Epidemiology quiz

Measures of Association II

Jennifer Deal, PhD Johns Hopkins University

The material in this video is subject to the copyright of the owners of the material and is being provided for educational purposes under rules of fair use for registered students in this course only. No additional copies of the copyrighted work may be made or distributed.

Review the Calculation of the OR in Unmatched Case-Control Studies

3

Relative Measures of Association

► Relative risk

► Odds ratio ► Unmatched case-control study ► Matched case-control study

4

Study of 10 Cases and 10 Unmatched Controls

► Table key ► E = exposed ► N = non-exposed

Cases Controls E N E E N N E N N E N N E N E E E N N N

Source: Gordis (6th ed.).

5

Set Up a 2x2 Table—1

► When exposure status is binary:

6

Set Up a 2x2 Table—2

► When exposure status is binary:

► Be alert! Sometimes the rows and columns are exchanged!

7

2xn Table ► When exposure status has several categories:

8

Example: Multiple Exposure Categories

► Source: Deal, J. A., Betz, J., Yaffe, K., et al. (2017). Hearing impairment and incident dementia and cognitive decline in older adults: The Health ABC Study. J Gerontol A Biol Sci Med Sci, 72(5), 703–709. http://dx.doi.org/10.1093/gerona/glw069

9

Study of 10 Cases and 10 Unmatched Controls—1

Cases Controls

E N

E E

N N

E N

N E

N N

E N

E E

E N

N N

10

Study of 10 Cases and 10 Unmatched Controls—2

Cases Controls

E N

E E

N N

E N

N E

N N

E N

E E

E N

N N

► 6 cases were exposed

► 3 controls were exposed

11

Study of 10 Cases and 10 Unmatched Controls—3

Cases Controls

E N

E E

N N

E N

N E

N N

E N

E E

E N

N N

► 4 cases were unexposed

► 7 controls were unexposed

12

Recall… Odds Ratio in a Case- Control Study

► 𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 = 𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂 𝑜𝑜𝑜𝑜 𝑎𝑎 ℎ𝑖𝑖𝑂𝑂𝑖𝑖𝑜𝑜𝑖𝑖𝑖𝑖 𝑜𝑜𝑜𝑜 𝑒𝑒𝑒𝑒𝑒𝑒𝑜𝑜𝑂𝑂𝑒𝑒𝑖𝑖𝑒𝑒 𝑖𝑖𝑖𝑖 𝑖𝑖ℎ𝑒𝑒 𝐶𝐶𝑎𝑎𝑂𝑂𝑒𝑒𝑂𝑂 𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂 𝑜𝑜𝑜𝑜 𝑎𝑎 ℎ𝑖𝑖𝑂𝑂𝑖𝑖𝑜𝑜𝑖𝑖𝑖𝑖 𝑜𝑜𝑜𝑜 𝑒𝑒𝑒𝑒𝑒𝑒𝑜𝑜𝑂𝑂𝑒𝑒𝑖𝑖𝑒𝑒 𝑖𝑖𝑖𝑖 𝑖𝑖ℎ𝑒𝑒 𝐶𝐶𝑜𝑜𝑖𝑖𝑖𝑖𝑖𝑖𝑜𝑜𝐶𝐶𝑂𝑂

► 𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 = 𝑎𝑎 𝑐𝑐 𝑏𝑏 𝑑𝑑

= 𝑎𝑎×𝑂𝑂 𝑏𝑏×𝑐𝑐

Turns out same in a cohort study

13

Study of 10 Cases and 10 Unmatched Controls—4

Cases Controls

E N

E E

N N

E N

N E

N N

E N

E E

E N

N N

► 𝑂𝑂𝑂𝑂 = 𝑎𝑎𝑂𝑂 𝑏𝑏𝑐𝑐

= 6×7 3×4

= 3.5

The material in this video is subject to the copyright of the owners of the material and is being provided for educational purposes under rules of fair use for registered students in this course only. No additional copies of the copyrighted work may be made or distributed.

Calculating the OR in Matched Case-Control Studies

2

Relative Measures of Association

► Relative risk

► Odds ratio ► Unmatched case-control study ► Matched case-control study

3

Recall… (1)

4

Recall… (2)

► What is the association between the exposure and the outcome independent of these factors?

5

Matched Case-Control Studies

► How can we calculate the OR in a case-control study in which the cases and controls have been matched on one or more factors?

► Matching puts cases and controls into pairs

6

Possible Types of Case-Control Pairs—1

► With respect to their exposure status: ► Concordant pairs

1. Pairs in which both the case and control were exposed 2. Pairs in which neither the case nor the control was exposed

7

Possible Types of Case-Control Pairs—2

► With respect to their exposure status: ► Concordant pairs

1. Pairs in which both the case and control were exposed 2. Pairs in which neither the case nor the control was exposed

► Discordant pairs 3. Pairs in which the case was exposed, but not the control 4. Pairs in which the control was exposed, but the case was not

8

Calculation of the OR for Matched Pairs

► ORmatched =

► 𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁 𝑜𝑜𝑜𝑜 𝒑𝒑𝒑𝒑𝒑𝒑𝒑𝒑𝒑𝒑 𝑖𝑖𝑖𝑖 𝑤𝑤𝑤𝑖𝑖𝑤𝑤𝑤 𝑡𝑡𝑤𝑁𝑁 𝑤𝑤𝑐𝑐𝑐𝑐𝑁𝑁 𝑤𝑤𝑐𝑐𝑐𝑐 𝑁𝑁𝑒𝑒𝑒𝑒𝑜𝑜𝑐𝑐𝑁𝑁𝑒𝑒 𝑐𝑐𝑖𝑖𝑒𝑒 𝑡𝑡𝑤𝑁𝑁 𝑤𝑤𝑜𝑜𝑖𝑖𝑡𝑡𝑁𝑁𝑜𝑜𝑐𝑐 𝑤𝑤𝑐𝑐𝑐𝑐 𝑖𝑖𝑜𝑜𝑡𝑡 𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁 𝑜𝑜𝑜𝑜 𝒑𝒑𝒑𝒑𝒑𝒑𝒑𝒑𝒑𝒑 𝑖𝑖𝑖𝑖 𝑤𝑤𝑤𝑖𝑖𝑤𝑤𝑤 𝑡𝑡𝑤𝑁𝑁 𝑤𝑤𝑜𝑜𝑖𝑖𝑡𝑡𝑁𝑁𝑜𝑜𝑐𝑐 𝑤𝑤𝑐𝑐𝑐𝑐 𝑁𝑁𝑒𝑒𝑒𝑒𝑜𝑜𝑐𝑐𝑁𝑁𝑒𝑒 𝑐𝑐𝑖𝑖𝑒𝑒 𝑡𝑡𝑤𝑁𝑁 𝑤𝑤𝑐𝑐𝑐𝑐𝑁𝑁 𝑤𝑤𝑐𝑐𝑐𝑐 𝑖𝑖𝑜𝑜𝑡𝑡

