BioSense: Using Health Data for Early Event Detection and Health Situational Awareness

1
BioSense Using Health Data for Early Event
Detection and Health Situational Awareness

• West Texas Preparedness Conference
• August 3, 2006
• Colleen A. Martin, MSPH
• Epidemiologist, SAIC
• BioSense Program
• Division of Emergency Preparedness and Response
• National Center for Public Health Informatics
• Centers for Disease Control and Prevention

2
Topics

• Purpose, vision, and approach
• Data sources
• Disease indicators
• Data analysis
• Users and jurisdictions
• BioIntelligence Center
• BioSense application interface
• Future application enhancements

3
BioSense Purpose
BioSense is a national program intended to
improve the nations capabilities for conducting
near real-time biosurveillance and health
situational awareness through access to existing
data from healthcare organizations.
4
BioSense Vision

• Provide local, state, and nationwide health
  situational awareness
• For suspect illness
• Cases of disease
• Before, during, and after a health event
• Help confirm or refute the existence of an event
• Monitor disease outbreak
• Size
• Location
• Rate of spread

5
BioSense Approach

• Real-time delivery of healthcare data to BioSense
  from
• Hospitals
• Laboratories
• Ambulatory care settings
• Other health data sources
• Electronic views, analytics, and reports for
• National public health, CDC
• State and local public health
• Contributing healthcare organizations

6
Early Event Detection and Health Situational
Awareness

• Great promise in using health related data,
  analysis, and visualization for initial event
  detection
• Established value in public health decision
  makers knowing
• Is there really something going on?
• Where is it?
• How big is it?
• Is it spreading?
• Is our response working?

7
Enables Public Health Access to
Jurisdiction-Specific Information

• Developing a nationwide capability for
  simultaneous access to biosurveillance data by
  all levels of public health
• Able to provide cross-jurisdictional information
  during a public health event

8
Other Essential BioSense Concepts

• Based on PHIN standards for vocabulary,
  messaging, security (consistent with NHIN,
  HITSP)
• A national system provides economies of scale
• Advances local public-private partnerships
• Considers and protects patient privacy
• Program includes rigorous evaluation
• User input sought for system development, ongoing
  improvement

9
Clinical Data of Interest

• Foundational demographics, chief complaint,
  discharge diagnoses, disposition, hospital
  utilization
• Clinical vitals, triage notes, working
  diagnosis, discharge summary
• Laboratory orders, microbiology results
• Pharmacy medication orders
• Radiology orders, interpretation results

10
Implementation Targets and Status National
Healthcare Data Sources
In 2006
11
Implementation Targets and StatusHospitals /
Healthcare Systems

12
Potential Map of BioSense Hospitals, 2006
Includes real-time data from participating
private and public healthcare systems, and VA
hospitals
13
Initial Data 2003 - Present

• Outpatient clinics
• ICD-9 diagnosis codes and CPT procedure codes
• Department of Defense (DoD) outpatient Medical
  Treatment Facilities (n 355)
• Veterans Affairs (VA) outpatient medical centers
  and clinics (n 849)
• Laboratory Corporation of America (LabCorp)
• Lab orders (local and LOINC codes)
• Reason for lab orders (ICD-9 codes)

14
Hospital Data Dec. 2005 - Present

• Clinically rich data in real time from hospitals
  in major US cities
• Foundational data types chief complaint,
  diagnosis, demographics, hospital census
• Additional data types ED clinical (e.g. vital
  signs), laboratory, radiology, pharmacy
• Hospital patient class emergency department,
  inpatient, outpatient
• 37 hospitals transmitting foundational data

15
Disease Indicators

• Chief Complaint / Diagnosis
• Syndrome (N11)
• Sub-syndrome (N78)
• Preprocessing
• Key word search for free text chief complaints
  and diagnoses
• ICD-9 codes grouping for coded diagnoses

16
Free-Text Processing Methods

• First phase key word search
• Two key word lists
• Chief complaint to sub-syndrome
• Free-text diagnosis to sub-syndrome
• Free text approximate the ICD-9 sub-syndromes
• Chief complaint
• Adapted EARS, NYC, Seattle key word lists
• Modified through sample data
• Diagnosis no prototype available
• Extracted terms from ICD-9 code description
• Expanded key word list using UMLS synonyms
• Modified through sample data

17
Free-Text Processing Enhancements

• Search methods
• Exact match, partial match, word starts-with,
  numeric input, regular expression
• Negation handling do not match key words within
  a 7-word window after NOT
• Second phase more sophisticated natural language
  processing methods

18
Syndromes (N 11)

• 11 syndromes
• Botulism-like, Fever, Gastrointestinal,
  Hemorrhagic illness, Localized cutaneous lesion,
  Lymphadenitis, Neurological, Rash, Respiratory,
  Severe illness/death, Specific infection
• http//www.bt.cdc.gov/surveillance/syndromedef/ind
  ex.asp
• Monitor critical bioterrorism-associated and
  natural infectious disease outbreaks

19
Sub-syndromes (N 78)

• Objectives
• Monitor bioterrorism-associated and
  naturally-occurring disease
• More granular than syndrome
• Examples
• Botulism-like Paralysis, Speech disturbance,
  Dysphagia
• Gastrointestinal Abdominal pain, Anorexia,
  Diarrhea, Food poisoning, Intestinal infections,
  Nausea and vomiting
• Others Injury, Allergy, etc.

