Completely agree. It’s a horrible terminology that we are unfortunately stuck with for now.
Indeed, interventional data are essential. However, in the story of the HER2 “predictive” biomarker we describe in section 3.4 of the Cancers review above, important information was also derived from the “observational” correlative data.
2 Likes
I love that you brought up Buyse’s work here.
If mathematicians can rework the very foundations of the tower of mathematics, I think oncologists can at least rectify unhelpful language of this sort.
1 Like
Along with curing cancer. A gargantuan task to put it mildly. I’ll thus give my colleagues a temporary pass on this terminology
1 Like
I work largely in paediatric cancer trials, where sample sizes are almost always constrained. Yet outcomes used are very much the standard ones used in most cancer trials - event-free survival times, binary tumour responses, and the abomination of “best response” over a treatment period. In other words, we’re discarding loads of information in a setting where we should be using it as efficiently as we can.
I’ve been going on about this to my colleagues for some time (with limited progress). I really like the idea of using unified longitudinal ordinal outcomes (and Bayesian methods, naturally), but my question is what might an outcome look like? I imagine death would be the worst state, alive and cancer-free the best state, with others in between that might vary depending on the nature of the trial. For example, metastasis, recurrence (if treatment intended to cure), progression (though that’s vague and hard to define and detect), further treatment (e.g. surgery), organ preservation in some cases.
I guess one down side is that outcomes would be much more trial-specific and wouldn’t be common across many trials. Not sure that’s actually a benefit, to be honest, though.
4 Likes
Focusing only on efficacy for simplicity (no toxicity/quality of life): Death is the worst outcome and then different levels of tumor burden (as measured radiologically and then converted to standard low resolution endpoints such as response rate and progression-free survival) in ordinal scale adjusted for baseline.
Simple, feasible, and uses information we already have available.
1 Like
At a trial meeting today I tried talking to the group about the issue that “progression-free survival time” only measures the time to the first event, and ignores everything that happens subsequently. Have to say there was general incomprehension, I think people generally just default to the way outcomes in cancer trials have always been done.
So from that anecdote there is some work to do.
4 Likes