using&simulaon&to&evaluate& surveillance&and&control
TRANSCRIPT
Using simula+on to evaluate surveillance and control strategies for waterborne infec+ous disease outbreaks in Montreal
Workshop on Syndromic Surveillance of Health and Climate-‐Related Impacts, March 17, 2014
David Buckeridge, MD PhD FRCPC Canada Research Chair in Public Health Informa+cs
The Surveillance Lab, McGill Clinical and Health Informa+cs Department of Epidemiology and Biosta+s+cs, McGill University
Agence de la santé et des services sociaux de Montréal, Direc+on de santé publique
The Need to Evaluate Surveillance
• Current surveillance systems have limita+ons – Insensi+ve – Slow – Inflexible
• Many poten+al improvements – Automate laboratory-‐based surveillance – Point-‐of-‐care diagnos+cs – Surveillance of health-‐care u+liza+on
• Evidence is need to priori+ze investment
Problem Formula+on
• Detec+on is necessary to ini+ate interven+ons • Benefits of detec+on are realized through effects of interven+ons
– Reduced morbidity and mortality – Decreased costs of other types
• Interven+on strategies are outbreak-‐specific • Outcomes depend not only on +meliness of detec+on • Not a linear process as illustrated!
Detection Intervention Decision making
Outbreak
Mo+va+on for Simula+on
• Significant morbidity and cost associated with waterborne disease outbreaks
• Public health preven+on and control strategies must be evaluated
• Limited historical data – Provide few insights on rela+ve benefits of alterna+ve strategies
– No control over outbreak condi+ons (and o\en li]le knowledge)
Model Descrip+on
Exposure
Infection infected or not
Census
Population Model
Water Drinking
Enquête OD
Mobility
Disease Progression
onset & duration of symptoms, mortality
Clinical Care and Reporting
Health-Care Seeking Behavior
onset & duration of symptoms, mortality
Water Distribution
System Structure
Pathogen Dispersion
synthetic population
timing, amount,
treatment
agent location
over time
space-time distribution of pathogen
number of ingested
organisms
Exposure
Infection infected or not
Census
Population Model
Water Drinking
Enquête OD
Mobility
Disease Progression
onset & duration of symptoms, mortality
Clinical Care and Reporting
Health-Care Seeking Behavior
onset & duration of symptoms, mortality
Water Distribution
System Structure
Pathogen Dispersion
synthetic population
timing, amount,
treatment
agent location
over time
space-time distribution of pathogen
number of ingested
organisms
Exposure
Infection infected or not
Census
Population Model
Water Drinking
Enquête OD
Mobility
Disease Progression
onset & duration of symptoms, mortality
Clinical Care and Reporting
Health-Care Seeking Behavior
onset & duration of symptoms, mortality
Water Distribution
System Structure
Pathogen Dispersion
synthetic population
timing, amount,
treatment
agent location
over time
space-time distribution of pathogen
number of ingested
organisms
Okhmatovskaia A, Verma AD, Barbeau B, Carriere A, Pasquet R, Buckeridge DL (2010). A Simula+on Model of Waterborne Gastro-‐Intes+nal Disease Outbreaks: Descrip+on and Ini+al Evalua+on. AMIA Annual Symposium.
Model Descrip+on
Enquête OD
Mobility
timing, amount,
treatment
Water Drinking
agent location
over time
Census
Population Model
synthetic population
Water Distribution
System Structure
Pathogen Dispersion
space-time distribution of pathogen Water
Distribution System
Structure
Pathogen Dispersion
space-time distribution of pathogen Water
Distribution System
Structure
Pathogen Dispersion
space-time distribution of pathogen
Pathogen Dispersion
• Team at Ecole Polytechnique (Benoit Barbeau) uses EPANET hydraulic model – Describes 2,500 km of Montreal’s municipal water distribu+on system from Atwater plant
– Simulates dissemina+on of pathogen through water system
– Generates concentra+ons of C. parvum oocysts in tap water at different loca+ons every hour
Linking Water to Regions
Model Descrip+on
Exposure
Census
Population Model
Water Drinking
Enquête OD
Mobility
Water Distribution
System Structure
Pathogen Dispersion
synthetic population
timing, amount,
treatment
agent location
over time
space-time distribution of pathogen
Census
Population Model
Water Drinking
Enquête OD
Mobility
Water Distribution
System Structure
Pathogen Dispersion
synthetic population
timing, amount,
treatment
agent location
over time
space-time distribution of pathogen
Census
Population Model
Water Drinking
Enquête OD
Mobility
Water Distribution
System Structure
Pathogen Dispersion
synthetic population
timing, amount,
treatment
agent location
over time
space-time distribution of pathogen
Infection infected or not
number of ingested
organisms
Disease Progression
onset & duration of symptoms, mortality
Disease
Symptomatic Illness
Complications
Infected
¬ Infected
Start: Exposed to C.parvum in drinking water
Health Care Utilization
Dose-response relation: (Teunis et. al, 1999)
pinf = 1 – e-rD
D – number of oocysts r – infectivity of oocysts
Disease Parameters
Name Unit Value Distribu1on Cita1on
Symptoms, General Prob 0.6112 U(0.5, 0.7) DuPont 1995
Symptoms, Immune compromised
Prob 0.7897 U(0.71, 0.85) Pozio 1997, …
Infec+vity per oocyst 0.1 U(0.05, 0.15) EPA 2003
