DAY 4 Considerations for Interpretation Confounding 1.

DAY 4

Considerations for Interpretation

Confounding

1

Introduction

Understanding what the data can tell us and cannot tell us

2

3

Approach to understanding the data

If we see an association, what other possible explanations can there be besides causation?

- Confounding

Confound

“To confuse”

4

Webster’s Online Dictionary www.webster-dictionary.org

“To mistake for another”

5

Guilt by association

Not Observed

James CauseXObservedBroken Window

George

Guilt by association

Scenario 1 – Individual level

Hugo

Lennie

HugoNot Observed

Empty Beer Bottles

Lennie Cause

Association Cause

XObserved

Scenario 1 – Individual level

Scenario 2 – Population level

Try Alcohol Not Try Alcohol

6Yes R-Rated

Movies

No R-Rated Movies

90

b

d

a

c

PrevalencePrevalence

Ratio

Scenario 2 – Population level

People who _____________ are ______times as likely

to __________ as people who ____________.

Trying Alcohol Watching

R-Rated Movies

“The study establishes a clear progression that

begins with watching R-rated

movies and leads to teenagers trying

alcohol.”

“If we can keep our teenagers from

watching R-rated movies, then we

can go a long way towards preventing

teenagers from trying alcohol.”

Cause

Scenario 2 – Population level

?

Trying Alcohol Watching

R-Rated Movies

Cause

Association Cause

X

Scenario 2 – Population level

13

A situation in which an association between an exposure and an outcome is observed as a result of the influence of a third, unobserved, factor, called a confounder.

The true effect of an exposure on an outcome is distorted because it is mixed up with another factor that can influence the outcome.

The confounder is both associated with the observed exposure and is a possible cause of the outcome.

Confounding

Unobserved Exposure

Try alcoholWatch

R-rated movies

Limited Adult

Supervision

Watch out for lurking variables!

Unobserved Exposure

Try alcoholWatch R-rated movies

Willingness to

Break the Rules

Unobserved Exposure

Try alcoholWatch R-rated movies

Poor Self-Control

Unobserved Exposure

Try alcoholWatch R-rated movies

Parental attitudes

Unobserved Exposure

Try alcoholWatch R-rated movies

Risk taker

Lack of Supervision or Poor Self- Control or Risk taking or . . .

Try alcoholWatch

R-rated moviesX

… and you avoided or eliminated the hypothesized cause, what would happen to the outcome?

?

If the association was due to confounding …

Cause

Association Cause

X

Confounder

Playing Volleyball

Skin Cancer

Cause

Association Cause

X

Confounder

Drinking Alcohol

Lung Cancer

Cause

Association Cause

X

Confounder

Eating Pretzels

Auto Accidents

Cause

Association Cause

X

Confounder

Watching TV Acne

Cause

Association Cause

X

Confounder

Drinking Coffee

Cancer

Cause

Association Cause

X

Confounder

Driving Motorcycles

Hepatitis C

Cause

Association Cause

X

Confounder

Playing Baseball

Oral Cancer

Cause

Association Cause

X

Confounder

Sand Skin

Cancer

Association is not necessarily causation.

Association is not necessarily causation.

Association is not necessarily causation.

Association is not necessarily causation.

Association is not necessarily causation.

Association is not necessarily causation.

Association is not necessarily causation.

Association is not necessarily causation.

Always Remember

Supplementary Lesson about Confounding

The confounder is associated with the observed exposure (but not affected by the exposure).

The confounder is a possible cause of the outcome.

Two criteria for Confounding

Possible Confounder

OutcomeExposure

31

• We suspect a factor to be a potential confounder based on intuition and evidence that it is associated with the exposure under study and a cause of the outcome under study

• But is this true in the data at hand?

• We can assess this by calculating the strength of association between: 1) exposure and the suspected confounder; and 2) between suspected confounder and outcome.

Possible Confounder

OutcomeExposure

PR = ?PR = ?

PR = ?

32

400

1000

40 %400 600 1000 a b

c d

Exercise less

Exercise more

TotalIncreased BP

No increased BP Prevalence

200

1000

20 %200 800 1000

Prevalence Ratio

2.0

÷ a a+b

c c + d

People who exercise less are more likely to have high blood pressure.

People who exercise less are 2.0 times as likely to have

increased blood pressure as those who exercise more.

2000

Cause

Association Cause

X

Confounder?

Exercise less

Increased blood pressure

Smoking

Increased blood

pressure

Exercise Less

PR = ?PR = ?

PR = 2.0

Assessment of smoking as a potential confounder

See if it meets two criteria for confounding

1. Associated

with

the exposure?

2. A possible cause

of the outcome?

35

450

1000

45.0 %450 550 1000 a b

c d

Less exercise

More exercise

TotalSmokersNon-smokers Prevalence

175

1000

17.5 %175 825 1000

Prevalence Ratio

2.6

People who exercise less are 2.6 times as likely to be

smokers as those who exercise more.

