Norman Bailey

advertisement
Decision-making for prevention and
control under economic constraints
John Edmunds
London School of Hygiene & Tropical Medicine
john.edmunds@lshtm.ac.uk
Overview
• Economic decision making
– Example
– Wider practice
• Future research directions (personal and
partial view)
• Conclude
Concluding remarks
.....Practical decision-making will in general
require an integration of the epidemiologically
based modelling with some form of socioeconomic modelling. This means that the usual
type of interdisciplinary OR team must be well
acquainted with modern scenario analysis
applied to public health activities. But it must
be emphasised that such scenario analyses
must be closely geared to epidemiological
models that have been fitted to local data (i.e.
not merely using plausible parameter values)
as strongly urged in this review.
Norman Bailey
Concluding remarks
.....Practical decision-making will in general
require an integration of the epidemiologically
based modelling with some form of socioeconomic modelling. This means that the usual
type of interdisciplinary OR team must be well
acquainted with modern scenario analysis
applied to public health activities. But it must
be emphasised that such scenario analyses
must be closely geared to epidemiological
models that have been fitted to local data (i.e.
not merely using plausible parameter values)
as strongly urged in this review.
Norman Bailey
Epidemic model
integrated within
economic evaluation
Multidisciplinary
collaborations
Sensitivity analyses
Model fitting &
statistical inference
confidence in predictions
Assessing seasonal flu options:
Marc Baguelin et al.
• Seasonal flu vaccine
• Current strategy (since 2000) is to target high risk
individuals & everyone over the age of 65
• Prior to 2000, strategy was to target those who were
high risk only
• Question: should we extend to low risk groups?
–
–
–
–
–
–
–
<5 years
50-64 years
5-16 years
<5 & 50-64 years
<5 & 5-16 years
<5 & 5-16 & 50-64 years
<64 years
Increasing cost £14.2m
£92.7m
Elements
Epidemic
model
Burden
of
disease
Economic
analysis
Schematic of approach
Epidemic parameters
Data
• Reproduction number
• Incubation period
• Infectious period
• Susceptibility profile
• Mixing patterns
• ……
• RCGP
• Swabbing
• Serology
Epidemic projections
Vaccine
assumptions
• Coverage
• By age & risk
• By year & strain
• Efficacy
Outcomes
Costs
• Risk and age:
• CFR
• Hospit.
• QALY loss
•…
• Hospitalisation
• Vaccine
• Delivery
•…
Projections in relevant
units & CEA
Building a picture of what
would have happened if....
• Epidemiology of flu has been disturbed by vaccination
for many years
• Attempt to reconstruct epidemiology of flu
– Detailed understanding of what happened (how many
cases, including how many prevented by vaccination)
– Estimate what would have happened if we had followed
alternative policies
Building a picture of what
would have happened if....
Burden of disease
Need:
• Clinical cases by strain, age group & risk group
• GP consultations by strain, age and risk group
• Hospitalisaitons by strain, age and risk group
• Deaths by strain, age and risk group
BUT
• Data non-specific e.g. all-cause deaths
• Use regression approaches to estimate
– Regress weekly laboratory confirmed cases of infection (RSV, flu, etc) against
outcome of interest (e.g. hospitalisations for respiratory illness)
– Risk group specific data only routinely available through HES
Burden of disease:
Cromer et al. (submitted)
Average weekly laboratory
reports by pathogen
<5 years old
• Regress weekly lab
reports against GP
consultation,
hospitalisation & death
data
– Age & risk-group specific
• Estimate of cases of
outcome attributable to
influenza (& other
causes)
>65 years old
Estimating cost-effectiveness
of alternative policies
• Sample set of reconstructed epidemics
– Number of infections over time
– For each epidemic, alternative vaccination scenarios are generated
• Link infections to outcomes via risk ratios, unit costs, etc.
• Monte-Carlo simulations, sampling over distributions for:
– risk ratios
– economic parameters
– QALY parameters
• Calculate summary statistics, e.g. ICER or Net Benefit
• Provides assessment of counterfactual
– But fitted to observed data (2000/1 to 2008/9)
What might have happened
H3N2-LR
H3N2-HR
H1N1-LR
H3N2-LR
Each box=
1000 epidemics
H1N1-HR
B-LR
B-HR
Results:
extending vaccination
Net benefit of extending vaccination to
different low-risk groups by coverage
Cost-effectiveness
Sensitivity & scenario analyses performed on:
• Mortality rates
• Coverage in low and high risk groups
• Cost of vaccinating
• Time frame of analysis
• Discount rates
Validation of model against other serological data
Concluding remarks
.....Practical decision-making will in general
require an integration of the epidemiologically
based modelling with some form of socioeconomic modelling. This means that the usual
type of interdisciplinary OR team must be well
acquainted with modern scenario analysis
applied to public health activities. But it must
be emphasised that such scenario analyses
must be closely geared to epidemiological
models that have been fitted to local data (i.e.
not merely using plausible parameter values)
as strongly urged in this review.
Norman Bailey
Epidemic model
integrated within
economic evaluation
Multidisciplinary
collaborations
Sensitivity analyses
Model fitting &
statistical inference
confidence in predictions
Case study: flu
Types of models used in
economic analyses
Systematic reviews of economic analyses
Thiry et al.
2003
Welte et al.
2005
Low et al.
2007
Newell
et al. 2007
Vargas-Palacios
(submitted)
A few problems
• Queue fever
Simple prophylactic interventions with limitless queue capacity
(e.g., vaccination)
λi
δ*r*I
S(t)
I(t)
D(t)
(1-δ)*r*I
λq
Q(t)
R(t)
min(nμ,Q(t)μ)
• Non-linearity in costs
– Marginal costs of expansion
– Timing of costs
• Seasonality
Queue inserted diverts
susceptibles
Simple treatment interventions with limitless queue capacity
and isolation of infecteds (e.g., quarantine + treatment)
λi
δ*r*I
S(t)
I(t)
D(t)
λq
(1-δ)*r*I
Q(t)
R(t)
min(nμ,Q(t)μ)
• Behaviour change
– Declining incidence
– Balancing epidemiological &
economic impact
Queue inserted diverts infecteds
into treatment post-transmission
Concluding remarks
• Economic evaluations of ID control programmes would often fail
“Bailey’s tests”, particularly:
– Epidemiological model integrated within economic evaluation
– Model fitting
• Worryingly, the bigger the problem the weaker the methods
– Fit to what?
– Insufficient data to develop an epidemic model?
• There is clearly a need to engage more with economists and public
health officials
– Multidisciplinary collaboration test
Concluding concluding
remarks
“We need to think carefully about how to persuade them of the
value of our work and how to understand what it is that constrains
their world and decisions”
Brian Williams 2013
Acknowledgements
•
•
•
•
•
•
•
Marc Baguelin, PHE/LSHTM
Stefan Flasche, LSHTM
Anton Camacho, LSHTM
Deborah Cromer, UNSW
Mark Jit, PHE/LSHTM
Liz Miller, PHE
Jonathan Weiss, LSHTM
The end
Download