Milo Puhan discusses the need for individualized risk stratification to determine if a drug will provide benefit or harm for a specific patient. Currently, risks are often oversimplified by looking at single outcomes from clinical trials. Puhan advocates combining a patient's individual risk factors and characteristics, the determinants of treatment outcomes, and the patient's values and preferences using modern technology to provide a multidimensional assessment of absolute treatment effects tailored to that individual. This personalized approach would help clinicians and patients make more informed decisions about whether a treatment is likely to provide net benefit or harm.

1. Milo Puhan, MD, PhD
Professor of Epidemiology & Public Health
University of Zurich, Switzerland
Johns Hopkins University, USA
Individual Risk Stratification
BRN – AJRCCM Workshop
Personalized Respiratory Medicine
Barcelona, June 13, 2014

2. Will this drug be beneficial for a specific individual?
-20% COPD exacerbations
Insomnia + 300%
Anxiety + 200%
Depression + 180%
Diarrhea + 300%
Nausea + 300%
Weight loss + 400%
Headache + 200%
Dizziness + 300%
Excess of

3. How to apply these results to individual patients?
-20% COPD exacerbations
Acute pancreatitis + 500% ?
Insomnia + 300%
Anxiety + 200%
Depression + 180%
Suicide + 500% ?
Diarrhea + 300%
Nausea + 300%
Weight loss + 400%
Headache + 200%
Dizziness + 300%
Excess of
Patient 1
Lung function ↓
≤ 1 exacerbation/y
Chronic GI problems
Mild depression
Patient 2
Lung function ↓↓
≥2 exacerbations/y
No GI problems
Mild depression
Don‘t use drug Worth a try?

4. Individuals have unique combinations of characteristics
Patient 1 ↓ ↓ ↓ ↑ Exacerb.↓ Exacerb.↓ ↑ ↑
Patient 2 ↓↓ ↑ ↓ ↓ Exacerb.↓ Exacerb. ↑ ↓ ↑
Patient 3 ↓↓ ↑ ↑ ↓ Exacerb.↑ Exacerb. ↑ ↑ ↑
Patient …

5. Individuals have unique combinations of characteristics
Patient 1 ↓ ↓ ↓ ↑ Exacerb
↓
Exacerb
↓
↑ ↑
Patient 2 ↓↓ ↑ ↓ ↓ Exacerb
↓
Exacerb
↓
↓ ↑
Patient 3 ↓↓ ↑ ↑ ↓ Exacerb
↑
Exacerb
↑
↑ ↑
Patient
…

6. Current practice of estimating benefits vs. harms
Evidence on
treatment effects
Patient values &
preferences
Determinants of outcome risks &
treatment effects
Exacerbations
Favors
treatment
Favors
control
0.5 0.7 1 1.5 2
Favors
treatment
Favors
control
Nausea
0.001 0.1 1 10 1000
Favors
treatment
Favors
control
Diarrhea
0.1 0.2 0.5 1 2 5 10
RCT control
group event
rates (e.g. 20%
in 1 year)
NNT: 25
NNH: 5
 May be worth a try
 Much uncertainty
since oversimplified:
Single outcome risks
Few outcomes
Implicit preferences

7. Evidence on
treatment effects
Patient values &
preferences
<1
1-5
6-10
11-15
16-20
21-30
61-80
>80
<1 1-3 3-5 6-10 11-15 16-20 21-25 26+
Determinants of outcome risks &
treatment effects
Combine 3 key ingredients and use modern technology

8. How can we get there?
Personalized through risk
stratification and preferences
Based on your individual characteristics and what
is important for you
<1
1-5
6-10
11-15
16-20
21-30
61-80
>80
<1 1-3 3-5 6-10 11-15 16-20 21-25 26+
The chance that you will
benefit from drug X is
very low
(nine out of ten patients
like you will experience
more harm than benefit)
Your risk for a hospital admission
because of COPD
Your risk for
experiencing
severe side
effects
Evidence on
treatment effects
Patient values &
preferences
<1
1-5
6-10
11-15
16-20
21-30
61-80
>80
<1 1-3 3-5 6-10 11-15 16-20 21-25 26+
Determinantsof outcome risks &
treatment effects

9. Lab research
Synthesis and
multidimensional
assessment
Epidemiological &
Clinical research
Well-defined translational paths
Exacerbations
Favors
treatment
Favors
control
0.5 0.7 1 1.5 2
<1
1-5
6-10
11-15
16-20
21-30
61-80
>80
<1 1-3 3-5 6-10 11-15 16-20 21-25 26+

10. Multidimensional
assessments
- A quickly
emerging field
RCTs
Registries
Observational studies
Surveys (preferences)
Observational studies
(prognosis)
Various methods
Outcome risks
Estimates of effect
Absolute effects
Benefit harm
analysis
Importance of
Outcomes
Lab research
Synthesis and
multidimensional
assessment
Epidemiological &
Clinical research
Guo et al. Value Health 2010, 13(5):657–666.
Puhan et al. BMC Med Res Meth 2012, 12:173
Boyd, et al. AHRQ 2012 .12(13)-EHC150-EF
FDA PDUFA V Plan (FY 2013-2017)

11. Approaches for benefit harm assessment
Lab research
Synthesis and
multidimensional
assessment
Epidemiological &
Clinical research
Decision making
context?

12. Simulation to consider statistical uncertainty of
effects and incidence rates
<1
1-5
6-10
11-15
16-20
21-30
61-80
>80
<1 1-3 3-5 6-10 11-15 16-20 21-25 26+
Lab research
Synthesis and
multidimensional
assessment
Epidemiological &
Clinical research

13. Based on your individual characteristics and what
is important for you
<1
1-5
6-10
11-15
16-20
21-30
61-80
>80
<1 1-3 3-5 6-10 11-15 16-20 21-25 26+
The chance that you will
benefit from drug X is
very low
(nine out of ten patients
like you will experience
more harm than benefit)
Your risk for a hospital admission
because of COPD
Your risk for
experiencing
severe side
effects

