ImetelChat

MG thank you for the work you have been doing with AI in trying to help us understand the implications of the Kuykendall abstract. I think we need to back up a bit to see what the abstract is trying to tell us. First off there are a couple of discrepancies that got us into interpretation trouble, partially due to AI trying to help but maybe leading us a little (or a lot) astray. For example use of the Flatiron Registry was problematic and this was entirely my fault for not going back and check the source. Clearly the Moffett registry and the Flatiron registry are different and OS assessment does not start with the onset of treatment as we initially were lead to believe. That has to come off the table. The Dr. K data is clear, BAT OS determination clock starts when Rux discontinuation starts, to death or censored at last follow-up. That is not the same as meeting criteria for entry into Impact when the clock for BAT OS starts for the purposes of the study. This is important since after Rux discontinuation in the Dr. K data lots can happen before the patient eventually gets enrolled in Impact (1. go to imetelstat or BAT soon after washout 2. go to other lines of therapy that take time and lead either to death at some point or even other lines of therapy all of which take time and would be measured in the mOS in Dr. K's abstract: time from cessation of Rux to death. A second flaw in the AI from today's note is the statement: "Pre-Enrollment "Lead-Time" Bias
Because the inclusion criteria require patients to enter immediately following ruxolitinib failure, the macro trend of early ruxolitinib discontinuation dictates the exact physiological profile of patients at the trial gate." This refers to the enrollment in Impact according to the Gemini analysis does not appear to be correct since we know patients can cycle into other treatment lines including other treatment lines and still meet inclusion criteria. I think STA was trying to get at this earlier. Anyway I do think we have to careful to not take AI at its word. So I want to try to take a fresh look at what the abstract is telling us and what kind of faith we might put into the Impact study design as a result of it.

I admit when I first read the abstract I drew a sharp breath and thought could the BAT have improved that much? But now we know that 1. Time to onset of Rux is much shorter 2. Time on Rux is much shorter 3. The effect of these factors on BAT are part of the longer mOS data. 3. We don't know what happens after Rux discontinuation other than transition to other therapy lines that are not specifically defined one of which could be randomization to Imetelstat or BAT as defined by the study requirements. 4. We don't know if any lower risk MF patients are included which could certainly add to the OS time. All of these factors could lead to a longer measured BAT for the purposes of the Dr. K data but a very different mOS number is likely once a patient enters the BAT arm of Impact which has strict entry requirements.

From this I conclude that the only way to have an accurate comparator arm in the IMpact study is for the study to be the multi center randomized trial that we have ongoing. The BAT mOS will likely be much shorter than what we see in the Dr. K publication for the reasons outlined. At least some of these considerations may also apply to the Imetelstat arm which may break the expected OS positively for that arm also. As far as projected numbers for BAT and the Imetelstat arm, I am going to stop guessing. Hope this helps in some way. But I think this is what the abstract is trying to tell us. bp

BP, we have achieved analysis paralysis. We can't have 100% confidence that any of the AI analysis is flawless. Nor can we be sure that our own cofirmation bias is not blinding us. OTH, we reviewed the data thouroughly and to my layman's eye, I think we can say with some confidence that, if Imetelstat's MOA is confirmed with IMPactMF there is a pretty good chance that the trial will be halted for efficacy. Trying to predict the mOS for BAT and IMET patients is too open to potential errors and or ommisions, however the trial seems well constructed for validating Imetelstat's effficacy.

I agree and I think it was well worth the process you led to get to this conclusion, I personally found it very helpful.

I would like to add a final wrinkle to our discussion. We know that crossover was designed into the trial because it is morally and ethically important to allow patients to get access to Rytelo if their clinician thinks that the patient will benefit. We also know now that the effect of such crossover isn't going to be material.

What we haven't discussed is the idea of discontinuation in the BAT arm. By this I mean that we are learning through the recent abstract that BAT improvements include alternative JAK inhibitors to Jakafi. These alternatives are verboten to patients in the BAT arm of IMPact. At what point would a clinician recommend that a BAT patient leave the trial so that they could try any of the newer options? It would seem ethically wrong to not keep these patients aware of their options, even if it meant them leaving the trial.

