Download Clinical Applications of Risk Prediction Models

Clinical Applications of Risk
Prediction Models
Laura Esserman, M.D., M.B.A.
Professor of Surgery and Radiology
Director, UCSF Carol Franc Buck Breast Care Center
Agenda
•Current Clinical Climate for Prevention
•Potential for Risk Tools to Refine Risk, motivate interventions
•Framework for Decision Aids: the need for tools that provide
information in a decision ready context
•How risk models can be integrated into clinical consultations
•Insights from using decision aids, models
Current Clinical Decision Making
Calculate
5-year
Gail Risk
Gail Risk
Below
1.67%
Gail Risk
Above
1.67%
Rush-Port, Vogel et al.
Screening
Offered
Tamoxifen
(50% risk
reduction)
5-20% take
Tamoxifen
80-95% choose
screening
The
Gail
Model
Does Not
Identify
a Identify
Truly High
The
Gail
Model
Does
Not
Risk Group of Women
a Truly High Risk Group of Women
100%
90%
75%
65%
44%
50%
33%
25%
14%
4%
0%
45-49
50-54
55-59
60-65
65-70
All Ages
Percent of Nurses Health Study Above the High-Risk
Percent
High-Risk Cutoff
Cutoff Point
Point
(5 yr Gail Score of 1.67%)
1.67%)
Rockhill et
Rockhill
et al.
al.
60
Cardiovascular
Breast Cancer
Lung Cancer
50
40
30
20
10
Age
Phillips, et al, NEJM, Vol. 340, No. 2, 1999
80
70
60
50
40
30
20
10
0
0
Percent of All Deaths
What should compel Providers to be concerned
with prevention
Age and Competing Causes of Death
High Risk Patients Don’t Choose Tamoxifen
2/43 high risk patients chose to take Tamoxifen for breast cancer
prevention
Educational sessions had no influence
Fear of side effects
*Rush Port E, et al Ann Surg Oncol, Vol.8, No. 7, 2001
Decision Making in the Clinical Setting
Breast Cancer Prevention Decisions are complex
High
Disagreement on
Preferences
Chaos
Zone of Complexity
Order
Low
Low
High
Uncertainty about Outcomes
What compels women at
high risk to consider an
intervention?
1. Evidence that their risk is significant compared to
others
2. Evidence that there is an intervention that will
help THEM specifically
3. Evidence that the intervention will not have
significant side effects
4. Evidence that the intervention is working
Improving the signal-to-noise ratio
Decision Analysis
Decision aid strives to provide the basic elements of a decision:
frame, alternatives, information, preferences and logic
Adult Learning
Decision aids should let women choose what they want to learn
– What are people ready to receive?
– Layers of complexity (start simple, detail is optional)
Cognitive Science (Tufte)
Decision aid should use graphical formats that require the least
amount of cognitive processing
– Train people on small number of formats, stick to them
Risk Communication
Relative risk presentations are confusing, misleading, and bias
patients toward intervention
Clinically Accessible Biomarkers
Biomarkers
Risk
Discrimination
Detection
Tool
Cost
Targeted
intervention
Atypia
++
rFNA
Ductal Lavage
Open Bx
++
++
+++
+Tamoxifen, ?AIs
Breast Density
++/+++
Mammo
MRI
++
+++
?Soy, Tam?
Serum
Estradiol
+
Blood Test
++
Tamoxifen, Raloxifen
Serum
Testosterone
+
Blood Test
++
Tamoxifen, Raloxifen
LCIS
++
Bx
MRI
++
+++
+Tam
DCIS
+++
Mammo
MRI, Bx
++
+++
? Tam ?AI ?Statins
?IGFR1 ?
BRCA 1,2
mutations
++++
Blood Test
+++
Propylactic surgery,
Tam (BRCA2),
oophorectomy
Prevention Decision Model
Carol Franc Buck Breast Care Center | UCSF Medical Center
©2002
Elissa Ozanne, Laura Esserman
Anna Bella Smith
5yr Gail Score: 2.1%
Lifetime Gail Score: 17.3%
Main
Learning About Your Risk
What is my risk of breast cancer?
Prevention Options
What can I do to lower my risk?
Getting Perspective
How does my risk compare to other women?
Risks and Benefits
Tests to learn more about breast cancer risks
and benefits of therapies
Anna Bella Smith
5yr Gail Score: 2.1%
Lifetime Gail Score: 17.3%
Main
Prevention Decision Model :
Learning About Your Risk:
What is my risk of breast cancer?
Anna Bella Smith
5yr Gail Score: 2.1%
Lifetime Gail Score: 17.3%
Next
Main
Prevention Decision Model : Learning About Your Risk
My Breast Cancer Risk Over Time
Age
51
Number of years of HRT usage
Menopausal Status
GAIL MODEL:
The Gail Risk Assessment Model is a
none
statistical model for estimating the risk
Post
of developing breast cancer in women
undergoing annual screening.This tool
was developed to assist in providing
Gail Risk
women with a realistic and
5 year
2.1%
Lifetime
17.3%
individualized risk estimate of short
and long term breast cancer risk.
CLAUS MODEL:
The Claus model estimates the
Claus Risk
By age 39
NA
By age 49
NA
By age 59
2.7%
By age 69
6.5%
By age 79
9.7%
Anna Bella Smith
5yr Gail Score: 2.1%
Lifetime Gail Score: 17.3%
probability that a woman will develop
breast cancer based on her family
history of cancer. This includes the
number of first and second-degree
relatives with breast cancer and the
Source: Fisher B, et al, JNCI, vol 90, No. 18, 1998
Anna Bella Smith
age of cancer onset.
Next
Main
4
Prevention Decision Model :
Getting Perspective:
How does my risk compare to other women?
Anna Bella Smith
5yr Gail Score: 2.1%
Lifetime Gail Score: 17.3%
Next
Main
Prevention Decision Model :
Getting Perspective
14%
140
12%
120
10%
100
8%
80
6%
60
Top 25%
4%
40
2%
1.67%
FDA threshold for
tamoxifen use.
20
Middle 50%
Bottom 25%
0%
<30
0
30
35
40
45
50
55
60
65
70
75
80
Age
EXAMPLE:
A 65 year old women with a five year Gail Score of 3% would fall
somewhere in the top 25% of this distribution.
Anna Bella Smith
5yr Gail Score: 2.1%
Lifetime Gail Score: 17.3%
Next
Main
Source: B.Rockhill, NHS data
1000
Risk of Diagnosis
100%
Rate/1000 Women
What Does My Gail Score Mean? What is My Risk Compared to Others?
1000
90%
900
80%
800
70%
700
60%
600
50%
500
40%
400
30%
300
Avg. Risk of BC Diagnosis
100%
20%
10%
0
2.8%
Within 10 yrs
Anna Bella Smith
5yr Gail Score: 2.1%
Lifetime Gail Score: 17.3%
6.2%
200
12.0%
9.6%
Rate/1000 women
Average Risk of Breast Cancer Diagnosis for Women (Age 50~60)
100
0
Within 20 yrs
Prev
Within 30 yrs
20~30 | 30~40 | 40~50
Lifetime Risk
50~60
60~70
Next
Main
Source: Surveillance, Epidemiology, and End Results (SEER) Cance r Statistics Review 1973 - 1998.
Prevention Decision Model : Getting Perspective
19
Average Chances of NOT Being Diagnosed with Breast Cancer (Age 50~60)
93.8%
Chances of not being diagnosed
90%
88.0%
900
80%
800
70%
700
60%
600
50%
500
40%
400
30%
300
20%
10%
0
2.8%
Within 10 yrs
Anna Bella Smith
1000
90.4%
5yr Gail Score: 2.1%
Lifetime Gail Score: 17.3%
6.2%
9.6%
Rate/1000 women
97.2%
100%
200
12.0%
100
0
Within 20 yrs
Prev
Within 30 yrs
20~30 | 30~40 | 40~50
Lifetime Risk
50~60
60~70
Next
Main
Source: Surveillance, Epidemiology, and End Results (SEER) Cance r Statistics Review 1973 - 1998.
Prevention Decision Model : Getting Perspective
20
Prevention Decision Model : Getting Perspective
50 year old woman has…
Source: Journal of the National Cancer Institute, Vol. 94, No. 11, June 5, 2002.
