Implications of Real-Time Versus Batch Reporting for Surveillance

Transcription

1 Implications of Real-Time Versus Batch Reporting for Surveillance Joe Lombardo JHU/APL Center of Excellence In Public Health Informatics

2 Impressions from Mike Stoto Presentation 1) The statistical algorithms used to perform the evaluation indicate that syndromic surveillance can only be used to detect outbreaks that are large. Little mention of the data, how data is binned into syndrome groupings, or algorithms used in the evaluation? 2) Implies that syndromic surveillance is only an automation process relieving the health departments of their role in performing surveillance.

3 Impressions from Marci Layton Presentation NYC Health Department has been able to effectively used information technology to assist in monitoring the health of their population. Will still perform syndromic surveillance, but value has not been fully determined.

4 Syndromic Surveillance Definition? From Wikipedia: Syndromic Surveillance is the analysis of medical data to detect or anticipate disease outbreaks. Its potential for use again bioterrorism is being investigated. What is the purpose of an advanced disease surveillance system? (Pat Kelley)

5 Advanced Disease Surveillance 1. The purpose of leveraging information technology to support disease surveillance is to provide tools to public health so that they can better use their time and talents in monitoring the health of the populations they serve. 2. The surveillance monitor / epidemiologist is the most important component of an advanced disease surveillance system. They can not be replaced by statistics. 3. The most useful systems use not only pre diagnostic indicators, but also lab results.

6 Situational Awareness vs. Syndromic Surveillance Syndromic Surveillance: -detection of abnormal patterns in time or space Situational Awareness: -detection of abnormal patterns in time or space -classification -tracking -response -monitor outcome It is all about how the data is used that is acquired for surveillance!

7 Original question asked? Implications of Real-Time Versus Batch Reporting for Surveillance

8 Definition of Real-Time vs. Batched IEEE Computer Society, Technical Committee on Real-Time Systems Real-time systems are systems in which its temporal properties are essential for reliability and correctness; the example applications include embedded systems, control systems, monitoring systems, and multimedia systems. Encarta - process items as batch: to process or assemble items as a batch or in batches - computer programs processed together: a set of programs or jobs processed on a computer at one time

9 Definition of Batched Reporting Batched reporting public health surveillance? a) Batched collection of health indicator data b) Batched processing of indicator data c) Reporting to health surveillance monitors that one or more rules have been triggered at a periodic time interval d) Sending reports for reportable diseases in a group at some specific reporting interval e) Etc.

10 Electronic Health Monitoring Components Health Condition Indicator ED Chief Complaint EMS Runs School Health Claims Military PHIN MS Secure Comms Archiving Cleansing Archive Text Parsing Signal Processing Web Server User Interfaces Query Interface Raw Cluster Alert Flag Lists CAP DMIS Messages Real-time HL-7 PHIN MS XML, etc. Secure FTP VPN, etc.

11 Electronic Health Monitoring Components Health Condition Indicator ED Chief Complaint EMS Runs School Health Claims EMR PHIN MS Secure Comms Archiving Cleansing Archive Text Parsing Signal Processing Web Server User Interfaces Query Interface Raw Cluster Alert Flag Lists CAP DMIS Messages Real-time HL-7, XML, etc. Secure FTP VPN, etc.

12 Indicator Health Condition Indicator ED Chief Complaint EMS Runs School Health Claims Military PHIN MS Secure Comms Archiving Cleansing Archive Text Parsing Signal Processing Web Server User Interfaces Query Interface Raw Cluster Alert Flag Lists CAP DMIS Messages Real-time HL-7 PHIN MS XML, etc. Secure FTP VPN, etc.

13 Sources Real-time data capture: - Over the counter medication sales - HL7 hospital Electronic Medical Records - EMS automated entry Potential latency in data capture: - Absentee - Claims

14 Archiving and Preprocessing Health Condition Indicator ED Chief Complaint EMS Runs School Health Claims Military PHIN MS Secure Comms Archiving Cleansing Archive Text Parsing Signal Processing Web Server User Interfaces Query Interface Raw Cluster Alert Flag Lists CAP DMIS Messages Real-time HL-7 PHIN MS XML, etc. Secure FTP VPN, etc.

15 Communications Virtual Private Network (VPN) to continually monitor all messages on a network: - used on hospital networks for sending real-time HL-7 messages - cost and maintenance File Transfer Protocol (FTP): - used to send batched files over in a burst at intervals from every 5 seconds to once every 24 hours.

