Title: Environmental Measurements for Complex System Models of Infection Transmission
1Environmental Measurements for Complex System
Models of Infection Transmission
- James S. Koopman MD MPH
- University of Michigan
- Dept. of Epidemiology
2Talk Outline
- Integrating Microbial Risk Assessment and
Transmission System Analysis to analyze
Environmental Transmission Systems - A path to productive ETS science
- ETS models (simple less simple) to help analyze
non-pharmacologic influenza interventions - Cumulative Dose-Response models
3Traditional Microbial Risk Assessment Models
4Traditional Infection Transmission System Models
- Diverse model types
- Deterministic compartmental
- Stochastic compartmental
- Individual event history
- Fixed network
- Dynamic network
Susceptible
Diverse natural histories of infection, encounter
processes, network linkage processes,
transmission probabilities by type of contact,
etc.
Infectious
Immune
5Traditional Infection Transmission System Models
Left out of traditional Infection Transmission
System Models
Diverse natural histories of infection, encounter
processes, network linkage processes,
transmission probabilities by type of contact,
etc.
Dead Pathogen in Environment
Susceptible
Microbial fate transport models plus
dose-response models
Infectious
Live Pathogen in Environment
Immune
6Environmental Transmission System (ETS) Models
Dead Pathogen in Environment
Susceptible
Infectious
Live Pathogen in Environment
Immune
7Why Model Pathogen Passage Through the
Environment?
- Many interventions act on environment we need
to now when, where, how to use interventions
cost-effectively - Much ignorance about modes of transmission
- Environmental measurement of pathogens should
provide powerful data for analyzing infection
transmission systems
8Are Simpler Models Better?
- Yes if inferences from them are robust to
realistic relaxation of assumptions - Simple models make unrealistic assumptions
- Inference robustness must be assessed by
comparing to more realistic model - Yes if their parameters are more estimable
- No if they ignore available data theory
9Why Environmental Transmission Models are Better
- Place of pathogen identification can be more
informative than contact history - Contact is hard define measure so contact
parameters are rarely separately estimable - Transmission systems are spatial place of
pathogen identification is more tightly related
to transmission than histories of places visited - Environmental pathogen data statistically
identifies more key parameters of ETS models than
history data does for contact models
10Path to Building ETS Science that Serves Public
Health
- Focus inference robustness assessment on key
public health decisions - Effects of interventions to block routes of
transmission in specific places (respiratory
hygiene, air or surface decontamination, air flow
change, masks, changing human flow, hand washing,
gloves, many more) - Cost-benefit decisions
- Models must encompass intervention effects
11Path to Building ETS Science that Serves Public
Health
- Focus on key public health decisions
- Build series of nested (dockable) ETS models with
increasing realism - Deterministic compartmental
- Stochastic compartmental
- Individual based
- Dynamic network
- Increasing spatial and behavioral detail
12Path to Building ETS Science that Serves Public
Health
- Focus on key public health decisions
- Build model series with increasing realism
- Find simplest model forms from which robust
inferences can be made specify most influential
parameters in these - Find problematic simplifying assumptions by
comparing models that relax assumptions - Assess sensitivity of outcomes of PH significance
to parameter values that may be extrinsically or
intrinsically estimated
13Path to Building ETS Science that Serves Public
Health
- Focus on key public health decisions
- Build model series with increasing realism
- Find simplest models for valid inferences
- Specify extrinsically estimable parameters
- Link up with a group of scientists enthusiastic
about conducting studies to estimate those
parameters in a Center like CAMRA
14Path to Building ETS Science that Serves Public
Health
- Focus on key public health decisions
- Build model series with increasing realism
- Find simplest models for valid inferences
- Specify extrinsically estimable parameters
- Find data that provides statistical
identifiability for parameters needing intrinsic
estimation - Test parameter identifiability with IBM data
- Find data that varies most with parameters having
large effects on public health decisions
15Path to Building ETS Science that Serves Public
Health
- Focus on key public health decisions
- Build model series with increasing realism
- Find simplest models for valid inferences
- Specify extrinsically estimable parameters
- Find data that identifies key parameters
- Develop new intrinsic parameter estimation
methods for key parameters - Least squares, MCMC, filtered likelihood
- Use simpler models for estimation test methods
on data from more detailed models
16Path to Building ETS Science that Serves Public
Health
- Focus on key public health decisions
- Build model series with increasing realism
- Find simplest models for valid inferences
- Specify extrinsically estimable parameters
- Find data that identifies key parameters
- Develop parameter estimation methods
- Use most detailed and realistic models available
to design studies - Maximize inference robustness for key PH decisions
17Path to Building ETS Science that Serves Public
Health
- Focus on key public health decisions
- Build model series with increasing realism
- Find simplest models for valid inferences
- Specify extrinsically estimable parameters
- Find data that identifies key parameters
- Develop parameter estimation methods
- Design studies for robust PH decisions
18Environmental Transmission System (ETS) Models
Dead Pathogen in Environment
Susceptible
Infectious
Live Pathogen in Environment
Immune
19Simple ETS Model
- pickup rate
- p linear dose response
- g recovery rate
- deposit rate
- m pathogen death rate
S
ED
I
EA
R
20Epidemics when pathogen die off is negligible
Higher Pickup Rate
Lower Pickup Rate
Fraction Infected
EA
I
21Simple ETS Behavior
- Identical to classic Kermack-McKendrick SIR model
when pickup rate is high - Density dependent infection process when pathogen
die off rate is high - Number of transmissions from infected individual
is proportional to population size - Frequency dependent when pathogen die off rate is
low - Number of transmissions from infected individual
doesnt change with population size
22Airborne vs. Hand-Fomite Transmission of Influenza
- Both routes transmit infection experimentally
- Role of either in transmission system is not
established - Mucosal ID50 dose is 10 times alveoli ID50 dose
- Contamination and exposure rates poorly studied
for either route - CDC is supporting numerous studies of mask
wearing vs. alcohol hand wipe effects as well as
studies of droplet size production to help
pandemic influenza control decisions
23Dormitory Study at University of Michigan
- Dorms randomized to masks, masks plus hand wipes,
or control - Students swabbed when symptomatic
- Environmental samples collected in air exposure
only spots and hand-touched spots - Higher CFU of both Staph aureus epidermidis
from untouched surfaces - Staph epidermidis much more frequent than aureus
on touched surfaces but both are equal on
untouched
24Analyzing Trials Without ETS
- Significance testing has lower power
- Best multi-level model capturing individual and
herd immunity effects makes poor predictions of
effects even within the range of attack rates
observed in dormitories (Using data generated by
simulations) - No ability to generalize to other venues
25Analyzing Trials with ETS
- A couple years away from developing needed
methods - Should provide better power to detect differences
in effect - Should provide some robust inferences of effects
in venues with differing characteristics
26Environmental Transmission Model of Influenza by
Wien
- Analysis by Larry Wien presented to Institute of
Medicine (still unpublished) concluded only
airborne was important - Relative contributions of air vs. fomite not
affected by human movement patterns or cumulative
dose response curves in this model - Our analysis suggests these are important factors
that should be taken into account
27Movement Effects on Airborne vs. Hand-fomite
Transmission
- Theaters have little human movement, stores have
more - Fomite contamination disseminates slowly through
space, airborne contamination disseminates more
rapidly - Movement increases both fomite contamination
dissemination and number of fomites contacted
28Individual Based Model of Infection Transmission
- N agents roaming in a venue of size M (2D-grid)
- Each cell has a certain level of air and fomite
contamination. - Agents move to different cells with a rate m
- An agent is infected following a Beta-Poisson
dose response. - Dose dose from air dose from fomite
- Once infected agents shed contamination
(aerosolized and deposited to surfaces) - Contamination between cells is transported by air
diffusion of by agents acting as carriers - Contamination dies out
29Individual Based Model of Infection Transmission
1) There are 3 people, 2 susceptible and one
infected
2) The infected person sneezes. Droplets
contaminate the surface and aerosolized pathogens
start to spread.
3) The air contamination diffuses. At the same
time the fomite and air contamination die-out at
their respective rates.
4) Contamination reaches a susceptible person.
5) The contamination of the air dilutes in the
whole area. The infected people move. One
susceptible person becomes infected.
6) The contamination of the air dilutes as it
disseminates. The infected people move.
7) The remaining susceptible person moves to the
cell with fomite contamination. The contamination
of the air is fully spread throughout the venue.
8) The person is infected. The concentration of
fomite contamination in that persons cell is
much larger than air contamination.
30Transmission model
31Results the role of movement
- Interventions do not only depend on the disease
or the venue. - The functional characteristics (the kind of venue
and consequently the patterns of movement in it)
play a crucial role in the transmission.
For disease parameters modeling influenza
- Air intervention is only effective if m is lower
than 1/180 (less than one movement in 3h) - Fomite intervention more effective when movement
is faster than one every hour. - Movement helps determine which transmission route
dominates.
32Mode of Transmission Dose Timing
- Airborne long slow low dose timing
- Hand-Fomite intermittent high doses
- Linear dose-response model gives both equal risk
for equal total dose - Realistic innate or acquired immune element
replenishment dynamics in our dose-response model
favor hand-fomite transmission - Control inferences of Wien are not robust to
realistic relaxation of linear dose-response
assumptions
33Model Development Cumulative Dose Response
- Continuous time Markov chain model
- Dose has less probability of infection if the
time of inoculation is longer. - Need for time-dependent dose-response experiments.
System state variables and parameters P of
pathogens I of immune particles D total
dose T total inoculation time gp growth rate of
pathogens mp natural death rate of pathogens dp
deactivation rate of pathogens ai Arrival rate
of immune particles mp natural death rate of
immune particles dp deactivation rate of immune
particles
34Model Development Cumulative Dose Response
Fast immune replenishment
Slow immune replenishment
- Fast immune replenishment (c0.1)
- Shorter dosing regimes shifts Dose-response
function to left (increased infectivity) - Slow immune replenishment (c0.001)
- Dose-response function independent of dosing time
periods
Blue to red transition represents longer/low
concentration dosing periods
35Model Development Cumulative Dose Response
- Dynamic response from dosing (pathogen growth in
time) - 22 of the trials ended up in infection (blue
lines) - 78 of the trial ended up in not infection (red
lines)
Infection (Immune particles depleted)
Constant Inoculation Period
Transient Period
No Infection (Immune particles grow back to
Equilibrium)
36Dose-Response Data
- Our model generates steeper than observed rises
given a one time dose - Inbred mouse curves for Lassa fever Yersinia
pestis rise much faster than outbred mouse curves
and can be fit - An immune element recruitment parameter also
flattens the curve and permits the model to fit
data
37Summary 1
- Infection transmission system science will be
transformed by environmental data models. - With a big contribution from CAMRA we will learn
- what modes of transmission play different roles
in different populations - what interventions affecting modes of
transmission in different places will have big
population effects - how to study transmission in hospitals for
emerging infections like SARS so that we can
predict which population interventions will be
most effective
38Summary 2
- Focusing on PH policy inference robustness will
help our science serve Public Health - Simple abstract and realistic ETS models should
work together to analyze data, project
intervention effects, assess inference
robustness - Cumulative dose-response models must be specified
for a science of ETS to advance
39Thanks