First, remember that progression free survival is a time to event, composite endpoint
It is typically the time until 1 of 4 things happen
Death, New lesions, Growth (without shrinkage) or Growth (from Nadir).
Watch my free lecture series to learn more:
As a result PFS is binned
It has a stair-step pattern b/c imaging is assessed not continuously, but at pre-specified time points
One can imagine that as a result: how often you assess the endpoint is important.
Very frequent assessment is rare in the messy reality of life, but might be common in trials
This can turn even trivial changes in tumor growth in stat. significant p values of <0.05!
aka $
We assessed how often PFS was checked in trials in our paper
Indeed it is checked frequently. Often less than 8 weeks is mandated for protocol specified scans.
Relentless scanning may be a tactic to find significant results for small gains
In many cases, the frequency of scanning in trials is more than guideline recommendations
Our study could not find a difference in the hazard ratio comparing different trials with different scan intervals, but this caveat is important to consider 👇👇
A future study should compare real world scan intervals to trial scan intervals. If anyone has access to data & is interested; email our lab, we would love to collaborate @vkprasadlab
Our new paper is now out in European Journal of Cancer
We analyze 55 cancer drugs that target genomic abnormalities & assess the evidence
Only 24% reported an improvement in survival 👇
A seductive mechanism of action apparently means low levels of evidence
[thread]
Modern oncology has several classes of drugs
Cytotoxic drugs
Checkpoint inhibitors
CAR-T therapies
General targeted drugs &
Drugs that target specific cancer genomic abnormalities
(Genome drugs!)
Genome drugs get outsized attention; Previously we found that (best case scenario) 13% of US cancer patients were eligible for these drugs; leaving 87% not eligible
Our new paper in @EJCI_News argues that Randomized trials are necessary in medicine & PH for interventions w putative benefit & at best MED to LG effect size.
Parachutes & smoking are not good counter examples
Some people argue that b/c we did not need RCTs to know smoking is harmful or Parachutes are life saving, we don't need them to test cloth masking, or the Impella, or some new cancer drug, or HCQ, or <insert ur favorite practice>
But this is based on misunderstanding
There is a huge range of things we can do to someone that might hurt them or save them, imagine the spectrum (below)
Now out in @EJCI_News
Logan Powell & I show where randomized trials necessary
When people say 'we don't need an RCT of smoking (to prove harm) or parachutes (to prove benefit)' does that apply to widespread medical interventions?
🧵 onlinelibrary.wiley.com/doi/10.1111/ec…
Led by Timothee Olivier, our new paper is now out at @JAMANetworkOpen
We analyze 12 years of FDA approvals, and do the hard work of sorting them into
New Mechanism of Action (MOA)
& New MOA for that tumor type
Vs next in class