Replies to post #3 on Galera Therapeutics Inc (GRTX)

[PlentyParanoid]

Oh my, I see some familiar aliases here. Good. I can then address this rant to you, Loanranger. In the past, you have shown remarkable tolerance of my messages, even long ones.

I have, or better said, have tried to do some checking of Galera's claims found in the report from GC4419 PIIb trial for SOM. I got outright giddy when I noticed that they have included SOM incidence and duration swim-lane chart. Hah, nitwits are providing me the means to do thorough checking of ALL REPORTED VALUES. I should have known better.

Obvious sign of trouble besides missing headcounts in table 2 is that in the said table Galera reports median duration of SOM for GC4419 90 mg dose as 1.5 days. If you check the swim-lane chart you see that for 90 mg dose there is total 35 SOM incidents. Headcount for intent to treat group for 90 mg dose happens to be 76. How can one get SOM median 1.5 days when less than 50 % of subjects did ever experience it? Of course, one can't. More misery can be found in the statistical methods section:

Duration in patients with unresolved SOM as of the last evaluation was imputed as the median duration among patients in the same treatment arm with at least that duration. For patients without observed SOM but with incomplete follow-up, duration was imputed as the median duration among patients in the same treatment arm who were free of SOM for at least that length of follow-up.

So, even if the values in table 2 are for ITT, as they should be, they are after ‘imputation’ of missing values, which I obviously don’t have access to. At this point you should exclaim (Please, do exclaim. It would make me feel better knowing you did):
“WHAAAT! Is this proper, legit, does it cut the mustard!”

Unfortunately, it does. FDA uses similar methods, especially with heavily censored data, to estimate what might be more proper values. Then, if their findings do indicate so, they may tell the applicate: “Nice try, but no cigar this time.” If FDA can use a method to approve or reject drug application, they hardly can object to applicant using it. Also, it is approved and frequent practice in academic circles.

The above means that my grand analysis plan went the usual way, into garbage. But I can squeeze something out of those swim-lanes. They probably are for subjects for which Galera had complete SOM history, before applying what I call backfilling. That would mean, according to Galera, that they should cover about 80 % of trial subjects. That would be good enough. Unfortunately, it quickly became obvious, mostly comparing with figures 2 and A1 (in appendix) that swim-lanes probably are not completely accurate even for those subjects. Oh darn!

Finally, I decided to use values that looked the most reliable (very relative term, in this context): SOM grade > 2 and grade 4 start times. And applied the least sensitive measure I could on them, that is: I calculated risk ratios.

Log-Rank risk ratio for SOM grade 4 start times:
0.56 with 95 % confidence interval [0.3 , 1.08] favoring GC4419
Log-Rank risk ratio for SOM grade > 2 start times:
0.67 with 95 % confidence interval [0.42 , 1.04] favoring GC4419
Risk ratio for SOM grade 4 incidence rates during IMR treatment (first 7 weeks):
0.53 with Miettinen-Nurminen 95 % confidence interval [0.28 , 0.98]

The last one favors GC4419 quite strongly, confidence interval does not include 1. Other confidence intervals do.

About above results I trust more numbers for grade 4 SOM. Those subjects are less likely to be lost for follow-up. They tend to be in hospital bed and in IV. Another thing: I have found that risk ratio below 0.7 in relatively sized trials is quite good indicator for successful phase 3. Of course, it is not infallible, especially with Galera data here.
So, here you are. Just remember to treat my results as you would a rabid cat – with extreme caution.

BTW: a link to the article in question. In case you don’t have it handy.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6881100/