Hi!
I am wondering if we can really have one of the 2 cause-specific death bias without the other.
Because we have 2 groups, couldn't we explain an apparent reduction in death because of X in group A compared to group B either by invoking equivalently:
Thx for the clarification!
Francois
Dear Francois and other students,
Sorry about the delay in responding. This came when I was attending and then I forgot about it.
I see your point that in some sense these two biases are complementary, but I think in practice they differ enough that the distinction is worth maintaining.
We include sticky diagnosis bias to be thorough, but I don't think it is a big problem very often. It occurs when the true cause of death is not known, which I think mainly happens in very old people.
Slippery linkage bias is a much bigger concern, because it can occur even when the cause of death is 100% accurately ascertained. In the Early Breast Cancer Trialists Collaborative Group individual level meta-analysis of 40 randomized trials cited in Chapter 6 (Lancet 2000;355:1757-70), the 20-year absolute risk reduction in breast cancer mortality of 4.8% (P=0.0001) was almost balanced by a 4.3% increase in mortality from other, chiefly vascular causes. (In fact, this effect is likely due to radiation of the heart, because it occurs more if the breast cancer was in the left breast than the right breast.)
As noted in Chapter 6, we need to be particularly concerned about slippery linkage bias when our treatments are treatments like chemotherapy, hormones or radiation, as opposed to surgery, because they may have longterm adverse effects that could get missed in trials with cause-specific mortality as the endpoint.
Tom