In the next ten years, an average
Risk of Diagnosis from:
Breast Cancer
2.8%
4.2
Risk of Death from:
0.75-1.0%
0.5 ~ 0.7%
Breast Cancer
Heart Attack
0.4 ~ 1.4%
Lung Cancer (smoker)
2.1 ~ 6.5%
Lung Cancer (non-smoker)
0.2 ~ 0.5%
Pneumonia (smoker)
0.1 ~ 0.2%
Accidents
0.2%
Other Risks this year alone:
Increase in breast cancer for each year of HRT use
1 ~ 2%
Injured in an automobile accident
8%
Visit the doctor about the flu
Anna Bella Smith
5yr Gail Score: 2.1%
Lifetime Gail Score: 17.3%
Prev
38%
20~30 | 30~40 | 40~50
50~60
60~70
Next
Main
21
Prevention Decision Model :
Prevention Options:
What can I do to lower my risk?
Lifestyle Changes
Chemoprevention
Surgery
Next
Prevention Decision Model : Preventative Measures
These moderate modifications are recommended for all women as potential
risk reduction strategies, in addition to vigilant surveillance.
-Weight control
-No cigarette smoking
-Decreased alcohol consumption
-Exercise
Click here to learn about Hormone Replacement Therapy and
Breast Cancer Risk.
Lifestyle Changes
Chemoprevention
Surgery
Next
Source: Ross D, 23rd annual San Antonio Breast Cancer Symposium, 2000:
Summary by Pritchard, KI Vogel VG, Cancer Journal for Clinicians, Vol. 50, No. 3, 2000
Lifestyle Changes
Prevention Decision Model : Prevention Options
Chemoprevention
Benefits and Risks of Tamoxifen Usage (Ages 35~49): 5 Year Estimates
1000
100%
Risks
200
20%
150
15%
100
10%
50
5%
3.35% 1.89%
1.34% 0.68%
0
Invasive
Breast Cancer
Non-Invasive
Breast Cancer
0.36% 0.14%
Fractures
0.07% 0.11% 0.36% 0.45%
Endometrial
Cancer
0%
Vascular Events
Placebo
Tamoxifen
35~49 | 50-60
Lifestyle Changes
Chemoprevention
| 60+
Surgery
Next
Source: Gail, et al, JNCI, vol 91, No. 3, 1999
Rate/1000
Benefits
Prevention Decision Model : Prevention Options
Chemoprevention
Benefits and Risks of Tamoxifen Usage (Age 50-60): 5 Year Estimates
1000
100%
Risks
200
20%
150
15%
100
10%
50
5%
3.1%
1.6%
1.34% 0.68%
1.9% 0.6%
0
Invasive
Breast Cancer
Non-Invasive
Breast Cancer
Fractures
0.4% 1.2%
Endometrial
Cancer
1.0% 1. 1%
0%
Vascular Events
Placebo
Tamoxifen
35~49 | 50-60 | 60+
Lifestyle Changes
Chemoprevention
Surgery
Next
Source: Gail, et al, JNCI, vol 91, No. 3, 1999
Rate/1000
Benefits
Prevention Decision Model : Prevention Options
Chemoprevention
Benefits and Risks of Tamoxifen Usage (Age 60+): 5 Year Estimates
1000
100%
Risks
200
20%
150
15%
100
10%
5.6%
50
3.67%
1.67%
1.34% 0.68%
1.9%
0
Invasive
Breast Cancer
Non-Invasive
Breast Cancer
Fractures
5%
3.8%
2.1%
3.1%
0.7%
Endometrial
Cancer
0%
Vascular Events
Placebo
Tamoxifen
35~49 | 50-60
Lifestyle Changes
Chemoprevention
| 60+
Surgery
Next
Source: Gail, et al, JNCI, vol 91, No. 3, 1999
Rate/1000
Benefits
Prevention Decision Model :
Risks and Benefits:
Genetic
Testing
Tests to learn more about breast
cancer risks and benefits of therapies
Ductal Lavage and
Fine Needle Aspiration
Serum
Estradiol
Next
Prevention Decision Model : Risks and Benefits
Ductal Lavage and Fine Needle Aspiration
Atypical Hyperplasia Predicts Benefit from Tamoxifen
1000
Expected Breast Cancer Risk Over Five Years
15%
200
50% relative risk
reduction with
tamoxifen
86% relative risk
reduction with
tamoxifen
150
10%
100
5.1%
5%
50
3.4%
1.7%
0.7%
0%
All Women
0
Atypical Hyperplasia
Women on tamoxifen had about 50% of the number
Women with atypical hyperplasia on
of breast cancers seen in the placebo group –
50% relative risk reduction.
tamoxifen had about 14% of the number
of breast cancers seen in the placebo
group – 86% relative risk reduction.
The absolute benefit is smaller - only 3.4% high-risk
women are expected to develop breast cancer as
compared to 1.7% in women using tamoxifen –
The absolute risk decreased from an
expected 5.1% to 0.7% - a 4.4%
1.7% absolute risk reduction over 5 years.
absolute risk reduction over 5 years.
Genetic
Testing
Ductal Lavage and
Fine Needle Aspiration
Serum
Estradiol
Placebo
Tamoxifen
Next
Source: Fisher B, et al, JNCI, Vol 90, No. 18, 1998
20%
Rate/1000 Women
Short term (~5 yr) Risk
of Breast Cancer
100%
Prevention Decision Model : Risks and Benefits
Ductal Lavage and Fine Needle Aspiration
1000
20%
Highest
Risk
Group
15%
15%
10%
5%
Lowest
Risk
Group
0%
200
150
100
Middle
Risk
Group
4%
Lowest risk group
For women with 5 yr Gail risk less
than 2%, risk decreases to below
1% over 3 years for both women
with AH and no AH.
Middle risk group
For women with 5 yr Gail risk
greater than 2% but with no AH,
risk is about 4% in 3 years.
50
0%
0
5 yr Gail Score < 2%
Rate/1000 Women
Short term (~3yr) Risk
of Breast Cancer
100%
5 yr Gail Score > 2%
Highest risk group
For women with 5yr Gail risk is
greater than 2% with the
presence of AH, risk is about 15%
in 3 years.
No Atypia
Atypia
Fabian JNCI 2001
Genetic
Testing
Ductal Lavage and
Fine Needle Aspiration
Serum
Estradiol
Next
Source: Sauter, 1997; Fabian CJ, et al, JNCI Vol. 92, No. 15, 2000
Learning from Atypical Hyperplasia (AH)
Prevention Decision Model : Risks and Benefits
Ductal Lavage and Fine Needle Aspiration
1000
20%
200
15%
15%
150
10%
5%
100
Each Less than 1%
50
4%
2.1%
2%
0%
0
Lowest Risk Group
5 yr Gail Score < 2%
Independent of AH findings
Middle Risk Group
5 yr Gail Score > 2%
No finding of AH
Highest Risk Group
5 yr Gail Score > 2%
Finding of AH
No Treatment
50% Risk Relative Reduction with Tamoxifen Use
86% Risk Relative Reduction with Tamoxifen Use
Genetic
Testing
Ductal Lavage and
Fine Needle Aspiration
Serum
Estradiol
Next
Source: Sauter, 1997; Fabian CJ, et al, JNCI Vol. 92, No. 15, 2000
100%
Rate/1000 Women
Short term (~3yr) Risk
of Breast Cancer
Atypical Hyperplasia and the Benefit from Tamoxifen
Prevention Decision Model : Risks and Benefits
Serum Estradiol
Learning From Serum Estradiol Level: Postmenopausal Women
20%
200
15%
150
10%
5%
0%
100
Lowest
Risk
Group
0.6%
0
Highest
Risk
Group
3%
1.2%
50
1.8%
0
>0 to <5
5 to 10
>10
Serum Estradiol Level (pmol/L)
Women with the highest estradiol level had about a three fold risk of breast cancer
as compared to the women with the lowest estradiol level.
Higher hormone levels in the blood are associated with a higher risk of breast cancer.
Genetic
Testing
Ductal Lavage and
Fine Needle Aspiration
Serum
Estradiol
Next
Source: Cummings S. et al, JAMA, 287: 22, 2002
1000
Short Term Breast Cancer Risk
Rate/1000 Women
Short term (~4yr) Risk
of Breast Cancer
100%
Prevention Decision Model : Risks and Benefits
Serum Estradiol
Learning From Serum Estradiol Level: Postmenopausal Women
20%
200
15%
150
10%
76% relative
risk reduction
From Raloxifen
5%
0%
0.6% 0.6%
0
1.2%
1.8%
0.4%
>0 to <5
0.8%
5 to 10
3%
0.7%
100
50
0
>10
Placebo
Serum Estradiol Level (pmol/L)
Raloxifene
Women with the highest estradiol levels on raloxifene had about 24% the number of
breast cancers seen in the placebo group. The absolute risk decreased from 3% to 0.7%.
As hormone levels in the blood is higher, the benefits of raloxifene increase. Side effects
of raloxifene are similar to those of tamoxifen but do not include endometrial events.
Genetic
Testing
Ductal Lavage and
Fine Needle Aspiration
Serum
Estradiol
Next
Source: Cummings S. et al, JAMA, 287: 22, 2002
1000
Short Term Breast Cancer Risk
Rate/1000 Women
Short term (~4yr) Risk
of Breast Cancer
100%
Prevention Decision Model : Risks and Benefits
Genetic Testing
Genetic Testing and the Benefit of Prevention Options
100%
1000
800
60%
600
50%
40%
20%
400
34%
20%
200
3.75%
6.4%
0%
0
Lower Risk Estimate
For Genetic Carriers
Higher Risk Estimate
For Genetic Carriers
No Treatment
50-70% Relative Risk Reduction from Oophorectomy
90-95% Risk Relative Reduction from Mastectomy
Genetic
Testing
Ductal Lavage and
Fine Needle Aspiration
Serum
Estradiol
Next
Source: ASCO Proceedings 2002
80%
Rate/1000 Women
Lifetime Risk
of Breast Cancer
85%
Insights
There is a critical need for dynamic models that enable us to
assess the impact of interventions– that is what patients want
Biomarkers that predict effectiveness of interventions will
increase willingness/motivation to accept interventions
There is a hierarchy of risk models
– e.g. BRCA trumps Gail
– Determines impact of and discussion about options,interventions
Risk that motivates patients to choose an intervention:
– 10-15% risk at 5 years
– Risk of recurrence after surgery for non-comedo DCIS
10-12% at 5 years, 20% risk at 10 years
– Maybe DCIS is the best opportunity for prevention?
Cost Benefit Model
Elissa Ozanne PhD; Laura Esserman MD MBA
Goals
Understand value of biomarkers for breast cancer risk
Evaluate cost effectiveness using atypia as an example
Methods
Markov model, evidence from clinical studies
Strategies Examined:
1.
2.
3.
4.
1. Screening: Routine screening (mammography) all women
2. Tamoxifen: Tamoxifen therapy for all women
3. Lavage: Attempt lavage, tam use if DL possible and atypia found
4. FNA: 4 quadrant FNA all women, tam use only for atypia
Ozanne, Esserman 2004, Cancer Epidemiology and Biomarkers, accepted
Sensitivity
•biomarker relative risk prediction increases cost effectiveness
• FNA and DL are more CE if atypia is a good predictor
•more effective intervention increases CE
• If biomarker predicts more effect of drug, CE increases
• inexpensive tests offer highly cost effective strategies
• If it is expensive/painful to get biomarker, treating
everyone50-70
Mammography
is more CE
•inexpensive interventions offer highly cost effective strategies
• Expensive effective interventions not very cost effective
What is the yearly hazard rate
for progression to cancer for . . .
Annual Hazard
DCIS
1-3%
Atypia
Gail Risk > 2
Gail Risk < 2
4%
1%
LCIS
family history
none
1-2%
0.5-1%
BRCA1/2
1-5%
5 yr Gail Risk >5
1-2%
60 yr old Gail <2
0.3-0.5%
CBC for pt with
Ca
0.5%
How do the treatments vary? . . .
Treatment
DCIS
Atypia
Gail Risk > 2
Gail Risk < 2
LCIS
family history
none
BCS
BCS + XRT
BCS + XRT+Tam
Mastectomy
Screen
Tam
Bilat Mastectomy
Screen
Tam
Bilat Mastectomy
BRCA1/2
Screen
Oophorectomy
Tam
Bilat Mastectomy
High Risk
Gail>1.7; Inv Ca
Screen
Consider Tam
What makes DCIS
treatment hard to change?
•
Perspective not optimal
•
Poor understanding of
Risk, timing of
progression
Prevention Paradigm
High Risk
Conditions
Atypia
Normal
cells
Breast
Cancer
DCIS
LCIS
Neoadjuvant Therapy?
Improvements
The Prevention Tool we developed is a physician decision
aid
evidence is organized using common outcome: Risk at
5,10 years
Patient Physician Aids should include more layering of
information
Decisions can be layered by side effects: serious vs. QOL
Trial of tool vs. not
desire for risk stratification
choice of interventions
Side Effects
Yes
Weigh risk vs benefit
Serious
No
Review side effects
Trial of medication
Sx
Weigh Sx vs. benefits
No Sx
Continue
A Good Decision Aid
Enables insight
Facilitates dialogue among providers, patients, families
Reduces confusion
Motivates change in approach based on personal preferences
Requires models that provide risk in perspective, and
enable tailoring of risk based on interventions