Confounding Flashcards
What is confounding
Confusion or mixing of effects; effect of the **exposure is mixed togethr with the effect of another variable, **leading to bias. Confounding is confusion, or mixing, of effects; the effect of the exposure is mixed together with the effect of another variable, leading to bias”
Conditions to be met for a variable to be a confounder
- Known risk factor of the disease e.g. Smoking (confounding is a risk factor for cancer, in Coffee (exposure) causes Pancreatic Cancer (outcome) scenario.
- Confounding is associated with the exposure but not a result of it . e.g. Smoking (confounder) is common among coffee drinkers but smoking is not a result of drinking coffee.
Classical approach to confounding
- Confounder is **casually or non-causally **associate with the exposure.
- Confounder is** causally associated with the outcome**
- Confounder is not on the causal pathway
Causal inference approach
- Confounder is a common cause of both exposure and outcome.
- Confounding occurs when we dont adjust for this variable.
- Confoundign is not on the causal pathway.
What happnes if we do not control for confounding?
- We may see an association when there is none
- We may see No associaiton when there is one
Types of confounding
- Positive (overestimation)
- Negative (underestimation)
- Qualitative (flips the direction of associaiton)
Positive confounding
Seeing association when there is none
Crude RR = 5
Adjusted RR = 1
1. Crude RR is overestimated (5 times away from the null value of 1).
2. In reality there is no association between exposure and outcome (adjusted for confounding RR = 1)
3. Positive confounding Crude and adjusted always More than null value (null = 1)
Partially Positive confounding
Seeing association when there is none
Crude = 2
Adjusted = 1.5
1. Crude overestimates exposure outcome relationship by 2 times but in actuality after adjustment it is only 1.5 times.
2. Crude is away from the null, adjusted is going towards the null value.
3. Both Crude and adjusted are greater than 1
Negative confounding
Seeing No association when there is one
Crude RR=1
Adjusted RR = 4
1. Underestimating the effect of the exposure outcome relationship. Infact there is 4times more likely for exposure to cause outcome but the Crude mentions only 1
2. Crude RR closer to 1, adjusted RR greater than 1.
Both still 1 or more.
Qualitative confounding
Qualitative confounding (extreme case): the confounding effect results in an inversion of the direction of the association.**Crude RR = 2.5 **meaning exposure causes the outcomes by 2.5 times
Lets consider exposure has a drug and outcome as death
Meaning exposure causes a BAD outcome
Adjusted RR = 0.5
- Flips the relationship of the Crude RR;
- Turns out exposure is not a “risk factor” to death but** infact a protective factor**.
- Adjusted RR is below the null value
Positive
Negative
Qualitative
Positive - Crude RR greater than null value (away from null e.g. 5) ; adjusted RR (closer to null e.g. 2)
Negative - (Crude RR closer to or equal to null e.g. 1 or 2) ; adjusted RR (away from the null value e.g. 4)
Qualitative - (Crude RR away from null value e.g 4 ); adjusted RR below the null value e.g. 0.5) We were sayign exposure likely to be associated wiht the outcomes 4 times whch is a Risk but turns out the exposure is actuallya protective factor.
Confounding vs Selection Bias
- Confounding freely occurs in Nature
- Selection Bias is a result of incorrect participation selection in the study
Mediator in a causal pathway
Third variable on the causal pathway . Exposure ——(mediator)——–»»»Outcome
Eg. Unprotected sex ———–»> Cervical caner; in this scenario HPV comes in between the two variables.. Unprotected sex can lead to contracting HPV which causes Cervical cancer
Metformin compliance————–»»lowers blood sugar ; Metformin in the blood stream is the mediator .. following metformin compliance metformin enters the blood stream eventually reducing blood sugar.
A mediator variable does not “cause” the exposure. It is not causally linked.
Essence of confounding
A confounder must differ in its distribution across categories
*
e.g., distribution of confounding variable differs in exposed and unexposed groups in a cohort study
*
e.g., distribution of confounding variable differs in experimental and control arms in an intervention study
*
e.g., distribution of confounding variable differs in cases and controls in a case-control study
If there is no difference in the distribution of a potential confounding variable, then confounding does not happen
does confoundign exist in nature?
yes, but selection bias is introduced when sampling