Incuba+on, Mean Days 7 U(5,9) MacKenzie 1994, …
Incuba+on, Variance Days 6 U(2,10) MacKenzie 1994, …
Dura+on, Mean Days 5 Beta(1,7,4,2) MacKenzie 1994, …
Dura+on, Variance Days 6 U(4,8) Assump+on
Dura+on, Scale Immune compromised
None 6.5 U(3,10) Calibrated
Mortality, Complicated Prob 0.44 U(0.2,0.68) Pozio 1997
Single Parameter Sensi+vity Analysis
492.7
451.9
308.0
568.8
563.6
657.7
Households Drinking Tap Water
Water Treatment Probability
Infectivity of Oocysts
308.0 513.8 657.7Number of People
(in thousands)
Number Infected
301.2
276.3
257.0
188.2
347.6
344.5
359.6
402.0
Households Drinking Tap Water
Water Treatment Probability
Symptoms Probability
Infectivity of Oocysts
188.2 314.0 402.0Number of People
(in thousands)
Number Symptomatic (Incident)
Model Descrip+on
Exposure
Census
Population Model
Water Drinking
Enquête OD
Mobility
Water Distribution
System Structure
Pathogen Dispersion
synthetic population
timing, amount,
treatment
agent location
over time
space-time distribution of pathogen
Census
Population Model
Water Drinking
Enquête OD
Mobility
Water Distribution
System Structure
Pathogen Dispersion
synthetic population
timing, amount,
treatment
agent location
over time
space-time distribution of pathogen
Census
Population Model
Water Drinking
Enquête OD
Mobility
Water Distribution
System Structure
Pathogen Dispersion
synthetic population
timing, amount,
treatment
agent location
over time
space-time distribution of pathogen
Infection infected or not
number of ingested
organisms
Disease Progression
onset & duration of symptoms, mortality
Clinical Care and Reporting
Health-Care Seeking Behavior
onset & duration of symptoms, mortality
Single Parameter Sensi+vity Analysis
22.2
22.2
21.5
21.3
19.6
18.2
13.3
12.1
16.0
10.2
17.5
22.8
23.5
23.0
24.7
24.4
25.5
28.5
32.3
42.8
41.8
58.0
Immunocompromised Population
Seek Care Probability Weekend Coefficient
Seek ED Probability (Weekend)
Households Drinking Tap Water
Water Treatment Probability
Symptoms Probability
Infectivity of Oocysts
Seek ED Probability (Weekday)
Seek Care Probability Slope
Seek Care Probability (Day 1)
Illness Duration Mean
10.2 22.3 58.0Number of People
(in thousands)
Number Visiting Emergency Department
Model Descrip+on
Exposure
Census
Population Model
Water Drinking
Enquête OD
Mobility
Water Distribution
System Structure
Pathogen Dispersion
synthetic population
timing, amount,
treatment
agent location
over time
space-time distribution of pathogen
Census
Population Model
Water Drinking
Enquête OD
Mobility
Water Distribution
System Structure
Pathogen Dispersion
synthetic population
timing, amount,
treatment
agent location
over time
space-time distribution of pathogen
Census
Population Model
Water Drinking
Enquête OD
Mobility
Water Distribution
System Structure
Pathogen Dispersion
synthetic population
timing, amount,
treatment
agent location
over time
space-time distribution of pathogen
Infection infected or not
number of ingested
organisms
Disease Progression
onset & duration of symptoms, mortality
Clinical Care and Reporting
Health-Care Seeking Behavior
onset & duration of symptoms, mortality
Boil Water Advisory Study
• Geographical seing: Island of Montreal • Specific pathogen: Cryptosporidium parvum • Scenario: Failure of Atwater treatment plan • Research ques+ons:
– Primary: How is the effec+veness of a boil-‐water advisory influenced by the dura+on of the contamina+on and the +ming of the advisory?
– Secondary: What surveillance methods are likely to be associated with effec+ve early detec+on?
Scenarios and Outcomes
• Atwater Treatment Plant – Main plant in Montreal – Capacity 680,000 m3 d-‐1
• Scenarios (165) – Failure (5): 3 to 20 days – Oocyst (3): 0.01 to 1 per L – BWA (11): 0 to 28 days
• Outcomes – Morbidity, Mortality – Healthcare U+liza+on – Cost
Simula+on, Uncertainty & Computa+on
• Simula+on model is rela+vely complex – Use many (~30) parameters
– Computa+onally expensive
• Many simula+on runs – Required to evaluate sensi+vity of findings – Cumbersome and +me consuming to perform
3 Experimental parameters
[165]
30 Uncertain parameters
[1000] 165,000 Runs
Applica+on of SnAP
• Model – Stochas+c – 2 million agents
• Colosse (U Laval) – 1K nodes, 8K cores – 24 TB memory
– 1 PB data storage
Randomness and Uncertainty
X ~ N(0, 1)
X ~ N(μ, σ)
μ ~ U(-‐1, 1)
σ ~ U(0, 2)
X
Frequency
-6 -4 -2 0 2 4 6
050
100
150
200
X
Frequency
-6 -4 -2 0 2 4 6
050
100
150
200
250
300
LHS Samples Efficiently
Hoare et al. Theore+cal Biology and Medical Modelling 2008 5:4
Random Full Factorial
La+n Hypercube • Define N-‐dimensional parameter space
• Par++on each parameter distribu+on into equal probability intervals
• Uniformly draw from each bin so that each bin is sampled exactly once
• Efficient, unbiased coverage of space
Mul+ple Parameter Sensi+vity Analysis
Effec+veness of Boil Water Advisory
Propor+on of Infec+ons
Averted
Ini+al Results and Next Steps
• So far – Poten+al benefit of surveillance increases as dura+on of exposure increases
– Recognizing space—+me ‘signature’ of plant failure may allow early detec+on
• Next Steps – Apply detec+on methods to simulated laboratory and healthcare u+liza+on data
– Analyze results in terms of morbidity and cost
surveillance.mcgill.ca