9

Study of 10 Cases, Each with 1 Matched Control

20 total participants; 10 case-control pairs:

Pair Cases ↔ Controls

1 E ↔ N

2 E ↔ E

3 N ↔ N

4 E ↔ N

5 N ↔ E

6 N ↔ N

7 E ↔ N

8 E ↔ E

9 E ↔ N

10 N ↔ N

E = exposed; N = non-exposed

10

Count Concordant and Discordant Pairs—1

Cases ↔ Controls

E ↔ N

E ↔ E

N ↔ N

E ↔ N

N ↔ E

N ↔ N

E ↔ N

E ↔ E

E ↔ N

N ↔ N

► Concordant pairs ► 2 pairs—both case and control

exposed ►

► Discordant pairs ►

►

11

Count Concordant and Discordant Pairs—2

Cases ↔ Controls

E ↔ N

E ↔ E

N ↔ N

E ↔ N

N ↔ E

N ↔ N

E ↔ N

E ↔ E

E ↔ N

N ↔ N

► Concordant pairs ► 2 pairs—both case and control

exposed ► 3 pairs—neither case nor

control exposed

► Discordant pairs ►

►

12

Count Concordant and Discordant Pairs—3

Cases ↔ Controls

E ↔ N

E ↔ E

N ↔ N

E ↔ N

N ↔ E

N ↔ N

E ↔ N

E ↔ E

E ↔ N

N ↔ N

► Concordant pairs ► 2 pairs—both case and control

exposed ► 3 pairs—neither case nor

control exposed

► Discordant pairs ► 4 pairs—case exposed and

control was not ►

13

Count Concordant and Discordant Pairs—4

Cases ↔ Controls

E ↔ N

E ↔ E

N ↔ N

E ↔ N

N ↔ E

N ↔ N

E ↔ N

E ↔ E

E ↔ N

N ↔ N

► Concordant pairs ► 2 pairs—both case and control

exposed ► 3 pairs—neither case nor

control exposed

► Discordant pairs ► 4 pairs—case exposed and

control was not ► 1 pair—control exposed and

case was not

14

Set Up the 2x2 Table

2x2 table for a matched case-control study:

15

Fill in the 2x2 Table—1

► Concordant pairs ► 2 pairs—both case and control

exposed ► 3 pairs—neither case nor control

exposed

16

Fill in the 2x2 Table—2

► Concordant pairs ► 2 pairs—both case and control

exposed ► 3 pairs—neither case nor control

exposed

► Discordant pairs ► 4 pairs—case exposed and control was

not ► 1 pair—control exposed and case was

not

17

Calculate the OR for a Matched Case- Control Study— 1

► In a matched pairs study, we calculate the OR as the ratio of the discordant pairs

► 𝑶𝑶𝑶𝑶𝑶𝑶𝒑𝒑 𝒑𝒑𝒑𝒑𝒓𝒓𝒑𝒑𝒓𝒓 (𝒎𝒎𝒑𝒑𝒓𝒓𝒎𝒎𝒎𝒎𝒎𝒎𝑶𝑶

𝒑𝒑𝒑𝒑𝒑𝒑𝒑𝒑𝒑𝒑) = 𝑏𝑏 𝑐𝑐

Discordant pairs

18

Calculate the OR for a Matched Case- Control Study— 2

19

Calculate the OR for a Matched Case- Control Study— 3

► 𝑶𝑶𝑶𝑶 𝒎𝒎𝒑𝒑𝒓𝒓𝒎𝒎𝒎𝒎𝒎𝒎𝑶𝑶 = 𝑁𝑁 𝑤𝑤

= 4 1

= 4

20

Calculate the OR for a Matched Case- Control Study— 4

► 𝑶𝑶𝑶𝑶 𝒎𝒎𝒑𝒑𝒓𝒓𝒎𝒎𝒎𝒎𝒎𝒎𝑶𝑶 = 𝑁𝑁 𝑤𝑤

= 4 1

= 4

► Interpretation is the same as the OR calculated in an unmatched case-control study: ► Exposed persons have 4 times the odds of the disease

compared with unexposed persons

The material in this video is subject to the copyright of the owners of the material and is being provided for educational purposes under rules of fair use for registered students in this course only. No additional copies of the copyrighted work may be made or distributed.

When Does the OR ≈ RR?

2

Recall… Important Facts about the OR—1

3

Recall… Important Facts about the OR—2

4

Recall… Important Facts about the OR—3

► The odds ratio (OR) is therefore the same in a case-control study and in a cohort study

5

Recall… Important Facts about the OR—4

► The odds ratio (OR) is the same in a case-control study as in a cohort study

► That means a case-control study may provide similar insight into an association as a cohort study

6

Recall… Important Facts about the OR—5

► The odds ratio (OR) is the same in a case-control study as in a cohort study

► That means a case-control study may provide similar insight into an association as a cohort study ► Under some conditions, the OR calculated in a case-control study may be a reasonable

estimate of the RR that would have been calculated in a cohort study

7

OR vs. RR

► While both are measures of association, the calculations of the OR and the RR are different

► 𝑅𝑅𝑅𝑅 = 𝑎𝑎

𝑎𝑎+𝑏𝑏 𝑐𝑐

𝑐𝑐+𝑑𝑑

► 𝑂𝑂𝑅𝑅 = 𝑎𝑎𝑎𝑎 𝑏𝑏𝑏𝑏

8

Recall… Important Facts about the OR—6

► The odds ratio (OR) is therefore the same in a case-control study and in a cohort study

► That means a case-control study may provide similar insight into an association as a cohort study

► Under some conditions, the OR calculated in a case-control study may be a reasonable estimate of the RR that would have been calculated in a cohort study

► What conditions?

