Two quick observations relevant to the post 2019 cohort:
1.This RWD collection includes low DIPSS and intermediate 1 patients (IMpact only allows DIPSS Intermediate 2 and high risk)
2. Table 3 lists "next therapies" after Rux failure. Notably Imetelstat is not on that list. There is no information as to what options might follow the next therapies. It seems probable that after Rux failure followed by next therapies in the 2019-2025 time frame that at least some patients would be offered IMpact randomization at a leading MF treatment center. Interestingly, in the discussion factors that could affect the markedly increased BAT OS, this is not mentioned as a potential contributor as other factors are such as time to start of Rux and time on Rux therapy. The study does clarify how many patients are anemic (this is reflected in the DIPSS score anyway) and that as we have discussed BAT OS is clearly measured from the time of Rux cessation and does not include time on Rux therapy the way the Flatiron Registry appeared to do.
The RWD numbers we are seeing (close to 40 months) clearly overestimate what we might expect as a BAT comparator arm in IMpact.
The poster from Dr. K is published on the Geron web site
Forum rules
- Comments must be civil and on topic
- Back up claims with evidence/reasoning/sources (posting links is allowed)
- No commercials/harassment/spam
- Comments must be civil and on topic
- Back up claims with evidence/reasoning/sources (posting links is allowed)
- No commercials/harassment/spam
-
biopearl123
- Posts: 2591
- Joined: Fri Jul 20, 2018 5:13 pm
-
biopearl123
- Posts: 2591
- Joined: Fri Jul 20, 2018 5:13 pm
Re: The poster from Dr. K is published on the Geron web site
Post 2019 cohort in Dr. K abstract is 27 patients. Other study sites (like Poland, Russia, etc.) might not have similiar pre rux and rux duration numbers or access to newer JAKi due to reimbursement issues pre rolling into Imet. Another factor to consider when trying to get to a realistic BAT estimate.
-
biopearl123
- Posts: 2591
- Joined: Fri Jul 20, 2018 5:13 pm
Re: The poster from Dr. K is published on the Geron web site
New factors to consider when trying to devine a reasonable OS for the purposes of IMpact:
1. Dr. K’s study post 2019 includes 2/27 low risk and 6/27 Intermediate 1 patients. This fact alone removes 8 patients from the 27 with the longest life expectancy. New denominator is 19 with 10th patient required to cross the 50% death line. Those 10 will all be in the high risk and intermediate 2 DIPSS class.
2. Moffett is a leader in MF and may have different experience with stopping Rux as early as they do, not sure this practice pattern extends around the world like to Russian centers participating.
3. We do not know what happens after “next therapies”. These immediate next therapies are listed and do not include imetelstat and may include therapies not available to all institutions around the world.
This study attributes the long OS to evolutions in rux usage over the years but including intermediate 1 and low risk patients in the column for post 2019 may be an unacknowledged
Important factor.
Let me know what you think and if there are any flaws here. Regards, bp
1. Dr. K’s study post 2019 includes 2/27 low risk and 6/27 Intermediate 1 patients. This fact alone removes 8 patients from the 27 with the longest life expectancy. New denominator is 19 with 10th patient required to cross the 50% death line. Those 10 will all be in the high risk and intermediate 2 DIPSS class.
2. Moffett is a leader in MF and may have different experience with stopping Rux as early as they do, not sure this practice pattern extends around the world like to Russian centers participating.
3. We do not know what happens after “next therapies”. These immediate next therapies are listed and do not include imetelstat and may include therapies not available to all institutions around the world.
This study attributes the long OS to evolutions in rux usage over the years but including intermediate 1 and low risk patients in the column for post 2019 may be an unacknowledged
Important factor.
Let me know what you think and if there are any flaws here. Regards, bp
-
mistergern
- Posts: 221
- Joined: Sat Mar 19, 2016 3:48 pm
Re: The poster from Dr. K is published on the Geron web site
BP, I thought this was a good time to summarize our findings so I asked the Claude with whom I have shared all available info to first address your points concerning the K study and then to do an overall summary.