16 Archiving and Preprocessing Archiving and pre processing of data should be performed as the data is received. The process includes: - cleansing of the data to remove errors incurred while entering the data - identifying and correcting errors which occurred while transmitting the data files - converting text unstructured data into structured data * chief complaint * clinic notes * radiology reports * etc.

17 Signal Processing for Detection Health Condition Indicator ED Chief Complaint EMS Runs School Health Claims Military PHIN MS Secure Comms Archiving Cleansing Archive Text Parsing Signal Processing Web Server User Interfaces Query Interface Raw Cluster Alert Flag Lists CAP DMIS Messages Real-time HL-7 PHIN MS XML, etc. Secure FTP VPN, etc.

18 Batch vs Real-Time Processing for Automated Surveillance Batch and Real-time : overused terms from information technology referring to the rate of availability and processing of data. The desired rate of data processing should depend on best surveillance practices - Still to be determined - But constrained by available resources

19 / Signal Processing Batched Processing: Most surveillance systems accumulate data for a specific time period before processing. - processing can take several hours depending on the quantity of the data - spatial algorithms can take a long time to run Real-time Processing: There are two forms of real-time processing that are used to support surveillance systems. - processing data as it is received - operator requested processing to support review and analysis Some systems process data in both batched and real-time modes!

20 User Interaction with the System Health Condition Indicator ED Chief Complaint EMS Runs School Health Claims Military PHIN MS Secure Comms Archiving Cleansing Archive Text Parsing Signal Processing Web Server User Interfaces Query Interface Raw Cluster Alert Flag Lists CAP DMIS Messages Real-time HL-7 PHIN MS XML, etc. Secure FTP VPN, etc.

21 User Interfaces Most modern disease surveillance systems provide to outputs of their algorithms to their users as soon as their signal processing stage is complete. Interfaces include: - flags to indicate that a threshold has been exceeded or an anomalous condition has been found - presentation of data both temporally and spatially - listings of cases that have caused a flag Some systems permit the user to process data in real-time based on requests from the user. Reporting relies on the user logging into the system and viewing the alerts or data!

22 Automatic Notification of Alert Health Condition Indicator ED Chief Complaint EMS Runs School Health Claims Military PHIN MS Secure Comms Archiving Cleansing Archive Text Parsing Signal Processing Web Server User Interfaces Query Interface Raw Cluster Alert Flag Lists CAP DMIS Messages Real-time HL-7 PHIN MS XML, etc. Secure FTP VPN, etc.

23 Automated Alerting of Users When Flags Occur Users are automatically notified when a threshold is exceeded or a rule has been satisfied. - s to users PDA s or Blackberry s - paging - faxes - CAP messages Automatic alerting would occur when the signal processing stage is Complete. Automatic alerting typically contains many false positives depending on system complexity or rule design. This function could be considered reporting!

24 Daily Surveillance: Two Modes of Operation Mode 1: Routine monitoring for various PH problems Manageable alert rate is paramount Large, growing multiple testing problem: data sources x regions x syndromes x age groups Does PH infrastructure want real-time processing? Could daily resources handle it? Mode 2: Focused monitoring of a perceived PH threat Rarely needed, but when it is, timely information is essential Desired data rate should be driven by realistic, target response capability

25 Critical Time Issues BioSensors BioSurveillance Clinical Diagnosis Percent Casualties Avoided (Average) Attack 120% 100% 80% 60% 40% 20% 0% 100% 99% 71% 29% 12% 12% 7% D0 D2 D3 D4 D5 D6 D7 D9 D12 5% Attack: 100k People Anthrax LD50 33k Expected Fatalities in a Week $26.2B Billion Economic Losses (Including Lost Potential Earnings) Up to 481 People Die/Hour (196/hr Average Over the Week) Future Earnings Lost Through Deaths, $325,000 per Hour ($133k/hr Averaged Over Week) Lives Saved: 59% Response Day Initiated Delay No Lag One Day Lag Two Day Lag Model from Kaufman, et al, Emerging Infectious Diseases

26 Summary 1. Reduction in mortality / morbidity from a bioterrorist incident requires timely reporting and response by public health authorities. 2. The benefits of real-time data collection are only realized if the other components of a surveillance system are real-time as well. 3. A surveillance system may be intended to have multiple purposes requiring real-time data collection for future possible use in a crisis. - batched routine surveillance - real-time crisis / situational awareness 4. Requirements for surveillance to perform routine health monitoring versus reportable diseases may vary requiring different time urgencies.