ImetelChat

Claude/Fable is here. Let’s hope it is not a fabulist.

BP, Fable sounds dangerously smart. I tried to use him yesterday but I kept being pushed back to Opus (ostensibly because Fable was not cleared to address the issues I presented). Regardless, Claude has done a pretty good job. AI is very good within boundaries that are well defined but they have their own programmed cognitive bias which can cause them not to be 100%s reliable. AI hallucinations are common as well. They are very good with complex formulas and research so they've been very helpful with IMPactMF. Claude and I will probably put together a full scale end-end analysis of IMPactMF when I get the time. We recently determine that the Final projection date of the 2nd half of 2028 is a significant tell on the overall death rate - which further confirms that patients are living a great deal longer.

Patients in both arms? Has that changed your opinion of the interim analysis?

STA - I asked Claude to break it down to Hypothetical (guesses) in terms of how many BAT patients from each sub cohort may be alive on 9/26.

The inputs I'm using (all estimates):

BAT_mOS_ BAT_Deaths 9/2026 Pct_BAT Dead Imetelstat Deaths Imetelstat_Pct_of_213 Rough_Implied_HR
14 months (historical benchmark) 87 81% 25 12% 0.15-0.25
24 months (moderate era-drift) 70 65% 42 20% 0.40-0.50
30 months (aggressive era-drift) 63 59% 49 23% 0.50-0.58
40 months (breakeven) 50 47% 62 29% 0.60-0.62

The above tables were based on this enrollment timing. Enrollment: started ~Q1 2021, 50% enrolled by 11/23, ~80% by 2/25, complete ~Q3 2025. So enrollment ran ~4.5 years, front-loaded-ish but with a long tail.

Given that this unofficial - everything here is non validated other than the trial parameters and Geron's announcements, a fair bet is that ~ 60% (+ or - 10) will be dead on 9/26. So that leaves 44 BAT patients alive on 9/26.

By the September 2026 interim, the BAT group is mostly spent — but not evenly. The patients still alive are overwhelmingly the late enrollees, the ones who joined the trial in 2024 and 2025 and simply haven't been in it long enough to reach the end of their road. The early enrollees — the 2021 and 2022 patients — are nearly all gone by now, because they've been followed for four to five years against a disease that kills in roughly two and a half (even under the generous 30-month assumption). So of the ~44 BAT patients still breathing after the interim, the large majority sit in the two youngest enrollment groups.

That tells you exactly where future BAT deaths come from, and how few there are. Those ~44 survivors die off slowly — only about two or three each quarter — trickling down toward the trial's 2028 final. Over the two years between the interim and the final, the BAT arm contributes maybe twenty more deaths total.

Which means the math forces a conclusion about the other arm: the trial needs roughly 48 additional deaths to reach its final analysis, and if BAT only supplies about twenty of them, the imetelstat arm — twice the size — has to supply the rest at a far slower per-person pace. The future deaths in this trial are increasingly an imetelstat-arm story.

So hypothetically at a 30 mOS for Bat patients ~20 more will die in the 2 years following the interim and 28 Imetelstat patients will die. Ultimately ~ 85 of the total deaths will come from BATs and 75 from Imet. All conjecture at 30 mOS for Bats - I welcome corrections to these hypothetical numbers.

Thank you. So how does this analysis of the deaths alter any previous understanding of the statistics at interim?

From Seeking Alpha:

...in December of 2024 the company was projecting this: "The planned interim analysis is expected in early 2026 and final analysis in early 2027". How does that factor into the analysis? By that date, 262 patients were already actively enrolled in the trial. We seemed to have gone from 1 year between interim and final to now two years. That has to be important data.

Does this information help inform the model?

STA - Claude's response - This is a genuinely sharp catch, and yes — it materially informs the model. A *changing* projection carries more information than any single static date, because the direction of the change tells you which way reality is diverging from the original plan. Let me work through what the shift actually implies, including the part that cuts against the bull read, because that's where the real signal is.

**What changed:** In Dec 2024, with 262 enrolled, Geron guided interim "early 2026" / final "early 2027" — a **~1-year** interim-to-final gap. Now it's interim "before end of 2026" / final "H2 2028" — a **~2-year** gap. So *both* dates slipped, and the gap between them *doubled*. Those are two distinct facts and they point in partly different directions.

**Fact 1 — both dates slipped later.** The interim moved from "early 2026" to "late 2026," and the final from "early 2027" to "late 2028." That means **deaths are accruing more slowly than the December 2024 model assumed.** Full stop. Whatever the cause, the trial is producing events slower than projected a year and a half ago. Slow accrual is the thing we've said all along is consistent with a durable imetelstat arm — so on its face this is mildly bull-confirming: the original model expected people to die faster than they are.

