Response Rate Tricks in Oncology: RAS inhibitors on the spot again...

Response Rate is a measure of anti-cancer drug activity. It's more reliable in early phase trials than PFS estimates ... yet to be analyzed with great scrutiny!

In early phase trials, one question is drug safety, obviously. Another question is drug’s activity, which is its ability to shrink the cancer according to some thresholds. They are best captured by what is called “response rate”.

Response Rate… the RECIST criteria

The response rate is the percentage of patients who will present a “response” (i.e. tumor shrinkage), at some point under a given therapy.

• first: a patient must “respond” according to RECIST criteria

Overall Response Rate (ORR) is the percentage of patients that will present either a CR or a PR according to these criteria.

• second: the response needs confirmation

A Novel RAS Inhibitor - Is it a “revolution”?

Recently, results from a novel generation RAS inhibitor compound - RMC-6236 - were presented. In patients in second line metastatic pancreatic cancer, PFS were reported as much higher than “benchmark median PFS” (panel below).

Selection bias: The problem with that interpretation is that it fails to account for the fact that patients enrolled in early phase trials are healthier and have more indolent disease biology than patients enrolling in larger phase 3 trials, and even more so when compared to real-life patients.

In other words, PFS metrics from those studies don’t tell you anything about the drug's efficacy. How it would compare against standard of care therapy in a randomized trial is unknown.

A metric that is not perfect but reflects drug activity more accurately is the response rate.

Response Rate Tricks

Here is a key figure from the entire presentation by the drug company (available here). The waterfall plot shows the best percentage change in tumor dimensions from baseline. In other words, bars falling below 30% (those on the right) represent patients with a response. The more the plot falls early and deeply, the better the drug is. Does this waterfall plot look good? The ORR reported (20%, 26%, 27%, etc.) are not very high but seem to fare better compared to benchmark data (9%).

As seen, they report different type of ORRs (overall response rates).

• ORR 14 weeks exclude many patients (to allow 2 potential scans)

• ORR 20 weeks exclude even more patients (to allow 3 scans)

• by doing this: you litteraly decrease the denominator and artificially increase your response rate.

Let’s do some maths: the total number of patients included was 127, only 92 are represented on the waterfall plot. Of those 5 are not displayed “due to lack of past-baselin target lesion assessment”. Also, amoung the 22 patients “responding” in the waterfall plot, 5 have unconfirmed response.

In other words, only 17 patients, out of 127 included patients, had a confirmed response… ORR is now … 13.4%! Not so far from “benchmark”, no?

Concluding thoughts

We have studied extensively the development of sotorasib, the first KRAS G12C inhibitor to be FDA approved, which was a failure of drug regulation (here is our first analysis of the CodeBreaK200 results, here is an overview of the regulatory failure). Sotorasib had a 37.1% ORR in the early phase CodeBreaK 100 trial, which later decreased to 28.1% in CodeBreaK 200.

• don’t be fooled by PFS estimates in early phase trials, they are driven by selection bias, with highly selected patients not mirroring real-life patients

• response rates (RR) are arbitrary metrics which are not directly related to patients’ outcomes

• arbitrary rules can be applied to present RR in waterfall plots

• always check for the real denominator

• always check for unconfirmed responses

• RR can decrease from early to later phase trials

If you like this post, become a free or paid subscriber of the drugdevletter.com for more content.