Regular ecancer blogger Dr Bishal Gyawali takes a look at the IMpassion130 trial in the form of a conversation between two doctors, A and B:

A: Which is the only drug in cancer history to have its accelerated approval revoked?

B: Bevacizumab.

A: Which tumour type was that?

B: Triple Negative Breast Cancer.

A: Why was it revoked?

B: Because it didn’t improve OS, although it did improve PFS by  nearly 6 months!

A: By the way, what’s the recent immunotherapy trial that people think is gamechanger in breast cancer?

B: You mean Impassion130? The atezolizumab plus abraxane trial?

A: Yes, that one. You think it’s a great regimen?

B: Yes, I think it’s revolutionary.

A: Which tumour type was that?

B: Triple negative breast cancer.

A: Improved PFS?

B: Yes, by 1.7 months.

A: Only?

B: No, but it improved by 2.5 months in PD-L1 positive subgroup.

A: Ok, but still, 2.5 months doesn’t seem great. You said even bevacizumab improved by 6 months!

B: But immunotherapies don’t improve PFS much.

A: Then why did you make it the primary endpoint?

B: We also added OS as primary endpoint later.

A: So OS was improved?

B: Actually, there was no significant difference in OS.

A: What?

B: But it did improve OS in PD-L1 positive subgroup of patients.

A: How do you know that?

B: What do you mean? We tested for …

A: Wait.. I thought, based on the protocol, you could test for OS difference in PD-L1 positive subgroup only when you found a difference in the overall population, right?

B: Well, you are right, but…

A: So basically, if we have to put if truthfully, we can only talk about the overall population now. And for the overall population, this regimen improved PFS but not OS among patients with triple negative breast cancer, correct?

B: Umm, I guess so.

A: Bevacizumab did the same-improved PFS but not OS, right?

B: Yes. But….

A: I know. The signal in PD-L1 subgroup of patients looks very impressive. But we did that analysis violating our own protocol, and now we are heavily confused. I’d put my enthusiasm on hold until the data are mature. I’d follow my protocol and look into PD-L1 subgroup only if there was signal from the overall population in the mature OS data. If not, maybe worth testing in PDL1 subgroup specifically?

B: But…..

The author has no conflicts of interest. The fictional characters A and B discuss about an actual trial IMPassion130 published here.