14. Age 45-54 Age 55-64 Age 65-74 Age 75-84
10-year risk of ischemic stroke: 0%-5% 10-year risk of ischemic stroke: 0%-5% 10-year risk of ischemic stroke: 0%-5% 10-year risk of ischemic stroke: 0%-5%
10-year risk of severe GI bleeds40%-50% 10-year risk of severe GI bleeds40%-50% 10-year risk of severe GI bleeds40%-50% 10-year risk of severe GI bleeds40%-50%
30%-40% 30%-40% 30%-40% 30%-40%
20%-30% 20%-30% 20%-30% 20%-30%
15%-20% 15%-20% 15%-20% 15%-20%
10%-15% 10%-15% 10%-15% 10%-15%
5%-10% 5%-10% 5%-10% 5%-10%
1%-5% 1%-5% 1%-5% 1%-5%
0%-1% 0%-1% 0%-1% 0%-1%
0%-1% 1%-3% 3%-5% 5%-10% 10%-15% 15%-20% 20%-25% 25%-30% 0%-1% 1%-3% 3%-5% 5%-10% 10%-15% 15%-20% 20%-25% 25%-30% 0%-1% 1%-3% 3%-5% 5%-10% 10%-15% 15%-20% 20%-25% 25%-30% 0%-1% 1%-3% 3%-5% 5%-10% 10%-15% 15%-20% 20%-25% 25%-30%
10-year risk of MI 10-year risk of MI 10-year risk of MI 10-year risk of MI
10-year risk of ischemic stroke: 5%-10% 10-year risk of ischemic stroke: 5%-10% 10-year risk of ischemic stroke: 5%-10% 10-year risk of ischemic stroke: 5%-10%
10-year risk of severe GI bleeds40%-50% 10-year risk of severe GI bleeds40%-50% 10-year risk of severe GI bleeds40%-50% 10-year risk of severe GI bleeds40%-50%
30%-40% 30%-40% 30%-40% 30%-40%
20%-30% 20%-30% 20%-30% 20%-30%
15%-20% 15%-20% 15%-20% 15%-20%
10%-15% 10%-15% 10%-15% 10%-15%
5%-10% 5%-10% 5%-10% 5%-10%
1%-5% 1%-5% 1%-5% 1%-5%
0%-1% 0%-1% 0%-1% 0%-1%
0%-1% 1%-3% 3%-5% 5%-10% 10%-15% 15%-20% 20%-25% 25%-30% 0%-1% 1%-3% 3%-5% 5%-10% 10%-15% 15%-20% 20%-25% 25%-30% 0%-1% 1%-3% 3%-5% 5%-10% 10%-15% 15%-20% 20%-25% 25%-30% 0%-1% 1%-3% 3%-5% 5%-10% 10%-15% 15%-20% 20%-25% 25%-30%
10-year risk of MI 10-year risk of MI 10-year risk of MI 10-year risk of MI
10-year risk of ischemic stroke: 10%-15% 10-year risk of ischemic stroke: 10%-15% 10-year risk of ischemic stroke: 10%-15% 10-year risk of ischemic stroke: 10%-15%
10-year risk of severe GI bleeds40%-50% 10-year risk of severe GI bleeds40%-50% 10-year risk of severe GI bleeds40%-50% 10-year risk of severe GI bleeds40%-50%
30%-40% 30%-40% 30%-40% 30%-40%
20%-30% 20%-30% 20%-30% 20%-30%
15%-20% 15%-20% 15%-20% 15%-20%
10%-15% 10%-15% 10%-15% 10%-15%
5%-10% 5%-10% 5%-10% 5%-10%
1%-5% 1%-5% 1%-5% 1%-5%
0%-1% 0%-1% 0%-1% 0%-1%
0%-1% 1%-3% 3%-5% 5%-10% 10%-15% 15%-20% 20%-25% 25%-30% 0%-1% 1%-3% 3%-5% 5%-10% 10%-15% 15%-20% 20%-25% 25%-30% 0%-1% 1%-3% 3%-5% 5%-10% 10%-15% 15%-20% 20%-25% 25%-30% 0%-1% 1%-3% 3%-5% 5%-10% 10%-15% 15%-20% 20%-25% 25%-30%
10-year risk of MI 10-year risk of MI 10-year risk of MI 10-year risk of MI
10-year risk of ischemic stroke: 15%-20% 10-year risk of ischemic stroke: 15%-20% 10-year risk of ischemic stroke: 15%-20% 10-year risk of ischemic stroke: 15%-20%
10-year risk of severe GI bleeds40%-50% 10-year risk of severe GI bleeds40%-50% 10-year risk of severe GI bleeds40%-50% 10-year risk of severe GI bleeds40%-50%
30%-40% 30%-40% 30%-40% 30%-40%
20%-30% 20%-30% 20%-30% 20%-30%
15%-20% 15%-20% 15%-20% 15%-20%
10%-15% 10%-15% 10%-15% 10%-15%
5%-10% 5%-10% 5%-10% 5%-10%
1%-5% 1%-5% 1%-5% 1%-5%
0%-1% 0%-1% 0%-1% 0%-1%
0%-1% 1%-3% 3%-5% 5%-10% 10%-15% 15%-20% 20%-25% 25%-30% 0%-1% 1%-3% 3%-5% 5%-10% 10%-15% 15%-20% 20%-25% 25%-30% 0%-1% 1%-3% 3%-5% 5%-10% 10%-15% 15%-20% 20%-25% 25%-30% 0%-1% 1%-3% 3%-5% 5%-10% 10%-15% 15%-20% 20%-25% 25%-30%
10-year risk of MI 10-year risk of MI 10-year risk of MI 10-year risk of MI
10-year risk of ischemic stroke: 20%-25% 10-year risk of ischemic stroke: 20%-25% 10-year risk of ischemic stroke: 20%-25% 10-year risk of ischemic stroke: 20%-25%
10-year risk of severe GI bleeds40%-50% 10-year risk of severe GI bleeds40%-50% 10-year risk of severe GI bleeds40%-50% 10-year risk of severe GI bleeds40%-50%
30%-40% 30%-40% 30%-40% 30%-40%
20%-30% 20%-30% 20%-30% 20%-30%
15%-20% 15%-20% 15%-20% 15%-20%
10%-15% 10%-15% 10%-15% 10%-15%
5%-10% 5%-10% 5%-10% 5%-10%
1%-5% 1%-5% 1%-5% 1%-5%
0%-1% 0%-1% 0%-1% 0%-1%
0%-1% 1%-3% 3%-5% 5%-10% 10%-15% 15%-20% 20%-25% 25%-30% 0%-1% 1%-3% 3%-5% 5%-10% 10%-15% 15%-20% 20%-25% 25%-30% 0%-1% 1%-3% 3%-5% 5%-10% 10%-15% 15%-20% 20%-25% 25%-30% 0%-1% 1%-3% 3%-5% 5%-10% 10%-15% 15%-20% 20%-25% 25%-30%
10-year risk of MI 10-year risk of MI 10-year risk of MI 10-year risk of MI
10-year risk of ischemic stroke: 25%-30% 10-year risk of ischemic stroke: 25%-30% 10-year risk of ischemic stroke: 25%-30% 10-year risk of ischemic stroke: 25%-30%
10-year risk of severe GI bleeds40%-50% 10-year risk of severe GI bleeds40%-50% 10-year risk of severe GI bleeds40%-50% 10-year risk of severe GI bleeds40%-50%
30%-40% 30%-40% 30%-40% 30%-40%
20%-30% 20%-30% 20%-30% 20%-30%
15%-20% 15%-20% 15%-20% 15%-20%
10%-15% 10%-15% 10%-15% 10%-15%
5%-10% 5%-10% 5%-10% 5%-10%
1%-5% 1%-5% 1%-5% 1%-5%
0%-1% 0%-1% 0%-1% 0%-1%
0%-1% 1%-3% 3%-5% 5%-10% 10%-15% 15%-20% 20%-25% 25%-30% 0%-1% 1%-3% 3%-5% 5%-10% 10%-15% 15%-20% 20%-25% 25%-30% 0%-1% 1%-3% 3%-5% 5%-10% 10%-15% 15%-20% 20%-25% 25%-30% 0%-1% 1%-3% 3%-5% 5%-10% 10%-15% 15%-20% 20%-25% 25%-30%
10-year risk of MI 10-year risk of MI 10-year risk of MI 10-year risk of MI
Lab research
Synthesis and
multidimensional
assessment
Epidemiological &
Clinical research

