IMPACT-BAM
(Better Ageing Model)
Sara Ahmadi-Abhari
Martin Shipley
Eric Brunner
Maria Guzman-Castillo
Piotr Bandosz
Simon Capewell
Martin O’Flaherty
This Talk
• Why modelling
– Why modelling dementia and CVD?
• IMPACT BAM
– Exploring the dynamics of CVD and Cognitive
decline in the English population
– IMPACT BAM as a decision support tool
• What’s next?
Models are everywhere
What drives the trends?
Explaining the decline in CHD mortality using the IMPACT Model
48%
55%
Spain 1988-2005 (IMPACT)
Italy 1980-2000 (IMPACT)
50%
43%
37%
Slovakia 1993-2008 (IMPACT)
Czech 1985-2007 (IMPACT)
Poland 1991-2005 (IMPACT )
Sweden 1986-2002 (IMPACT)
Finland 1972-1992 (Vartiainen)
Finland 1982-1997 (IMPACT)
46%
40%
40%
43%
40%
43%
47%
43%
44%
52%
42%
46%
New Zealand 1982-1993 (IMPACT )
New Zealand 1974-1981 (Beaglehole)
Scotland 1975-94 (IMPACT)
Scotland 2000-2010 (IMPACT)
USA 1968-1976 (Goldman)
USA 1980-1990 (Hunink)
USA 1980-2000 (IMPACT)
Canada 1994-2005 (IMPACT)
Ireland 1985-2000 (IMPACT)
England 2000-2007 (IMPACT)
England & Wales 1981-2000 (IMPACT)
Holland 1978-1985 (Bots)
0%
10%
20%
30%
9%
5%
8%
5%
5%
41%
52%
54%
70%
72%
76%
72%
25%
23%
24%
23%
Iceland 1981-2006 (IMPACT )
1%
5%
51%
40%
5%
4%
50%
60%
51%
39%
54%
50%
50%
48%
48%
34%
52%
44%
40%
50%
60%
70%
9%
18%
7%
7%
3%
9%
8%
14%
6%
10%
80%
90%
100%
Treatments
Risk factors
Unknown
Why Simulation modelling in
Epidemiology?
•
A disease in the real world is complex
•
•
Primary evidence hard to obtain
•
•
CVD and dementia share determinants but differ in timescales
Evidence base incomplete for dementia epidemiology.
Decision are being made right now
•
NICE 2015
Interesting evolving paradigm:
Joint prevention of CVD and Dementia
Risk Factor
Evidence
Association with AD
Smoking
3 SRs, 8-19 cohorts
Increase risk
Alcohol intake
2 SRs, 15 -23 cohorts
Protective
Physical Activity
SR, 16 cohorts
Protective
Fish, Omega-3
SR, 8 cohorts
4 out of 8 studies protective
Blood pressure
SRs, 8-15 cohorts
Hypertension in midlife increase risk
No association/protective at old age
Obesity
SRs, 8-15 cohorts
Obesity in midlife increase risk
Total Cholesterol
SRs, 18 cohorts
High cholesterol in midlife increase risk
Statins
SRs, observational studies
Inconsistent effect, might be protective
Diabetes
SRs, 8-11 cohorts
Increases risk independently of
vascular disease
Qiu et al Journal of Alzheimer’s Disease 32 (2012) 721–731
Figure 1: IMPACT-BAM Markov model
IMPACT-BAM Calibration:
Comparing model estimates with observed estimates
from independent sources
Model forecasts:
from 2006 to 2011 for age-specific dementia prevalence vs. CFAS 2011
MEN
WOMEN
Validation
CVD deaths : observed vs predicted
Why simulation models?
Optimal Window
RCT unfeasible
Ongoing CVD trials window
FINGER
Effect of
Intervention
preDIVA
Dementia
Incidence
MAPT
Dementia
Incidence
after
intervention
A different approach:
integrate different sources of evidence to
support decision making on interventions jointly
preventing CVD and dementia
Mid-age
60
70
Old-age
Based on Richard J Neurol Sci 2012
Population level INTERVENTION SCENARIO
INTERVENTION
CHANGE IN RELEVANT
SET OF RISK FACTORS
CHANGE IN RELEVANT RISKS
POLICY SCENARIOS
EPIDEMIOLOGY ENGINE
Policy modelling platform
PRIMARY PREVENTION
SECONDARY/TERTIARY
PREVENTION
Dementia – Population over 65
Number of Cases in 2014 ~ 727,690
2040
Baseline
(Current Trends)
Scenario 1 :
0.1 mmHg reduction
Scenario 2:
1.3 mmHg reduction
Scenario 3 :
3 mmHg reduction in
midlife cohorts
Number of
Cases
1,746,399
Model Predicted Change
in Number of Cases
+1,018,714
Prevalence
Change
+ 4.1%
1,743,964
- 2,435
- 0.1 %
1,714,816
- 31,583
- 0.2 %
1,680,506
-65,893
- 0.5 %
* All estimates are accompanied by uncertainty analysis.
