Epidemiology quiz

Bias

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.

Introduction

3

Recall a Goal of Epidemiologic Research

► Goal: To identify exposure and disease associations

4

Possible Explanations for Associations

► Cause

► Chance

► Bias

► Confounding

5

What Is Bias?

► “Systematic deviation of results or inferences from truth… An error in the conception and design of a study—or in the collection, analysis, interpretations, reporting, publication, or review of data— leading to results or conclusions that are systematically (as opposed to randomly) different from the truth.”

Source: Porta, M. (Ed.). (2014). A dictionary of epidemiology (6th ed.). Oxford University Press.

6

Role of Bias in Epidemiology

► In observational epidemiologic studies ► Bias exists

► After the results of a study are obtained, identify type, likelihood, direction and magnitude of bias ► Assess the influence of bias on the inferences, including causal inferences, that may be

drawn from the study

► Best approach is to prevent or minimize bias in the first place ► Design ► Analysis

7

Recall… Populations Lecture

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.

Information Bias, Non-differential Misclassification in Association Studies

2

Recall…

3

Information Bias

► Information bias occurs when the means for obtaining information about the participants in the study are inadequate so that, as a result, some of the information gathered regarding exposures and/or disease outcomes is incorrect

4

Role of Measurement—1

5

Role of Measurement —2

► Self-report

► Sphygmomanometer

► Self-report

► Hemoglobin A1C

► Fasting glucose levels

6

What Is Bias?

► “Systematic deviation of results or inferences from truth… An error in the conception and design of a study—or in the collection, analysis, interpretations, reporting, publication, or review of data— leading to results or conclusions that are systematically (as opposed to randomly) different from the truth.”

Source: Porta, M. (Ed.). (2014). A dictionary of epidemiology (6th ed.). Oxford University Press.

7

Components of Measurement Error

► Measured value = True value + Error

► Error = Bias* + Random† error

* Systematic component of the error

† Random component of the error

8

Role of Measurement —3

► Validity of the measure

► Systematically reading higher (or lower) than true value

► Inaccurate calibration

► Improper cuff size

9

Consequences of Imperfect Measurement

► Categorial variable: Misclassification

► Continuous variable: Measurement error

10

Possible Explanations for Associations

► Cause

► Chance

► Bias ► Information bias

► Confounding

11

Non-differential Bias

► Misclassification or measurement error in the exposure and/or outcome

► Extent of the error does not differ between the groups being compared

12

Non-differential Bias in a Case-Control Study

► The extent of the misclassification or measurement error of the exposure does not differ between cases and controls

13

Non-differential Bias in a Cohort Study

► The extent of the misclassification or measurement error of the outcome does not differ between exposed and unexposed

14

Example of Non-differential Misclassification in a Case-Control Study—1

15

Example of Non-differential Misclassification in a Case-Control Study—2

► What if 10% of exposed cases and 10% of exposed controls are misclassified as unexposed?

16

Example of Non-differential Misclassification in a Case-Control Study—3

► What if 10% of exposed cases and 10% of exposed controls are misclassified as unexposed?

17

Example of Non-differential Misclassification in a Case-Control Study—4

► What if 10% of exposed cases and 10% of exposed controls are misclassified as unexposed?

18

Example of Non-differential Misclassification in a Case-Control Study—5

► 80 exposed cases × 0.10 = 8 cases ► Eight exposed cases are misclassified as unexposed

19

Example of Non-differential Misclassification in a Case-Control Study—6

► 80 exposed cases × 0.10 = 8 cases ► Eight exposed cases are misclassified as unexposed

20

Example of Non-differential Misclassification in a Case-Control Study—7

► 80 exposed cases × 0.10 = 8 cases ► Eight exposed cases are misclassified as unexposed

21

Example of Non-differential Misclassification in a Case-Control Study—8

► What if 10% of exposed cases and 10% of exposed controls are misclassified as unexposed?

22

Example of Non-differential Misclassification in a Case-Control Study—9

► 50 exposed controls × 0.10 = 5 controls ► Five exposed controls are misclassified as unexposed

23

Example of Non-differential Misclassification in a Case-Control Study—10

► 50 exposed controls × 0.10 = 5 controls ► Five exposed controls are misclassified as unexposed

24

Example of Non-differential Misclassification in a Case-Control Study—11

► 𝑂𝑂𝑂𝑂 = 𝑎𝑎𝑎𝑎 𝑏𝑏𝑏𝑏

= 80×50 50×20

= 4.0

25

Example of Non-differential Misclassification in a Case-Control Study—12

► 𝑂𝑂𝑂𝑂 = 𝑎𝑎𝑎𝑎 𝑏𝑏𝑏𝑏

= 80×50 50×20

= 4.0 ► 𝑂𝑂𝑂𝑂 = 𝑎𝑎𝑎𝑎 𝑏𝑏𝑏𝑏

= 72×55 45×28

= 3.1

26

Example of Non-differential Misclassification in a Case-Control Study—13

► 𝑂𝑂𝑂𝑂 = 𝑎𝑎𝑎𝑎 𝑏𝑏𝑏𝑏

= 80×50 50×20

= 4.0 ► 𝑂𝑂𝑂𝑂 = 𝑎𝑎𝑎𝑎 𝑏𝑏𝑏𝑏

= 72×55 45×28

= 3.1

27

Consequence of Non-differential Bias

► Bias associations toward the “null” ► Association observed is weaker than the truth

► If the true OR* > 1, the biased OR* < true OR* ► For example, true OR = 2.0 and biased OR = 1.5

► If the true OR* < 1, the biased OR* > true OR* ► For example, true OR = 0.7 and biased OR = 0.9

* Same is true for the RR in a cohort study

28

Example of Non-differential Misclassification in a Case-Control Study—14

► 𝑂𝑂𝑂𝑂 = 𝑎𝑎𝑎𝑎 𝑏𝑏𝑏𝑏

= 80×50 50×20

= 4.0 ► 𝑂𝑂𝑂𝑂 = 𝑎𝑎𝑎𝑎 𝑏𝑏𝑏𝑏

= 72×55 45×28

= 3.1

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.

Information Bias, Differential Misclassification in Association Studies

2

Possible Explanations for Associations

► Cause

► Chance

► Bias 1. Information bias 2. Differential

► Confounding

3

Differential Bias

► Misclassification or measurement error in the exposure and/or outcome

► Extent of the error differs between the groups being compared

4

Differential Bias in a Case-Control Study

► The extent of the misclassification or measurement error of the exposure

differs between cases and controls

5

Differential Bias in a Cohort Study

► The extent of the misclassification or measurement error of the outcome differs between exposed and unexposed

6

Example of Differential Misclassification in a Case-Control Study—1

► What if 10% of exposed cases are misclassified as unexposed (all exposed controls are correctly classified as exposed)?

7

Example of Differential Misclassification in a Case-Control Study—2

► What if 10% of exposed cases are misclassified as unexposed (all exposed controls are correctly classified as exposed)?

8

Example of Differential Misclassification in a Case-Control Study—3

► 𝑂𝑂𝑂𝑂 = 𝑎𝑎𝑎𝑎 𝑏𝑏𝑏𝑏

= 80×50 50×20

= 4.0 ► 𝑂𝑂𝑂𝑂 = 𝑎𝑎𝑎𝑎 𝑏𝑏𝑏𝑏

= 72×50 50×28

= 2.6

9

Example of Differential Misclassification in a Case-Control Study—4

► 𝑂𝑂𝑂𝑂 = 𝑎𝑎𝑎𝑎 𝑏𝑏𝑏𝑏

= 80×50 50×20

= 4.0 ► 𝑂𝑂𝑂𝑂 = 𝑎𝑎𝑎𝑎 𝑏𝑏𝑏𝑏

= 72×50 50×28

= 2.6

10

Consequences of Differential Bias—1

► Differential bias can bias measures of association in any direction ► Can lead to apparent association even if one does not really exist ► Can lead to an apparent lack of association even if one does exist

11

Consequences of Differential Bias—2

► Differential bias can bias measures of association in any direction ► Can lead to apparent association even if one does not really exist ► Can lead to an apparent lack of association even if one does exist

► The magnitude can be substantial, even with small differences in sensitivity or specificity

12

Example of Differential Misclassification in a Case-Control Study—5

► 𝑂𝑂𝑂𝑂 = 𝑎𝑎𝑎𝑎 𝑏𝑏𝑏𝑏

= 80×50 50×20

= 4.0 ► 𝑂𝑂𝑂𝑂 = 𝑎𝑎𝑎𝑎 𝑏𝑏𝑏𝑏

= 72×50 50×28

= 2.6

13

You Practice! Differential Misclassification of Outcome in a Cohort Study—1

► True RR:

14

You Practice! Non-differential Misclassification of Outcome in a Cohort Study

► True RR:

► What if 20% of exposed who develop the disease are misclassified as *not* developing disease (but all persons who are unexposed have their disease status correctly measured)?