If this is happening, then it could account for the delays in interim analysis due to the loss (to the trial) of early enrolled BAT patients. We are counting on the BAT patients to make up an outsized portion of deaths, but if there are fewer of them in the trial currently than anticipated then the data is weaker at interim.

STA, good point - I asked Gemini to address - do not take this as absolute truth - AI's hallucinate facts - verify -but this looks correct.

What happens if BAT patients drop out of the trial to try some other treatment?

If a patient on the Best Available Therapy (BAT) arm decides to stop taking their assigned treatment to try another option, the way their data is handled depends on a critical distinction in clinical trial operations: discontinuing the study drug versus withdrawing completely from trial tracking.

The primary Overall Survival (OS) analysis handles these scenarios through specific statistical rules.

1. The Intent-to-Treat (ITT) Reality (Most Common)
In almost all cases, when a BAT patient "drops out" of their assigned treatment (usually due to lack of efficacy or side effects) and switches to an alternative commercial therapy, they do not drop out of the trial itself.

Under the strict Intent-to-Treat (ITT) framework that governs the primary endpoint of IMpactMF:

The Rule: Once randomized to the BAT arm, the patient belongs to the BAT arm forever, regardless of what they do next.

The Action: The trial investigators will stop administering the protocol's non-JAKi therapy, but they are required by the protocol to continue tracking that patient’s survival status via phone calls or medical record checks for the rest of their life.

The Mathematical Effect: If that patient goes on to receive an effective, off-protocol real-world therapy (like an unpermitted commercial JAKi) and survives for another three years, those three years of survival are credited entirely to the BAT control arm. This creates "subsequent therapy contamination," which naturally dilutes the Hazard Ratio (HR) by artificially lifting the control curve.

2. Complete Withdrawal of Consent (Censoring)
If a patient becomes so frustrated with being randomized to the traditional BAT arm that they explicitly state, "I am completely withdrawing from the study; you no longer have permission to track my medical records or contact me," the statistical treatment changes entirely:

The Rule: The patient’s data line is truncated.

The Action: They are censored at the exact date of their last documented clinical contact.

The Mathematical Effect: They do not count as a survival "event" (death), nor do they count as a long-term survivor. They simply vanish from the denominator of active risk from that date forward.

The Informative Censoring Risk: If a high number of BAT patients completely withdraw consent early because they feel the control arm options are inferior, it creates "informative censoring." This can introduce bias into the Kaplan-Meier survival curves, which regulatory bodies like the FDA scrutinize heavily during review.

3. How the Statistical Analysis Plan (SAP) Controls for This
Because Geron's statisticians know that real-world patients will inevitably drop out of traditional non-JAKi BAT therapies to pursue newer drugs or clinical trials, the Statistical Analysis Plan (SAP) includes pre-specified sensitivity analyses to protect the trial's integrity:

Censoring at Next Line of Therapy: The SAP includes a secondary analysis where a BAT patient is mathematically censored the exact day they start a non-permitted subsequent therapy (such as a commercial JAKi). This strips out the artificial survival extension caused by modern salvage therapies.

Inverse Probability of Censoring Weighting (IPCW): This advanced statistical model recreates what the survival curve would look like by mathematically weighting the remaining patients who stayed on protocol to fill the gaps left by those who switched therapies.

Summary
If a BAT patient stops their assigned therapy to try something else, the ITT principle ensures their ultimate survival data is still captured and credited to the control arm, which can dilute imetelstat’s apparent advantage. Geron relies heavily on secondary, pre-specified sensitivity analyses to mathematically strip away this post-progression contamination and present the FDA with a clean picture of imetelstat's true monotherapy benefit.

Please then ask if these statistical corrections will occur at interim or if they are only relevant at trial completion. Also, is it valid to assume that discontinuation could explain the continued pushing back of the interim analysis?