2000

1. Is the potential confounder associated with the exposure?

(Is smoking associated with exercise?)

Smoking

Increased Blood

Pressure

Exercise Less

PR = 2.6PR = ?

PR = 2.0

37

475

625

76.0 %475 150 625 a b

c d

Smokers

Non-smokers

TotalIncreased BP

No increased BP Prevalence

125

1250

10.0 %125 1250 1375

Prevalence Ratio

7.6

Smokers are 7.6 times as likely to have increased blood

pressure compared to non-smokers.

2000

2. Is the potential confounder a possible cause of the outcome?

(Is smoking a possible cause of increased blood pressure?)

Smoking

Increased Blood

Pressure

Exercise Less

PR = 2.6PR = 7.6

PR = 2.0

Is there Possible Confounding?

39

Marijuana Use and Getting Tattoos

People who ____________________________________________

are ______ times as likely to _______________________________

compared to people who __________________________________ 40

50

200

25 %50 150 200 a b

c d

MarijuanaUser

Not a marijuana User

TotalTattoos

No Tattoos Prevalence

30

200

15 %30 170 200

Prevalence Ratio

1.7

÷ a a+b

c c + d

People who smoke marijuana are more likely to get tattoos than people who do not smoke marijuana.

People who use marijuana are 1.7 times as likely to get tattooed compared to people who do not use marijuana. 41

50

200

25 %50 150 200 a b

c d

MarijuanaUser

Not a marijuana User

TotalTattoos

No Tattoos Prevalence

30

200

15 %30 170 200

Prevalence Ratio

1.7

People who smoke marijuana are more likely to get tattoos than people who do not smoke marijuana.

Cause

Association Cause

X

Confounder?

Use Marijuana

Get Tattoos

Access to Money

Get TattoosUse Marijuana

PR = ?PR = ?

PR = 1.7

Assessment of “access to money” as a potential confounder

See if it meets two criteria for confounding

1. Associated

with

the exposure?

2. A possible cause

of the outcome?

44

150

200

75.0 %150 50 200 a b

c d

Total Prevalence

50

200

25.0 %50 150 200

Prevalence Ratio

3.0

People who use marijuana are 3.0 times as likely to have

access to money than non-users.

400

MarijuanaUser

Not a marijuana User

Access to Money

No Access to Money

1. Is the potential confounder associated with the exposure?

(Is access to money associated with marijuana use?)

Access to Money

Get TattoosUse Marijuana

PR = 3.0PR = ?

PR = 1.7

46

60

140 42.8%

60 140 200 a b

c d

Total Prevalence

20

180

11.1 %20 180 200

Prevalence Ratio

3.9

People with access to money are 3.9 times as likely to get

tattoos than people without access to money.

400

Access to Money

No Access to Money

Tattoos No Tattoos

2. Is the potential confounder a possible cause of the outcome?

(Is access to money a possible cause of getting tattoos?

Access to Money

Get TattoosUse Marijuana

PR = 3.0PR = 3.9

PR = 1.7

Is there Possible Confounding?

Possible Confounder

OutcomeExposure

PR = 0.97 PR = 1.10

PR = 1.8

Associated

with th

e exposure?

A possible cause

of the outcome?

Another Example - Is there Possible Confounding?

Day 4

First Morning Breakout

Assessing Possibility of Confounding

49

50

Use the 54 cards in the deck to explore the potential for confounding in this data set

Deck Worksheet – pages 3-5

Less Frequent Breakfast

High Physical Activity

High SleepPR = ?

PR = ? PR = ?Associa

ted

with th

e exposure?

A possible cause

of the outcome?

Assess Possibility of Confounding by Frequency of Breakfast

Assessing confounding in your study

Assembly to brainstorm ideas . . .

52

Day 4

Second Morning Breakout

Possible confounders you should examine?

53

Possible Confounder(s)?

Your Outcome

Your Exposure

Discuss possible confounding in your study

Label your possible confounding triangle on large post-it

Breakout Deliverable for Assembly/Gallery

walk

55

Large post-it with your team’s confounding triangle

(During gallery walk after the break, use small post-its to provide constructive comments and

suggestions on other teams’ triangles).

56

Study Proposal: Section 7a

7. Considerations for Interpretation

7a. Confounding

Describe how a confounding variable could influence your hypothesized association

Develop one or two additional study questions to help assess possible confounding

If you have time, start this . . .

57

Breakout Session Tips

• Create one or two questions for your survey that will assess possible confounding.

• Your question(s) must be school appropriate

• Limit your answer options to two answers, such as Yes/No, Male/Female, or other 2-choice answers.

Anything else?

58

Reminder

• It should be written so that its intended audience understands it.

• It should mean the same thing to everyone who reads it.

• The answer options must categorize and cover the entire range of possible behavior

• The time period to consider when answering a particular question must be appropriate (long or short enough)

Criteria for a good survey question