Educational Tool: what-if scenarios, e-attending
35. 35 B
    ioinformatics, LLC
    P
    recision
    Summary •  EASE: Making data meaningful, accessible, and actionable •  Representing clinical experience using a graphical metaphor •  Proof-of-concept for delivering analytics at point-of-care Secure Cloud Deployment Elasticsearch query builder Intuitive, real-time user interface •  Evidence-Based, Personalized Medicine
36. 36 B
    ioinformatics, LLC
    P
    recision
    Future Directions •  Text data (contour graph analogy) •  Intuitive analytics, statistical/machine learning •  Optimization of schema and performance •  Usability trials
37. 37 B
    ioinformatics, LLC
    P
    recision
    Acknowledgement •  Simon Yates, Sybaris, Toronto, Ontario •  Jenna Lovely, PharmD, BCPS, Mayo Clinic, Rochester, MN •  David Larson, MD, MBA, Mayo Clinic, Rochester, MN •  Many staff members on the EASE Program •  IT staff of Mayo Clinic
38. 38 B
    ioinformatics, LLC
    P
    recision
    Questions ? •  Email: li.peter@precisionbioinformatics.com