Confounding and Bias

Question 1

if we survey NYC and top baseball team is Yankees, can we see it is the US’s fave team? Why not?

Answer 1

No, bc of selection bias

Question 2

what 2 studies have forward directionality?

Answer 2

cohort
RCT

Question 3

selection bias in cohort/RCT

Answer 3

ppl dropout or are lost to follow up (may be unequal in each group)

Question 4

selection bias in case control

Answer 4

hospital controls may have higher alcohol intake (or a diff characteristic) than community controls

Question 5

internal validity

Answer 5

proper group selection + lack of error in measurement

Question 6

for internal validity, you must have accurate measurement of __, __, and __

Answer 6

for internal validity, you must have accurate measurement of exposure, outcome, and association btwn the two

Question 7

external validity aka

Answer 7

generalizability

Question 8

external validity

Answer 8

ability to generalize beyond a set of observations to a universal statement

Question 9

confounding

Answer 9

an association btwn exposure and outcome is observed, BUT only as a result of a third variable

Question 10

crude and adjusted measures of effect are not equal when __ is present

Answer 10

crude and adjusted measures of effect are not equal when a confounder is present

Question 11

to be a confounder, the extraneous factor must:

Answer 11

1) be a risk factor of the disease
2) be associated with the exposure
3) not be an intermediate step in causal pathway btwn exposure and disease

Question 12

3 ways to prevent confounding

Answer 12

1) randomization
2) restriction
3) matching

Question 13

randomization attempts to have equal __ of __ in groups

Answer 13

randomization attempts to have equal distribution of confounders in groups

Question 14

pros of randomization

Answer 14

1. convenient
2. cheap
3. straightforward data analysis

Question 15

cons of randomization

Answer 15

1. need control over exposure and ability to assign subjects to study groups
2. need large samples

Question 16

randomization: how many conditions of confounding are not met?

Answer 16

1 of 2

Question 17

restriction aims to prevent __ of __ in study groups

Answer 17

restriction aims to prevent variation of confounder in study groups

Question 18

restriction provides complete control over __

Answer 18

restriction provides complete control over KNOWN confounders

Question 19

restriction pros

Answer 19

1. conceptually easy
2. handles difficult to quantify variables
3. can be used in analysis phase

Question 20

restriction cons

Answer 20

1. limits eligible subjects
2. inefficient to screen and then not use people
3. residual confounding if categories are not narrow enough
4. limits generalizability
5. can’t assess interaction (can’t stratify by age when age is narrow)

Question 21

restriction: how many conditions of confounding are not met?

Answer 21

2

Question 22

restriction is good for when confounder is __, but __

Answer 22

restriction is good for when confounder is quantitative, but hard to measure

Question 23

restriction exmaple

Answer 23

sexual behavior and HIV
injectable drugs is a confounder, quantitative, but hard to measure
SO, just don’t include individuals who use injectable drugs

Question 24

matching 2 types

Answer 24

1. frequency matching
2. individual matching

Question 25

frequency matching

Answer 25

number of cases with match characteristics are equalized

Question 26

individual matching

Answer 26

pairing of 1 or more controls to each case based on sex, race, etc.

Question 27

cohort study matching

Answer 27

unexposed black matched to exposed black

Question 28

case-control matching

Answer 28

control age 50 to case age 50

Question 29

matching pros

Answer 29

1. less ppl required than unmatched studies with same hypothesis

Question 30

matching cons

Answer 30

1. expensive b/c extensive searching and record keeping is required

Question 30

matching cons

Answer 30

1. expensive b/c extensive searching and record keeping is required

Question 31

2 types of analysis to control confounding

Answer 31

1. stratification 2. multivariate techniques

Question 32

stratification holds ___ constant

Answer 32

stratification holds **confounder** constant

Question 33

pros of stratification

Answer 33

1. direct and logical strategy 2. minimum assumptions required 3. straightforward 4. you may find an interaction

Question 34

cons of stratification

Answer 34

1. strata may not have equal numbers 2. multiple ways to form strata when using continuous variables 3. hard with multiple confounders 4. categorization causes information loss

Question 34

cons of stratification

Answer 34

1. strata may not have equal numbers 2. multiple ways to form strata when using continuous variables 3. hard with multiple confounders 4. categorization causes information loss

Question 35

multivariate statistics uses __ to estimate what association would be if __ wasn't associated with __

Answer 35

multivariate statistics uses **stats** to estimate what association would be if **confounder** wasn't associated with **exposure**

Question 36

pros of multivariate statistics

Answer 36

1. continuous variables to do not need to be converted to categorical 2. can control for multiple exposure variables at same time

Question 37

cons of multivariate statistics

Answer 37

1. potential for misuse

Question 38

confounding issues

Answer 38

1. can conclude a relationship when there's not one 2. can conclude there's not a relationship when there is

Question 39

confounding is NOT __

Answer 39

confounding is NOT **an error**

Question 40

is an error to __

Answer 40

is an error to **NOT control for confounding**

Question 41

is an error to __

Answer 41

is an error to **NOT control for confounding**

Question 42

bias =

Answer 42

systematic design error that mistakes exposure's effect on disease

Question 43

what kind of studies is bias a problem in?

Answer 43

ALL studies

Question 44

selection bias is flaws in

Answer 44

selecting participants

Question 45

information bias is flaws in

Answer 45

gathering info

Question 46

selection bias: relationship btwn disease and exposure is different from in those who

Answer 46

selection bias: relationship btwn disease and exposure is different from in those who **were eligible but didn't**

Question 47

in study looking at cancer and coffee, the study sample over-represents...

Answer 47

in study looking at cancer and coffee, the study sample over-represents **controls who don't drink coffee (d)**

Question 48

cohort study selection bias: lung cancer and abestos who is over-represented?

Answer 48

unexposed with disease lots of unexposed sick people at risk for lung cancers

Question 49

exclusion bias

Answer 49

if you apply different eligibility criteria to cases and controls EX: Reserpine and breast cancer

Question 50

non-response bias is a type of __ bias

Answer 50

non-response bias is a type of **selection** bias

Question 51

non-response bias

Answer 51

responses may be higher in diseased or exposed could show associated when tehre is none

Question 52

how to limit non-response bias?

Answer 52

1. get people to respond 2. characterize non-respondents as much as possible (they may be more likely to have certain traits)

Question 53

types of information bias

Answer 53

interview bias surveillance bias recall bias reporting bias

Question 54

misclassification bias

Answer 54

misclassify unexposed as exposed is more common in cases

Question 55

misclassification bias dilutes __ and __

Answer 55

misclassification bias dilutes **RR** and **OR**

Question 56

interviewer/abstractor bias

Answer 56

interviewer probes cases more thoroughly for an exposure

Question 57

prevarication (lying) bias

Answer 57

participants have ulterior motives

Question 58

similar misclassification in cases and controls pushses association towards

Answer 58

1

Question 59

5 ways to prevent bias

Answer 59

1. careful attention to sampling 2. minimize non-reponse 3. standardize measurements 4. training and quality control 5. blinding