20
Hospital Census Data

• Hospital level
• Occupancy rate
• Admission count
• Discharge count
• Death count
• Unit level
• Occupied beds
• Available beds

21
BioSense Patient ID

• Unique for a patient to a single facility or a
  healthcare system
• No personally identifiable information
• Example 123451234
• Only hospital personnel can link with patient
  identifiers

22
BioSense Data Analysis

• Purpose determine if an increase has occurred in
  a disease indicator
• Currently CuSum method (CDC EARS)
• C1 current day compared with previous 7-day
  moving average
• C2 current day compared with 7-day moving
  average, with 2-day lag
• C3 sum of C2 values for previous 2 days
• Units of analysis facility, MRA, state
• In development more flexible and sophisticated
  methods (e.g., regression)

23
Users and Jurisdictions

• Users
• BioSense administrator for each facility and
  applicable state/local health department - grants
  rights to personnel in their jurisdiction
• Data available simultaneously to CDC and other
  users
• Jurisdictions
• Lowest level of data is the facility (e.g.
  hospital)
• Rights to view data granted to the facility,
  local/state health departments where facility
  located, and CDC
• Cross-jurisdictional views a possibility

24
CDC BioIntelligence Center Functions

• Monitor, analyze, and interpret BioSense and
  other relevant data
• Support state and local personnel
• Work closely with DEOC
• Assist with system troubleshooting and enhancement

25
BioSense Application Interface

• Core functionalities
• Automated analysis for statistical anomalies
• Time series display
• Drill down on individual patient records
• Visualize by location and time
• Descriptive data
• Continuing development guided by analysis, and
  user input

26
BioSense System Screen Shots
27
Beta Version
28
Data Access Agreement
29
Home Page
30
Chief Complaints and Diagnoses(Demonstration
Data)
31
Chief Complaint / Diagnosis Options
32
Statistical Anomalies(Demonstration Data)
33
Statistical Anomalies Options
34
Time Series(Demonstration Data)
35
Time Series Options
36
Select Data
37
Patient List(Demonstration Data)
38
Patient Map(Demonstration Data)
39
Patient Map Options
40
Describe(Demonstration Data)
41
Describe Options
42
Application Near-Future Plans

• Additional data analysis methods (regression,
  spatial-temporal methods)
• Custom event creator (define/modify
  syndromes/sub-syndromes)
• Custom time series graphs
• Custom tables
• Stratified analyses
• Additional data types (laboratory, radiology,
  pharmacy)
• Signal fusion and dashboard
• All data sources in one application

43
August Release Patient Class Stratification

• Outpatient
• Reason for visit
• Working diagnosis
• Final diagnosis
• Emergency department
• Chief Complaint
• Working Diagnosis
• Final Diagnosis
• Inpatient
• Reason for admit
• Working diagnosis
• Final diagnosis

Therefore, for each disease indicator, there are
9 potential analysis units
44
August Release Patient Class Stratification
45
August Release National Map
46
August Release Data Analysis

• Modified CuSum 2 approach compare weekdays to
  weekdays and weekends to weekends
• Empiric recurrence interval
• Uses frequency distribution of observed
  stratified by expected values
• Top 1 observed for a given expected value is
  equivalent to an empiric p-value of
  approximately 0.01 and recurrence interval of 100
  days

47
Near Future Laboratory Data

• Lab Orders
• Classified as culture, specific test, stain,
  identification, susceptibility test
• Cultures further categorized by specimen (e.g.,
  blood, CSF)
• Lab Results
• Classified as positive/negative if positive,
  which organism stain, positive/negative

48
Near Future Radiology Data

• Orders and results for chest X-rays, extremity
  films
• Possible disease indicators from chest X-ray
  infiltrate, pneumonia, widened mediastinum, etc
• Possible disease indicators from extremity films
  fracture, dislocation, etc
• Requires free-text parsing

49
Near Future Pharmacy Data

• Physician pharmacy orders
• Classification according to high-level
  pharmaceutical classes
• E.g., antibacterial, antiviral, HTN, anti-TB
• Antibacterial further classified as Penicillin,
  Aminoglycoside, etc

50
Conclusions

• Federal, state, and local public health and
  clinical partners access to existing health data
• Surveillance capabilities for naturally occurring
  and bioterrorism events
• Goal to add substantial data sources
• Beta phase with continual application enhancement
• Platform for monitoring multi- jurisdictional
  events

51
Thank You!

• Colleen A. Martin, MSPH
• Cmartin5_at_cdc.gov
• 404-639-7612