ImetelChat

https://www.tandfonline.com/doi/full/10 ... 361#d1e285

MG your diagnosis of “paralysis by analysis” is apt but I am not yet ready to stop. No curare for us!

This in response to my questions at to what percentage of patients at enrollment were anemic:

For the Phase 3 IMpact-MF trial, the company has not publicly released a complete baseline characteristics table showing the exact percentage of patients who were anemic at study entry.

However, we can make a reasonably informed estimate from the population enrolled:

* Patients had intermediate-2 or high-risk myelofibrosis.
* All were relapsed/refractory to JAK inhibitor therapy.
* Many had already discontinued ruxolitinib or another JAK inhibitor due to lack/loss of response.
* Anemia is extremely common in this setting, especially after prolonged JAK inhibitor exposure.

Based on comparable post-JAK MF studies:

* Roughly 70–90% of patients likely entered IMpact with anemia (Hb <10 g/dL).
* Approximately 30–50% were probably transfusion-dependent at baseline.

This aligns with what was seen in:

* The Phase 2 IMbark study (the direct predecessor to IMpact).
* The MOMENTUM and SIMPLIFY-2 populations of JAK-experienced MF patients.

For your IMpact modeling, the important implication is that the trial population was not a relatively healthy post-rux cohort. It was a heavily pretreated group with substantial baseline cytopenias, meaning anemia-related mortality risk was already high before randomization. That tends to support BAT survival estimates in the low-to-mid 20-month range rather than the much longer survivals seen in broader post-rux real-world registries that include less ill patients.

My confidence that baseline anemia exceeded 75% of IMpact patients is about 85–90%. My confidence that transfusion dependence was around one-third to one-half of patients is about 70–80%. The exact percentages will likely become known only when the full baseline characteristics are published.

BP, I followed up an used my AI minions to analyze the Amenia and paper and the result aligned with yours. All of my models seem to concur that the probability of the trial being halted is about 70%. There seems to be consensus that 34-39 mOS for the general RR population won't apply to the IMPactMF trial but the change in Rux treatment regime will likely increase the mOS for both branches. The crux of the situation is that without the actual trial results we will just be guessing. In the final analysis - if we replicate the IMBark trial results - the trial will be halted - so far the odds seem to favor that outcome.

Mr G, thank you for doing that. I think the data we are finding is converging on a number that might (only in retrospect!!) turn out to be reliable. I went back to an earlier presentation linked to the Flatiron information and was able to confirm that in that study (to be clear NOT IMpact) the following appears to be true and I read the poster and confirmed:

In this Kuykendall/Palandri/Fillbrunn article, OS is defined from ruxolitinib initiation, not from MF diagnosis and not from rux discontinuation.

Exact key language from the related poster/article indexing:

“Index date: Date of ruxolitinib treatment initiation.”
“OS was defined as the time from the index date to the date of death.”

So the reported survival numbers—e.g., 37.4 months in baseline anemic patients vs 64.9 months in non-anemic patients—are survival after starting ruxolitinib.

Bottom line: this is not post-rux failure OS and is not directly comparable to IMpact-MF, where the survival clock starts after patients are already relapsed/refractory or ineligible for JAK inhibitor therapy and then randomized.

Pretty clear that it would help greatly to know the number of patients that have anemia at the time of randomization into IMpact (which we do not at present know). Also clear that for patients with anemia (presumes a substantial number of MF + anemia randomized to IMpact) and for our purposes of speculative discussion we can back out the time from Rux initiation to the time of Rux failure (roughly 15 mo in the updated post 2019 chart in the Kuykendall abstract to get to a more realistic guess of BAT in IMpact. This comes to 37.4-15=22.7 months and seems to comport with your work acknowledging that the latest Kuykendall numbers show 39.8 mos (39.8-15=24 months). This seems pretty generous and reasonable for any projections we do acknowledging that it does not account for all of the other factors we looked at e.g. “dilution” from lower classes of severity, time to cycle through other therapies (note momelitinib only available after 2023 and plays into STA’s issue about further non IMpact therapy for the BAT arm should a patient choose),

BP - My analysis consistently supports 22-26 mOS for BAT and 39-41 mOS for Imetelstat (seems reasonable but could be dead wrong). Hopefully we will get more data next week at ASCO.

Bet we have to wait for ASH or interim whichever comes first. Other forums are still suggesting we are still down the rabbit hole.

It definitely is paralysis by analysis. We’ll see the real deal data soon.

Check the Oncology Nurses summation… plain and simple, this drug is helping patients. It will be approved for use, it’s a when not an if and trying to figure out when is not possible, at this juncture.

Please remind me, if BAT has an mOS of 26 months does the model still suggest that the interim analysis favors ending the trial early for efficacy?

STA, I asked ChatGPT and Claude to evaluate the 26 mOS scenario - this is Claude's response after reviewing ChatGPT's response:

This is good work and the correction at the end is the important part. Let me run it cleanly with my own model rather than just endorsing ChatGPT's numbers.Here's my own answer, and it both confirms and corrects ChatGPT's work.

**But ChatGPT over-corrected in the other direction.** Their final table swung to implied HR of 0.41-0.44 by assuming both strong late-weighting AND treating the crude HR as requiring imetelstat mOS of 59-72 months. My model shows the BAT death count is less sensitive to enrollment weighting than their final table suggested.

**My numbers at BAT mOS = 26 months:**

Under uniform enrollment: ~62 BAT deaths, 50 IME needed, crude HR ≈ 0.40.

Under strong late-weighting (accelerating enrollment, your scenario): ~58 BAT deaths, 54 IME needed, crude HR ≈ 0.46-0.47.

The exponential and Weibull models give nearly identical results here — within 1-2 deaths.

**So the honest range at BAT mOS = 26 is crude HR ≈ 0.40-0.47**, not 0.28 and not quite as high as ChatGPT's final 0.41-0.44 midpoint either. It sits in between.

**The key point both ChatGPT's analysis and I agree on:**

Every scenario clears the boundary. Z-scores range from 3.74 to 4.62, all well above 2.34. At BAT mOS of 26 months the trial crosses the efficacy boundary regardless of enrollment weighting or survival model choice.

**The one thing I'd add for honesty:**

The crude HR here is not requiring imetelstat to hit 90+ months as ChatGPT's first pass implied. At ~54 IME deaths out of 213, the implied imetelstat mOS is more like 55-65 months on a hazard-equivalent basis — still high, still implying strong performance, but the crude-HR-to-median translation is where these models get shakiest. I trust the HR estimate more than the implied median, because the median extrapolation depends heavily on curve shape we can't observe.

Bottom line: at BAT 26 months the boundary is crossed comfortably across all reasonable assumptions. Your enrollment-weighting correction matters at the margin but doesn't change the conclusion.

Definitely trust Claude over Chat. I work 9-5 dealing with these models and can pretty much guarantee you Claude is superior…

Both of those Hazard Ratios would likely deliver a halt for efficacy. I remain suspect of that scenario, although it will be so so nice… either result should deliver a surge in this truly despicable SP. look at Revolution Medicines wirh their pancreatic breakthrough for pancreatic cancer… should move wirh 2 spikes, one at continuation one at final… although the GERN stock recent history portends to much more volatility… although I think it bears a general parallel to expect for a Ph III for a life extending, safe therapy in oncology…

I do think an mOS for BAT probably is a reasonable 26 for the purposes of hypothetical projections. This does put OS projections for Imetelstat into numbers never before reported or even contemplated. Plenty of reason for caution and continued digging.