How can we exploit IMPACT BAM
• Use as a platform to engage with epidemiologist
– Finding gaps
– Increasing precision
• Use as a platform to engage decision makers
–
–
–
–
Forecast future morbidity and costs
Effectiveness
Cost-effectiveness
Equity?
• Better characterize uncertainty around key parameters
– Trends of INCIDENCE of cognitive decline at the population
level?
Closing Remarks
Models should start to be seen as an integral part of
epidemiology enquiry
– For discovery
– For evidence synthesis
– For decision support when evidence is not complete
(Brighton Declaration, 2014)
• IMPACT BAM is at the interface of epidemiology
and public health, facilitating a dialogue between
scientist and decision-makers.
Why modelling?
1. Explain (very distinct from predict)
2. Guide data collection
3. Illuminate core dynamics
4. Suggest dynamical analogies
5. Discover new questions
6. Promote a scientific habit of mind
7. Bound (bracket) outcomes to plausible ranges
8. Illuminate core uncertainties.
9. Offer crisis options in near-real time
10. Demonstrate tradeoffs / suggest efficiencies
11. Challenge the robustness of prevailing theory through perturbations
12. Expose prevailing wisdom as incompatible with available data
13. Train practitioners
14. Discipline the policy dialogue
15. Educate the general public
16. Reveal the apparently simple (complex) to be complex (simple)"
Epstein, 2008
Systolic Blood Pressure
Key Model Assumptions
• We base our definition of cognitive impairment and functional decline on
ELSA and Whitehall studies.
• Modelling unit: a birth cohort, assumption that within cohort individuals are
homogeneous (e.g. baseline risk)
• CVD incidence and mortality decline continue
• CI incidence assumed stable from 2014 estimate
• Early life influences before 35 years ignored.
• Demographic assumptions:
• Future population projections based on ONS projections (baseline fertility,
mortality and migration)
Modelling interventions on systolic BP
Scenarios of
SBP
reduction
Description
Baseline
No PH intervention
Scenario 1
Reduce SBP by 0.1 mmHg in whole population 2015-2040
Salt intake reduced by health promotion, food labelling
(Collins, Mason et al. 2014)
Scenario 2
Reduce SBP by 1.3 mmHg in whole population 2015-2040
Salt content reduced by mandatory reformulation of processed
food
(Collins, Mason et al. 2014)
Scenario 3
Reduce SBP by 3 mmHg in midlife cohorts (35-65 years) 20152040.
Cardiovascular Disease – Population over 65
Number of Cases in 2014 ~2,433,100
2040
Baseline
(Current Trends)
Scenario 1 :
0.1 mmHg reduction
Scenario 2:
1.3 mmHg reduction
Scenario 3 :
3 mmHg reduction
midlife cohorts
Number of
Cases
Model Predicted Change
in Number of Cases
Prevalence
Change
2,123,605
- 309,519
-10.5%
2,120,956
-2,640
- 0.1%
2,088,728
- 34,877
- 0.3%
2,036,203
-87,402
- 0.6 %
* All estimates are accompanied by uncertainty analysis.
CVD deaths – Population over 65
Number of Deaths in 2014 ~67,880
2040
Baseline
(Current Trends)
Scenario 1 :
0.1 mmHg reduction
Scenario 2:
1.3 mmHg reduction
Scenario 3 :
3 mmHg reduction in
midlife cohorts
Number of
Deaths
Model Predicted Change
in Number of Deaths
Mortality
Change
25,806
- 42,072
- 0.49%
25,730
- 76
- 0.01 %
24,810
- 996
-0.01 %
23,519
- 2,287
- 0.02 %
* All estimates are accompanied by uncertainty analysis.
Disability – Population over 65
Number of Cases in 2014 ~1,333,300
2040
Baseline
(Current Trends)
Scenario 1 :
0.1 mmHg reduction
Scenario 2:
1.3 mmHg reduction
Scenario 3 :
3 mmHg reduction in
midlife cohorts
Number of
Cases
Model Predicted Change
in Number of Cases
Prevalence
Change
2,256,411
+923,129
+ 1.4%
2,253,597
- 2,814
- 0.1%
2,221,419
- 34,992
- 0.3 %
2,179,170
- 77,241
-0.6 %
* All estimates are accompanied by uncertainty analysis.
Reserves
Cardiovascular Disease – Population over 65
Number of Cases in 2014 ~ 2,477,621
2040
Number of
Cases
Model Predicted
Change in Number
of Cases
Prevalence
Change
Baseline
(Current Trends)
2,186,286
-291,335
- 3.6 %
Scenario 1:
1-unit reduction in BMI
2,131,759
-54,527
- 0.2 %
Scenario 2:
2 units reduction in BMI
2,079,071
-107,215
- 0.4 %
* All estimates are accompanied by uncertainty analysis.