On the Kuykendall Post-2019 RWD and Its Limits as an IMpact BAT Comparator
The following analysis was developed with parameters set by Mistergern: be scientific, objective, and truthful; flag conjecture clearly; and don't shy from a bullish or bearish read where the evidence warrants one. I'll mark which claims are grounded and which are inference.
BP has raised several observations about the post-2019 real-world cohort in Dr. Kuykendall's poster. Working through them, most hold up, and together they support a clear conclusion: the ~34-40 month real-world OS figures materially overstate what should be expected from the IMpact BAT control arm. That claim is well-grounded, not conjecture. The reasons:
1. Risk-class mismatch (grounded, and the strongest point). The post-2019 cohort is n=27 and includes 2 low-risk and 6 intermediate-1 patients by DIPSS. IMpactMF enrolls only intermediate-2 and high-risk patients. Lower-risk patients carry inherently longer life expectancy, so their inclusion lifts the cohort's median above what an all-intermediate-2/high-risk population would show. This is a direct, structural reason the figure cannot be read across to IMpact's BAT arm. It is not in dispute.
Caution on how far to push this: it's tempting to recalculate by stripping the 8 lower-risk patients and re-deriving a median from the remaining 19. I'd resist that. At n=19 — and the poster already lists the cohort's upper confidence bound as "not estimable" — any recomputed median would carry confidence intervals so wide as to be uninformative. The qualitative point is strong; the precise recalculation is not supportable and would invite fair criticism. State the direction, not a new number.
2. Measurement window (grounded). This poster measures OS from ruxolitinib cessation, which is the correct frame for an IMpact comparator (IMpact's clock starts at randomization, post-JAKi failure). This is distinct from the Flatiron registry analysis, which appeared to measure from rux initiation and therefore embedded the entire rux treatment duration into its survival figures. Conflating the two produces badly inflated comparators. Keeping them separate is essential.
3. Geographic and practice-pattern variation (reasoned inference, not established). BP notes Moffitt is a leading MF center whose early-rux-discontinuation practice and access to newer JAK inhibitors (fedratinib, pacritinib) may not extend to all IMpact sites, some of which are in regions with different treatment patterns or reimbursement constraints. This is a plausible and reasonable inference — but I'll flag it as inference. We don't have IMpact's site-by-site BAT data, and the direction of the effect, while logically toward lower BAT OS at less-resourced sites, is not quantified.
4. The "rolled into IMpact" hypothesis (explicit conjecture). BP raises the possibility that some post-rux patients in this era were offered IMpact randomization, and that imetelstat notably does not appear in the poster's Table 3 "next therapies." This is genuinely interesting but it is conjecture, and BP appropriately frames it that way. We don't know the downstream therapy sequence after the listed next-line treatments, and we can't confirm overlap with IMpact enrollment. Worth raising as an open question; not worth leaning on.
What this does and does not tell us:
What it tells us (grounded): the contemporary RWD figures near 40 months are not a valid estimate of the IMpact BAT arm. They include lower-risk patients, reflect a leading center's practice patterns, and benefit from commercial JAKi sequencing that IMpact's protocol bans on the control arm. The IMpact BAT comparator should be expected to land well below those figures.
What it does not tell us (the honest limit): the actual IMpact BAT arm median is unknown to anyone outside the DSMB. The published matched comparator most applicable to the IMpact population — the propensity-weighted IMbark analysis — put BAT at roughly 15-17 months. The contemporary era effect plausibly lifts that somewhat, but how much is genuinely uncertain. Anyone claiming to know the precise IMpact BAT number — high or low — is guessing.
Bottom line: BP's core point is correct and worth stating with confidence — the headline RWD survival figures overstate the IMpact comparator, primarily because of risk-class contamination and the JAKi-sequencing the trial prohibits. Where I'd hold the line is against converting that valid directional argument into a specific BAT estimate. The argument is strong precisely because it's structural; it weakens the moment it reaches for precision the data can't support.
On the Kuykendall Post-2019 RWD and Its Limits as an IMpact BAT Comparator
The following analysis was developed with parameters set by Mistergern: be scientific, objective, and truthful; flag conjecture clearly; and don't shy from a bullish or bearish read where the evidence warrants one. I'll mark which claims are grounded and which are inference.