9

When Does the OR Approximate the RR?

► Cases are representative, with regard to history of exposure, of all people with the disease in the source population

► Controls are representative, with regard to history of exposure, of all people without disease in the source population

► Risk of disease is low—“Rare disease assumption”

10

When the Risk of Disease Is Low

► a + b ≅ b

► c + d ≅ d

11

When the Risk of Disease Is Low—1

►𝑅𝑅𝑅𝑅 = 𝑎𝑎

𝑎𝑎+𝒃𝒃 𝑐𝑐

𝑐𝑐+𝒅𝒅 ≅

𝑎𝑎 𝑏𝑏 𝑐𝑐 𝑑𝑑

12

When the Risk of Disease Is Low—2

►𝑅𝑅𝑅𝑅 = 𝑎𝑎

𝑎𝑎+𝒃𝒃 𝑐𝑐

𝑐𝑐+𝒅𝒅 ≅

𝑎𝑎 𝑏𝑏 𝑐𝑐 𝑑𝑑

= 𝑎𝑎×𝑎𝑎 𝑏𝑏×𝑏𝑏

= 𝑂𝑂𝑅𝑅!

13

Example 1: Disease Risk Is Low—1

► Disease risk is low: ► 100 ÷ 10,000 = 1%

14

Example 1: Disease Risk Is Low—2

► Disease risk is low: ► 100 ÷ 10,000 = 1%

► 𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹 𝒓𝒓𝑹𝑹𝒓𝒓𝒓𝒓 = 200

10,000 100

10,000 = 2.00

15

Example 1: Disease Risk Is Low—3

► Disease risk is low: ► 100 ÷ 10,000 = 1%

► 𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹 𝒓𝒓𝑹𝑹𝒓𝒓𝒓𝒓 = 200 ÷10,000 100 ÷10,000

= 2.00

► 𝑶𝑶𝒅𝒅𝒅𝒅𝒓𝒓 𝒓𝒓𝑹𝑹𝑹𝑹𝑹𝑹𝒓𝒓 = 200 × 9,900 9,800 × 100

= 2.02

16

Example 1: Disease Risk Is Low—4

► Disease risk is low: ► 100 ÷ 10,000 = 1%

► 𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹 𝒓𝒓𝑹𝑹𝒓𝒓𝒓𝒓 = 200 ÷10,000 100 ÷10,000

= 2.00

► 𝑶𝑶𝒅𝒅𝒅𝒅𝒓𝒓 𝒓𝒓𝑹𝑹𝑹𝑹𝑹𝑹𝒓𝒓 = 200 × 9,900 9,800 × 100

= 2.02

► OR ≈ RR

17

Example 2: Disease Risk Is High—1

► Disease risk is high: ► 25 ÷ 100 = 25%

18

Example 2: Disease Risk Is High—2

► Disease risk is high: ► 25 ÷ 100 = 25%

► 𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹 𝒓𝒓𝑹𝑹𝒓𝒓𝒓𝒓 = 50 ÷100 25 ÷100

= 2.00

19

Example 2: Disease Risk Is High—3

► Disease risk is high: ► 25 ÷ 100 = 25%

► 𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹 𝒓𝒓𝑹𝑹𝒓𝒓𝒓𝒓 = 50 ÷100 25 ÷100

= 2.00

► 𝑶𝑶𝒅𝒅𝒅𝒅𝒓𝒓 𝒓𝒓𝑹𝑹𝑹𝑹𝑹𝑹𝒓𝒓 = 50 × 75 50 × 25

= 3.00

20

Example 2: Disease Risk Is High—4

► Disease risk is high: ► 25 ÷ 100 = 25%

► 𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹 𝒓𝒓𝑹𝑹𝒓𝒓𝒓𝒓 = 50 ÷100 25 ÷100

= 2.00

► 𝑶𝑶𝒅𝒅𝒅𝒅𝒓𝒓 𝒓𝒓𝑹𝑹𝑹𝑹𝑹𝑹𝒓𝒓 = 50 × 75 50 × 25

= 3.00

► OR ≠ RR

21

Evaluating Rare Disease Assumption

► Can we test this assumption in a case-control study?

► No! Because we cannot directly estimate risk

► So, “rare disease assumption” is evaluated in the source population for the cases

22

The OR Yields a More Extreme Estimate of the RR—1

► Recall odds and P: ► What is the odds of exposure? ► The odds of exposure is the proportion

exposed divided by the proportion not exposed: ● 𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂 = 𝑃𝑃

1−𝑃𝑃 P = the proportion with exposure

23

The OR Yields a More Extreme Estimate of the RR—2

► Recall odds and P: ► What is the odds of exposure? ► The odds of exposure is the proportion

exposed divided by the proportion not exposed: ● 𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂 = 𝑃𝑃

1−𝑃𝑃 P = the proportion with exposure

► 𝑂𝑂𝑅𝑅 = 𝑃𝑃1

1−𝑃𝑃1 𝑃𝑃0

1−𝑃𝑃0

► P1 = probability of disease in the exposed

► P0 = probability of disease in the non- exposed

24

The OR Yields a More Extreme Estimate of the RR—3

► 𝑂𝑂𝑅𝑅 = 𝑃𝑃1

1−𝑃𝑃1 𝑃𝑃0

1−𝑃𝑃0

► P1 = probability of disease in the exposed

► P0 = probability of disease in the non- exposed

► Rearrange:

𝑂𝑂𝑅𝑅 = 𝑃𝑃1 × 1 − 𝑃𝑃0 𝑃𝑃0 × 1 − 𝑃𝑃1

25

Built-in biasRelative risk

The OR Yields a More Extreme Estimate of the RR—4

► 𝑂𝑂𝑅𝑅 = 𝑃𝑃1

1−𝑃𝑃1 𝑃𝑃0

1−𝑃𝑃0

► P1 = probability of disease in the exposed

► P0 = probability of disease in the non- exposed

► Rearrange:

𝑂𝑂𝑅𝑅 = 𝑃𝑃1 × 1 − 𝑃𝑃0 𝑃𝑃0 × 1 − 𝑃𝑃1

26

The OR Yields a More Extreme Estimate of the RR—5

► The OR yields a more extreme estimate of the RR: ► If RR > 1, then OR > RR ► If RR < 1, then OR < RR

