BioSense Data Analyses and Anomaly Characterization

1
BioSense Data Analyses and Anomaly
Characterization

• Gabriel Rainisch, MPH
• Colleen A. Martin, MSPH
• Division of Emergency Preparedness and Response
• National Center for Public Health Informatics
• Centers for Disease Control and Prevention
• "The findings and conclusions in this
  presentation are those of the authors and do not
  necessarily represent the views of the Centers
  for Disease Control and Prevention."

2
Purpose Outline

• Purpose Summary of recent analyses using
  BioSense data
• To inform alerting monitoring protocols
• Improving the utility of BioSense
• Outline
• BioSense overview
• Descriptive analyses
• Cluster detection methods and anomaly
  characterization
• Epidemiologic studies

3
BioSense Data Sources

• Department of Defense (DoD n466) and Veterans
  Affairs (VA n905) outpatient medical facilities
• Daily data
• ICD-9 diagnosis codes and CPT procedure codes
• Hospital real-time data (n34)
• Chief complaints, diagnoses, demographics
• Patient class emergency department, inpatient,
  outpatient

4
Disease Indicators Syndromes (N 11)

• 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/in
  dex.asp)

• Monitor critical bioterrorism associated and
  natural infectious disease outbreaks

5
Disease Indicators Sub-syndromes (N 78)

• Newly developed
• Monitor infectious, chronic, and other disease
  indicators
• More granular than the 11 syndromes
• Require ongoing evaluation and enhancement
• 46 sub-syndromes map to syndromes
• Syndrome Gastrointestinal
• Sub-syndromes included diarrhea, abdominal pain,
  gastrointestinal hemorrhage
• 32 sub-syndromes do not map to syndromes
• Examples Allergy, Injury, Excessive heat

6
Analyses
7
Descriptive Analyses

• Understand and compare syndrome data from
  long-term sources (VA/DoD) vs. newer (Hospital
  Real-Time)
• Explore distribution of new disease indicators
  (sub-syndromes)
• Understand the capabilities for disease
  monitoring using both existing and new data and
  indicators

8
Visits Assigned to a Syndrome, Feb-June 2006(in
thousands)

• Hospital facilities on average receive 7 times
  the mean monthly visit volume of VA/DoD
  facilities
• The percent of total visits meeting a syndrome
  definition is similar for all data sources

9
Percent of Visits Assigned to a Syndrome

• Respiratory and Gastrointestinal combined,
  comprise 60-75 of binned visits
• The only marked difference between data sources
  is for the Gastrointestinal syndrome (Hospital,
  30 DoD 18 VA, 11)

10
Percent of Visits Assigned to a SyndromeHospital
Real-Time Data
11
Hospital Real-Time Final Diagnoses20 Most
Common Sub-Syndromes (Feb-Aug 2006)
12
Sub-syndrome Demographics

• Age effect
• Older ages more chronic conditions
• Younger ages more infectious, acute conditions
• Variation among sub-syndrome distributions by
  hospital system
• Differences in patient population (e.g. ages)
• Variation in patient class mix
• outpatient clinics, emergency department,
  inpatient

13
Descriptive Analyses

• Newer Hospital real-time data
• Expands capabilities for disease monitoring
• Is similar to VA/DoD data in many respects
• Combined analyses appear to be feasible

14
Cluster Detection Anomaly Characterization

• Modified CuSum C2 Statistic W2
• 7-day average with 2-day lag
• Weekdays compared with weekdays weekend days
  with weekend days
• Calculate a recurrence interval
• Reciprocal of a p-value
• Recurrence interval (RI) of days between two
  observed similar residual values
• Better analytic method that produces anomalies at
  an appropriate frequency

15
Anomaly Characterization

• What characteristics identify anomalies most
  likely to be of potential importance for further
  evaluation?

16
Ranking Data Anomalies

• Metrics already in use
• Recurrence Interval (RI)
• Observed vs. previous max count
• Observed / expected (relative risk)
• Observed expected (residual)
• of consecutive days flagged
• Other anomalies at the same facility/area
• Metrics slated for evaluation
• Severity of illness (e.g. pneumonia vs. cough)
• Homogeneity of chief complaints/diagnoses, age,
  gender, etc.

17
BioSense Epidemiologic Studies

• Influenza
• Neurological Syndrome
• Respiratory Syncytial Virus (RSV)
• Injuries following Hurricane Katrina
• Heat Related Illnesses

18
Heat Related Illness

• Goal Explore utility of BioSense data for heat
  related illness surveillance
• Hospital Real-Time Data (n 34 hospitals)
• Study period June 1st August 10th 2006
• Sub-syndrome Heat, Excessive
• Diagnosis 1 ICD-9 Code E900.0 (Excessive heat
  due to weather conditions)
• Chief complaint suggestive of heat injury

19
Heat Related Illness Visits, BioSense
Application, ICD-9 Diagnosis (n139 Visits)
20
Heat-Related Illness (HRI) Diagnosis, June 1-
August 10, 2006

• Several HRI diagnoses were identified that are
  not being used in BioSense
• Visit counts
• All HRI diagnoses 217 visits
• HRI diagnoses identified by current BioSense
  methods 139 visits
• Additional HRI diagnoses should be included in
  BioSense

21
Heat-Related Illness (HRI) Chief Complaints, June
1- August 10, 2006

• 217 visits with HRI diagnoses
• 194 visits with HRI diagnosis but not an HRI
  chief complaint
• 125 of these visits had chief complaints. Such
  as
• Malaise and fatigue 24
• Syncope and collapse 19
• Chest pain 19
• Dizziness 17
• Dehydration 14

22
HRI Indicators

• Breakdown of 297 HRI visits
• Male Female 6634
• Age
• 0-3 yrs. 2
• 4-11 yrs. 4
• 12-19 yrs. 12
• 20-49 yrs. 45
• 50 yrs. 37
• 19 of patients with an HRI chief complaint also
  mapped to the Cerebrovascular Disease
  sub-syndrome

23
HRI Analyses

• Text parsing creates challenges
• HEAT MURMUR
• IRREGULAR HEAT BEAT
  
• rapid heat beat
• HEAT PROBLEMS ???
• Next steps
• Correlate with temperature data
• Examine distribution of chronic conditions among
  historical visits

24
Summary

• Descriptive analyses
• Hospital Real-Time data adds new capabilities
• Disease indicator patterns similar to VA/DoD data
• Cluster detection methods modified methods and
  evaluation of ranking criteria can help
  prioritize anomalies for further investigation
• Epidemiologic studies ongoing to correct
  problems, expand capacities, and improve utility
  of BioSense

25
Acknowledgements

• Jerry Tokars, MD, MPH
• Colleen Martin, MSPH
• Data Quality Programming Team
• Roseanne English
• Paul McMurray, MDS
• Felicita David, MS
• Laura Hall, MPH

26
For more information

• Gabriel Rainisch GRainisch_at_cdc.gov
• Colleen Martin CMartin5_at_cdc.gov_at_cdc.gov
• BioSense Website http//www.cdc.gov/biosense
• Technical Help Desk 1-800-532-9929
• Secure Data Network question
• BioSense technical support
• BioSense E-mail Address BioSenseHelp_at_cdc.gov
• General questions
• Additional user requests
• Problem reporting
• Suggestion for enhancements
• All feedback!