BP has raised several observations about the post-2019 real-world cohort in Dr. Kuykendall's poster. Working through them, most hold up, and together they support a clear conclusion: the ~34-40 month real-world OS figures materially overstate what should be expected from the IMpact BAT control arm. That claim is well-grounded, not conjecture. The reasons:
1. Risk-class mismatch (grounded, and the strongest point). The post-2019 cohort is n=27 and includes 2 low-risk and 6 intermediate-1 patients by DIPSS. IMpactMF enrolls only intermediate-2 and high-risk patients. Lower-risk patients carry inherently longer life expectancy, so their inclusion lifts the cohort's median above what an all-intermediate-2/high-risk population would show. This is a direct, structural reason the figure cannot be read across to IMpact's BAT arm. It is not in dispute.
Caution on how far to push this: it's tempting to recalculate by stripping the 8 lower-risk patients and re-deriving a median from the remaining 19. I'd resist that. At n=19 — and the poster already lists the cohort's upper confidence bound as "not estimable" — any recomputed median would carry confidence intervals so wide as to be uninformative. The qualitative point is strong; the precise recalculation is not supportable and would invite fair criticism. State the direction, not a new number.
2. Measurement window (grounded). This poster measures OS from ruxolitinib cessation, which is the correct frame for an IMpact comparator (IMpact's clock starts at randomization, post-JAKi failure). This is distinct from the Flatiron registry analysis, which appeared to measure from rux initiation and therefore embedded the entire rux treatment duration into its survival figures. Conflating the two produces badly inflated comparators. Keeping them separate is essential.
3. Geographic and practice-pattern variation (reasoned inference, not established). BP notes Moffitt is a leading MF center whose early-rux-discontinuation practice and access to newer JAK inhibitors (fedratinib, pacritinib) may not extend to all IMpact sites, some of which are in regions with different treatment patterns or reimbursement constraints. This is a plausible and reasonable inference — but I'll flag it as inference. We don't have IMpact's site-by-site BAT data, and the direction of the effect, while logically toward lower BAT OS at less-resourced sites, is not quantified.
4. The "rolled into IMpact" hypothesis (explicit conjecture). BP raises the possibility that some post-rux patients in this era were offered IMpact randomization, and that imetelstat notably does not appear in the poster's Table 3 "next therapies." This is genuinely interesting but it is conjecture, and BP appropriately frames it that way. We don't know the downstream therapy sequence after the listed next-line treatments, and we can't confirm overlap with IMpact enrollment. Worth raising as an open question; not worth leaning on.
What this does and does not tell us:
What it tells us (grounded): the contemporary RWD figures near 40 months are not a valid estimate of the IMpact BAT arm. They include lower-risk patients, reflect a leading center's practice patterns, and benefit from commercial JAKi sequencing that IMpact's protocol bans on the control arm. The IMpact BAT comparator should be expected to land well below those figures.
What it does not tell us (the honest limit): the actual IMpact BAT arm median is unknown to anyone outside the DSMB. The published matched comparator most applicable to the IMpact population — the propensity-weighted IMbark analysis — put BAT at roughly 15-17 months. The contemporary era effect plausibly lifts that somewhat, but how much is genuinely uncertain. Anyone claiming to know the precise IMpact BAT number — high or low — is guessing.
Bottom line: BP's core point is correct and worth stating with confidence — the headline RWD survival figures overstate the IMpact comparator, primarily because of risk-class contamination and the JAKi-sequencing the trial prohibits. Where I'd hold the line is against converting that valid directional argument into a specific BAT estimate. The argument is strong precisely because it's structural; it weakens the moment it reaches for precision the data can't support.
-
mistergern
- Posts: 221
- Joined: Sat Mar 19, 2016 3:48 pm
Re: The poster from Dr. K is published on the Geron web site
BP - Here is Claude's latest summary:
IMpactMF Interim — A Full Synthesis of What We Know, What We Infer, and What Remains Unknown
Same parameters from Mistergern: scientific, objective, conjecture flagged explicitly, willing to take a side where evidence warrants. This builds on the RWD discussion above and lays out the complete picture for community scrutiny. Where I cite a fact, I'll note its source quality. Where I reason, I'll say so.
