Headline of the day
This is my favorite headline about the demise of drotrecogin alpha (a.k.a. activated protein C, Xigris), which was withdrawn from the market yesterday after the PROWESS-SHOCK study results revealed it to be no better than placebo in the treatment of severe sepsis. The headline, of course, implies that Xigris once was effective, it just isn’t any more…and from the quotes in this story, it appears that Lilly is arguing that improvements in sepsis management have now made it impossible to detect the incremental benefit that Xigris provided (you know, back when we were super-terrible at treating sepsis).Fair enough, but that explanation ignores the controversial history of this drug and the role of Eli Lilly in the surviving sepsis campaign. It also ignores one of the most important lessons we should learn from this debacle: the perils of the subgroup analysis. Eli, our in-house methodologist, may wish to chime in, but you can start here with a nice review that points out why subgroup analyses of clinical trials should be treated with extreme caution, and used only to generate hypotheses, not for clinical decision making (or drug approval!).
Hard for me to speak out against another entity named Eli (Lilly). However, subgroup analyses can be extremely misleading. I think of it this way:
ReplyDelete1) Benefit of an RCT is the randomization
2) Randomization, if done correctly, results in measured and UNMEASURED confounders being equally distributed between exposed and unexposed groups (Xigris and no-Xigris in this case). Thus RCTs have high internal validity.
3) When you look at subgroups, you lose the benefit of randomization. You can control for measured confounders, but not the sneaky unmeasured confounders.
4) Sub-group analyses are, thus, no longer RCTs. They have become more like observation trials. They have lower internal (causal) validity.
5) Don't rely on subgroup analyses.
How about this for a reason that "Xigris (is) no longer effective": Antibiotics kill bacteria releasing cytokines and all sorts of badness, thus, when we have no more effective antibiotics secondary to MDROs cytokines won't be released; Xigris only works when cytokines are present, thus Xigris won't work any more but it ain't Xigris's fault. How 'bout them apples?