Oct 4

Question 1

Describe Selection Bias in RCTs

Answer 1

• with respect to randomization; if we don’t adequately generate the random sequences (can have runs; if you’re making a random sequence of 10 sometimes it might end up that you have 9 assigned to A and 1 assigned to B so you get a run of people entering your study getting assigned to A- if they are early morning people, different reponse to people, etc. this could bring in selection bias)
• allocation concealment; we don’t know who is assigned to what until the moment they are assigned (compromisation in this process ie. research assistant administering therapy- if they know what the next assignment is and you think the next person is going to do really well on this treatment you might give it but if you think the next person won’t do well on this treatment you might say oh let’s just give it to the next person

Question 2

Describe Performance Bias in RCTs

Answer 2

• differences that occur due to knowledge of intervention allocation in either the resercher or the participant
• results in differences in the care received by the intervention and control groups in a trial that are over and above the intervention that are being compared
• performance bias: whether or not we’ve done a good job of blinding
• if blinding has been done properly, this usually means we have taken care of performance bias so not an issue

Question 3

What is attrition bias in RCTs

Answer 3

• also in prospective cohort studies
• incomplete outcome data/losses to follow up
• anytime you have lots of people dropping out, it compromises your ability to say something about the treatment
• this becomes more important when the drop out is differential (one group has higher dropout rate than others. if that is related to treatment or outcomes, that is a problem)

Question 4

What is detection bias in RCTs

Answer 4

• similar to performance bias except this is the way we report events
• want know the people assessing the outcome- are they biased about deciding who is a case/control, who has developed outcome/hasn’t
• if we know the treatment someone was assigned to, you might be more likely to say one over the other

Question 5

What bias are case control studies more susceptible to than other studies?

Answer 5

• recall bias
• we select people once they are already a case and ask them to remember exposure histories
• if exposure histories happened a long time ago they may be recalled differently in cases than controls

Question 6

What is a common bias seen in cohort studies?

Answer 6

• selection bias because there could be different rates of loss to follow up in the exposed and unexposed groups
• whatever factors that lead people to be exposed may lead them to not be good at participating in studies (or vice versa)
• people that stay in study and tend to be interested in their health are different than people who do not

Question 7

Describe random error

Answer 7

-reduces precision of the estimates (OR, RR, etc.)

Question 8

Describe systematic error

Answer 8

• more of an issue of getting the wrong answer
• reduces validity or accuracy of the estimates
• eg. scale routinely weighing people 10lbs less than they really are

Question 9

Fill in the blanks

Answer 9

Figure A: Accurate and precise, no error (can do a number of studies and each study gets close to the truth)

Figure B: precise, but not accurate. Systematic error. Lots of studies cluster in the same area (precise) but they are far from the truth. All studies are misestimating truth by about the same amount.

Figure C: Accurate, but not precise. Random error. Number of studies all aren’t going to get same answer due to random variability in populations they study, outcomes they assess, but dots are more spread out than in Figure A where everything was true.

Figure D: Neither accurate nor precise. Systematic and random error. All studies get wrong answers.

Question 10

What is internal validity?

Answer 10

• we do a study and it measures what we intend it to measure
• gives unbiased estimate of true effect in that study
• study was designed properly, used the best methods, collected data in best way we could, analysis was high quality

Question 11

What is external validity?

Answer 11

• refers to generalization
• could get a specific population and do a well controlled study but all of the people that contribute to that study might not be representative of the entire population

Question 12

What is confounding?

Answer 12

• refers to the “mixing” of the effect of the exposure with the effect of another factor (the confounder)
• distorts the estimated measure of association

Question 13

What is a confounder?

Answer 13

• a variable that is: a risk factor for the disease of interest (independent of exposure), associated with the exposure of interest (without being a consequence of the exposure)
• not in the causal pathway

exposure variable –> variable in causal pathway (happens as the result of an exposure that then leads to disease) –> diesease variable

Question 14

Is persistent coughing a confounder with smoking and SUI?

Answer 14

• first criterion satisfied: coughing causes an increase in intra-abdominal pressure which leads to SUI (risk factor)
• any coughing may lead to SUI regardless of smoking status
• however, smoking increases frequency/intensity of coughing so smokers may have more bouts of SUI but it’s happening because coughing is the mechanism by which smoking affects SUI
• smoking –> coughing –> SUI (in causal pathway) therfore is not a confounder

Question 15

What studies to confounders affect?

Answer 15

• all observational study designs
• without randomization, there is no way to be certain that possible confounders are distributed evenly across the groups
• uneven distribution of confounders causes biased ORs and RRs

Question 16

What stages can we handle confounding at?

Answer 16

-design or analysis

Question 17

What is the best way to control confounding?

Answer 17

• randomization with RCT
• balances known and unknown confounders
• can’t always randomize in observational study because we don’t do anything to people we just watch

Question 18

What is restriction?

Answer 18

• limit entrance of participants into people that fall into one category of the other of the confounder
• reduces available participants
• if we don’t actually know association between the confounder and the outcome, we may not be able to choose the correct cut off point between categories of confounders (if it’s improperly defined can lead to residual confounding)

Question 19

What is matching?

Answer 19

• way to handle confounding
• only include controls with comparable characteristics to cases on more than one or one variable
• problems: time consuming and costly to find comparison controls, cannot evaluate effects of matched variables on disease

Question 20

What is multivariable regression analysis?

Answer 20

• look at associations and add confounding variables to association models to help adjust for different confounders
• if you’ve measured confounder properly in study, you can deal with it in the analysis phase

Question 21

What is stratification?

Answer 21

• separate ORs or RRs by different levels of a confounder
• take weighted averages of each stratum to make sure we get overall effect
• can become difficult with many stratum (many confounders)
• cannot be done when the confounder is a continuous variable that cannot be categorized (eg. age or BP- decide not to put people in groups)

Question 22

How do we identify confounders?

Answer 22

• literature search is important (when we read previous studies, gives guidance as to which potential confounders to collect data on)
• biological plausability (this always changes, not all confounders known, results unknowingly biased- things that we thought were confounders have been proven wrong)

Question 23

What can a confounder do?

Answer 23

-wipe out an association, create an invalid association, reverse the direction of an association, change the degree of association

Question 24

What is effect modification?

Answer 24

• tells us the effect of the primary exposure on outcome differs depending on the level of a third variable
• biological phenomenon
• think effect of exposure on outcome differs according to some factor
• confounding is a bias such that we get the wrong answer, effect modification shows exposure has different impact in different circumstances
• when the magnitude of the effect of the primary exposure on an outcome (association) differs depending on the level of a third variable

Question 25

How do you design 2x2 tables in case control versus cohort study

Answer 25

• in case control, you are looking at outcomes then going back and assessing exposure
• try to put outcome on the y axis then exposure on the x axis
• if we do cohort study starting with exposure then looking for outcome, exposure on y axis and outcome on x axis

Question 26

What are stratified analyses?

Answer 26

• stratum specific estimates
• take weighted average
• controlled for the confounder
• Mantel Haenszel Estimator= measure average effect of exposure across all strata (accounts for confounding and EM)
• when we have confounding, crude odds ratio doesn’t match any stratum specific estimates

Question 27

How do you tell if there is a confounder versus EM with MH estimator?

Answer 27

• if MH estimator is equal to 1 or less than crude odds ratio and within each stratum odds ratio is also 1 we would say we have developed a confounder
• if MH and crude odds ratio are identical but we have different stratum specific odds ratios then we have EM

Question 28

Can something be both a confounder and effect modifier? Give an example.

Answer 28

• yes
• age is classic example
• for all diseases, age is a risk factor
• second, the effect of an exposure of disease differs between young and old people often
• eg: think of the effect of a fall on a 4 year old versus a 94 year old. In both cases, age is associated with increased risk of falls and the probability of a bad outcome of fallse. Age also impacts how serious the consequences of the fall will be. So same exposure (fall) will have different result based on age.

Question 29

In a scenario of both confounding and effect modification, some of the observed effect of an exposure on an outcome is due to….

Answer 29

the confounding effect of the third variable, but this effect is not completely eliminated within each stratum

-Mantel Haenszel average of stratum specific estimates is somewhere between stratum specific ORs and less than the crude OR

Question 30

In effect modification, the observed effect of an exposure on an outcome depends on…

Answer 30

the level of the effect-modifying variable

-Mantel Haenszel average of stratum specific estimates is somewhere between the stratum specific OR and equal to the crude OR

Question 31

In pure confounding, _______ of an exposure on an outcome are due to ______

Answer 31

In pure confounding, all of the observed effect of an exposure on an outcome are due to the confounding variable

-OR=1.00

Question 32

In a scenario in which the third variable is neither a confounder nor an effect modifier, the association is ____

Answer 32

the same no matter how we look at it!

• stratum specific estimates are equal
• Mantel Haenszel average of stratum specific estimates is equal to the stratum specific odds ratios and equal to the crude OR (but it’s not 1)