The structural facts (grounded, primary sources):
IMpactMF is a Phase 3, 2:1 randomized, open-label trial of imetelstat vs. non-JAKi best available therapy in ~320 patients with DIPSS intermediate-2 or high-risk myelofibrosis, relapsed/refractory to JAK inhibitors, with overall survival as the primary endpoint. First patient dosed April 13, 2021. Fifty percent enrollment December 2023. Full enrollment of 320 completed September 2025. Interim guided for H2 2026. (Geron press releases, ClinicalTrials.gov NCT04576156.)
The statistical design, from Geron's 10-K: powered at >85% to detect HR 0.60 (one-sided alpha 0.025); final analysis after >50% of enrolled patients have died (~160 deaths); interim after ~70% of final-analysis events (~112 deaths), with interim alpha spend ~0.01. This resolves a contradiction that circulated publicly: the interim is NOT at a 35%-information-fraction OBF look. It is at ~70% information fraction (112 of ~160 events), where an O'Brien-Fleming boundary naturally produces a critical threshold near Z=2.34, p≈0.01 — matching the disclosed alpha spend. (10-K is a primary source; this is the single most important clarification in the whole analysis.)
What the boundary actually requires (grounded math):
At ~112 events with 2:1 randomization, the standard error on ln(HR) is √(4.5/112) ≈ 0.20. To cross Z=2.34 the observed HR must be roughly ≤0.62. An observed HR of 0.56 produces Z≈2.89 — clearing comfortably. This means the trial does not need to replicate IMbark's extraordinary HR (~0.51 vs. matched RWD) to halt early. It needs roughly 0.62 or better. That's a meaningful and underappreciated point, and it's grounded in the disclosed design.
The BAT comparator (mix of grounded and inferred):
The BAT comparator (mix of grounded and inferred):
Grounded: the propensity-weighted IMbark-matched analysis (the population most comparable to IMpact) put BAT median OS at ~15-17 months (unweighted 15.4; ATO 17.0; sIPTW 16.1; HR ~0.51 across all three). Per the RWD discussion above, the contemporary ~34-40 month figures overstate the IMpact comparator due to risk-class contamination and banned JAKi sequencing.
Inferred: the true IMpact BAT arm likely sits somewhat above the historical 15-17 months due to era effects (earlier rux pivoting producing healthier patients at entry), but well below 34-40. A defensible inferred range is roughly 15-24 months. This is reasoned, not known. No one outside the DSMB has the actual figure.
The enrollment-timing arithmetic (grounded method, inferred inputs):
Using the verified enrollment shape (first patient 4/2021, 50% by late 2023, 100% by 9/2025) and a survival model checked at a ~9/2026 interim, the early-enrolled cohort has 34-65 months of follow-up by interim and the later cohort 12-34 months. Across BAT median OS assumptions from 15 to ~33 months, the model produces BAT deaths sufficient that the imetelstat arm needs only a modest death count to reach the 112 trigger — yielding crude HRs that clear the Z=2.34 boundary in every scenario up to roughly 37 months BAT OS.
Important honesty flags on this model: it uses a Weibull survival shape (k=0.85) that is an assumption, not measured; it assumes a survival-curve form for both arms; and "crude HR" derived from death-count ratios is an approximation of the formal Cox/log-rank HR the trial will actually report. These caveats could shift the precise numbers. They do not appear to change the directional conclusion, but a reader should know the model rests on assumptions.
Crossover (unverified source — flag clearly):
Geron's CMO reportedly stated at the Stifel conference (May 2026) that crossover impact is minimal — capped by protocol, requiring a washout period and sponsor approval, with physician reluctance to cross already-ill patients. I have not seen a transcript; this is secondhand from community reports. If accurate, it removes the largest dilution risk we'd otherwise model. If inaccurate or overstated, crossover could compress the observed ITT HR. This is a load-bearing input that deserves verification from the actual transcript before anyone relies on it.
The IMbark evidence base (grounded, high quality):
IMbark (Phase 2, 9.4 mg/kg) showed median OS ~28-31 months across data cuts, with seven orthogonal supporting endpoints — OS, bone marrow fibrosis reversal, VAF reduction in driver mutations, telomerase target engagement, dose-response consistency, biomarker-survival correlation, and activity in poor-prognosis subgroups. Peer-reviewed (JCO 2021), consistent across six years of conference scrutiny, with independent laboratory mechanistic validation. The internal low-dose comparator arm strengthens it from a single-arm study toward an internally controlled dose-response experiment. This is unusually strong Phase 2 data. The legitimate limitation: n=59 in the high-dose arm, no concurrent randomized OS control — which is precisely what IMpact exists to resolve.
IMpactMF Interim — A Full Synthesis of What We Know, What We Infer, and What Remains Unknown
Same parameters from Mistergern: scientific, objective, conjecture flagged explicitly, willing to take a side where evidence warrants. This builds on the RWD discussion above and lays out the complete picture for community scrutiny. Where I cite a fact, I'll note its source quality. Where I reason, I'll say so.
The structural facts (grounded, primary sources):
IMpactMF is a Phase 3, 2:1 randomized, open-label trial of imetelstat vs. non-JAKi best available therapy in ~320 patients with DIPSS intermediate-2 or high-risk myelofibrosis, relapsed/refractory to JAK inhibitors, with overall survival as the primary endpoint. First patient dosed April 13, 2021. Fifty percent enrollment December 2023. Full enrollment of 320 completed September 2025. Interim guided for H2 2026. (Geron press releases, ClinicalTrials.gov NCT04576156.)
The statistical design, from Geron's 10-K: powered at >85% to detect HR 0.60 (one-sided alpha 0.025); final analysis after >50% of enrolled patients have died (~160 deaths); interim after ~70% of final-analysis events (~112 deaths), with interim alpha spend ~0.01. This resolves a contradiction that circulated publicly: the interim is NOT at a 35%-information-fraction OBF look. It is at ~70% information fraction (112 of ~160 events), where an O'Brien-Fleming boundary naturally produces a critical threshold near Z=2.34, p≈0.01 — matching the disclosed alpha spend. (10-K is a primary source; this is the single most important clarification in the whole analysis.)
What the boundary actually requires (grounded math):
At ~112 events with 2:1 randomization, the standard error on ln(HR) is √(4.5/112) ≈ 0.20. To cross Z=2.34 the observed HR must be roughly ≤0.62. An observed HR of 0.56 produces Z≈2.89 — clearing comfortably. This means the trial does not need to replicate IMbark's extraordinary HR (~0.51 vs. matched RWD) to halt early. It needs roughly 0.62 or better. That's a meaningful and underappreciated point, and it's grounded in the disclosed design.
The BAT comparator (mix of grounded and inferred):
The BAT comparator (mix of grounded and inferred):
Grounded: the propensity-weighted IMbark-matched analysis (the population most comparable to IMpact) put BAT median OS at ~15-17 months (unweighted 15.4; ATO 17.0; sIPTW 16.1; HR ~0.51 across all three). Per the RWD discussion above, the contemporary ~34-40 month figures overstate the IMpact comparator due to risk-class contamination and banned JAKi sequencing.
Inferred: the true IMpact BAT arm likely sits somewhat above the historical 15-17 months due to era effects (earlier rux pivoting producing healthier patients at entry), but well below 34-40. A defensible inferred range is roughly 15-24 months. This is reasoned, not known. No one outside the DSMB has the actual figure.
The enrollment-timing arithmetic (grounded method, inferred inputs):
Using the verified enrollment shape (first patient 4/2021, 50% by late 2023, 100% by 9/2025) and a survival model checked at a ~9/2026 interim, the early-enrolled cohort has 34-65 months of follow-up by interim and the later cohort 12-34 months. Across BAT median OS assumptions from 15 to ~33 months, the model produces BAT deaths sufficient that the imetelstat arm needs only a modest death count to reach the 112 trigger — yielding crude HRs that clear the Z=2.34 boundary in every scenario up to roughly 37 months BAT OS.
Important honesty flags on this model: it uses a Weibull survival shape (k=0.85) that is an assumption, not measured; it assumes a survival-curve form for both arms; and "crude HR" derived from death-count ratios is an approximation of the formal Cox/log-rank HR the trial will actually report. These caveats could shift the precise numbers. They do not appear to change the directional conclusion, but a reader should know the model rests on assumptions.
Crossover (unverified source — flag clearly):
Geron's CMO reportedly stated at the Stifel conference (May 2026) that crossover impact is minimal — capped by protocol, requiring a washout period and sponsor approval, with physician reluctance to cross already-ill patients. I have not seen a transcript; this is secondhand from community reports. If accurate, it removes the largest dilution risk we'd otherwise model. If inaccurate or overstated, crossover could compress the observed ITT HR. This is a load-bearing input that deserves verification from the actual transcript before anyone relies on it.
The IMbark evidence base (grounded, high quality):
IMbark (Phase 2, 9.4 mg/kg) showed median OS ~28-31 months across data cuts, with seven orthogonal supporting endpoints — OS, bone marrow fibrosis reversal, VAF reduction in driver mutations, telomerase target engagement, dose-response consistency, biomarker-survival correlation, and activity in poor-prognosis subgroups. Peer-reviewed (JCO 2021), consistent across six years of conference scrutiny, with independent laboratory mechanistic validation. The internal low-dose comparator arm strengthens it from a single-arm study toward an internally controlled dose-response experiment. This is unusually strong Phase 2 data. The legitimate limitation: n=59 in the high-dose arm, no concurrent randomized OS control — which is precisely what IMpact exists to resolve.
-
mistergern
- Posts: 221
- Joined: Sat Mar 19, 2016 3:48 pm
Re: The poster from Dr. K is published on the Geron web site
Claude continued:
The DSMB context (mostly grounded):
The trial has run ~65 months through regular DSMB safety reviews without interruption, hold, or modification. Reportedly the same DSMB members as Phase 2 (secondhand, from the Stifel report). Sustained uninterrupted conduct is meaningful evidence against a hidden safety or futility problem — though it is not direct evidence of efficacy.
The disease-modification and Phase-2-to-Phase-3 translation question (genuinely contested):
The bull mechanistic argument — that earlier rux pivoting gives imetelstat a healthier marrow substrate to work on, expanding its effect — is plausible but unquantified and untested. I'll flag this as the weakest link in the strongly bullish cases circulating. It's reasonable that the era effect helps imetelstat. It's equally arguable that it lifts both arms proportionally, leaving the HR roughly unchanged, or even that earlier/milder discontinuers are less clonally driven and thus less ideal for a telomerase inhibitor. The direction is not established. The robust version of the bull case does not need imetelstat to exceed IMbark — it only needs IMbark-level efficacy against a depleted BAT arm.
Overall read (taking a position, as requested):
Weighing all of the above, an early efficacy halt at the H2 2026 interim is more likely than not. My honest estimate is roughly 65-75%, with the residual being either a continue-to-final outcome (HR positive but DSMB elects data maturity, ~15-20%) or a genuine negative/futility surprise (~10-15%, driven not by any identified red flag but by irreducible Phase 2→3 translation risk).
The single most likely specific scenario: BAT arm ~17-20 months, imetelstat ~30 months reproducing IMbark, observed HR ~0.50-0.56, clearing the boundary, halt announced late 2026.
The honest limits of this entire analysis — please read before weighting it:
The load-bearing uncertainties are (1) the actual imetelstat Phase 3 arm performance, unknowable until interim; (2) the true BAT arm OS, inferred not known; (3) the crossover characterization, secondhand and unverified; and (4) DSMB discretion to continue even if the boundary is crossed. Convergence among multiple AI analyses on similar numbers is NOT independent validation — we are all drawing from the same public well and can share the same blind spots. The probability estimate is a reasoned synthesis, not a forecast with the reliability that a single number implies. