Upload
others
View
6
Download
0
Embed Size (px)
Citation preview
Using Epidemiology for Data-Driven Decision-Making in Tuberculosis Programs February 24, 2016
How Can Mathematical Modeling Help Us Strategize for TB Prevention and Control
1
How can mathematical modeling help
us strategize for TB prevention and control?
Using Epidemiology for Data-Driven Decision-Making in TB Programs
RTMCC pre-meeting training, Denver, CO, February 2016
Emily Kendall, MD
Learning Objectives
• Apply a basic transmission model to a decision
about TB control strategy
• Interpret model results, with consideration of
assumptions and limitations
• Consider how to effectively engage models and
modelers in decision-making
A health department’s dilemma
• You oversee TB control for a local/regional health department.
• Your shrinking budget allows you only to treat diagnosed cases and screen their closest contacts.
• You receive a grant to identify and treat latent TB in your area.
Options include:
1. Expand contact investigation for local cases
2. Screen foreign-born individuals
How will you choose?
Using Epidemiology for Data-Driven Decision-Making in Tuberculosis Programs February 24, 2016
How Can Mathematical Modeling Help Us Strategize for TB Prevention and Control
2
Whom should we screen?
• Assume (for now) that the same resources can screen the same number in either population
120 Foreign born individuals 120 casual contacts of local cases
One simple answer:
Focus on the population with more latent TB
Foreign born population: 1/3 latently infected
Casual contacts of local cases: 1/6 latently infected
Susceptible
Latently infected
… but not all latent TB becomes active TB
from Esmail et al, Philosophical Transactions B 2014,
based on Ferebee Bibl Tuberc 1970
When most local contacts
would be screened and treated
When most foreign-born
individuals would be screened and treated
Using Epidemiology for Data-Driven Decision-Making in Tuberculosis Programs February 24, 2016
How Can Mathematical Modeling Help Us Strategize for TB Prevention and Control
3
Foreign born individuals: infection events are more remote
Casual contacts of local cases: mostly recent infections = greater reactivation risk
Susceptible
Latently infection (recently infected)
Latent infection (remotely infected)
Another simple answer: Find and treat the
individuals at highest risk for reactivation
Casual contacts of local cases: mostly recent infections = greater reactivation risk
8
Susceptible
Latently infection (recently infected)
Latent infection (remotely infected)
Foreign born individuals: infection events are more remote
Susceptible
Future active cases
Estimate
individual cumulative lifetime risk
Foreign born: 1/3 latently infection
5% lifetime reactivation risk remaining
= expect 2 individuals to develop active TB
Local contacts: 1/6 latently infected
10% lifetime reactivation risk remaining
= expect 2 individuals to develop active TB
Latently infected, never reactivate
Using Epidemiology for Data-Driven Decision-Making in Tuberculosis Programs February 24, 2016
How Can Mathematical Modeling Help Us Strategize for TB Prevention and Control
4
But that’s still not the whole story.
1. We may care more about shorter-term impact.
2. And, we still haven’t accounted for the people that
these incident cases may infect
… Or if/when secondarily-infected individuals may develop
active TB…
… Or whom else those secondary cases may infect…
A role for modeling
• Epidemics are complex
• In particular, characteristics of TB limit intuition:
– Long time scales
– Complex natural history
– Airborne transmission difficult to pinpoint
– Imperfect treatment efficacy
Susceptible
Early Latent
TB
Active TB
Recovered
Rapid
progression
Reactivation
Late Latent
TB
Treatment
Infection Preventive
therapy
A role for modeling
• Mathematical models help simplify complex systems into something we can explore.
• They translate epi data into a decision-making framework.
– “If your assumptions are correct, these are the implications.”
• They help conceptualize important questions.
– What are the key drivers of impact, cost, etc.?
– If we change our assumptions, how much will our outcomes change?
– What data do we most need to collect?
Inputs
• Data • Assumptions
Outputs
• Implications • Relationships
** Model’s can’t replace data or tell us that our assumptions
are correct
Using Epidemiology for Data-Driven Decision-Making in Tuberculosis Programs February 24, 2016
How Can Mathematical Modeling Help Us Strategize for TB Prevention and Control
5
A role for modeling
A dynamic transmission model can help us choose our target population here.
account for changes
over time
(e.g. short-term vs.
long-term reactivation
risk, plus changing
size of epidemic)
account for secondary
cases
(e.g. preventing one
reactivation may
prevent transmitted
cases too)
Local contacts: Incident active cases are more likely to occur sooner
Foreign born: Future reactivations are spread more evenly over remaining lifetimes
Year 1
First let’s add time (but consider only the latent infections in our screening population)
Susceptible
Latently infected
Active cases
Recovered
Year 3
First let’s add time (but consider only the latent infections in our screening population)
Local contacts: Incident active cases are more likely to occur sooner
Foreign born: Future reactivations are spread more evenly over remaining lifetimes
Susceptible
Latently infected
Active cases
Recovered
Using Epidemiology for Data-Driven Decision-Making in Tuberculosis Programs February 24, 2016
How Can Mathematical Modeling Help Us Strategize for TB Prevention and Control
6
Year
5
Local contacts: Incident active cases are more likely to occur sooner
Foreign born: Future reactivations are spread more evenly over remaining lifetimes
Susceptible
Latently infected
Active cases
Recovered
First let’s add time (but consider only the latent infections in our screening population)
Year
15
Individual risks are equivalent over lifetimes, but not
over a shorter time horizon.
And we still need to add transmission…
Local contacts: Incident active cases are more likely to occur sooner
Foreign born: Future reactivations are spread more evenly over remaining lifetimes
Susceptible
Latently infected
Active cases
Recovered
First let’s add time (but consider only the latent infections in our screening population)
And now also add transmission
• Additional input data/assumptions:
– # secondary infections per active TB case.
– Time-varying risk of progression after infection
• for the individuals we may screen,
• and also for those they may infect if not treated.
Using Epidemiology for Data-Driven Decision-Making in Tuberculosis Programs February 24, 2016
How Can Mathematical Modeling Help Us Strategize for TB Prevention and Control
7
Local contacts Foreign born
Year 1 – no intervention
Susceptible
Latently infected
Active cases
Recovered
And now also add transmission
Local contacts Foreign born
Year 2 – no intervention
Susceptible
Latently infected
Active cases
Recovered
And now also add transmission
Local contacts Foreign born
Year 3 – no intervention
Susceptible
Latently infected
Active cases
Recovered
`
And now also add transmission
Using Epidemiology for Data-Driven Decision-Making in Tuberculosis Programs February 24, 2016
How Can Mathematical Modeling Help Us Strategize for TB Prevention and Control
8
Local contacts Foreign born
Year 4 – no intervention
Susceptible
Latently infected
Active cases
Recovered
And now also add transmission
Local contacts Foreign born
Year 5 – no intervention
Susceptible
Latently infected
Active cases
Recovered
And now also add transmission
Interpretation of model projections
• Our no-transmission analysis underestimated TB
incidence (not a surprise)
• Timing also mattered in transmission dynamics
Cases in local contacts occur earlier
Transmission occurs sooner
Epidemic grows faster
Using Epidemiology for Data-Driven Decision-Making in Tuberculosis Programs February 24, 2016
How Can Mathematical Modeling Help Us Strategize for TB Prevention and Control
9
Now, simulate preventive interventions
• Underlying epidemiologic data and assumptions, again:
– LTBI prevalence
– Reactivation rates
– Transmission intensity
– [No] overlap of target populations
• Additional assumptions about the interventions:
– Our ability to reach and diagnose each target population
– Expected effectiveness of LTBI treatment
– Timing/duration of intervention – For now, assume one-time intervention at year 0
Say, 50% effectively
diagnosed + treated
Year 5 outcome:
Screening of foreign-born Total Year 5 prevalence falls from 4 to 3, And cumulative incidence from 11 to 10
Susceptible
Latently infected
Active cases
Recovered
Treated for LTBI
Year 5 outcome:
Expanded local contact tracing Total Year 5 prevalence falls from 4 to 2, And cumulative incidence from 11 to 4
Susceptible
Latently infected
Active cases
Recovered
Treated for LTBI
Using Epidemiology for Data-Driven Decision-Making in Tuberculosis Programs February 24, 2016
How Can Mathematical Modeling Help Us Strategize for TB Prevention and Control
10
Example results - projected impact
For this single run:
Uncertainty analysis:
• Parameter/data uncertainty
• Random effects
Sensitivity analysis:
• What estimates drive the uncertainty in our predictions?
– Small change in parameter x large change in outcome?
• Which assumptions most influence our outcomes?
– Different model structure vastly different result?
We may want to refine these inputs, e.g. through more data collection
Parameter uncertainty in this model:
• LTBI prevalences (in each group)
• Recentness of infection (in each group)
• Factors affecting reactivation (e.g. age, diabetes)
• # of secondary cases (disease burden, treatment delay, mixing)
• Effectiveness of preventive therapy
• Resource requirements
– For example, might we engage private providers in foreign-born
screening to reach more people with the same $$?
Using Epidemiology for Data-Driven Decision-Making in Tuberculosis Programs February 24, 2016
How Can Mathematical Modeling Help Us Strategize for TB Prevention and Control
11
A published example – Household contact tracing One-way sensitivity analysis: Compares influence of each model parameter on a
single outcome (here, TB incidence at 5 years) to identify critical data
Kasaie et al, AJRCCM 2014 189(7):845
Another published example -- Impact of a TB strain typing service via improved contact tracing and reduced
diagnostic delay
J Mears et al. Thorax doi:10.1136/thoraxjnl-2014-206480
Copyright © BMJ Publishing Group Ltd & British Thoracic Society. All rights reserved.
Structural uncertainty in this model
• Epidemiological assumptions – What if we included heterogeneity of …
• infection risk (age, HIV status) ?
• transmission risk (disease burden, time to treatment initiation)?
– What if we specified contact structure (households, social networks)?
• Specifics of intervention – Timing (duration or frequency) of intervention
– Different amounts or mixes of resource allocation
– Other types of interventions
33
Using Epidemiology for Data-Driven Decision-Making in Tuberculosis Programs February 24, 2016
How Can Mathematical Modeling Help Us Strategize for TB Prevention and Control
12
Aside about cost-effectiveness models
• Considers added benefit of an intervention per added cost, versus an alternative (often standard of care)
• Can stand alone…
• …or can be added to a transmission model
– Captures transmission-related cost savings
via changes in epidemic size
34
# people affected (e.g.
current incidence)
Costs with
intervention
e.g. the
intervention
itself
Costs without
intervention
e.g. treating
cases it might
prevent
Individual
well-being
with
intervention
Individual
well-being
without
intervention
Costs with
intervention
Costs without
intervention
Total well-being with Total well-being without
Using models (and modelers!)
in decision-making
• Define the question.
– Models must simplify. The best ones tailor assumptions to a specific situation.
– What factors will guide decisions? Short vs long-term impact, cost, etc.?
• Identify relevant data
– Including their degree of uncertainty.
• Engage modelers – early when possible.
– Model development process may identify knowledge gaps and clarify thinking.
Summary
• Models simplify complex systems
• They are useful for seeing relationships: If this, then that
• Outputs are only as accurate as the data and assumptions
that go in
• Using models can aid decision making
– Set priorities, clarify assumptions, identify data gaps