27

The OR Estimated in Nested Case-Control Studies

► Recall that a case-control study can be nested in a cohort study

► OR estimated in the nested case-control study will be an unbiased estimate of the RR ► “Rare disease assumption” not necessary

28

An Example Nested Case-Control Study

► Erlinger, T. P., Platz, E. A., Rifai, N., and Helzlsouer, K. J. (2004). C-reactive protein and the risk of incident colorectal cancer. JAMA, 291(5), 585– 590. http://dx.doi.org/10.1001/jama.291.5.585

► A JHSPH cohort: CLUE II

29

An Example Nested Case-Control Study: Study Design

► Study design: “Prospective, nested case-control study…”

Erlinger, T. P., Platz, E. A., Rifai, N., and Helzlsouer, K. J. (2004). C-reactive protein and the risk of incident colorectal cancer. JAMA, 291(5), 585–590. http://dx.doi.org/10.1001/jama.291.5.58

30

An Example Nested Case-Control Study: Cases and Controls

► Cases and controls: “A total of 172 colorectal cancer cases were identified through linkage with the Washington County and Maryland State Cancer registries. Up to 2 controls (n=342) were selected from the cohort for each case and matched by age, sex, race, and date of blood draw…”

Erlinger, T. P., Platz, E. A., Rifai, N., and Helzlsouer, K. J. (2004). C-reactive protein and the risk of incident colorectal cancer. JAMA, 291(5), 585–590. http://dx.doi.org/10.1001/jama.291.5.58

31

An Example Nested Case-Control Study: Measure of Association

► Measure of association “Odds ratio (OR) of incident colon and rectal cancer”

Erlinger, T. P., Platz, E. A., Rifai, N., and Helzlsouer, K. J. (2004). C-reactive protein and the risk of incident colorectal cancer. JAMA, 291(5), 585–590. http://dx.doi.org/10.1001/jama.291.5.58

32

An Example Nested Case-Control Study: Interpretation of the OR in the Context of a Cohort Study

► Interpretation of the OR in the context of a cohort study: “The risk of colon cancer was higher in persons in the highest vs lowest quartile of CRP (OR, 2.55; 95% confidence interval [CI], 1.34–4.88; P for trend = .002). In smokers, the corresponding association was stronger (OR, 3.51; 95% CI, 1.64–7.51; P for trend<.001). …”

Erlinger, T. P., Platz, E. A., Rifai, N., and Helzlsouer, K. J. (2004). C-reactive protein and the risk of incident colorectal cancer. JAMA, 291(5), 585–590. http://dx.doi.org/10.1001/jama.291.5.58

33

Other Details Related to the Odds and ORs

► Do we ever calculate an odds difference? ► By convention, no

► Do we ever calculate the OR in a cohort study? ► Rarely ► Usually calculate risk ratios, rate ratios, or hazard ratios

► Can a rate (rather than risk) be used to calculate an odds? ► No, because a rate is not a probability (P)

The material in this video is subject to the copyright of the owners of the material and is being provided for educational purposes under rules of fair use for registered students in this course only. No additional copies of the copyrighted work may be made or distributed.

Attributable Risk and Population Attributable Risk

2

Excess Risk and Disease Prevention—Recall…

3

Attributable Risk (AR)

► Risk attributable to exposure in the exposed

► (𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝑅𝑅𝑖𝑖 𝑡𝑡𝑡𝑡𝑡 𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆) − (𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝑅𝑅𝑖𝑖 𝑡𝑡𝑡𝑡𝑡 𝑖𝑖𝑛𝑛𝑖𝑖𝑡𝑡𝑛𝑛𝑛𝑛𝑛𝑛𝑅𝑅𝑡𝑡𝑛𝑛)

4

Attributable Risk Percent

► Proportion of risk attributable to exposure in the exposed

► 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝑅𝑅𝑖𝑖 𝑡𝑡𝑡𝑡𝑡 𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆 − 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝑅𝑅𝑖𝑖 𝑡𝑡𝑡𝑡𝑡 𝑖𝑖𝑛𝑛𝑖𝑖𝑡𝑡𝑛𝑛𝑛𝑛𝑛𝑛𝑅𝑅𝑡𝑡𝑛𝑛

𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝑅𝑅𝑖𝑖 𝑡𝑡𝑡𝑡𝑡 𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆 × 100%

5

Smoking and Coronary Heart Disease

► Cohort study—over 1 year of follow-up with no loss to follow-up (closed cohort)

Smoking status Total number of

participants Develop CHD

Do not develop CHD

CHD incidence per 1,000 people per

year

Smoke 3,000 84 2,916 28.0

Does not smoke 5,000 87 4,913 17.4

6

Attributable Risk—1

► Risk of CHD attributable to smoking in smokers

► 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝑅𝑅𝑖𝑖 𝒆𝒆𝒔𝒔𝒆𝒆𝒔𝒔𝒆𝒆𝒔𝒔𝒆𝒆 − 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝑅𝑅𝑖𝑖 𝑖𝑖𝑛𝑛𝑖𝑖𝑅𝑅𝑛𝑛𝑛𝑛𝑅𝑅𝑡𝑡𝑛𝑛𝑅𝑅

► 𝐴𝐴𝑅𝑅 = 28.0 𝑛𝑛𝑡𝑡𝑛𝑛 1,000 − 17.4 𝑛𝑛𝑡𝑡𝑛𝑛 1,000 = 10.6 𝑛𝑛𝑡𝑡𝑛𝑛 1,000 𝑛𝑛𝑡𝑡𝑛𝑛 𝑦𝑦𝑡𝑡𝑦𝑦𝑛𝑛

7

Attributable Risk—2

► Risk of CHD attributable to smoking in smokers

► 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝑅𝑅𝑖𝑖 𝒆𝒆𝒔𝒔𝒆𝒆𝒔𝒔𝒆𝒆𝒔𝒔𝒆𝒆 − 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝑅𝑅𝑖𝑖 𝑖𝑖𝑛𝑛𝑖𝑖𝑅𝑅𝑛𝑛𝑛𝑛𝑅𝑅𝑡𝑡𝑛𝑛𝑅𝑅

► 𝐴𝐴𝑅𝑅 = 28.0 𝑛𝑛𝑡𝑡𝑛𝑛 1,000 − 17.4 𝑛𝑛𝑡𝑡𝑛𝑛 1,000 = 10.6 𝒆𝒆𝒆𝒆𝒔𝒔 𝟏𝟏, 𝟎𝟎𝟎𝟎𝟎𝟎 𝒆𝒆𝒆𝒆𝒔𝒔 𝒚𝒚𝒆𝒆𝒚𝒚𝒔𝒔

► Must specify the units and the time interval!

