No Slide Title

Download Report

Transcript No Slide Title

Detection Systems: A tutorial
RODS: http://www.health.pitt.edu/rods
Main Point
• In addition to expertise about algorithms,
you need know-how about:
• How to put it all together
• Requirements
• Getting the data
• Legal
• Negotiations
• Technical
• Storing the data
• User interfaces
Step 1: Be clear (and explicit) about
what it is you are trying to detect!
• Which organism(s)
• What route(s) of transmission
• What size(s) outbreak
E.g., large
bioaerosol release of
anthrax outdoors
Example System: RODS
RODS Initial Goal-Detect Anthrax
Release
WHAT IF …
Worst case
• 100,000 people are exposed
• Onset of illnesses occurs over days 1-7
• The costs are
Cumulative Cost (Billions)
20G
16G
12G
• Treatment of sick
Best case
• Prophylaxis of4G healthy (exposed and unexposed)
• Future earnings
lost through deaths, valued at
0
1
2
3
4
5
6
approximately0 $790,000
per
Days After the Release
8G
• Mass treatment occurs on days 0, 1, 2, 3, 4,
5, 6 or 7… and it has 90% efficacy
Key
Day After Exposure on Which Response Began
0
5
1
6
2
7
3
never
4
Kaufmann, The economic impact of a bioterrorist attack: Are prevention and postattack
intervention programs justifiable EID 3(2):83-94, 1997.
Medical Computer
Scientist
(aka Medical
Informatician)
Medical
Informatics
System Being
Demo’d
Computer
Scientist
Real-time monitoring of
chief complaints
February 5, 2002
How RODS Works
Gender
Age
Home Zip
MSH|^~\&||xxx||RODS|200307181731||ADT^A04|2003071XXXX
XXXX|P|2.3<CR>
PID|||||||^05|M|||^^^^15301|||||<CR>
Date and time of registration
PV1||E|||||||||||||||||98765432||||||||||||||||||||||
||200307181731||<CR>
DG1||||CARBON MONOXIDE EXPOSURE<CR>
IN1||||||||||||||||||||||||||||||||||||||||||||^^^^<C
Chief complaint
R>
HL7 admission/discharge/transfer message
about a patient registration in an emergency
department
How RODS Works
Health Dept
Health Care System #1
Hosp 1
Hosp 2
Hosp 3
Hosp N
Detection
Detection
algorithms
algorithm
Message
Router*
Health Care System #2
Internet
HL7
Naive
Listener
Bayes
Alerts
DB
Health Care System #3
.
.
.
Health Care System #n
GIS
* In a more advanced deployment, the health department would have a health-system resident
computer between the message router and the Internet to do data linkage and ELR
Web
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
Key Points
• Central hub for all data
communication in hospital
• Real-time data communication
• Mission critical
• The vast majority of hospitals are
familiar with using it to transmit data
• Getting data from it means you
are on hospitals’ critical path
• You will thus get data in realtime reliably with minimal
downtime
RODS HL7 Listener
Key Points
Health Dept
Detection
Detection
algorithms
algorithm
Message
router
Internet*
•
Alerts
•
•
HL7
Listener
Naive
Bayes
DB
Web
•
•
GIS
•
Maintains connection to
hospitals
Receives HL7 messages
Free (typical market price
$50K-$160K)
Robust and reliable
(multiple installations
running since 1999)
Written in Java™ so it
will run on any platform
Extremely high capacity
(multi-threaded)
* Usually the connection uses a Virtual Private Network (VPN) running over the Internet. As a
result, it is free (there are no charges for leased lines).
Naive Bayes Text Classifier
Message
Health Dept
Detection
router
algorithm
HL7
Naive
Listener
Bayes
DB
Alerts
Web
GIS
P(Respiratory|carbon
monoxide exposure)= .9
P(Botulinic|CME)= .001
“carbon
monoxide
exposure”
(chief complaint
of patient)
P(Constitutional|CME)= .01
P(GI|CME) = .05
Bayes
Classifier
P(Hemorrhagic|CME)= .001
P(Neurologic|CME)= .001
P(Rash|CME)= .001
P(None|CME)= .036
Respiratory
(Syndromic
classification
of patient)
Database
Message
Key points
Health Dept
Detection
router
algorithm
HL7
Naive
Listener
Bayes
•
Alerts
•
DB
Web
GIS
Designed since 1999 to store & retrieve realtime data
Optimized for data retrieval:
• User interfaces
• Detection algorithms
• Multiple levels of drilldown (age, gender,
syndrome, spatial regions, etc, etc.)
119
107
8/23/2004 12:00 PM
8/23/2004 1:00 PM
Health System Resident Component (HSRC)
Hospital
Message
Router
Case detection
occurs here
HSRC
Internet
Health
Department
Outbreak detection
occurs here
Key Points
• Case detection – linking patient data at
hospital:
• Results of micro cultures and
serology tests
• Orders for tests
• Results of chest radiographs
• Patients that have two or more data
characteristics such as chief
complaint of fever and pneumonia on
chest x-ray
• Necessary if you want more data in real
time besides chief complaints
OTC Data
Key Points
• National Retail Data Monitor (NRDM) is
the source of OTC data
• Research has shown that several
categories are useful
Also,
Cough and cold
Antidiarrheal
Pediatric electrolytes
Pediatric fever …
More info:
Hogan et al.
See case studies at https://www.rods.pitt.edu/nrdm
RODS User Interfaces
Key points
Health Dept
