1.5 Confound & Bias

Question 1

Describe Internal Validity

Answer 1

Do the results reflect the true situation in the sample. Have alternative explanations for the results been dealt with appropriately

Question 2

Describe External Validity

Answer 2

Can we generalise the results to the broader population of interest. Is the sample representative of the broader population?

Question 3

What is Bias?

Answer 3

Distorts the causal association between exposure and outcome

Question 4

Describe Selection Bias

Answer 4

Systematic difference between people included and excluded in a study. (e.g. differences between people who volunteer for a psych study than those in the general population)
Leads to: non-generalisasble results (external validity) because our sample isn’t representative

Question 5

What measures can we use to control Selection Bias?

Answer 5

Random selection/allocation
Clearly defined case-definition
Clearly defined eligibility (inclusion/exclusion criteria)
Clearly defined target population
Strategis to ensure high participation rates
Strategies to maximise retention of participants

Question 6

Describe Measurement Bias

Answer 6

Errors in measurement, leads to misclassification of exposure status and/or disease status.

Question 7

What are the two types of error measurement bias can lead to?

Answer 7

Random Error and Systematic Error

Question 8

Describe Random Error

Answer 8

It is associated with the precision of the instrument used for measurement. There is a random variation in measurement. Repeated measurement can help reduce random error. Imprecise. Should not overly effect results because true random error should ‘cancel out’
e.g. Calculating weight on old analog scales and not digital.

Question 9

Describe Systematic Error

Answer 9

Fault or accuracy problems with the measurement (e.g. scales not correctly calibrated). Innacurate. Can under or over estimate the true population mean value.

Question 10

Name the two types of errors in classification

Answer 10

Differential

Non Differential

Question 11

Describe ‘Non-Differential’ Errors

Answer 11

Misclassification errors are the same in all the groups being compared. e.g. all measure with the same tape measure and scale.

Question 12

Describe ‘Differential’ Errors

Answer 12

Misclassification errors different in groups being compared. i.e. misclassified one group. e.g. Self-report BMI. Those with higher BMI likely to underestimate their BMI but people with normal weight will report correctly.

Question 13

How can we control measure bias?

Answer 13

Clearly defined exposure / outcome (case-definition)

Quality of measurement device (recall bias, calibration, standardised interview, structured questionnaires)

Question 14

Define Confounding

Answer 14

The distortion of the effect between exposure and outcome due to the association of the exposure with other factors that influences the outcome

Question 15

What is a confounder?

Answer 15

A variable that can be used to decrease confounding bias when properly adjusted for.

Question 16

What are the criteria for a confounder?

Answer 16

Associated with the outcome
Associated with the exposure
Not be an intermediate between exposure and outcome (i.e not lie on the causal pathway)

Question 17

Give an example of confounding.

Answer 17

An association between alcohol intake and lung cancer is found. However drinkers are more likely to smoker and there is a known association between smoking and lung cancer. Thus smoking is confounding the association between alcohol and lung cancer.

Question 18

What is the rule for identifying confounders?

Answer 18

If the difference between crude and adjusted effect +/- 10% it is a confounder

Question 19

What is an effect modifier?

Answer 19

A variable that differentially (positively or negatively) modifies the observed effect of a risk factor on disease status.

Question 20

Give an example of an effect modifier

Answer 20

Breast Cancer occurs in both men and women but at a 800fold increased rate in women. Thus gender interacts with other risk factors for breast cancer.

Question 21

How do we determine if a variable is a confounder or an effect modifier?

Answer 21

Calculate the Crude RR.
Stratify and calculate the stratum specific RR.
If the stratum specific RRs are DIFFERENT to each other, then there is an effect modification and we therefore use the stratum specific RR.
However, if the stratum specific RRs are SIMILAR to each other then there is no effect modification. If the Crude RR is similar to the Adjusted RR then there is no major confounding and we use the Crude RR. If the crude RR is different to the adjusted by =/- 10% then there is major confounding and we use the Adjusted RR.