15. Moderate or severe
exacerbation
Acute pancreatitis
Insomnia
Anxiety
Depression
Suicide (completed)
Diarrhea
Nausea
Weight loss
Headache
Dizziness
Benefits and harms of roflumilast
Yu T, Fain K, Boyd CM, Singh S, Weiss CO, Li T, Varadhan R, Puhan MA. Thorax 2014, doi: 10.1136
FDApivotal and
safety pool
COPD observational studies
Placebo groups of RCTs
Survey (preferences)
Observationalstudies
(clustering events)
Outcome risks
Estimates of effect
Absolute effects
Benefit harm
analysis
Importance of
Outcomes
Lab research
Synthesisand
multidimensional
assessment
Epidemiological &
Clinical research

16. Net benefit-harm index* per 10,000 patients treated over 1 year by patient profiles
Type of analysis
Patients' projected 1-year risk of having ≥1 moderate or severe
COPD exacerbation
30% 60% 90%
Men Women Men Women Men Women
Age
<65 ≥65 <65 ≥65 <65 ≥65 <65 ≥65 <65 ≥65 <65 ≥65
Analysis I:
Equal weights
0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Analysis II:
Weights based on importance of
outcomes
0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Analysis III (main analysis):
Weights based on importance and
co-occurrence of harm outcomes
0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Probability that roflumilast overall beneficial
Yu T, Fain K, Boyd CM, Singh S, Weiss CO, Li T, Varadhan R, Puhan MA. Thorax 2014, doi: 10.1136
Lab research
Synthesisand
multidimensional
assessment
Epidemiological &
Clinical research

17. Probability that
roflumilast is
beneficial
Prevention of severe exacerbations (requiring admission)
Lab research
Synthesis and
multidimensional
assessment
Epidemiological &
Clinical research
Yu T, et al. Thorax 2013, doi: 10.1136

18. Lab research
Synthesis and
multidimensional
assessment
Epidemiological &
Clinical research
Well-defined translational paths
Exacerbations
Favors
treatment
Favors
control
0.5 0.7 1 1.5 2
<1
1-5
6-10
11-15
16-20
21-30
61-80
>80
<1 1-3 3-5 6-10 11-15 16-20 21-25 26+

19. Great importance of risk prediction
Lab research
Synthesis and
multidimensional
assessment
Epidemiological &
Clinical research
<1
1-5
6-10
11-15
16-20
21-30
61-80
>80
<1 1-3 3-5 6-10 11-15 16-20 21-25 26+
Mortality: BODE, ADO, HADO, Briggs
Exacerbations: DOSE, SAFE, Briggs
Health status: HADO, Siebeling
Harm outcomes: COPD observational
studies, RCT control groups, registries

20. Prediction models of the future?
Rigorous evaluation is critical
Mean predicted risk: 10.5%
Mean observed risk: 8.2%
HL: p<0.0001
0
10
20
30
40
Observed risk
of acute renal
failure in %
0 10 20 30 40
Predicted risk of acute
renal failure in %
0.00
0.25
0.50
0.75
1.00
Sensitivity
0.00 0.25 0.50 0.75 1.00
1 - Specificity
Area under curve
0.76 (95% CI 0.71-0.81)
<1
1-5
6-10
11-15
16-20
21-30
31-40
>40
10 year
risk for
severe GI
bleed
10 year risk for MI
<1 1-3 3-5 6-10 11-15 16-20 21-25 26+
Current COPD
models
Current COPD
models
+ biomarkers
Current COPD
models
+ genetics
Only genetics
Lab research
Synthesis and
multidimensional
assessment
Epidemiological &
Clinical research

21. Model update with new predictors
(e.g. biomarkers) and correct re-classification
<1
1-5
6-10
11-15
16-20
21-30
31-40
>40
10 year
risk for
severe GI
bleed
10 year risk for MI
<1 1-3 3-5 6-10 11-15 16-20 21-25 26+
Manolio TA. N Engl J Med 2010;363:166-176.
Lab research
Synthesis and
multidimensional
assessment
Epidemiological &
Clinical research

22. Thomsen et al. JAMA. 2013;309(22):2353-2361
Lab research
Synthesis and
multidimensional
assessment
Epidemiological &
Clinical research
High CRP,
fibrinogen and/or
leukocyte count
0: 0.9%
1: 1.8%
2: 3.2%
3: 8.1%
Only biomarkers Added value of biomarkers
From uni- to multivariate associations

23. Lab research
Synthesis and
multidimensional
assessment
Epidemiological &
Clinical research
From associations of 3 biomarkers to risk prediction

24. From risk prediction to benefit harm prediction
Probability that
roflumilast is
beneficial
Lab research
Synthesis and
multidimensional
assessment
Epidemiological &
Clinical research

25. Testing
EBM 2.0: Focus on the individual
Exploring what is
important
Decision making
Based on your individual characteristics and what
is important for you
<1
1-5
6-10
11-15
16-20
21-30
61-80
>80
<1 1-3 3-5 6-10 11-15 16-20 21-25 26+
The chance that you will
benefit from drug X is
very low
(nine out of ten patients
like you will experience
more harm than benefit)
Your risk for a hospital admission
because of COPD
Your risk for
experiencing
severe side
effects
Lab research
Synthesis and
multidimensional
assessment
Epidemiological &
Clinical research

26.  Maximizes benefit – minimizes
harms for individuals
 Links individual characteristics
(from molecular to composite
level) to treatment oucomes
Individual Risk Stratification
Personalized through risk
stratification and preferences
Based on your individual characteristics and what
is important for you
<1
1-5
6-10
11-15
16-20
21-30
61-80
>80
<1 1-3 3-5 6-10 11-15 16-20 21-25 26+
The chance that you will
benefit from drug X is
very low
(nine out of ten patients
like you will experience
more harm than benefit)
Your risk for a hospital admission
because of COPD
Your risk for
experiencing
severe side
effects
Evidence on
treatment effects
Patient values &
preferences
<1
1-5
6-10
11-15
16-20
21-30
61-80
>80
<1 1-3 3-5 6-10 11-15 16-20 21-25 26+
Determinantsof outcome risks &
treatment effects
Well-defined translational approach to make evidence-
based personalized respiratory medicine happen
Lab research
Synthesis and
multidimensional
assessment
Epidemiological &
Clinical research