Feldmand. Module 4 - Causation, bias & confounding Flashcards
Explanations for an Association
• If we determine an association between an exposure and an outcome exists (i.e., there is a significant OR or RR), there are a number of possibilities to explain the association:
– It’s REAL
– SELECTION bias (who gets into the study) accounts for it
– INFORMATION bias (how information is collected for the study) accounts for it
– It’s CONFOUNDING the association between another risk factor and disease
– It’s due to error in conducting the study
– It’s due to CHANCE
Bias
• 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
• Two major categories
1. Selection bias
2. Information bias
Selection Bias
• Can occur whenever identification of individual subjects for inclusion results in a mistaken estimate of the measure of effect
• More simply, selection bias is a problem with who is in the study
• Some examples:
– Detection (aka surveillance or diagnostic) bias – persons followed more closely by health care providers because of some exposure are more likely to be diagnosed as a case
– Self-selection bias – subjects differentially self-refer
– Non-response bias – subjects differentially do not participate
– Inappropriate comparison group
– comparison group does not appropriately represent the population from which cases arose
Information Bias
• Systematic error in the collection of exposure or outcome data that results in a mistaken estimate of an exposure’s effect on the risk of disease
• More simply, a problem with the information you collect
• Some examples
– Questionnaire faults
– Interviewer bias: An interviewer interjects his or her bias into interview
– Bias from surrogate interviews: e.g., parent or family member
– Respondent errors: recall bias and other issues with recall
– Bias from abstracting records
– Misclassification bias: when either exposure or disease outcome is misclassified (cases are misclassified as controls or controls as cases, etc.)
Confounding
- The distortion of an exposure-disease association by the effect of some third factor – the confounder
- The association between the exposure and outcome is distorted by the association between the exposure and the confounder, and the association between the disease and the confounder
- Confounding results in a mistaken estimate of an exposure’s effect on the risk of disease
- Unlike most other types of bias, confounding can sometimes be eliminated during data analysis
To be a confounder, a variable must be:
• To be a confounder, a variable must be
– Associated with the outcome (i.e., is a risk factor for the disease)
– Associated with the exposure, but is not a result of the exposure (i.e., is not on the causal pathway from exposure to outcome)
Confounding
- An apparently strong association between exposure and outcome observed in a ‘crude’ analysis can be partially or even wholly due to confounding
- Less commonly, confounding can mask an association
- In epidemiology, confounding is a nuisance and one tries to eliminate the effects of (‘control for’) confounding
How to Control for Confounding
IN THE STUDY DESIGN: 1 - 3:
1. Randomization – not possible in observational studies
2. Restriction – if you restrict your study population to one group, then the variable that might be a confounder is no longer an issue
3. Matching – You can no longer assess the variable on which you matched as a risk factor
IN THE ANALYSIS: (4-5)
4. Stratification – relatively simple to do
5. Multivariate analysis – requires modeling with computer software
Effect Modification
• Aka interaction
• The association between exposure and disease
is different for different levels of a third variable
• Inotherwords,theeffectoftheexposureonthe
disease is modified by the third variable
• Effect modification is a finding to be reported, not a bias to be eliminated; it is a “natural phenomenon” that exists independently of the study design
Causation
- Having determined that an association between exposure and outcome is real, the next step is to consider if it is causal - does the exposure cause disease?
- Numerous criteria and theories on causality
- Causal criteria as proposed by US Dept of Health, Education and Welfare: Smoking and Health: Report of the Advisory Committee to the Surgeon General. 1964.
Guidelines for Assessing Causality
- Temporal relationship–If a factor is believed to be the cause of a disease, then exposure to that factor must occur before disease develops
- Strength of association–Measured by the relative risk or odds ratio. The stronger the association, the more likely that the relation is causal
- Dose-response relationship – As the dose of exposure increases, the risk of disease also increases. If a dose-response relationship is present, it is strong evidence of a causal relationship. However, the absence of a dose- response relationship does not rule out a causal relationship.
- Replication of findings –Expect to see the same findings in different studies and in different populations
- Biologic plausibility – The findings should be consistent with existing biologic knowledge
- Consideration of alternate explanations – Important to report that alternative explanations (bias or confounding) have been considered
- Cessation of exposure – If a factor is a cause of disease, we would expect the risk of disease to decline when exposure to the factor is reduced or eliminated
- Consistency with other knowledge – Causal relationships have findings consistent with other data
- Specificity of association – When a certain exposure is associated with only one disease
Relatively weak criterion; cigarette manufacturers use it to argue that diseases attributed to cigarette smoking do not meet the causal criteria because smoking has been linked many diseases
When specificity is found, however, it can provide additional support for a causal inference Absence of specificity does not negate a causal relationship
