Thread by @iskander

Fine, here we go.

First let's start with some baseline expectations. Pancreatic cancer is scary, 5y survival is ~10%.

So if you see a trial with an apparent survival asymptote at ~50% then certainly *something* is happening, right?

Well, if you look at the # at risk, we're down to *1* patient by 4y, so the confidence intervals by the end of this curve should be widening to include 0%. So let's say the 5y OS in a bigger cohort would end up 30-40%. Still has to be some kind of therapeutic activity, right?

Well, not so fast. This is a highly selected cohort!

These patients had resectable local disease (better 5y OS) and had to remain without evidence of disease for a while to get their vaccine (3/19 patients dropped, leaving ones with better odds of complete resection)

ALSO, they had to have unusually high TMB to meet the 5+ neoantigen threshold -- PDAC has an exome small variant TMB ~1-2/megabase, so total ~30-60 mutations, of which a small minority would predicted as generating a neoantigens

...and what does high TMB mean in PDAC (like in many other cancers)? That's right, better response to checkpoint blockade () -- see the dose of atezolizumab at week 6.ascopubs.org/doi/full/10.12…

Fine, maybe the 16 patients on the trial had baseline better chances than the general PDAC population *but* they stratify by response to the vaccine -- doesn't that mean it worked?

First of all -- let's be clear about something that's obvious to some and very non-obvious to others: these aren't randomly assigned arms of an RCT. We're looking at all 16 patients in a single arm and splitting their relapse KM curves by magnitude of immune response to vaccine

Got it? Not an experiment +/- vaccine but rather a post-hoc stratification that risks getting causality mixed up.

People *want* to believe that immunological responses to the vaccines are keeping cancer away but...

what if the cohort is so small that splitting it into two smaller groups risks pulling in very unbalanced prognostic factors, either by chance or by virtue of healthier patients responding to the neoantigen vaccine more easily?

So, how balanced *are* the prognostic factors between these two groups?

Well, looking at the previous paper on this trial () you'll notice a few thingsnature.com/articles/s4158…

Immunological responders *also* had:

smaller tumors

more favorable pathological staging

less lymph node involvement

favorable location of the tumor in the pancreas (head vs. body/tail).

This *alone* potentially explains all subsequent differences in RFS curves

An interesting residual question is why did patients with less aggressive or earlier-stage tumors also tend to have stronger T-cell responses to the neoantigen vaccine (and confusingly worse T-cell responses to SARS-CoV-2 vaccination).

It's an interesting scientific mystery...

...but also beside the point of whether this trial deserves all the attention it's getting as a breakthrough in the treatment of pancreatic cancer.

If you want to know whether this therapeutic concept has legs, there's a much better trial to look at...

Moderna's honest-to-goodness randomized controlled trial of aPD1 +/- their mRNA neoantigen vaccine in a cancer type: "Individualised neoantigen therapy mRNA-4157 (V940) plus pembrolizumab versus pembrolizumab monotherapy in resected melanoma (KEYNOTE-942)"

In this trial you see an actual albeit modest effect between randomly assigned arms with and without treatment.

It's honestly the only evidence I've seen that neoAg vaccination has any therapeutic activity and deserves all the attention until BioNTech does an RCT.

Lost some words in there before “cancer type”, meant to say we have a much higher a priori expectation that this might work in melanoma than PDAC both bc of the higher mutational load and experience with other immunotherapies working.

@breichholf This would also explain the small effect size: they’re putting immune pressure on some tumors through, on average, 0-2 targets — the ones with zero don’t benefit, ones with 1 can be escaped from, 2+ probably more durable

@denis_mongin I do, however, have to block *a lot* of people on here -- it's a nasty mess of hate and propaganda, so I understand people not wanting to let this place colonize their consciousness

@yummocatvibes Looked back at the patient characteristics table, it 6:2 yes vs no splenectomy in the responders and 3:5, so entirely stark but still >2x more frequent. A whole cloud of potent confounding!

@varma_ashwin97 @wgibson Of course, the numbers are very stochastic but if even a subset of patients had strong T-cell responses to ~1 presented neoantigen, then a subset of those could get effective immune surveillance, even if transient/evolutionarily unstable.