Week 5 Bias and Confounding

Question 1

What is internal Validity?

Answer 1

The extend to which the causal correlation stated is warranted

Internal validity is necessary for external validity

The presence of Systematic errors (e.g. bias and confounding) reduce internal validity

These factors lead to either an over- or under-estimation of the true association between exposure and outcome

Question 2

What does systematic error affect?

Answer 2

It affects internal validity. It can be avoided via careful study design, and compensated for (to a lesser degree) in the analysis. It does not change with increases in study size because the error is SYSTEMATIC

Question 3

What does random error affect?

Answer 3

It affects the reliability and precision of the estimate. Can be somewhat reduced with increasing sample size

Question 4

What is Bias?

Answer 4

Any systematic error in the design, conduct, or analysis of a study that results in a mistaken estimate of an exposure’s effect on the risk of disease

It is important to be able to spot systematic errors in a study

Question 5

Selection Bias

Answer 5

Occurs as a result of using improper procedures to select participants

Also occurs as a result of the kinds of factors that influence continued participation in the study

Selection Bias = when the difference between exposure and diseased observed in the study participants, is different from the difference in the target population

I.e. the participants fail to represent the population

Question 6

Example source of Selection Bias

Answer 6

E.g.
Individuals who volunteer may possess different characteristics to the average person in the target population

In general, the kinds of people who respond to a study can differ from those who do not in terms of demographic, cultural background, SES, lifestyle, behavioural and medical characteristics

Thus it is important to identify non-responders as much as possible***

This can occur in case control and cohort studies

Question 7

Selection Bias: Ascertainment or detection bias

Answer 7

Occurs when there are different inclusion/exclusion criteria between the cases and controls.

Cases and controls should always be matched as closely as possible to eliminate bias

Question 8

Selection bias: Loss to follow up

Answer 8

The reasons who participants might be lost to follow up may cause bias in the study outcomes (reducing internal validity)

E.g. If testing a new drug, the sickest people might withdraw, leaving the healthier people in the study. This will lead to a false outcome - bias.

E.g. smokers might be harder to follow up on (due to other associated variables such as lower SES, etc.) and they might be the ones that are developing the condition you’re interested in (e.g. COPD). Thus, loss of smokers may lead to a Type II error…

Question 9

How can we control Selection Bias?

Answer 9

Clearly defined selection criteria that does not differ between case and control groups

Trying to gain some information about non-responders (e.g. trying a phone call questionare, or visint with a questionnaire, etc)

Using external comparisons (e.g. national statistics)

Question 10

What is Information Bias?

Answer 10

Arises when study variables such as exposure and disease/outcome measurements are inaccurately measured or classified

Errors in the gathered information may lead to incorrect conclusions (either Type I or Type II error) due to information bias

Question 11

What are the types of Information Biases?

Answer 11

Misclassification Bias

Differential Misclassification (non-randon) Bias

Non-differential Misclassification (random) Bias

Recall Bias

Question 12

What is misclassification Bias?

Answer 12

Information Bias

Occurs when either the ‘exposure’ or ‘existence of disease’ is incorrectly measured, resulting in the participant being wrongly classified

E.d. Incorrectly concluding a ‘case’ participant as being exposed more than controls, when they weren’t really

Question 13

What is differential (random) misclassification?

Answer 13

Information Bias

(or non-random)

When misclassification occurs unequally throughout the case and control groups.
E.g. if looking at effect of smoking on LC, a differential misclassification is when 20% of true smokers were incorrectly classified as non-smokers in the study, and only 3% of non-smokers were incorrectly identified as smokers in the study

Differential misclassification always results in an underestimation of the true association

Question 14

Non-random Misclassification?

Answer 14

Information Bias

When misclassification occurs in the same proportion in each group.

Results in either an over- or under- estimate of the true association

= an underestimate if fewer cases are considered exposed, or if fewer exposed are considered diseased (failure to recognise causation or association)

= an overestimate is more cases are considered exposed, or if more exposed are considered diseased

Question 15

Recall Bias?

Answer 15

Information Bias

non-random

Occurs when there is a difference in how accurately an exposure is recalled by cases and controls

E.g. in a study of congenital malformations, mothers of malformed babies may recall past exposures more thoroughly than mothers of healthy babies.
The adverse pregnancy would have served as a stimulus for the mother to consider potential causes, whereas mothers of healthy babies may not have had reason to consider such things

Question 16

Interviewer of observer bias?

Answer 16

Information bias

Non random

Occurs when one group is followed more closely than the other.

E.g. diagnosis of emphysema is more likely to be made and reported in exposed individuals, because clinicians are more likely to look for evidence.

Result = a misclassification of outcome ‘cases’ that occurs differently in smokers and non-smokers - therefore it is non-random because fewer unexposed individuals are being systematically underdiagnosed

Question 17

Reporting Bias?

Answer 17

Information Bias

Random

For various reasons, including social desirability, participants may be unwilling to report an exposure accurately

May result in under-reporting of alcohol consumption, number of cigarettes they smoke, the illicit drugs they use, etc

Question 18

How to Avoid Bias?

Answer 18

Use well standardised protocols

Train interviewers and use standardized data collection methods to eliminate differences between groups

Use other sources to verify information - e.g. biological evidence, asking spouse

Careful attention to design of the study - use a smaller pilot study to identify potential problems

Question 19

What is confounding?

Answer 19

COnfounding is the confusing, or mixing of effects

Also a form of bias that can over- or under- estimate effect

Question 20

What are the criteria of confounders?

Answer 20

1) the confounder must be a known risk factor for the disease
2) it must be associated with the main exposure, but not as a result of the exposure
3) It must have an effect and be imbalanced between exposed and non-exposed groups

E.g. the prevalence of down syndrome in terms of birth order may make it look like 5th-born children are much more likely to have it than first-born children.
However, this is a confounder, because in reality, the causal factor is maternal age.

Question 21

How to avoid confounding?

Answer 21

Through Study Design:

Randomisation

Restriction: subjects with known confounders are not allowed to participate

Matching: subjects in one study group with a confounder could be matched with the same confounder in the other group

Collect additional information

In Study Analysis:

Statistically adjust: