Syndromic Surveillance in Montreal An Overview of Practice

Syndromic Surveillance in Montreal: An Overview of Practice and Research David Buckeridge, MD Ph. D Epidemiology and Biostatistics, Mc. Gill University Surveillance Team, Montreal Public Health QPHI Surveillance Meeting KFL&A Public Health, Kingston, ON June 13 th, 2008

Syndromic Surveillance in Montreal (ou, Vigie Multirisque) Counts, Native coding 1. Identifying schemes, ISDS cases individual consensus syndromes Individual Event Definitions Event Detection Algorithm Telehealth 911 Calls Hospital Reportable Data Describing Population Routine Sa. TScan, Daily review of 2. Detecting population alerts for shared patterns 3. analysis results, Conveying information for action addresses not clear protocol Population Pattern Definitions Event Reports Intervention Guidelines Pattern Report Pattern Detection Algorithm Population Under Surveillance Intervention Decision Algorithm Knowledge Public Health Action

Vigie Multirisque: Data Sources v Emergency Departments w Currently: All 22 ED in Montreal via web form, total counts, no diagnosis or chief complaint w Future: Automated feeds under development, triage code and level, chief complaint, postal code v EMS Dispatch and Billing v Long-Term Care v Tele Health v Reportable Diseases

Vigie Multirisque: Dashboard

Vigie Multirisque: Analysis

Vigie Multirisque: Analysis

Vigie Multirisque: Descriptive

Surveillance Research

Syndromic Surveillance Research 1. Identifying individual cases 2. Detecting population patterns Individual Event Definitions Event Detection Algorithm Subsets of admin data for ILI surveillance Data Describing Population 3. Conveying information for action Population Pattern Definitions Event Reports Intervention Guidelines Pattern Report Pattern Detection Algorithm Population Under Surveillance Intervention Decision Algorithm Knowledge Public Health Action

Looking for the Leading ILI Indicator in Billing Data

Syndromic Surveillance Research Accuracy of ICD codes 1. Identifyingand 2. Detecting population syndromes in individual cases patterns ambulatory practice Individual Event Definitions Event Detection Algorithm Subsets of admin data for ILI surveillance Data Describing Population 3. Conveying information for action Population Pattern Definitions Event Reports Intervention Guidelines Pattern Report Pattern Detection Algorithm Population Under Surveillance Intervention Decision Algorithm Knowledge Public Health Action

Syndromic Surveillance Research 1. Selecting the best algorithm Accuracy of ICD 2. codes 1. Identifyingand 2. Detecting population syndromes in individual cases patterns 3. ambulatory practice Individual Event Definitions Event Detection Algorithm Subsets of admin data for ILI surveillance Data Describing Population 3. Conveying information for action Population Pattern Definitions Event Reports Intervention Guidelines Pattern Report Pattern Detection Algorithm Population Under Surveillance Intervention Decision Algorithm Knowledge Public Health Action

Building the Knowledge-Base for Algorithm Selection 1. Model the aberrancy detection process 3. Use machine learning to identify and model the determinants of detection 2. Evaluate modeled algorithms using high throughput software

Syndromic Surveillance Research 1. Selecting the best algorithm Accuracy of ICD 2. Looking for codes 1. Identifyingand 2. Detecting population connected cases syndromes in individual cases patterns 3. ambulatory practice Individual Event Definitions Event Detection Algorithm Subsets of admin data for ILI surveillance Data Describing Population 3. Conveying information for action Population Pattern Definitions Event Reports Intervention Guidelines Pattern Report Pattern Detection Algorithm Population Under Surveillance Intervention Decision Algorithm Knowledge Public Health Action

System Architecture Current Case Management System Web Client Firefox, Explorer DCIMI Client Web-based Cartography Software Statistical Analysis Server Mapping and Web Server Python, R -Server, Sa. TScan Apache + PHP, Map. Server + Map. Script Oracle Forms DCIMI Database Oracle Spatial Database Post. Gre. SQL / Post. GIS DB

Organizing Data by Person, Place and Time Spatial Database Post. Gre. SQL / Post. GIS DB Episode Contact Onset Date Disease Type … Person MADO Name Birthdate … Situation Role (Home, Work, School, …) Active Date … Place Address X, Y Place Type (Residence, Workplace) …

Address Validation and Correction in a Public Health System

Dracones – Query Form Person Time Place

Dracones – Sa. TScan Results

Syndromic Surveillance Research 1. Selecting the best algorithm Accuracy of ICD 2. Looking for codes 1. Identifyingand 2. Detecting population connected cases syndromes in individual cases patterns 3. Spatial TB clusters ambulatory practice Individual Event Definitions Event Detection Algorithm Subsets of admin data for ILI surveillance Data Describing Population 3. Optimal Conveying information decision making for action after an alarm Population Pattern Definitions Event Reports Intervention Guidelines Pattern Report Pattern Detection Algorithm Population Under Surveillance Intervention Decision Algorithm Knowledge Public Health Action

Using Surveillance Information to Manage Outbreaks Effectively v Much research on the statistical accuracy of aberrancy detection algorithms v Little attention to what happens next w Some attempts to describe response protocols (e. g. , flow chart, wait a day) w No quantitative modeling of response v Rational response is important w Small window to obtain benefit w Surveillance information uncertain

The Traditional Surveillance Alert Response Model Environmental Data Knowledge Intervention Alert Wait Yes Detection Method No Review Records No Investigate Yes No Yes Confirm No Alert No Outbreak No Intervention

Identifying an Optimal Policy v The goal is to identify a policy, or a mapping from a belief state (probability distribution over states) to actions v The belief state, provides the same information as maintaining the complete history v Value iteration is used to solve POMDP

Applying a POMDP to Surveillance S - True outbreak state {No Outbreak, D 1, …. } O - Output from detection algorithm {0, 1} A - Possible public health actions T(s, a, s’) - Impact of actions given the state R(s, a) - Costs of actions and outbreak states (Izadi M & Buckeridge DL, 2007) Action Do nothing Review records Investigate cases Declare outbreak Transition

POMDP Policy Dominates Ad Hoc Policy

Syndromic Surveillance Research 1. Selecting the best algorithm Accuracy of ICD 2. Looking for codes 1. Identifyingand 2. Detecting population connected cases syndromes in individual cases patterns 3. Spatial TB clusters ambulatory practice Individual Event Definitions Event Detection Algorithm Subsets of admin data for ILI surveillance Data Describing Population 3. Optimal Conveying information decision making for action after an alarm Population Pattern Definitions Event Reports Intervention Guidelines Pattern Report Pattern Detection Algorithm Population Under Surveillance Intervention Decision Evaluating Syndromic. Algorithm Surveillance in Public Health Practice: Detecting Knowledge Waterborne Outbreaks Public Health Action

Automated and ‘Traditional’ Surveillance for Waterborne Outbreaks Syndromic Surveillance S Infectious (Asympto matic) O O Latent Infected O Infectious (Symptom atic) Historical Telehealth and ED Data Telehealth S Analysis by Public Health S ED R R Outpatient Stool Test R R R Analysis by Public Health S, R Outbreak Detection Historical Case Reports Dispersion Exposure Disease Health Care Utilization Reportable Disease Surveillance

Modeling Dispersion of Microorganisms Dispersion

Modeling Infection: Mobility

Mobility-Weighted Infection Probability by Home Address

Modeling Disease, Visits, Testing, Reporting to Public Health

Evaluating the Effect of Surveillance Enhancements

For more information…