Epidemiology: Understanding Bias and Systematic Error

Epidemiological Errors and Biases

Fundamentals of Epidemiological Error

• Bias: A systematic error that results in an incorrect estimate of the association between exposure and disease.
• Random Error: Fluctuations around a true value due to chance. These can be reduced by increasing the sample size.
• Systematic Error: A flaw in the study design or conduct that pulls results away from the truth in a specific direction. Increasing the sample size does not fix this.

Understanding Selection Bias

Selection bias occurs when the relationship between exposure and disease is different for those who participate in the study versus those who theoretically would be eligible but do not participate.

• Non-response Bias: This occurs when those who choose to participate differ significantly from those who decline (e.g., healthier people are more likely to volunteer for a fitness study).
• Detection Bias: This occurs when exposed individuals are monitored more closely by doctors, leading to a higher rate of diagnosis than the unexposed.
• Loss to Follow-up: A major issue in cohort studies where participants drop out. If those who drop out have different risks or outcomes than those who stay, the results are biased.

Misclassification and Information Bias

Misclassification bias occurs when the measurement of exposure or outcome is inaccurate.

• Recall Bias: Common in case-control studies; "Cases" (sick people) usually rummage through their memories more intensely for "why" they got sick compared to healthy controls.
• Non-differential Misclassification: The inaccuracy is the same across all groups (e.g., a faulty scale that adds 2 lbs to everyone). This usually pulls the result toward the "null" (making it look like there is no effect).
• Differential Misclassification: The inaccuracy differs between groups (e.g., researchers probe the "exposed" group more intensely for symptoms than the "unexposed" group). This can inflate or underestimate the true effect.

Confounding Bias and Causal Pathways

Confounding is a "mixing of effects." A third variable (the confounder) makes it look like the exposure is causing the outcome, but it is actually the third variable doing the work.

Three Criteria for a Confounding Variable

1. It must be associated with the Exposure.
2. It must be a risk factor for the Outcome (independent of exposure).
3. It must NOT be on the causal pathway (it is not a step between the exposure and the outcome).