Detection
algorithms
Message
router
HL7
Listener
Naive
Bayes
• Web-browser based, no
additional installation of
client software required
Main Screen
• Tested with different
Epiplot
browsers (IE, Mozilla)
• Secure connections, data is
Mapplot
encrypted
• Public-health user feedback
incorporated
Alerts
DB
GIS
Web-browser Interfaces (Main)
All visits
Respiratory
GI
Rash
Constitutional
Hemorrhagic
Neurological
Botulinic
Copyright University of Pittsburgh 2002
Houston
Epiplot
Pediatric
electrolytes
ED visits for GI
complaints
All ages
Pediatric
Mapplot
Drill down to investigate “red zone”
Algorithms
Bayesian
BARD
Health Dept
Message
router
HL7
Listener
Naive
Bayes
Suite
Key Points
•
Spatial WSARE •
Scan
•
DB
GIS
Web
•
Goal: early, sensitive detection
with low false alarm rates
New approaches &
improvements to existing
approaches
Different algorithms usually fill
different needs/ requirements
Technical reports and journal
articles available upon request
RODS Open Source Software Deployments
Jurisdiction
Hospitals (real
time/total)
Agreements
Connections
System
Setup
Platforms
Pennsylvania
50/53
RODS
RODS
RODS
Utah
27/27
RODS
RODS
Ohio
10/13
PHRT
RODS
3/3
PH
RODS
Solaris/
Oracle
(All these
jurisdictions are
using the
Pittsburgh server
facilities)
2500 visits/day
PH
PH
PHRA
Unix/Oracle
0/190+
PH
PH
PHRT
Unix/Oracle
1/1, 2 more
pending
Aug ‘04
PHRT
PHRT
PHRT
Windows/Oracle
6 pending Aug ‘04
PHRA
PHRA
PHRA
Windows/MSSQL
pending
PHRA
PHRA
PHRA
Windows/MSSQL
(1 from U Miss)
PH
n/a
PHRT
Linux/Oracle
Atlantic City, NJ
Michigan
Taiwan
Houston, TX
El Paso, TX
Los Angeles, CA
Mississippi (OTC)
RODS, RODS Laboratory personnel; PH, public health personnel; PHRT, public health personnel with
formal RODS training; PHRA, public health personnel with informal RODS assistance
More Data Means More Specific Case Detection
What we do now for
Anthrax
Chief
Chief
Chief
Chief
Complaint
Complaint
Complaint
Complaint
Respiratory
Respiratory
Respiratory
Respiratory
Syndrome
Syndrome
Syndrome
Syndrome
Spatial and temporal
analysis to detect
overdensity of cases in a
zip code or larger region
Future
Chief
Chief
Chief
Pneumonia
Chief
Complaint
Chief
Complaint
Chief
Complaint
Chief
Complaint
on X-ray Complaint
Chief
Complaint
Complaint
Complaint
Chief
Chief
Chief
Temperature
Chief
Complaint
Complaint
Complaint
Complaint
Respiratory
Respiratory
Respiratory
Respiratory
SARS
Syndrome
Syndrome
Syndrome
Syndrome
Spatial and temporal
analysis to detect small
number of cases in a
hospital or hotel
Beijing Fever Clinics
Fever Clinics
• 61 Clinics
• Record 31 variables per patient
• Report centrally once a day (by 10 am)
Fever Clinic Data
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
Time of visit (met receptionist)
Medical record number
Name
Gender
Age
Occupation
Work place
Contact telephone number
Foreigners, incluign Hong kong Taiwan
Outside of Beijing (other provinces)
First visit:
Repeat visit:
Onset date:
Temperature
Cough
Chest pain
Contact/exposure history
Travel history
Respiratory/observation case
This hospital
Transfer
This hospital
Transfer
Upper repirtoary
Acute tohsiliitic
Larynhgopharygitis/sore throat
Other diagnosis
Epi report YES
Epi report NO
Street township
Treating physciain
One Year of Daily Counts
Before the Technology…
• Getting data-provider buy in
• Present a clear, concise proposal
• Anticipate questions, issues
• Minimize the work they have to do
• Data-sharing agreements
•
•
•
•
•
What data and how transmitted
Allowable uses
How confidentiality is maintained
Who can terminate and why
Disclaimers/audits/compliance
Key Points
• We have done these things,
and trained and assisted our
public health partners to do
these things, with approx. 100
hospitals in over 5 states
• We have a highly polished
pitch for getting buy in
• We have a standard datasharing agreement reviewed
by dozens of lawyers in over
5 states
These non-technical tasks are critical to success!
Public Health Law – Wisconsin and
Utah
Utah
Wisconsin
Act 109 a pharmacist or pharmacy
shall report:
Act 26-23b-105. A pharmacist
shall report:
•
•
an unusual increase in the
number of prescriptions filled for
antimicrobials;
•
any prescription that treats a
disease that has bioterrorism
potential if that prescription is
unusual or in excess of the
expected frequency; and
•
an unusual increase in the
number of requests for
information about or sales of
over-the-counter
pharmaceuticals …
An unusual increase in the number of
prescriptions dispensed or
nonprescription drug products sold
for the treatment of medical
conditions specified by DHFS by rule.
•
An unusual increase in the number of
prescriptions dispensed that are
antibiotic drugs.
•
The dispensing of a prescription for
the treatment of a disease that is
relatively uncommon or may be
associated with bioterrorism
Michigan too!
Some Systems You May Have Heard
Of
•
•
•
•
•
Essence
Red bat
Biosense
NYC
…