Estimating the Burden of DiseaseExamining the impact of changing risk factors on colorectal cancer incidence and mortality

Karen M. Kuntz, ScD

Cancer Risk Prediction Models: A Workshop on Development, Evaluation, and Application

National Cancer Institute

May 20-21, 2004

** Results presented are preliminary.

Decision-Analytic Models

• Analytical structures that represent key elements of a disease

• Goal: evaluate policies in terms of costs and health benefits (not estimation)

• Cohort models vs. population-based model

• Risk functions often incorporated

Age-standardized incidence and mortality

Cases

Deaths

0

20

40

60

80

100

120

1975 1980 1985 1990 1995 2000

Year

CR

C E

ven

ts p

er

10

0,0

00

CRC Risk Factors

• Body mass index (BMI)• Smoking• Folate intake (multivitamin use)• Physical activity• Red meat consumption• Fruit and vegetable consumption• Aspirin use• Hormone replacement therapy (HRT)

Individual Risk Functions

• Pr(CRC | BMI, smoking, MV use, etc.)– Annual risk– 10-year probability

• Estimate from cohort studies– Nurses’ Health Study (NHS)– Health Professionals’ Follow-up Study (HPFS)

NHS & HPFS Data Multivariate logistic regression of NHS/HPFS data

provide information about the relationship between risk factors and diagnosed (but not underlying) CRC

Diagnosis freeDetected

CRC

Aggregate CRC risk function

Stage-Specific Risk Functions

Goal: decompose the aggregate function into stage-specific risk functions

Disease freeUndetected

CRC

Risk function1

f(age, aspirin use, etc.)

Aggregate CRC risk function

Detected CRC

Adenoma

Risk function2

f(age, activity, etc.)

• Establish “observed relationship” between risk factor and diagnosed CRC

• Simulate incidence of CRC in hypothetical cohort that is matched to study cohort

• Use regression analysis to examine simulated relationship between risk factor and diagnosed CRC

• Calibrate ORs of simulated data analysis to those of cohort analysis

Our Approach

Example: 50 yo white woman

BMI = 25 kg/m2

Non-smoker

MV user

5 met-hr/wk

2 sv/wk red meat

5 sv/dy fruit/veg

No aspirin use

No HRT use

Lifetime CRC risk:

4.8%

Example: 50 yo white woman BMI = 35 kg/m2

Smoker No MV use

5 met-hr/wk

2 sv/wk red meat

5 sv/dy fruit/veg

No aspirin use

No HRT use

Lifetime CRC risk:

9.7%

CISNET Model

CRCModel

Risk factortrends

Screeningbehavior

Diffusion ofnew treatments

Calendar Time

1970 1975 1980 1985 1990

CRCincidence &

mortality

Age-standardized incidence

US Population

0

20

40

60

80

100

120

1975 1980 1985 1990 1995 2000 2005

Year

CR

C C

ase

s p

er

100,0

00

Age-standardized incidence

US population

Model population

0

20

40

60

80

100

120

1975 1980 1985 1990 1995 2000 2005

Year

CR

C C

ase

s p

er

10

0,0

00

Age-standardized incidence

Flat trends since 1970

8574

Risk factor trends

0

20

40

60

80

100

120

1975 1980 1985 1990 1995 2000 2005

Year

CR

C C

ase

s p

er

10

0,0

00

Age-standardized incidence

Healthy weight in 1970

7174

0

20

40

60

80

100

120

1975 1980 1985 1990 1995 2000 2005

Year

CR

C C

ase

s p

er

10

0,0

00

Age-standardized incidence

No smoking in 19706374

0

20

40

60

80

100

120

1975 1980 1985 1990 1995 2000 2005

Year

CR

C C

ase

s p

er

10

0,0

00

Age-standardized incidence

All MV users in 197056

74

0

20

40

60

80

100

120

1975 1980 1985 1990 1995 2000 2005

Year

CR

C C

ase

s p

er

10

0,0

00

Age-standardized incidence

Worst case

Best case27

185

74

0

20

40

60

80

100

120

140

160

180

200

1975 1980 1985 1990 1995 2000 2005

Year

CR

C C

ase

s p

er

10

0,0

00

US Population

Age-standardized mortality

0

10

20

30

40

50

60

1975 1980 1985 1990 1995 2000 2005

Year

CR

C D

eath

s p

er

10

0,0

00

US population

Model population

Age-standardized mortality

0

10

20

30

40

50

60

1975 1980 1985 1990 1995 2000 2005

Year

CR

C D

eath

s p

er

10

0,0

00

Age-standardized mortality

Risk factor trends34

Flat trends since 1970

39

0

10

20

30

40

50

60

1975 1980 1985 1990 1995 2000 2005

Year

CR

C D

eath

s p

er

10

0,0

00

Age-standardized mortality

Worst case

Best case14

79

34

0

20

40

60

80

1975 1980 1985 1990 1995 2000 2005

Year

CR

C D

eath

s p

er

10

0,0

00

Concluding Remarks

• Trends in risk factors over the past 35 years account for a 13% decrease in both CRC incidence and mortality compared to “flat trends”

• Population-based simulation models provide an important tool for evaluating the impact of changing risk factors