confounding factor

[cage92]

hey i have problem in understanding some concept in confounding
when doing stratification since the smoking is the confounding why it does not cause bladder cancer ?the diagram show that either in presence or not no cancer is present

---

Members don't see this ad.

 

[Missorleans]

Because then it would be an effect modifier.

A confounder makes it appear that there's an association in crude analysis, then when you stratify you find that there actually is none, in either level of stratification.

An effect modifier makes it look like there's a significant association in crude analysis, but then when you stratify by another exposure, you find that not all stratification levels have a significant association. In other words, there is a significant association, but it's all due to this effect modifier, not the exposure you were initially looking at.

 

[cage92]

but why the confounder in staratification even in presence of alcohol like the first study has no effect? how it exert its effect above and not below

 

[SBR249]

but why the confounder in staratification even in presence of alcohol like the first study has no effect? how it exert its effect above and not below

You misunderstood that diagram. Under confounding, it's saying that in crude analysis, there appears to be an association between alcohol use and bladder cancer. However when you stratify the sample population based on smoking status, there's no association between alcohol use and bladder cancer in either the smokers or non-smokers.

The basic principle is that you'd have 2 paired control/test groups:

A) smoker+drinker :: smoker+non-drinker (<- no difference in bladder cancer rate WITHIN the pair = no association)
B) non-smoker+drinker :: non-smoker+non-drinker (<- no difference in bladder cancer rate WITHIN the pair = no association)

^There IS a difference in bladder cancer rate between A and B (which are the different "strata").

So when you look at each pair independently (aka when you've stratified your sample population), the differences in bladder cancer rates between members of the SAME pair disappears. However, there may be significant differences in bladder cancer rates between DIFFERENT pairs, the implication being the difference between different pairs is the confounding factor. In the example above, the confounding factor is smoking status.

Edit: in short to answer your question directly, the confounder has no effect in the bottom portion of the diagram because it is either present or absent in both control and test groups which is what stratification is designed to do -> separate cases based on a third variable aside from the dependent and independent variables.

 

[cage92]

so now i am getting the idea but correct me if i am wrong since smoking is the confounder so now we stratify in a way to detect if alcohol can cause bladder cancer in smoker patient? which cant so the idea is that the main cause of cancer is smoking?

 

[SBR249]

Yes, you do the stratification to allow you to determine the true contributions due to alcohol. If alcohol did lead to increased additive risk, you'd see the association being preserved for all the strata. So the idea is that because of the confounding factor, your crude analysis was comparing a heterogenous population that masked the true cause and the "association" that you got was not really due to the causal factors that you tested.

 

[cage92]

thannk you very much