Introduction to Cancer Epidemiology Epidemiology and Molecular Pathology of Cancer: Bootcamp course Tuesday, 3 January 2012.

Introduction to Cancer Epidemiology

Epidemiology and Molecular Pathology of Cancer: Bootcamp course

Tuesday, 3 January 2012

Learning Objectives

To define causality in epidemiological research

To clarify causal association vs. statistical association

To give an introduction to study designs in epidemiology

To present patterns of global burden of cancer

Causality and causal inference

Causation

Definition: A cause of disease is an event, condition or characteristic that preceded the disease and without which the disease would not have occurred, or would not have occurred at that time.

Rothman and Greenland, Am J Public Health 2005

Causation A disease can be caused by more than one causal

mechanism

Causal mechanism involves joint actions of component causes

Necessary, sufficient and component causes: Smoking and lung cancer

Rothman and Greenland, Am J Public Health 2005

smokingGenetic Suscept-ibility

smokingPassive smoke

genetics

Air pollu-tion

genetics

gender

Causation

Most causes are neither necessary nor sufficient

However, elimination of the cause may result in elimination of substantial proportion of disease

Estimating causal associations is paramount in epidemiological research and a prerequisite for prevention

Rothman and Greenland, Am J Public Health 2005

Causation and Causal inference Follow group of exposed

(smoking) individuals over time and observe outcome

What would have happened to same group if not been exposed to smoking?

If the two outcomes differ causal effect

If the two outcomes same no causal effect

“Time Machine”

Ysmoking YNo smoking

Michael 1 0

Jennifer 1 0

Linda 0 0

Jeremy 0 0

Axel 1 0

Sophia 1 1

Elisa 1 0

Hernan, J Epidemiol Comm Health 2004

Counterfactual outcomes of lung cancer among individuals

Causal inference Causal inference – scientific reasoning that allows one to arrive

at the conclusion that something is or is likely to be cause

Goal in epidemiology is to approximate counterfactual to estimate causal effects of exposure on disease risk

Study design to approximate the counterfactual

Unexposed group should be a proxy of counterfactual experience for the exposed group

Causal versus Statistical associations

Causal vs. statistical association

Statistical associations are what we measure in epidemiological study or randomized trial Relative measures: Odds ratios, rate ratios, hazard ratios Absolute measures: Risk difference, rate difference,

number needed to treat/screen

“Men who drink coffee regularly have a 60 percent lower risk of lethal prostate cancer compared to men who don’t drink coffee”

Measures of association

Statistical associations Provide estimate of the size of the association

E.g. Compared to healthy weight individuals, does obesity influence risk of postmenopausal breast cancer by a little or alot?

Informs direction of the effect Does the exposure increase risk of disease or

decrease risk of disease compared to not being exposed

E.g. Compared to nonusers, individuals who take aspirin are at lower risk of colorectal cancer

Aim to approximate causal associations 0

0.2

0.4

0.6

0.8

1

1.2

1.4

<1 per month 1-3 per month 1-2 per week >2 weeks

Relative risk

Total prostate cancer

Advanced cancer

REF

NULL

Lower risk

Increased risk

Causal vs. statistical association

Statistical associations can arise: E D Causal association

D E Reverse causation/recall bias

C E D Confounding

Statistical associations can also arise due to misclassification, missing data, selection bias

Confounding example

Physical activity

Lung cancer?

Smoking

+_

Women with vigorous physical activity had 80 percent lower risk of lung cancer compared to women who did not exercise

Confounding

• In epidemiology, nonrandom allocation of the exposure

• It is a mixing of effects. Association between exposure and disease is distorted because it is mixed with the effect of another factor.

• The result of confounding is to distort the true causal association between an exposure and disease

• The direction of the distortion can be either toward the null or away from the null.