8

Attributable Risk Percent (AR%)

► Proportion of risk of CHD attributable to smoking in the smokers

► 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝑅𝑅𝑖𝑖 𝒆𝒆𝒔𝒔𝒆𝒆𝒔𝒔𝒆𝒆𝒔𝒔𝒆𝒆 − 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝑅𝑅𝑖𝑖 𝑖𝑖𝑛𝑛𝑖𝑖𝑅𝑅𝑛𝑛𝑛𝑛𝑅𝑅𝑡𝑡𝑛𝑛𝑅𝑅

𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝑅𝑅𝑖𝑖 𝒆𝒆𝒔𝒔𝒆𝒆𝒔𝒔𝒆𝒆𝒔𝒔𝒆𝒆

► 𝐴𝐴𝑅𝑅𝐴 = 28.0−17.4 28.0

= 10.6 28.0

= 0.379 = 37.9%

9

Population Attributable Risk (PAR)—1

► The population attributable risk is similar to the attributable risk, except that it pertains to the entire population (some exposed, some not exposed), not just the exposed

10

Population Attributable Risk (PAR)—2

► The population attributable risk is similar to the attributable risk, except that it pertains to the entire population (some exposed, some not exposed), not just the exposed

► With the AR, we ask: What is the risk in the exposed that is attributable to the exposure? How much risk could be prevented in the exposed if they were not exposed?

11

Population Attributable Risk (PAR)—3

► The population attributable risk is similar to the attributable risk, except that it pertains to the entire population (some exposed, some not exposed), not just the exposed

► With the AR, we ask: What is the risk in the exposed that is attributable to the exposure? How much risk could be prevented in the exposed if they were not exposed?

► With the PAR, we ask: What is the risk in the total population that is attributable to the exposure? How much risk in the total population could be prevented if no one in the population were exposed?

12

Population Attributable Risk (PAR)—4

► Risk attributable to exposure in the total population

► 𝑃𝑃𝐴𝐴𝑅𝑅 = 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝑅𝑅𝑖𝑖 𝒕𝒕𝒆𝒆𝒕𝒕𝒚𝒚𝒕𝒕 𝒆𝒆𝒆𝒆𝒆𝒆𝒑𝒑𝒕𝒕𝒚𝒚𝒕𝒕𝒑𝒑𝒆𝒆𝒑𝒑 − 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝑅𝑅𝑖𝑖 𝑡𝑡𝑡𝑡𝑡 𝒑𝒑𝒆𝒆𝒑𝒑𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆 ► Total population = exposed plus the nonexposed ► Nonexposed is still the same comparison group!

13

Population Attributable Risk Percent (PAR%)

► Proportion of risk attributable to exposure in the total population

► 𝑃𝑃𝐴𝐴𝑅𝑅𝐴 = 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝑅𝑅𝑖𝑖 𝒕𝒕𝒆𝒆𝒕𝒕𝒚𝒚𝒕𝒕 𝒆𝒆𝒆𝒆𝒆𝒆𝒑𝒑𝒕𝒕𝒚𝒚𝒕𝒕𝒑𝒑𝒆𝒆𝒑𝒑 − 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝑅𝑅𝑖𝑖 𝑡𝑡𝑡𝑡𝑡 𝑖𝑖𝑛𝑛𝑖𝑖𝑡𝑡𝑛𝑛𝑛𝑛𝑛𝑛𝑅𝑅𝑡𝑡𝑛𝑛 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝑅𝑅𝑖𝑖 𝒕𝒕𝒆𝒆𝒕𝒕𝒚𝒚𝒕𝒕 𝒆𝒆𝒆𝒆𝒆𝒆𝒑𝒑𝒕𝒕𝒚𝒚𝒕𝒕𝒑𝒑𝒆𝒆𝒑𝒑

14

Interpretation of the PAR%

► Interpretation: ► Proportion of the disease risk in the population that is due to the exposure ► Proportion of the disease risk in the population that could have been avoided if no one

in the population had been exposed

15

Calculating Risk in the Total Population—1

► If the risk in the total population is not given, it can be calculated if we know: 1. The risk among the exposed 2. The risk among non-exposed 3. The proportion of the total population that is exposed

16

Calculating Risk in the Total Population—2

17

Calculating Risk in the Total Group—Back to the Smoking and CHD Risk Example…

► If the risk of CHD in the total group is unknown, it can be calculated if we know: 1. The CHD risk among the smokers 2. The CHD risk among non-smokers 3. The proportion of the total group that smokes

18

Information Needed for the PAR

► From an earlier slide on the cohort study on smoking and CHD risk: ► CHD risk in smokers = 28.0 per 1,000 per year ► CHD risk in non-smokers = 17.4 per 1,000 per year ► The proportion of smokers in the total group is 37.5%:

𝟑𝟑,𝟎𝟎𝟎𝟎𝟎𝟎 𝟑𝟑,𝟎𝟎𝟎𝟎𝟎𝟎+𝟓𝟓,𝟎𝟎𝟎𝟎𝟎𝟎

► Thus: ► The proportion of non-smokers in the total group is 62.5%:

𝟓𝟓,𝟎𝟎𝟎𝟎𝟎𝟎 𝟑𝟑,𝟎𝟎𝟎𝟎𝟎𝟎+𝟓𝟓,𝟎𝟎𝟎𝟎𝟎𝟎

19

Risk in the Total Group

► 28.0 𝑛𝑛𝑡𝑡𝑛𝑛 1,000 × 0.375 + 17.4 𝑛𝑛𝑡𝑡𝑛𝑛 1,000 × 0.625 = 21.375 𝑛𝑛𝑡𝑡𝑛𝑛 1,000 𝑛𝑛𝑡𝑡𝑛𝑛 𝑦𝑦𝑡𝑡𝑦𝑦𝑛𝑛

20

Population Attributable Risk (PAR)

► CHD risk attributable to smoking in the total population (PAR)

