Epi Class 9 Flashcards
Confounding
distorts or hides a relationship
When two exposures are related and that makes it look like exposure A causes the disease when really it is exposure B that causes the disease.
Effect modification
real effect is different in different populations
When different subpopulations with different biological responses are incorrectly grouped (means the RR or OR does not reflect the impact of effect modification)
No 3rd Variable Effect
OR1 = OR2 = ORcrude (all equal)
Types of Bias
Selection bias
Misclassification bias
Information bias
Selection bias
error due to systematic differences between those selected for study and those not selected for study
Misclassification bias
inaccuracies in methods of data acquisition may misclassify subjects
Information bias
bias in the way that information is collected from study participants (recall bias, interviewer bias, non-response bias)
Interaction
When the presence of two risk factors at the same time causes different outcomes than the presence of either one.
Can make both risk factors appear more risky or less risky than either really is.
Interaction can be additive or multiplicative
Confounding test steps
- Confirm that the potential confounder is associated with the exposure
- Confirm that the potential confounder is associated with the outcome
- Calculate crude OR between exposure and outcome
- Stratify by the potential confounder and calculate OR (or RR) for each stratum
Mantel-Haenszel (MH) analysis
creates a summary measure that is somewhere between OR1 and OR2 and adjusts for the sample size in each stratum
Effect Modification test steps
- Confirm that the potential effect modifier is associated with the exposure
- Confirm that the potential effect modifier is associated with the outcome
- Calculate crude OR between exposure and outcome
- Stratify by the potential effect modifier and calculate OR (or RR) for each stratum
The 3rd variable is an effect modifier if:
OR1 ≠ OR2 ≠ crude OR
The 3rd variable is a confounder if:
- Stratified ORs are equal: OR1 = OR2
AND - Stratified ORs ≠ crude OR
Confounding Summary
Third variable effect:
distorts or hides a relationship
Evidence:
OR1 = OR2 ≠ ORcrude
Results of Breslow-Day test for homogeneity:
p > 0.05
- the strata (OR1 and OR2) are not different
Reporting: ORadjusted = ORmh
- use stratified analysis or multiple regression to find ORmh
Effect Modification Summary
Third variable effect:
Real effect is different in different populations
Evidence:
OR1 ≠ OR2 ≠ ORcrude (none equal)
Results of Breslow-Day test for homogeneity:
p < 0.05
- the strata (ORz and OR2) are different
Reporting:
Stratum-specific ORs: OR1 and OR2 listed separately
No Third Variable Effect Summary
Evidence:
OR1 = OR2 = ORcrude (all equal)
Results of Breslow-Day test for homogeneity:
p ≥ 0.05
- the strata (OR1 and OR2) are the same
Reporting:
ORcrude
4 Causal Relationships
- Necessary and sufficient
- Necessary but not sufficient
- Sufficient but not necessary
- Neither sufficient nor necessary
Necessary and Sufficient Causation
Rarely found
Example: exposure to an infectious agent causes disease in everyone with the exposure
Necessary but Not Sufficient Causation
Example: infectious agent must be contracted to get disease, but not everyone who is infected will become ill
Example: stages of carcinogenesis- each step is part of the causal pathway, but no one step is sufficient to cause cancer
Sufficient but Not Necessary Causation
Individual exposures are rarely sufficient, but multiple exposures could lead to disease
Example; radiation and benzene both can cause leukemia but are not necessary exposures
Neither Sufficient Nor Necessary Causation
Complex factors that contribute to chronic disease
Temporal relationship
exposure before disease.
temporal = time
Strength of association
OR, RR far from 1
Dose-response relationship
more exposure = higher risk of disease
Cessation of exposure
removing exposure reduces risk of disease
Specificity of the association
one exposure = one disease
How to reduce chance of bias
use large sample size to reduce chance of bias
Biologic plausibility
makes biological sense.
shark/ice cream
Replication of findings
other studies show the same thing
Consistency with other knowledge
Corresponds to other information published by the field