15

You Practice! Differential Misclassification of Outcome in a Cohort Study—2

► True RR:

16

You Practice! Differential Misclassification of Outcome in a Cohort Study—3

► True RR:

► Biased RR:

(Answers at the end of this lecture)

17

Examples of Differential Information Bias in a Case-Control Study—1

► Interviewer bias ► E.g., if interviewer is aware of disease status and the hypothesis under study, the

interviewer consciously or subconsciously may try to elicit more complete exposure information from the cases

18

Examples of Differential Information Bias in a Case-Control Study—2

► Interviewer bias ► E.g., if interviewer is aware of disease status and the hypothesis under study, the

interviewer consciously or subconsciously may try to elicit more complete exposure information from the cases

► Recall bias and reporting bias ► E.g., a case may be more likely to remember an earlier exposure than a control

because he/she is searching for an explanation for the disease ► E.g., a case may be less likely to report his/her exposure for fear of revealing that they

participated in a disease-causing behavior

19

Minimize Information Bias in a Case-Control Study

► Mask (blind) participants to study hypothesis

► Mask (blind) interviewers to case or control status of the participants

► Have trained interviewers interview same proportion of cases and controls

► Use objective measures of exposure

► Use of multiple control groups

20

Examples of Differential Information Bias in a Cohort Study—1

► Lack of masking to exposure status (e.g., medical record reviewer is aware of patients’ exposure status when assessing disease)

21

Examples of Differential Information Bias in a Cohort Study—2

► Lack of masking to exposure status (e.g., medical record reviewer is aware of patients’ exposure status when assessing disease)

► Physician’s diagnosis of disease depends on exposure (e.g., diagnostic bias)

22

Examples of Differential Information Bias in a Cohort Study—3

► Lack of masking to exposure status (e.g., medical record reviewer is aware of patients’ exposure status when assessing disease)

► Physician’s diagnosis of disease depends on exposure (e.g., diagnostic bias)

► Opportunity for diagnosis depends on exposure status (e.g., detection bias)

23

Minimize Information Bias in a Cohort Study

► Mask (blind) participants to the study hypothesis when relying on self-report of disease

► Mask (blind) medical record reviewers to exposure status of participants

► Use objective measures of disease and with confirmation

24

Summary of Information Bias

► Information bias is an over- or underestimation of an association caused by measurement error

► A non-differential measurement error usually biases the association towards the null

► A differential measurement error can lead to an over- or underestimation of an association

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.

Selection Bias in Association Studies

2

Possible Explanations for Associations—1

► Cause (listen online to causal inference lecture)

► Chance

► Bias ► Information bias ►

► Confounding

3

Possible Explanations for Associations—2

► Cause (listen online to causal inference lecture)

► Chance

► Bias ► Information bias ► Selection bias

► Confounding

4

Selection Bias

► Selection bias: distortion of a measure of association in study population from what would be observed in the source/target population due to the choice of participants

5

Recall… Populations Lecture

6

All Selection Bias Is Differential

► Bias ► Information bias

● Non-differential ● Differential

► Selection bias ● Differential

7

Selection Bias in a Case-Control Study—1

► Inclusion of an individual as a case control in the study depends differentially on exposure status

8

Recall…

9

Selection Bias in a Case-Control Study—2

► Distortion of a measure of association from what would be observed in the source population due to the way the cases and controls were selected in a case-control study

► The purpose of the controls in a case-control study is to provide an estimate of the prevalence of the exposure in the source population for the cases

10

Classic Example of Selection Bias

► MacMahon et al. (1981). Coffee and cancer of the pancreas. N Engl J Med, 304(11), 630– 633.

► Cases: Patients with histologic diagnosis of pancreatic cancer in any of 11 large hospitals in Boston and Rhode Island between October 1974 and August 1979

► Controls: Other patients under the care of the same physician as the cases with pancreatic cancer—patients with diseases known to be associated with smoking or alcohol consumption were excluded

11

Primary Study Results—1

► 𝑂𝑂𝑂𝑂 = 𝑎𝑎𝑎𝑎 𝑏𝑏𝑏𝑏

= 207×32 275×9

= 2.7

12

Primary Study Results—2

► 𝑂𝑂𝑂𝑂 = 𝑎𝑎𝑎𝑎 𝑏𝑏𝑏𝑏

= 207×32 275×9

= 2.7

► Drinking one or more cups of coffee per day is associated with 2.7 times the odds of pancreatic cancer compared to drinking 0 cups of coffee per day

13

Selection Bias Occurred in the Study Design— 1

► Relative to the source population for the cases, the control group included:

14

Selection Bias Occurred in the Study Design— 2

► Relative to the source population for the cases, the control group included: ► Other patients under the care of the same doctors at the

time of an interview with a case ● Most of the doctors were gastroenterologists whose

other patients were likely advised to stop drinking coffee for GI problems

15

Selection Bias Occurred in the Study Design— 3

► Relative to the source population for the cases, the control group included: ► Other patients under the care of the same doctors at the

time of an interview with a case ● Most of the doctors were gastroenterologists whose

other patients were likely advised to stop drinking coffee for GI problems

► Patients with diseases known to be associated with smoking or alcohol consumption were excluded ● Coffee drinking is associated with smoking and alcohol

drinking

16

Selection Bias Occurred in the Study Design— 4

► Relative to the source population for the cases, the control group included: ► Other patients under the care of the same doctors at the

time of an interview with a case ● Most of the doctors were gastroenterologists whose

other patients were likely advised to stop drinking coffee for GI problems

► Patients with diseases known to be associated with smoking or alcohol consumption were excluded ● Coffee drinking is associated with smoking and alcohol

drinking ► Thus, the sample of controls was depleted of coffee drinkers

● The controls did not have an exposure prevalence reflecting that of the source population for the cases

17

Subsequent Case-Control Study Using Population-Based Controls—1

► 𝑂𝑂𝑂𝑂 = 𝑎𝑎𝑎𝑎 𝑏𝑏𝑏𝑏

= 84×14 82×10

= 1.4

Source: Gold, E. B., Gordis, L., Diener, M. D., et al. (1985). Diet and other risk factors for cancer of the pancreas. Cancer Res, 55(2), 460–7.

18

Subsequent Case-Control Study Using Population-Based Controls—2

► 𝑂𝑂𝑂𝑂 = 𝑎𝑎𝑎𝑎 𝑏𝑏𝑏𝑏

= 84×14 82×10

= 1.4

► Estimated OR is much less than in the original study, and in prospective studies the RR=1

Source: Gold, E. B., Gordis, L., Diener, M. D., et al. (1985). Diet and other risk factors for cancer of the pancreas. Cancer Res, 55(2), 460–7.

19

Minimizing Selection Bias in a Case-Control Study

► Do not put eligibility/ineligibility criteria on the controls that are not also put on the cases ► Can distort the exposure prevalence in the controls relative to the source population

and thus result in selection bias ● E.g., patients with diseases known to be associated with smoking or alcohol

consumption were excluded from the control group in the MacMahon study, but they did not exclude smokers and alcohol drinkers from the cases

20

Selection Bias in a Cohort Study

► Inclusion in the cohort as exposed or unexposed depends differentially on disease status

21

Selection Bias in a Cohort Study—Selection into the Study

► Less likely to occur in prospective studies because disease has not yet developed when exposure assessed

► Can happen in a retrospective cohort study ► E.g., if archival exposure records are more likely to be located for individuals who were

exposed and later developed the disease

22

Selection Bias in a Cohort Study—Selection out of the Study

► E.g., loss to follow-up is differential by exposure status

23

Summary of Selection Bias

► Selection bias is an over- or underestimation of an association due to a systematic error in the way participants are selected into (or out of) a study

24

Important Notes about Selection—1

► “Selection” of the study population from a source population ≠ selection bias

25

Important Notes about Selection—2

► “Selection” of the study population from a source population ≠ selection bias

► For a study to be internally valid, it should be free of selection (and information) bias

26

Important Notes about Selection—3

► “Selection” of the study population from a source population ≠ selection bias

► For a study to be internally valid, it should be free of selection (and information) bias

► Sometimes a selected study population can help improve the internal validity of the study ► E.g., nurses participating in the Nurses’ Health Study ► But often not representative of the target population

27

Important Notes about Selection—4

► “Selection” of the study population from a source population ≠ selection bias

► For a study to be internally valid, it should be free of selection (and information) bias

► Sometimes a selected study population can help improve the internal validity of the study ► E.g., nurses participating in the Nurses’ Health Study ► But often not representative of the target population

► If not representative, the results may be internally valid, but not generalizable to the target population

28

Important Notes about Selection—5

► If not representative, the results may be internally valid, but not generalizable to the target population

► How to evaluate generalizability of the results?

► If not representative, how do the study and target populations differ?

29

Important Notes about Selection—6

► If not representative, the results may be internally valid, but not generalizable to the target population

► How to evaluate generalizability of the results?

► If not representative, how do the study and target populations differ?

► Do those factors modify the exposure-disease relationship? ► If yes—results are not generalizable to the target population ► If no—even though the study population is not representative of the target population,

the results are generalizable

30

Lessons Learned

► Defined information and selection bias

► Identified sources of these biases

► Described how each type of bias might impact study inferences

► Described approaches to minimize bias in association studies

31

You Practice! Answers— Differential Misclassification of Outcome in a Cohort Study

► True RR:

► Biased RR: