The Present, Near Term, and Future of Realtime Public Health Surveillance

1
The Present, Near Term, and Future of Real-time
Public Health Surveillance
(Pitt)
(CMU)

• Michael Wagner, MD, PhD
• Director, RODS Laboratory
• University of Pittsburgh
• http//www.health.pitt.edu/rods

2
The Present (Using RODS as an example)
3
HL7 admission/discharge/transfer message about a
patient registration in an emergency department
Data ED and Acute Care ADT
Gender
Age
Home Zip
MSH\xxxRODS200307181731ADTA042003071X
XXXXXXXP2.3ltCRgt PID05M15301
ltCRgt PV1E98765432
200307181731ltCRgt DG1CARBON
MONOXIDE EXPOSUREltCRgt IN1
ltCRgt
Date and time of registration
Chief complaint
4
Data OTC Sales

• 7500 products (UPC codes) used for
  self-treatment of infectious diseases
• 18 analytic classes at present (categories)

Antifever Pediatric (274) Antifever Adult
(1340) Bronchial Remedies (43) Chest Rubs
(78) Diarrhea Remedies (165) Electrolytes
Pediatric (75) Hydrocortisones (185) Thermometer
Pediatric (125) Thermometer Adult (313)
Cold Relief Adult Liquid (709 products) Cold
Relief Adult Tablet (2467) Cold Relief Pediatric
Liquid (323) Cold Relief Pediatric Tablet
(74) Cough Syrup Adult Liquid (592) Cough Syrup
Adult Tablet (32) Cough Syrup Pediatric Liquid
(24) Nasal Product Internal (371) Throat Lozenges
(364)
Numbers in parenthesis are the number of UPC
codes in the category
5
Analysis What we do now to detect Influenza,
Crypto, Anthrax
Chief Complaint
Chief Complaint
Chief Complaint
Chief Complaint
Chief Complaint
Chief Complaint
Chief Complaint
Chief Complaint
Chief Complaint
Chief Complaint
Chief Complaint
Chief Complaint
Chief Complaint
UPC code
BayesClassifier
Categorymapping
RespiratorySyndrome
RespiratorySyndrome
RespiratorySyndrome
RespiratorySyndrome
RespiratorySyndrome
RespiratorySyndrome
RespiratorySyndrome
RespiratorySyndrome
RespiratorySyndrome
RespiratorySyndrome
RespiratorySyndrome
RespiratorySyndrome
RespiratorySyndrome
Pediatric Electrolyte
BARD
Univariate
Algorithms that perform spatial and temporal
analysis to detect overdensities of cases in a
zip codes or larger regions
WSARE
Spatial Scan
6
Spatial Scanning of Electrolyte Sales

• Ultra fast version typically turns multi-day
  analysis into less than an hour.
• Searches all rectangular regions.
• Role in surveillance find clusters of illness
  with shapes such as oriented rectangles.

More info http//www.autonlab.org/autonweb/showPr
oject/4/
7
BARD (Bayesian Aerosol Release Detector)
Approach

• BARD automates the analysis done by Messelson et
  al of the meteorological conditions in Sverdlovsk
  in the days prior to the outbreak.
• The method uses a computational inversion of the
  well-known Gaussian dispersion model, although
  any dispersion model can be used.

Role in surveillance detect patterns of disease
activity consistent with the weather and the
incubation time of Anthrax (or other organism
that can be released as an aerosol)
More info BARD Tech report
8
WSARE (Whats Strange About Recent Events)
Approach
REPRESENTATIVE SURVEILLANCE DATA

• Search over hundreds of thousands of
  subpopulations
• For each subpopulation, use as good of a model
  as can be created to predict expected counts
• Compute p value, taking into account multiple
  testing

WSARE Approach Monitor hundreds of thousands of
subpopulations Pay close attention to effect of
multiple testing
All Historical Data
Todays Environment
Standard Approach Select in advance which
subpopulations to monitor (e.g., each county,
zip) Do not pay close attention to effect of
multiple testing
Todays Cases
What should be happening today?
Whats strange about today, considering its
environment? And how significant is this?
Evaluation
Detailed comparison on 2,000 simulated scenarios
and Western PA ED Data
Role in Surveillance
Detect small clusters of illness in healthcare
workers, age groups, workplaces
More info http//www.autonlab.org/autonweb/showPr
oject/4/
9
Evaluations
Performance of Bayesian parser
Example of a Detectability Analysis
Public Water Drinking Advisory
Antidiarrheal Sales
Case Study of OTC Monitoring
10
1
Case studies
Sensitivity
0.1 Affected
Days into outbreak
10
RODS Deployments
11
The Present (summarized)

• Data
• More widely collected
• Real-time HL7 transmission of chief complaints
  from emergency departments
• OTC data with 12-24 hour delay
• Web based disease reporting
• Less widely collected
• Call center
• Electronic lab reporting
• School absenteeism
• Algorithms
• Univariate
• Multivariate
• Spatial scanning
• Aerosol detection
• Deployments (not just RODS)
• Many jurisdictions
• Spotty in terms of data coverage
• Evaluation
• Understanding of methodology is good

12
The Near Term (next two years)
13
More Data Means Enables More Specific Case
Detection
What we do now
Future
Chief Complaint
Chief Complaint
Chief Complaint
Chief Complaint
Chief Complaint
Chief Complaint
Chief Complaint
Chief Complaint
Chief
Temperature
Chief Complaint
Chief Complaint
Chief Complaint
Pneumonia on X-ray
Chief Complaint
Chief Complaint
Chief Complaint
Chief Complaint
Chief Complaint
Chief Complaint
Chief Complaint
Chief Complaint
Chief Complaint
Chief Complaint
Chief Complaint
Chief Complaint
Chief Complaint
RespiratorySyndrome
RespiratorySyndrome
RespiratorySyndrome
RespiratorySyndrome
RespiratorySyndrome
RespiratorySyndrome
RespiratorySyndrome
RespiratorySyndrome
RespiratorySyndrome
RespiratorySyndrome
RespiratorySyndrome
RespiratorySyndrome
RespiratorySyndrome
SARS Syndrome
Spatial and temporal analysis to detect
overdensity of cases in a zip code or larger
region
Spatial and temporal analysis to detect small
number of cases in a hospital or hotel
14
One Year of Daily Counts of Febrile Illness in
Beijing
15
The Hospital Message Router
Electronic Health Record
Lab
Radiology
Message Router
Scheduling/ Registration
Transcription
Billing
Pharmacy
Hospital IT Infrastructure
aka Interface Engine aka Integration Engine
16
Algorithms that Process Multiple Data Streams
Red Cough Sales Blue ED Respiratory Visits
Signal
This is an anomaly
One Sigma
Cough Sales
2 Sigma
ED Respiratory Visits
17
Near term summarized (next 2 years)

• Data
• Clinical Laboratory results (Micro), radiology,
  temperatures, outpatient data military
• Integration of data from Biowatch sensors and
  military healthcare
• Algorithms
• That process multiple data streams
• That ntegrate water supply routes into
  surveillance
• For building monitoring
• More complete deployments in major cities and
  across border
• Evaluations of detectability, studies of behavior
  of sick individuals

18
The Future
19
Approaching the theoretical limits of
detectability (first case on day of infection ?)
Detecting small outbreaks and detecting bigger
outbreaks earlier

• Through better surveillance data
• More an better biosensors
• Earlier detection of patients with fever,
  constitutional symptoms
• Passiveembedded chips, smart toilets,
• Activeself reporting
• Widespread use of microchip arrays for diagnosis
• Environmental and intelligence data to set priors
• Decision support at the point of care
• and algorithms that can extract all information
  from knowledge and data
• Social networks
• Food distribution information

20
PANDA 2
Approach
Aggregate Observations

• PANDA2 is designed to be able to fuse ALL data
  and knowledge to achieve the very earliest
  detection.
• The method involves using causal networks that
  models the relationships between causes of
  disease outbreaks and the health state of
  individuals in the population, as well as
  aggregate observations about the population
  (e.g., levels of OTC sales)
• These causal networks represent prior knowledge
  about disease presentation as well as knowledge
  about the spatio temporal spread of outbreaks

One Patient
A city-wide model
More info Uncertainty in AI paper