Alzheimer Disease prevalence trend
forecasts
Increase in Dementia Burden
16
• Global numbers of AD doubling every 20
6
4
2
Hebert et al Neurology 2013;80;1778-1783
2050
0
2040
2005, Nepal 2010)
8
2030
• Australia: 3x increase in numbers (Jorm et al
10
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
• Germany: 2x increase (GerMent Health 2013)
12
Persons, millions
years (Prince 2013)
14
Key Model Assumptions
• Disease-free state: free of CVD or CI, other risk
captured by background mortality
• No overlap between states, overlap only captured
in the joint CVD-CIND state
• CVD not reversible (e.g. not cured)
• Disability can be improved/reversed
Key Model Assumptions
• We base our definition of cognitive impairment and functional decline on
ELSA and Whitehall studies.
• Modelling unit: a birth cohort, assumption that within cohort individuals are
homogeneous (e.g. baseline risk)
• Early life influences before 35 years ignored.
• Demographic assumptions:
• Future population projections based on ONS projections (baseline fertility,
mortality and migration)
IMPACT-BAM (Better Ageing Model)
Policy layer: forecasting impact of environment/behaviour change on health outcomes
Potential economics layer: cost-benefit analysis of policy options, including versus do-nothing
scenarios
Validation
CVD prevalence : Model versus HSE 2011
Validation
CVD deaths : observed vs predicted
Reduction in CVD cases
•
Baseline: In 2040 We expect to have ~1,350,000 cases of CVD in men and ~780,000 in
women
•
However by reducing SBP in the population, we could observe fewer cases as results of
the hypothetical interventions
Men
Scenarios on SBP
reduction
Scenario 1
(0.1 mmHg reduction )
Scenario 2
(1.3 mmHg reduction)
Scenario 3
(3 mmHg reduction)
Scenario 4
(3 mmHg reduction in midlife
cohorts)
Median
LUI
Women
UUI
Relative
change
Median
LUI
UUI
Relative
change
1,605
1,058
2,267
0.1%
1,250
579
2,267
0.2%
20,666
13,610
29,219
1.5%
16,054
7,430
29,151
2.1%
47,033
30,932
66,591
3.5%
36,406
16,842
66,247
4.7%
51,066
33,745
71,845
3.8%
40,645
18,989
71,462
5.2%
Reduction in CVD mortality
•
We could expect ~545,700 CVD deaths in men and ~508,000 in women
•
Expected reduction attributable to intervention:
Men
Scenarios on SBP
reduction
Scenario 1
(0.1 mmHg reduction )
Women
Median
LUI
UUI
Relative
change
Median
LUI
UUI
Relative
change
1,276
1,014
1,610
0.2%
1,333
872
2,035
0.3%
16,345
12,962
20,577
3.0%
16,954
11,113
25,973
3.3%
36,937
29,203
46,407
6.8%
38,104
25,011
58,592
7.5%
35,371
27,837
44,909
6.5%
36,435
23,863
56,560
7.2%
Scenario 2
(1.3 mmHg reduction)
Scenario 3
(3 mmHg reduction)
Scenario 4
(3 mmHg reduction in midlife
cohorts)
Reduction in Dementia cases
•
Baseline ~860,000 cases of dementia in men and ~891,000 in women
•
Expected reduction attributable to intervention
Men
Scenarios on SBP
reduction
Scenario 1
(0.1 mmHg reduction )
Scenario 2
(1.3 mmHg reduction)
Scenario 3
(3 mmHg reduction)
Scenario 4
(3 mmHg reduction in midlife
cohorts)
Median
LUI
Women
UUI
Relative
change
Median
LUI
UUI
Relative
change
1,162
567
1,670
0.1%
1,631
503
2,486
0.2%
15,025
7,319
21,601
1.8%
21,044
6,461
32,026
2.4%
34,392
16,698
49,425
4.0%
48,019
14,617
73,321
5.4%
30,781
15,057
46,156
3.6%
41,874
12,615
73,191
4.7%
Reduction in disability cases
•
Baseline: ~1,141,000 cases of disability in men and ~1,115,000 in women
•
Expected reduction attributable to intervention:
Men
Scenarios on SBP
reduction
Scenario 1
(0.1 mmHg reduction )
Scenario 2
(1.3 mmHg reduction)
Scenario 3
(3 mmHg reduction)
Scenario 4
(3 mmHg reduction in midlife
cohorts)
Median
LUI
Women
UUI
Relative
change
Median
LUI
UUI
Relative
change
1,242
621
1,891
0.1%
1,784
564
2,793
0.1%
16,076
8,032
24,460
1.4%
23,035
7,261
36,020
1.4%
36,859
18,347
56,029
3.2%
52,630
16,520
82,036
3.3%
35,644
17,897
54,158
3.1%
49,498
15,726
84,623
3.5%