► = 𝐶𝐶𝐶𝐶𝐶𝐶 𝑛𝑛𝑅𝑅𝑅𝑅𝑅𝑅 𝑅𝑅𝑖𝑖 𝑡𝑡𝑛𝑛𝑡𝑡𝑦𝑦𝑡𝑡 𝑔𝑔𝑛𝑛𝑛𝑛𝑔𝑔𝑛𝑛 − 𝐶𝐶𝐶𝐶𝐶𝐶 𝑛𝑛𝑅𝑅𝑅𝑅𝑅𝑅 𝑅𝑅𝑖𝑖 𝑡𝑡𝑡𝑡𝑡 𝑖𝑖𝑛𝑛𝑖𝑖𝑅𝑅𝑛𝑛𝑛𝑛𝑅𝑅𝑡𝑡𝑛𝑛𝑅𝑅

► = 21.375 𝑛𝑛𝑡𝑡𝑛𝑛 1,000 𝑛𝑛𝑡𝑡𝑛𝑛 𝑦𝑦𝑡𝑡𝑦𝑦𝑛𝑛 − 17.4 𝑛𝑛𝑡𝑡𝑛𝑛 1,000 𝑛𝑛𝑡𝑡𝑛𝑛 𝑦𝑦𝑡𝑡𝑦𝑦𝑛𝑛

► = 3.975 𝑛𝑛𝑡𝑡𝑛𝑛 1,000 𝑛𝑛𝑡𝑡𝑛𝑛 𝑦𝑦𝑡𝑡𝑦𝑦𝑛𝑛

21

Population Attributable Risk Percent

► Proportion of CHD risk attributable to smoking in the total group (PAR%)

► = 𝐶𝐶𝐶𝐶𝐶𝐶 𝑛𝑛𝑅𝑅𝑅𝑅𝑅𝑅 𝑅𝑅𝑖𝑖 𝑡𝑡𝑛𝑛𝑡𝑡𝑡𝑡𝑡𝑡 𝑔𝑔𝑛𝑛𝑛𝑛𝑔𝑔𝑛𝑛 − 𝐶𝐶𝐶𝐶𝐶𝐶 𝑛𝑛𝑅𝑅𝑅𝑅𝑅𝑅 𝑅𝑅𝑖𝑖 𝑡𝑡𝑡𝑡𝑡 𝑖𝑖𝑛𝑛𝑖𝑖𝑅𝑅𝑛𝑛𝑛𝑛𝑅𝑅𝑡𝑡𝑛𝑛𝑅𝑅 𝐶𝐶𝐶𝐶𝐶𝐶 𝑛𝑛𝑅𝑅𝑅𝑅𝑅𝑅 𝑅𝑅𝑖𝑖 𝑡𝑡𝑛𝑛𝑡𝑡𝑡𝑡𝑡𝑡 𝑔𝑔𝑛𝑛𝑛𝑛𝑔𝑔𝑛𝑛

► = 21.375 𝑛𝑛𝑡𝑡𝑛𝑛 1,000 𝑛𝑛𝑡𝑡𝑛𝑛 𝑦𝑦𝑡𝑡𝑡𝑡𝑛𝑛 − 17.4 𝑛𝑛𝑡𝑡𝑛𝑛 1,000 𝑛𝑛𝑡𝑡𝑛𝑛 𝑦𝑦𝑡𝑡𝑡𝑡𝑛𝑛 21.375 𝑛𝑛𝑡𝑡𝑛𝑛 1,000 𝑛𝑛𝑡𝑡𝑛𝑛 𝑦𝑦𝑡𝑡𝑡𝑡𝑛𝑛

► = 3.975 𝑛𝑛𝑡𝑡𝑛𝑛 1,000 𝑛𝑛𝑡𝑡𝑛𝑛 𝑦𝑦𝑡𝑡𝑡𝑡𝑛𝑛 21.375 𝑛𝑛𝑡𝑡𝑛𝑛 1,000 𝑛𝑛𝑡𝑡𝑛𝑛 𝑦𝑦𝑡𝑡𝑡𝑡𝑛𝑛

= 0.186 = 18.6%

22

Inferring to a Target Population—1

► We calculated the PAR and PAR% for the cohort study assuming that the exposure (smoking) prevalence was not set by the investigators

23

Inferring to a Target Population—2

► We calculated the PAR and PAR% for the cohort study assuming that the exposure (smoking) prevalence was not set by the investigators

► But sometimes investigators pick an equal number of exposed and unexposed individuals ► If so, don’t use that prevalence of exposure in calculating the PAR and PAR%: use the

prevalence from the target population

24

Inferring to a Target Population—3

► If we can reasonably make additional assumptions… ► The association is causal (not due to bias or confounding!) ► The association is the same in other populations (no effect modification!)

► Then, we can estimate the PAR and PAR% for a target population

25

Interpretation of This PAR%

► Smoking and CHD in the target population

► Interpretation: ► Proportion of the CHD risk in the target population that is due to smoking ► Proportion of the CHD risk in the target population that could be avoided if no one in

the population smoked ► Noting all caveats, including causality!

26

Levin’s Equation for the PAR%

► 𝑃𝑃𝐴𝐴𝑅𝑅𝐴 = 𝐸𝐸𝑛𝑛𝑛𝑛𝑛𝑛𝑅𝑅𝑔𝑔𝑛𝑛𝑡𝑡 𝑛𝑛𝑛𝑛𝑡𝑡𝑝𝑝𝑡𝑡𝑡𝑡𝑡𝑡𝑖𝑖𝑝𝑝𝑡𝑡 × 𝑅𝑅𝑅𝑅−1 𝐸𝐸𝑛𝑛𝑛𝑛𝑛𝑛𝑅𝑅𝑔𝑔𝑛𝑛𝑡𝑡 𝑛𝑛𝑛𝑛𝑡𝑡𝑝𝑝𝑡𝑡𝑡𝑡𝑡𝑡𝑖𝑖𝑝𝑝𝑡𝑡 × 𝑅𝑅𝑅𝑅−1 +1

► Exposure prevalence in the target population

► RR = relative risk ► If assumptions hold, conveniently, we can calculate the PAR% using the risk ratio, rate

ratio, hazard ratio, and even the odds ratio with Levin’s equation!

27

A Big AR% Does Not Necessarily Equate to a Big PAR%—1

RR Exposure

prevalence AR% PAR%

1.05 0.005 ~5% 0.02%

1.05 0.1 ~5% 0.5%

1.05 0.5 ~5% 2.4%

2 0.005 50% 0.5%

2 0.1 50% 9.1%

2 0.5 50% 33.3%

10 0.005 90% 4.3%

10 0.1 90% 47.4%

10 0.5 90% 81.8%

28

A Big AR% Does Not Necessarily Equate to a Big PAR%—2

29

AR, AR%, PAR, PAR%—1

AR PAR

𝐼𝐼𝑖𝑖𝐼𝐼𝑅𝑅𝑛𝑛𝑡𝑡𝑖𝑖𝐼𝐼𝑡𝑡 𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆 − 𝐼𝐼𝑖𝑖𝐼𝐼𝑅𝑅𝑛𝑛𝑡𝑡𝑖𝑖𝐼𝐼𝑡𝑡 𝑔𝑔𝑖𝑖𝑡𝑡𝑛𝑛𝑛𝑛𝑛𝑛𝑅𝑅𝑡𝑡𝑛𝑛

𝐼𝐼𝑖𝑖𝐼𝐼𝑅𝑅𝑛𝑛𝑡𝑡𝑖𝑖𝐼𝐼𝑡𝑡 𝒕𝒕𝒆𝒆𝒕𝒕𝒚𝒚𝒕𝒕 𝒆𝒆𝒆𝒆𝒆𝒆𝒑𝒑𝒕𝒕𝒚𝒚𝒕𝒕𝒑𝒑𝒆𝒆𝒑𝒑 − 𝐼𝐼𝑖𝑖𝐼𝐼𝑅𝑅𝑛𝑛𝑡𝑡𝑖𝑖𝐼𝐼𝑡𝑡 𝑔𝑔𝑖𝑖𝑡𝑡𝑛𝑛𝑛𝑛𝑛𝑛𝑅𝑅𝑡𝑡𝑛𝑛

► 𝑰𝑰𝒑𝒑𝑰𝑰𝒑𝒑𝒆𝒆𝒆𝒆𝒑𝒑𝑰𝑰𝒆𝒆 𝒕𝒕𝒆𝒆𝒕𝒕𝒚𝒚𝒕𝒕 𝒆𝒆𝒆𝒆𝒆𝒆𝒑𝒑𝒕𝒕𝒚𝒚𝒕𝒕𝒑𝒑𝒆𝒆𝒑𝒑 = 𝐼𝐼𝑖𝑖𝐼𝐼𝑅𝑅𝑛𝑛𝑡𝑡𝑖𝑖𝐼𝐼𝑡𝑡 𝑡𝑡𝑛𝑛𝑛𝑛𝑛𝑛𝑅𝑅𝑡𝑡𝑛𝑛 × 𝑃𝑃𝑛𝑛𝑛𝑛𝑃𝑃𝑦𝑦𝑃𝑃𝑅𝑅𝑡𝑡𝑅𝑅𝑡𝑡𝑦𝑦 𝑡𝑡𝑛𝑛𝑛𝑛𝑛𝑛𝑅𝑅𝑡𝑡𝑛𝑛 + 𝐼𝐼𝑖𝑖𝐼𝐼𝑅𝑅𝑛𝑛𝑡𝑡𝑖𝑖𝐼𝐼𝑡𝑡 𝑔𝑔𝑖𝑖𝑡𝑡𝑛𝑛𝑛𝑛𝑛𝑛𝑅𝑅𝑡𝑡𝑛𝑛 × 𝑃𝑃𝑛𝑛𝑛𝑛𝑃𝑃𝑦𝑦𝑃𝑃𝑅𝑅𝑡𝑡𝑅𝑅𝑡𝑡𝑦𝑦 𝑔𝑔𝑖𝑖𝑡𝑡𝑛𝑛𝑛𝑛𝑛𝑛𝑅𝑅𝑡𝑡𝑛𝑛

30

AR, AR%, PAR, PAR%—2

AR

𝐼𝐼𝑖𝑖𝐼𝐼𝑅𝑅𝑛𝑛𝑡𝑡𝑖𝑖𝐼𝐼𝑡𝑡 𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆 − 𝐼𝐼𝑖𝑖𝐼𝐼𝑅𝑅𝑛𝑛𝑡𝑡𝑖𝑖𝐼𝐼𝑡𝑡 𝑔𝑔𝑖𝑖𝑡𝑡𝑛𝑛𝑛𝑛𝑛𝑛𝑅𝑅𝑡𝑡𝑛𝑛

AR%

𝐼𝐼𝑖𝑖𝐼𝐼𝑅𝑅𝑛𝑛𝑡𝑡𝑖𝑖𝐼𝐼𝑡𝑡 𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆 − 𝐼𝐼𝑖𝑖𝐼𝐼𝑅𝑅𝑛𝑛𝑡𝑡𝑖𝑖𝐼𝐼𝑡𝑡 𝑔𝑔𝑖𝑖𝑡𝑡𝑛𝑛𝑛𝑛𝑛𝑛𝑅𝑅𝑡𝑡𝑛𝑛 𝐼𝐼𝑖𝑖𝐼𝐼𝑅𝑅𝑛𝑛𝑡𝑡𝑖𝑖𝐼𝐼𝑡𝑡 𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆

31

AR, AR%, PAR, PAR%—3

AR PAR

𝐼𝐼𝑖𝑖𝐼𝐼𝑅𝑅𝑛𝑛𝑡𝑡𝑖𝑖𝐼𝐼𝑡𝑡 𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆 − 𝐼𝐼𝑖𝑖𝐼𝐼𝑅𝑅𝑛𝑛𝑡𝑡𝑖𝑖𝐼𝐼𝑡𝑡 𝑔𝑔𝑖𝑖𝑡𝑡𝑛𝑛𝑛𝑛𝑛𝑛𝑅𝑅𝑡𝑡𝑛𝑛

