Transcript Computational Modeling of Pandemic Flu

Computational Modeling of Pandemic
Influenza Control Strategies
N.M. Ferguson, D.A. Cummings, C. Fraser, S. Cauchemez, S. Riley,
A. Meeyai, S. Iamsirithaworn, W.D. Wheaton, P.C. Cooley, D.S. Burke
Presented by Donald S. Burke, M.D.
Workshop on Pandemic Influenza Vaccines: Building a Platform
For Global Collaboration, Beijing, China, January 28-30, 2007
National Institutes of Health
National Institute of General Medical Sciences
Models of Infectious Disease Agent Studies
University
Of Washington
&
Los Alamos National Lab
University
of Pittsburgh
&
Imperial College
Virginia
Tech
University
Containment – what does it take
(in theory)?
• The spread of an infectious pathogen is characterised the basic
reproduction number, R0 – the average number of secondary cases
generated by a single case in an entirely susceptible population.
• Control policies optimally reduce transmission so that R0 <1 – since
at that level an epidemic cannot sustain itself.
• Hence control policies need to eliminate a fraction 1-1/ R0 of
transmission – i.e. 33% for R0 =1.5, 50% for R0 =2, 75% for R0 =4.
• This can be achieved by:
 Reducing contact (quarantine, increasing social distance).
 Reducing susceptibility (vaccination, antiviral prophylaxis).
Reducing infectiousness (antiviral treatment).
• Key issues are who is targeted, how much effort is needed, and
how fast do we need to act.
Large Scale Model: 85+ million individuals
Key Modeling Research Partner: Thailand
Dr. Kumnuan and colleagues of the Thai Bureau of
Epidemiology and the Field Epidemiology Training Program
Population Density
Social Contact Processes
Individuals in households assigned to schools/workplaces with distance function based on data
household
workplace
elementary school
secondary school
workplace
SE Asia Movies
Applying SE Asian containment
policy to US?
1000000
• 1 billion course stockpile runs out on day 130
– overwhelmed by new introductions from
overseas.
• Other options, but comparably draconian.
100000
Daily cases
• Social + 5km radial prophylaxis plus 100%
school and 50% workplace reactive closure.
10000
1000
100
10
1
0
30
60
90
Day
120 150 180
Realistic Objectives for USA Pandemic Control
1. Diminish overall disease and death
2. Delay epidemic peak
( buy time)
3. Flatten epidemic peak
( limit surge burden on healthcare infrastructure )
#2
Pandemic outbreak:
No intervention
#3
Daily
Cases
Pandemic outbreak:
With intervention
#1
Days since First Case
Possible mitigation measures
Aim: minimize morbidity/mortality
until vaccine available, using:
1. Antivirals
2. Case isolation
3. Household quarantine.
4. ‘Social distancing’ – e.g. school
closure
5. Travel restrictions
6. Vaccines
Pre-vaccination: A poorly matched vaccine
• Assume availability of a low-efficacy pre-pandemic
vaccine, given as soon as pandemic recognized.
• Assume 30% reduction in susceptibility. If infected,
50% reduced chance of being a clinical ‘case’ and
30% additional reduction in infectiousness (matched
vaccine would be expected to be 70-90% protective
in healthy adults)
• A 10% stockpile of pre-pandemic vaccine would
reduce attack rates from 29% to 25% for next day
treatment + school closure policy [R0=2].
• A 20% stockpile of pre-pandemic vaccine would
reduce attack rates to 20%.
• Targeting children <16 is best way of reducing
transmission, given limited stocks of limited efficacy
vaccine. Targeting >60s has worst impact.
Impact of mass vaccination: a well
matched vaccine
• Examine scenarios of vaccine being available in US from day 30, 60 or 90 of the global
epidemic (1st US case on day 47, peak or epidemic on day 113).
• Doses for 1% of population are manufactured per day – v. optimistic.
• Assume vaccine takes 21 days to confer 70% reduction in susceptibility.
