A clarification about "prevalent user bias" - I understand the slides on how including prevalent users is a form of selection bias and causes an emulation failure per Hernan.
I saw in the answer key to HW 1, Q3, it states: "There may be unmeasured
factors influencing multivitamin use and prostate cancer risk, such as multivitamin
use in prior decades, (which would affect baseline multivitamin use and future
prostate cancer risk)". Here, prior multivitamin use is treated as a confounder.
Can we conceive of it both ways? Am I missing a nuance between examples that makes it confounding in one situation and selection bias in another?
Thanks,
Jean
Hi Jean,
In response to your question:
"Can we conceive of it both ways? Am I missing a nuance between examples that makes it confounding in one situation and selection bias in another?"
It has taken a while to respond bc I and the TA's are not sure we understand your question or point of confusion.
However, upon further reflection, I think I might get where you are coming from.
Please note - In the paper for HW1, the outcome was incident prostate cancer in generally healthy cohort of men. In HW 2, where we raised the issue of prevalent user bias/selection bias, please note the difference in study populations and outcomes. Both studies have a vulnerability for confounding. The HW2 study has more of concern for prevalent user bias leading to a selection bias, because of the study population of interest (i.e., cohort of men with prostate cancer, examining progression). Can confounding and selection bias be happening simultaneously in the 2nd study, I think Yes, and it's harder to deal with the latter.
Hope that helps,
JMC