𝐼𝐼𝑖𝑖𝐼𝐼𝑅𝑅𝑛𝑛𝑡𝑡𝑖𝑖𝐼𝐼𝑡𝑡 𝒕𝒕𝒆𝒆𝒕𝒕𝒚𝒚𝒕𝒕 𝒆𝒆𝒆𝒆𝒆𝒆𝒑𝒑𝒕𝒕𝒚𝒚𝒕𝒕𝒑𝒑𝒆𝒆𝒑𝒑 − 𝐼𝐼𝑖𝑖𝐼𝐼𝑅𝑅𝑛𝑛𝑡𝑡𝑖𝑖𝐼𝐼𝑡𝑡 𝑔𝑔𝑖𝑖𝑡𝑡𝑛𝑛𝑛𝑛𝑛𝑛𝑅𝑅𝑡𝑡𝑛𝑛

AR% PAR%

𝐼𝐼𝑖𝑖𝐼𝐼𝑅𝑅𝑛𝑛𝑡𝑡𝑖𝑖𝐼𝐼𝑡𝑡 𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆 − 𝐼𝐼𝑖𝑖𝐼𝐼𝑅𝑅𝑛𝑛𝑡𝑡𝑖𝑖𝐼𝐼𝑡𝑡 𝑔𝑔𝑖𝑖𝑡𝑡𝑛𝑛𝑛𝑛𝑛𝑛𝑅𝑅𝑡𝑡𝑛𝑛 𝐼𝐼𝑖𝑖𝐼𝐼𝑅𝑅𝑛𝑛𝑡𝑡𝑖𝑖𝐼𝐼𝑡𝑡 𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆𝒆

𝐼𝐼𝑖𝑖𝐼𝐼𝑅𝑅𝑛𝑛𝑡𝑡𝑖𝑖𝐼𝐼𝑡𝑡 𝒕𝒕𝒆𝒆𝒕𝒕𝒚𝒚𝒕𝒕 𝒆𝒆𝒆𝒆𝒆𝒆. − 𝐼𝐼𝑖𝑖𝐼𝐼𝑅𝑅𝑛𝑛𝑡𝑡𝑖𝑖𝐼𝐼𝑡𝑡 𝑔𝑔𝑖𝑖𝑡𝑡𝑛𝑛𝑛𝑛𝑛𝑛𝑅𝑅𝑡𝑡𝑛𝑛 𝐼𝐼𝑖𝑖𝐼𝐼𝑅𝑅𝑛𝑛𝑡𝑡𝑖𝑖𝐼𝐼𝑡𝑡 𝒕𝒕𝒆𝒆𝒕𝒕𝒚𝒚𝒕𝒕 𝒆𝒆𝒆𝒆𝒆𝒆𝒑𝒑𝒕𝒕𝒚𝒚𝒕𝒕𝒑𝒑𝒆𝒆𝒑𝒑

Note: 𝑰𝑰𝒑𝒑𝑰𝑰𝒑𝒑𝒆𝒆𝒆𝒆𝒑𝒑𝑰𝑰𝒆𝒆 𝒕𝒕𝒆𝒆𝒕𝒕𝒚𝒚𝒕𝒕 𝒆𝒆𝒆𝒆𝒆𝒆. = 𝐼𝐼𝑖𝑖𝐼𝐼𝑅𝑅𝑛𝑛𝑡𝑡𝑖𝑖𝐼𝐼𝑡𝑡 𝑡𝑡𝑛𝑛𝑛𝑛𝑛𝑛𝑅𝑅𝑡𝑡𝑛𝑛 × 𝑃𝑃𝑛𝑛𝑛𝑛𝑃𝑃𝑦𝑦𝑃𝑃𝑅𝑅𝑡𝑡𝑅𝑅𝑡𝑡𝑦𝑦 𝑡𝑡𝑛𝑛𝑛𝑛𝑛𝑛𝑅𝑅𝑡𝑡𝑛𝑛 + ( )

𝐼𝐼𝑖𝑖𝐼𝐼𝑅𝑅𝑛𝑛𝑡𝑡𝑖𝑖𝐼𝐼𝑡𝑡 𝑔𝑔𝑖𝑖𝑡𝑡𝑛𝑛𝑛𝑛𝑛𝑛𝑅𝑅𝑡𝑡𝑛𝑛 × 𝑃𝑃𝑛𝑛𝑛𝑛𝑃𝑃𝑦𝑦𝑃𝑃𝑅𝑅𝑡𝑡𝑅𝑅𝑡𝑡𝑦𝑦 𝑔𝑔𝑖𝑖𝑡𝑡𝑛𝑛𝑛𝑛𝑛𝑛𝑅𝑅𝑡𝑡𝑛𝑛

32

PAR% Example—1

► Dementia prevention, intervention, and care ► “In this Commission, we have used

the best available evidence to make recommendations. When evidence is incomplete we have summarised the balance of evidence and explained its strengths and limitations. An overall limitation is that this evidence is generally focused on, and from, high-income countries and we have less evidence from LMICs.”

Source: Livingston et al. (2017). Dementia prevention, intervention, and care. Lancet, 390(10113), 2673-2734. https://doi.org/10.1016/S0140-6736(17)31363-6

33

PAR% Example—2

Source: Livingston et al. (2017). Dementia prevention, intervention, and care. Lancet, 390(10113), 2673-2734. https://doi.org/10.1016/S0140-6736(17)31363-6

34

PAR% Example—3

Source: Livingston et al. (2017). Dementia prevention, intervention, and care. Lancet, 390(10113), 2673-2734. https://doi.org/10.1016/S0140-6736(17)31363-6

35

Difference vs. Relative Measures of Association—1

► When should the absolute (ARs) versus relative measures of association (RRs) be calculated?

36

Difference vs. Relative Measures of Association—2

► When should the absolute (ARs) versus relative measures of association (RRs) be calculated?

► Difference measures (ARs) are calculated when describing the public health impact of an exposure ► Need to know the absolute excess disease risk to be able to prioritize interventions

37

Difference vs. Relative Measures of Association—3

► When should the absolute (ARs) versus relative measures of association (RRs) be calculated?

► Difference measures (ARs) are calculated when describing the public health impact of an exposure ► Need to know the absolute excess disease risk to be able to prioritize interventions

► Relative measures are calculated when evaluating etiologic questions ► Stronger associations imply greater contributions to the etiology of the outcome

38

Lessons Learned

► Reviewed the calculation of the OR for unmatched studies and calculated the OR for matched case-control studies

► Identified when the OR is a reasonable estimate of the RR

► Discussed and applied the concepts of attributable risk percent and the population attributable risk percent