1.8
1%
13%
31%
16%
R0
1.5
Day 30, children first
Day 60, children first
Day 90, children first
Day 60, random
0.0%
0.4%
1.6%
0.6%
0%
0.0%
2%
0.0%
19%
4%
0% 10% 20% 30% 40%
Cumulative attack rate (%)
0.7%
0.1%
0.0%
1.0%
2.0%
Peak daily attack rate (%)
 Strain-specific vaccination has v. limited impact on first wave of pandemic
unless available within 2 months. Vaccinating children first gives best impact.
Conclusions
Computational modelling and simulation can be a
useful tool to evaluate complex policy issues, such
as timing and impact of vaccines
Initial modelling results suggest that well matched
vaccines must be available within two months to have
an appreciable effect on the pandemic course
Key sensitivities
Many unknowns (transmissibility, natural history of infection,
severity of disease, compliance with treatment/controls…) !
We will need to rapidly collect data and refine model
projections of spread and effect of controls during the
first few weeks of a pandemic
END
Nature 437: 209-214 (8 September 2005)
Strategies for containing an emerging influenza
pandemic in Southeast Asia
Neil M. Ferguson, Derek A.T. Cummings, Simon
Cauchemez, Christophe Fraser, Steven Riley,
Aronrag Meeyai, Sopon Iamsirithaworn
and Donald S. Burke
Synthesis of an artificial population and using it to evaluate
intervention strategies: “epidemiology in silicon”
Age Distribution of Thai population
Distribution of Household Sizes
Distribution of School Sizes
Probability of traveling over a certain distance to work
Assumptions about influenza
biology and treatment
16
8
6
4
2
.
5
4.5
4
3.5
3
2.5
2
1.5
1
0.5
0
2
5000
Days
4000
1.5
3000
1
2000
0.5
1000
0
0
0
2
4
6
8
Days post infection
10
Viral load (data)
• …but also can be used in uninfected
people to reduce chance of being
infected (=prophylaxis).
10
0
• Antiviral drugs reduce symptoms and
infectiousness…
Frequency
• People most infectious very soon after
symptoms.
12
.
• Short incubation period – 1-2 days.
14
Infectiousness
• New analysis of best available data on
pandemic and inter-pandemic flu.
Mitigating the effects of a pandemic on the USA
Can we contain a pandemic?
• Contain = eliminate virus before spread is extensive.
• So long as stockpile is 3m courses or larger, - can contain R0~1.8.
• Need to detect outbreak at <50 cases, react to new cases in 2 days.
• Needs work to make feasible – but far more effective than anything else
USA model data: demography
Proportion
0.1
model
data
0.08
0.06
0.04
0.02
0
0
10
20
30
40
50
60
70
80
Age category
0.4
Proportion
• Model uses Landscan population
density data, and matches US
census data on age & household
structure.
model
data
0.3
0.2
0.1
0
1
2
3
4
5
Household size
6
7
8
USA model: School/workplace data
Using a data on all schools in US, plus data on workplace size/location.
Frequency .
1000
Data
Model
100
10
1
10
100
1000
10000
Fit of Zipf model
15000000
10000000
5000000
.
20000000
1
Number of workers
0.1
Probability
25000000
0.01
data
model
0.001
0.0001
0.00001
0
15
610
11
-2
0
21
-5
51 0
-1
10 00
12
25 5 0
1
50 -50
1- 0
10
0
10 0
00
+
Number of workers
Size
Workplace size category
0.000001
1
10
100
1000
Distance (km)
10000
Model validation
• Validation difficult – no
pandemic for 40 years.
• We do try and match past
pandemics’ rate of spread,
proportion affected etc.
• But data limited.
• Key result: R0 for pandemic
flu in 1.5-2.0 range.
• Try to adjust for changes in
populations, travel etc.
R0 = Basic reproduction number
= number of secondary cases per case at start of epidemic
Model of a USA pandemic
• Large urban centres affected first, followed by spread to less densely
populated areas.
R0 = 1.8 / 1.5
Daily cases
5,000,000
4,000,000
3,000,000
First US
case
2,000,000
1,000,000
0
0
30 60 90 120 150 180
Day of global outbreak
Up to 12% absenteeism at peak
Baseline epidemic
• ‘Realistic’ seeding using expected number of imported infections estimated from
simple global model and travel data.
• Peak ~65 days after first case for R0=1.8.
• Epidemic growth rate matches peak 1918 growth rate for R0=1.8.
• Timing may be pessimistic (no account of seasonal variation in transmission).
Daily cases
5,000,000
4,000,000
3,000,000
First US
case
2,000,000
1,000,000
R0=1.8 / 1.5
0
0
30 60 90 120 150 180
Day of global outbreak
• Indirect effect relies on treatment in
<24h since infectiousness peaks soon
after symptoms start.
• With a 24h delay, treatment of 90% of
cases reduces attack rate from 34% to
29% for HT scenario.
• 48h delay gives almost no reduction in
transmission and poorer clinical benefit.
.
2
5000
1.5
4000
3000
1
2000
0.5
1000
0
0
0
2
4
6
8
Days post infection
10
Viral load (data)
• 25% stockpile is just enough
assuming 50% of those infected seek
treatment – but may be higher.
Infectiousness
• Reduce severity of cases, but can
also reduce transmission (reducing
attack rates from 34% to 28%.
.
Mitigation: case treatment
School closure
• Reactive policy: After the first case in a
school, it is closed the next day for certain
period.
• After reopening, school closes again after
further cases.
• Children out of school have 50% increase
in household contacts, 25% increase in
community contacts.
• Main effect is to reduce peak height (by ~
40%).
• Cumulative attack rates reduce from 32%
to 29% when school closure added to nextday treatment policy and R0=2 scenario.
Household prophylaxis/quarantine
• Household prophylaxis= treatment of
everyone in house of case, not just case
herself.
• Combined with school closure and next-day
treatment can reduce clinical attack rate to
20% – but needs antiviral stockpile of 50% of
population (for R0=2.0).
• When 20% pre-vaccination (of <16’s) is
added, attack rate drops to 14%
• For R0=1.7, same policy can reduce attack
rates to 7% (from 28% baseline),
• Voluntary household quarantine potentially
boosts effectiveness – but would need
prophylaxis to be ethical.
Impact of matched mass vaccination
• Imagine vaccine available in US from day 0,
30, 60 or 90 of the global epidemic, and assume
1% of population vaccinated per day.
 Vaccination has very limited impact unless
available within 2 months.
 So need to stockpile in advance, even if
efficacy is limited because vaccine not perfectly
matched.
33
36
38
30
32
34
25
30
30
22
28
29
36
37
38
34
35
36
32
31
31
30
29
29
36
37
38
34
35
36
30
30
31
28
5.8
6.4
6.7
5.8
6.4
6.8
5.8
6.1
6.5
6
6
6.5
6.1
6.3
6.4
6
6.2
6.3
5.9
6.2
6.5
5.8
6.2
6.4
6.1
6.2
6.4
6
6.2
6.3
5.7
6.2
6.5
5.7
19
17
16
19
17
16
15
14
14
16
14
14
13
13
13
13
13
13
14
14
14
14
14
14
13
13
13
13
13
13
15
14
14
15
8.7
8.1
7.7
8.7
8.1
7.7
8
7.9
7.7
8
7.8
7.7
7.3
7.2
7.1
7.3
7.2
7
7.7
7.8
7.6
7.7
7.8
7.6
7.3
7.2
7.1
7.3
7.2
7
7.8
7.8
7.7
7.9
55
55
56
55
55
56
58
59
58
58
59
59
60
59
59
61
60
59
60
59
58
60
59
59
60
59
59
61
60
59
59
59
58
59
12
13
14
12
13
14
13
13
13
13
13
13
13
14
15
13
14
15
13
13
13
13
13
13
13
14
15
13
14
15
12
13
13
12
% attack rate in 60-85 age band
21
20
20
22
21
20
17
19
17
15
19
17
22
20
19
21
20
19
21
19
17
21
19
17
22
20
19
21
20
19
21
19
17
20
% attack rate in 20-60 age band
16
14
13
16
14
13
2.6
0.9
0.3
2.2
1.3
0.3
0.4
0.3
0.2
0.4
0.3
0.2
1.2
0.6
0.4
1.3
0.7
0.4
0.4
0.3
0.3
0.5
0.3
0.3
2.5
0.8
0.4
2.9
% attack rate in 15-20 age band
30
30
30
33
33
32
45
50
52
47
51
54
41
43
43
44
45
45
45
49
51
47
51
53
41
43
43
44
45
45
45
50
52
47
% attack rate in 5-15 age band
0.1
0.1
0.1
0.1
0.1
0.1
9.4
0.7
0.2
13
1.3
0.2
0.2
0.1
0.1
0.2
0.1
0.1
0.7
0.3
0.1
1
0.4
0.2
0.2
0.1
0.1
0.2
0.1
0.1
2.2
0.4
0.2
3.5
% of infections in 20-60 age band
% of infections in 60-85 age
band
0
0
0
0
0
0
20.2
20.2
20.2
20.2
20.2
20.2
20.2
20.2
19.6
20.2
20.2
20.2
12.2
12.9
13.4
12.3
13.0
13.8
12.3
12.8
13.2
12.3
13.1
13.5
12.6
13.1
13.5
12.6
% attack rate in 0-5 age band
0.5
0.6
0.6
0.7
0.7
0.8
0.9
1.1
2.0
0.9
1.0
2.0
2.1
2.6
3.0
2.1
2.9
3.3
0.3
0.8
1.5
0.3
0.7
1.5
1.3
1.9
2.3
1.3
2.2
2.7
0.1
0.5
1.3
0.1
% of infections in 15-20 age band
87
74
70
92
75
71
61
122
128
50
122
130
115
80
70
126
93
70
89
98
84
88
104
99
116
82
67
130
95
75
98
138
104
91
% of infections in 5-15 age band
48
41
36
51
42
36
17
55
58
17
49
51
49
38
37
52
47
35
35
37
33
37
39
42
50
40
35
54
49
39
30
42
39
28
% of infections in 0-5 age band
9.0
9.0
9.0
9.0
9.0
9.0
9.0
9.0
9.0
9.0
9.0
9.0
9.0
9.0
9.0
9.0
9.0
9.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
% of infections in community
0
0
0
0
0
0
17
24
46
17
21
46
46
59
66
47
65
74
8
20
36
6
18
37
31
47
55
32
53
64
2
13
33
2
% of infections in workplaces
1.5
2.4
3.4
2.0
3.3
4.6
0.01
0.06
0.48
0.01
0.04
0.45
0.5
1.4
2.1
0.5
1.5
2.5
0.16
0.38
0.92
0.14
0.36
0.92
0.5
1.2
2.1
0.5
1.4
2.5
0.03
0.15
0.62
0.03
% of infections in schools
Time to peak [NOT WELL
(days)
ESTIMATED]
32
39
44
42
52
59
0.3
4.3
18
0.3
2.9
19
17
27
34
19
32
41
4.2
11
21
3.7
10
23
17
27
34
19
32
41
1.3
7
19
1.1
% of infections in households
Cum. non-flu
ILI attack rate (%)
50
50
50
67
67
67
50
50
50
67
67
67
50
50
50
67
67
67
50
50
50
67
67
67
50
50
50
67
67
67
50
50
50
67
% of infections due to seeding
Antiviral
usage (%)
1.9
2.4
3
1.9
2.4
3
1.9
2.4
3
1.9
2.4
3
1.9
2.4
3
1.9
2.4
3
1.9
2.4
3
1.9
2.4
3
1.9
2.4
3
1.9
2.4
3
1.9
2.4
3
1.9
Average absenteeism due
to place closure (%)
Peak attack
rate (%)
N/A
N/A
N/A
N/A
N/A
N/A
Base
Base
Base
Base
Base
Base
Base
Base
Base
Base
Base
Base
Base
Base
Base
Base
Base
Base
Base
Base
Base
Base
Base
Base
0.01%
0.01%
0.01%
0.01%
Duration of
[NOT WELL
epidemic (days) ESTIMATED]
Average absenteeism due to
case withdrawal or quarantine
(%)
Cumul. attack rate by day 220 of
global epidemic (~day 180 of
US epidemic) (%)
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
None
Proportion of infections
becoming cases (%)
N/A
N/A
N/A
N/A
N/A
N/A
Area
Area
Area
Area
Area
Area
Area
Area
Area
Area
Area
Area
Area
Area
Area
Area
Area
Area
Area
Area
Area
Area
Area
Area
Area
Area
Area
Area
Other controls
School closure type
Treat all ILI as flu
N/A
N/A
N/A
N/A
N/A
N/A
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
R0
N/A
N/A
N/A
N/A
N/A
N/A
Yes
Yes
Yes
Yes
Yes
Yes
No
No
No
No
No
No
Yes
Yes
Yes
Yes
Yes
Yes
No
No
No
No
No
No
Yes
Yes
Yes
Yes
Threshold for closure
and social distancing
7
7
7
7
7
7
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
Generic/workplace social
distancing
Scenario
Combined strategies!
26
36
42
35
48
56
0.3
3.7
16
0.2
2.5
17
15
24
30
16
29
37
3.5
9.6
19
3
9.1
21
15
24
30
16
29
37
1.1
6.1
17
0.9
42
47
49
57
63
65
0.3
4.3
18
0.3
2.9
19
15
25
31
17
29
38
4.1
11
21
3.6
10
23
16
25
31
17
30
38
1.3
7
19
1.1
38
44
47
51
59
63
0.3
4.7
19
0.3
3.1
20
17
27
33
19
32
41
4.5
12
22
3.9
11
24
17
27
33
19
32
41
1.4
7.6
20
1.2
31
39
44
42
52
59
0.3
4.5
19
0.3
3.1
20
19
29
35
20
35
44
4.5
12
22
3.9
11
24
19
29
35
20
35
44
1.4
7.4
20
1.1
24
32
39
31
42
51
0.2
3.4
14
0.2
2.3
15
14
24
31
16
28
38
3.3
8.7
17
2.9
8.2
19
14
24
31
16
28
38
1
5.5
15
0.8
Working Conclusions
• Vaccine needed at start of pandemic to have significant impact.
• Treatment-only’ policy needs to be delivered very rapidly for optimal effect.
• School closure potentially effective at reducing peak of epidemic.
• Household prophylaxis can reduce attack rates by 1/3.
• Adding pre-pandemic vaccine with 30% efficacy to these policies substantially
increases impact – possible to reduce attack rate by 67-75%.
• Social distancing (namely reductions in non-household contact rates) can (of
course) be highly effective at controlling disease transmission, if high degree of
contact rate reduction assumed.
TALK OUTLINE
1.
2.
3.
4.
The “MIDAS” group
Prevention of Avian Influenza Emergence in SE Asia
Mitigation of a Pandemic Impact in the USA
Discussion
Modeling and Simulation to Guide Policy Decisions in the USA
“MIDAS”
Models of Infectious Disease Agents Studies
A collaborative network of scientists who
conduct research on the use of computational
and mathematical models to prepare the
United States and the World to respond to
outbreaks of infectious diseases.
The University of Pittsburgh
Models of Infectious Diseases Agents Study (MIDAS) Team
Pittsburgh
Don Burke Derek Cummings
Imperial
Neil Ferguson
Probability of eliminating an otherwise large epidemic ( using an
idealized policy of socially targeted antiviral prophylaxis ): Impact of
epidemic size when policy is implemented
1
Effect on epidemic timing
• Most policies do not slow epidemic substantially, but just reduce
magnitude – so need to be maintained throughout pandemic.
• Very intensive prophylaxis &/or social distancing measures can (in
theory) dramatically reduce attack rates – but feasibility a key issue.
• e.g. for R0=2.0 (in US):
None
Daily cases
5,000,000
Policy 4
4,000,000
Policy 13
3,000,000
Policy 19
Policy 21
2,000,000
Policy 27
1,000,000
0
0
30 60 90 120 150 180
Day of US outbreak
Simulated daily incidence (complete absence of controls) Blue line = mean;
gray zone = 95% envelope; colored lines = seven stochastic realizations