**Fact 2 — the gap doubled, and this is the more informative one.** Here's the subtle part. The interim is a fixed event count (~112 deaths); the final is a fixed event count (~160). The *gap* between them is the time to accrue those last ~48 deaths. If that gap stretched from ~1 year to ~2 years, the model now expects the **back-half death rate to be roughly half** what it thought in Dec 2024. The back half of accrual is dominated by the slower-dying arm — which, in a 2:1 trial where BAT is largely spent by interim, is the **imetelstat arm.** So a doubling of the interim-to-final gap is specifically a statement that *imetelstat-arm survival is being modeled as longer than it was a year ago.* That's the most bullish single inference available here: the company's own evolving event model is implying the treatment arm is more durable than they originally projected.

**But now the discipline, because there's a real alternative explanation you have to hold:** both dates slipping *and* the gap widening is **also** consistent with the bear case you yourself raised two days ago — **elevated BAT survival.** If contemporary BAT patients are living longer than the Dec-2024 model assumed (the Rux-era supportive-care drift), then *both* arms accrue deaths slower, both dates slip, and the gap can widen. In that world, the slow accrual isn't imetelstat heroics — it's that the whole population is outliving the original assumptions, which *compresses* the HR toward the boundary rather than widening it. So the timeline shift cannot, by itself, distinguish "imetelstat arm more durable than expected" (bullish) from "both arms living longer than expected, including BAT" (bearish for the *margin*). Same observable, two readings — the exact World A / World B ambiguity, now showing up in the *change* in guidance rather than the static date.

**What tips the interpretation, slightly:** the fact that the **interim also slipped** is the relevant tell. If only the *final* had moved out (gap widening from the back), that's cleanly "imetelstat arm durable." But the interim moving from early-2026 to late-2026 means even the *early* accrual — which includes the rapidly-dying BAT arm — ran slower than projected. BAT dying slower than the Dec-2024 model is direct evidence for *your own bear case*: effective BAT survival is running above the historical benchmark the original model likely used. So the interim slip is a quiet point *for* the elevated-BAT reading, while the gap-doubling is a point *for* the durable-imetelstat reading. They're both in the data, pulling opposite ways on the *margin* even as both confirm "slower than expected overall."

**The net deduction, honestly stated:**
1. The trial is accruing deaths slower than Geron's own Dec-2024 model — confirmed by both dates slipping. Real, and it raises the floor on overall survival in the trial.
2. The interim-to-final gap doubling implies the *back-half* (imetelstat-dominated) accrual is now modeled as much slower — consistent with a durable treatment arm.
3. **But** the interim slip itself implies even BAT may be outliving the original assumptions — which is your Rux-era bear case showing up in the timeline, and which pushes the implied HR *toward* the boundary, not away.
4. So the shift is best read as: **the whole population is surviving longer than the 2024 model expected, the effect is still there, but the data is now equally consistent with "blowout" and with "real-but-boundary-hugging."** It widens the *range* of outcomes rather than narrowing it toward the blowout.

So to your question — does it help inform the model? Yes, importantly: it tells you the Dec-2024 assumptions were too pessimistic on survival for *both* arms, which is why the dates moved. What it does *not* do is resolve the one thing that matters — whether the slower accrual is imetelstat pulling away (early stop) or both arms drifting up together (continue to 2028). If anything, an honest reading says the timeline change should *widen* your error bars, not tighten them toward success — because the same slip that's consistent with your bull case is also the fingerprint of the bear case you correctly identified. That's the intellectually honest deduction, and it's the one the board won't make because the gap-doubling *feels* purely bullish.

Yes, a definite glitch, BAT arm likely also living longer. This is supported by Registries that seems to confirm, and key MF center (Moffett) speculating it is due to earlier Rux start and shorter duration of therapy. But to go from assumed BAT of 15 month (IMbark) to 30 (pre 2019 Kuykendall) and then 40 (!) post 2019 Kuykendall is probably way more than the planners had in mind when the study was formulated and likely contributing to the long accrual times. These shifting mos timelines are really something and only this randomized study can sort them out. There are sooo many issues like registries that include low risk and DIPSS int-1 patients and many non US, non European sites that may show different BATs for a number of reasons. All of the K-M curves show an early plunge in deaths and then a plateau. I think it is worth it to look at death rate accruals in this time frame also. Dr. M has made a plea to enroll patients before their bone marrows burn out in the hope of some recovery. Slower BAT OS rates may allow for later Imet arm Mos to be apparent but could mean more waiting after interim for final analysis before the study is called. Few current stockholders would have the stomach for that.

BP, Ran your analysis past Claude and he agrees with your focus. I think we're at the stage where we have as much clarity as we're going to have unless new info is released. I believe it is highly likely that the mOS for both BAT and Imet patients has increased significantly. I have held this stock for this long because I believe in the MOA of Imetelstat which thus far this trial has not disproven. The fact that patients are living longer doesn't prove or disprove this thesis - but it is more comforting than the alternative.

It does make you wonder though how urgent hematologists will feel the need is for Imetelstat if they can simply follow Moffet's BAT and see such impressive mOS. Keeping in mind that Rytelo is an expensive drug it may come to pass that Rytelo is used as a salvage treatment for many years before moving up the paradigm. After all, we've seen this over the last two years in MDS.

The new analysis has actually shifted me from bullish to bearish on the interim analysis and knowing the market, a continuation will likely cause the share price to plummet. After all, I doubt that institutions are in this for another 2 years of uncertainty. In my opinion, if we have to go to 2028 (or later, there's plenty of time for them to continue to expand the timeline between now and then), I think we probably get a hostile takeover that leaves the shareholders holding the bag.

STA, good points but I do think trying to fit the Moffett data onto an IMpact BAT prediction has many weakness. If we just went with it the BAT arm would outperform the historical imet arm! So clearly lots of other objections baked into that data. Nonetheless Dr K cautions that BAT has improved considerably and it helps us to try to understand how.

SAT
You posted- "It does make you wonder though how urgent hematologists will feel the need is for Imetelstat if they can simply follow Moffet's BAT and see such impressive mOS. Keeping in mind that Rytelo is an expensive drug it may come to pass that Rytelo is used as a salvage treatment for many years before moving up the paradigm. After all, we've seen this over the last two years in MDS."
IS Rytello more expensive that the BAT? What is the BAT? I assume it has to be administered a longer time to produce longer OS. Does BAT require a lot of jerryrigging and changing? If so, it would be simpler to use the more effective drug Rytello right away. It's not like BAT people are magically living longer. They I assume are being actively treated.

I find your ongoing bearish comments are not too subtle.

FYI, I'm not trying to soft bash or be subtle. I'm trying to understand how quickly doctors will adopt Rytelo when it is approved for MF. If the market penetration is anything like MDS then we would have a problem.

If hematologists who are used to managing RUX have a patient who stops responding I think that there is a high likelihood that they would be more comfortable switching to another JAK inhibitor before going to Rytelo. Yes, we know that Rytelo can clear VAF and marrow but when mOS on subsequent JAK inhibitors starts showing mOS to the Phase 2 Imetelstat trial then I worry that adoption will be fast enough.

Both Rux and Fedratinib are both $200k+ annually, so perhaps price won't be as big of a factor. If you've been with Geron as long as I have, you were probably also guilty of thinking that MDS adoption would be quick because of the dearth of other treatments and the quality of life afforded by being transfusion free. That has not come to pass in a meaningful way after 2 years, so I think it's fair for me to be worried and to want to discuss that with other knowledgeable investors.

STA and Jay, Good points all around. There is certainly weirdness in the "new" MOS which will certainly wash out with a randomized controlled study. I have some definite thoughts about this and yes STA, we have beaten this to death but I think still may have missed something that speaks to your point about physician adaptation. If the "new"BAT is real, you are right, in this elderly population why not give them "disease control", (with modest improvements in life extension), make them comfortable and wait another 4 years that at least on the surface "new" BAT seems to confer for many. To Jay on the other hand it really comes down to will we find true "disease modification", the holy grail (obviously not my words). If the latter happens a cascade of things could occur (with a nod to chat GPT) which we have not really discussed. These might include changes in the way people live and people die. For example:1. infection 2. bone marrow failure 3. bleeding 4. cachexia 5. portal hypertension 6. Vascular events 7. AML . 8. Anemia control. 9. Better symptom control. 10. Smaller spleens. This is where disease modification really takes us.(If it is proven with significance.) So if just like the IMbark data and the BAT arm with it are a decade old (!!), and BAT is better for all the reasons STA has listed, the Imet arm gets to start in the same place and benefit from whatever is causing the BAT arm to do so much better. Dr. M himself has made the plea in one of his interviews not to wait until the bone marrow is burned out and has some chance to recover with imetelstat treatment. So it really comes down to what we want to believe at this monument in time which is incredibly biased and we must fight against in the pursuit of accuracy. After all we know a lot about new JAKi and improved care (to STA's point), while Imetelstat in MF is just a promise that may be kept or broken (to MG's point).

FWIW, I have been spending a lot of time trying to understand the subtleties of the differences between mOS and overall death counts and how long tails on these curves as they flatten out can prolong the mOS but early rapid decent of the curves appear to represent many deaths that occur early. This is why the study is designed to be triggered by an absolute death number rather than an mOS number. As to the Moffett data, small number, incudes DIPSS-1, long tail for mOS determination etc etc., I think we get it. Nonetheless, it's telling us something important worthy of note and it has relevance to the excellent points raised by both STA and Jay.