• Extent of confounding depends on pr[C], RR[D]|[C], RR[E]|C

95% Confidence Intervals

Range of plausible values consistent with data

If no bias or confounding

Surround the measure of association (point estimate)

Relative risk = 2.5, 95% Confidence Interval = 1.7 – 3.9

Size of confidence interval is based on size of cohort, number of outcomes, and prevalence of exposure

95% Confidence Intervals Relative risk = 2.5, 95% Confidence Interval = 1.7 – 4.1

UPPER 95% CI

Study design

Randomized Studies Investigator randomly assigns who gets exposure

Cannot directly observe individual effects Compare outcomes in exposed vs. unexposed

Placebo group is proxy for what would have happened to statin group if not exposed to statins

10,000 people

Statin N=5,000

Cancer? Cancer?

Statin N=5,000T ime

Cohort studies

Analagous to the experiment, but investigator does not assign exposure

Cohorts are groups of individuals followed over time

Cohorts are longitudinal and outcome assessed over time

E

Ē

Intervention study

E

Ē

Cohort study

Cohort study

Person

1

2

3

4

5

6

7

8

ExposedUnexposedCase

Time

Cohort = group defined by membership defining event

Once a member, always a member until death

Once defined and follow-up begins, no one is added

Cohort Study

Person

1

2

3

4

5

6

7

8

Exposed

Unexposed

Case

Time

Most exposures vary over time

Weight, diet, smoking, infections, blood pressure

Case control studies

An efficient and valid alternative to cohort study Case-control study attempts to observe a population more

efficiently Efficiency comes from use of control series in place of

complete assessment of cohort experience

Case control studies

Identify and enroll cases

Determine “cohort” that gave rise to cases

Cases give information about numerators of rates that would have calculated in cohort

Controls should be selected from the same cohort

Controls should estimate exposure in the population from where the cases came

Case Control: Risk set sampling

Time

Person

1

2

3

4

5

6

7

8

9

10

Exposed

Unexposed

Case

Case Control: Risk set sampling

Time

Person

1

2

3

4

5

6

7

8

9

10

Exposed

Unexposed

Case

1992 1994 1996 1998 2000 2002

Person

1

2

3

4

5

6

7

8

9

10

11

Exposed

Unexposed

Cancer case

Case Control: Case-cohort sampling

1992 1994 1996 1998 2000 2002

Person

1

2

3

4

5

6

7

8

9

10

11

Exposed

Unexposed

Cancer case

Case Control: Case-cohort sampling

1992 1994 1996 1998 2000 2002

Person

1

2

3

4

5

6

7

8

9

10

11

ExposedUnexposedCancer case

Cases

Case Control: “Traditional” sampling

Global burden of cancer

Global burden of cancer (2008)

12 million new cases of cancer worldwide

7.6 million cancer deaths 3rd leading cause of death annually

24.6 million persons alive with cancer (within 5 years of diagnosis)

In 2030 26 million new cases, 17 million deaths: WHY?

Global burden of cancer

Thun et al, 2010

US burden of cancer (2011): 303 million

Estimated 1,596,000 million people will be diagnosed with cancer in 2011 822,000 men; 744,000 women

Estimated 572,000 will die of cancer in 2011 (300,000 men; 272,000 women) 5-year relative survival 68%

American Cancer Society, 2011

Data from the International Agency For Research on Cancer (IARC) website

www.iarc.fr Cancer Epidemiology databases, Globocan 2002 and Cancer Incidence in Five

Countries (CI-VIII, IX)

What are the major cancers among men and women?

Cancer Incidence in the World (Men)

Cancer Incidence in the World (Women)

Cancer incidence rates in Africa, 2002

0 25 50 75 100 125 150

Kaposi sarcoma

Liver

Prostate

Esophagus

NHL

Stomach

Colon/rectum

Bladder

Cervix

Breast

Ovary

All sites

ASR per 100,000

Females

Males

Cervix UteriAge-standardized incidence rate per 100,000

Stomach, MalesAge-Standardized incidence rate per 100,000

Stomach, FemalesAge-standardized incidence rate per 100,000

Testis,Age-Standardized incidence rate per 100,000

Trends in Stomach Cancer Mortality over time: Men