STA - Gemini's reply to your question: Please then ask if these statistical corrections will occur at interim or if they are only relevant at trial completion. Also, is it valid to assume that discontinuation could explain the continued pushing back of the interim analysis?

If a patient on the Best Available Therapy (BAT) arm decides to stop taking their assigned treatment to try another option, the way their data is handled depends on a critical distinction in clinical trial operations: **discontinuing the study drug** versus **withdrawing completely from trial tracking**.

The primary Overall Survival (OS) analysis handles these scenarios through specific statistical rules.

---

### 1. The Intent-to-Treat (ITT) Reality (Most Common)

In almost all cases, when a BAT patient "drops out" of their assigned treatment (usually due to lack of efficacy or side effects) and switches to an alternative commercial therapy, they do **not** drop out of the trial itself.

Under the strict **Intent-to-Treat (ITT)** framework that governs the primary endpoint of IMpactMF:

* **The Rule:** Once randomized to the BAT arm, the patient belongs to the BAT arm forever, regardless of what they do next.
* **The Action:** The trial investigators will stop administering the protocol's non-JAKi therapy, but they are required by the protocol to continue tracking that patient’s survival status via phone calls or medical record checks for the rest of their life.
* **The Mathematical Effect:** If that patient goes on to receive an effective, off-protocol real-world therapy (like an unpermitted commercial JAKi) and survives for another three years, those three years of survival are credited entirely to the **BAT control arm**. This creates "subsequent therapy contamination," which naturally dilutes the Hazard Ratio ($\text{HR}$) by artificially lifting the control curve.

---

### 2. Complete Withdrawal of Consent (Censoring)

If a patient becomes so frustrated with being randomized to the traditional BAT arm that they explicitly state, *"I am completely withdrawing from the study; you no longer have permission to track my medical records or contact me,"* the statistical treatment changes entirely:

* **The Rule:** The patient’s data line is truncated.
* **The Action:** They are **censored** at the exact date of their last documented clinical contact.
* **The Mathematical Effect:** They do not count as a survival "event" (death), nor do they count as a long-term survivor. They simply vanish from the denominator of active risk from that date forward.

> **The Informative Censoring Risk:** If a high number of BAT patients completely withdraw consent early because they feel the control arm options are inferior, it creates "informative censoring." This can introduce bias into the Kaplan-Meier survival curves, which regulatory bodies like the FDA scrutinize heavily during review.

---

### 3. How the Statistical Analysis Plan (SAP) Controls for This

Because Geron's statisticians know that real-world patients will inevitably drop out of traditional non-JAKi BAT therapies to pursue newer drugs or clinical trials, the **Statistical Analysis Plan (SAP)** includes pre-specified sensitivity analyses to protect the trial's integrity:

* **Censoring at Next Line of Therapy:** The SAP includes a secondary analysis where a BAT patient is mathematically censored the exact day they start a non-permitted subsequent therapy (such as a commercial JAKi). This strips out the artificial survival extension caused by modern salvage therapies.
* **Inverse Probability of Censoring Weighting (IPCW):** This advanced statistical model recreates what the survival curve *would* look like by mathematically weighting the remaining patients who stayed on protocol to fill the gaps left by those who switched therapies.

### Summary

If a BAT patient stops their assigned therapy to try something else, the ITT principle ensures their ultimate survival data is still captured and credited to the control arm, which can dilute imetelstat’s apparent advantage. Geron relies heavily on secondary, pre-specified sensitivity analyses to mathematically strip away this post-progression contamination and present the FDA with a clean picture of imetelstat's true monotherapy benefit.

I wonder if the clinician explains the consequences to the BAT patient. It would be nice to know that in the interest of expediting approval of a potentially life saving drug that these patients opt for complete withdrawal from the trial.

Beautiful clarity of this data from the perspective of an oncology nurse :

https://www.oncnursingnews.com/view/ime ... -Author(s)

STA, I have little doubt that if the interim read out is positive, all of the BAT patients will be allowed to cross over. As to what a patient in the BAT arm might do now is another question. Some number probably already tried and failed new JAKi before entry into the study so there really might not be anywhere else to go.