Bias and Confounding

Question 1

What is bias ?

Answer 1

Systematic error in study design, data collection, analysis, or interpretation that leads to incorrect conclusions

Bias can result in overestimation, underestimation, or complete misrepresentation of associations between exposures and outcomes

Bias will systematically skew your results in 1 direction

Question 2

Describe selection bias and the different types:

Answer 2

The association between the exposure and the disease is different for those who participate compared to those who should be theoretically eligible for inclusion in the study

Sampling bias - occurs when the whole population is not selected or when a random sample of subjects has not been taken

Participation bias - subjects who participate may differ in some way to those who do not participate, this who participate tend to be healthier meaning results aren’t representative

Referral bias - selective referral of exposed cases, selected cases may have a stronger association with the exposure than all cases in the population

Diagnostic bias

Self selection bias - people refer themselves to the study

Prevalence/ incidence bias - prevalent cases may have changed their lifestyle once they have been diagnosed, represent survivors of the condition and they may not be typical of all patients

Survival bias - occurs when only survivors of a highly lethal disease i.e. oesophageal or pancreatic cancer, enter the study

Question 3

Explain how to minimise biases in epidemiological studies:

Answer 3

Selection bias mitigation:

• ensure random sample is taken to ensure a representative study population
• implement high follow-up rates (>80%) to reduce attrition bias
• utilise matching and stratification in case-control studies to balance confounder

Information bias reduction:

• Use standardised protocols and objective measurements
• Implement blinding (single, double, or triple-blind studies) to prevent interviewer and observer bias

Try to gain a high participation rate

Must be done in the study design stage as selection bias

Question 4

Describe the role of confounders in observational studies:

Answer 4

A confounder is a third variable that distorts the true relationship between an exposure and an outcome

Confounding can cause;

• underestimation of the association
• overestimation of the association
• reverse the direction of the association (Simpson’s paradox)

Criteria for confounding:

• associated with the exposure
• associated with the outcome
• not an intermediate step in the causal pathway

Question 5

List how to control for confounding in the design and analysis of observational studies:

Answer 5

Randomisation - distributes confounders evenly between groups in large samples, eliminates both known and unknown confounders

Matching - ensures equal
representation of subjects with known confounders in study groups. It has to be coupled with matched analysis

Restriction - only include participants within a specific range of the confounder, effective but reduces sample size and therefore statistical power

Statistical Adjustment during analysis:

Question 6

Explain the concept of unknown / unmeasured confounders + residual confounding:

Answer 6

= confounders that are not identified or cannot be measured, leading to residual confounding

Residual confounding - the distortion that remains in the association between a disease and an exposure after controlling for confounding in the design and/or analysis of a study

Methods to address unknown confounders;

• randomisation
• sensitivity analysis
• negative control studies

Question 7

Describe effect modification and how it differs from
confounding:

Answer 7

Effect modification - situation where the effect of an exposure on an outcome varies across different levels of a third variable

Unlike confounding, effect modification is a real biological interaction, not a bias

Stratified analysis shows different effect sizes across subgroups

Question 8

Describe the difference between bias and chance:

Answer 8

Bias:

• systematic error
• errors will not cancel each other out whatever the sample size
• inaccurate results

Chance:

• random error
• errors will cancel each other if sample size is large enough
• imprecise results

Question 9

Describe information bias :

Answer 9

Bias in obtaining information from participants

Non-differential misclassification - measure of association biased towards the null, occurs equally in exposed and unexposed

Differential misclassification - degree of misclassification differs systematically between exposed and unexposed

Question 10

Describe the different types of information bias:

Answer 10

Observer/ interviewer bias - Researchers unintentionally influence participants’ responses

Recall bias - differences in memory between cases and controls (e.g., cancer patients recall past exposures more accurately than healthy individuals)

Reporting bias - subjects not willing to report an exposure accurately because of social desirability, sensitivity

Surveillance/ detection bias - non-random misclassification, one group followed more closely than the other

Misclassification bias - incorrect categorisation of exposure or disease status (can be differential or non differential)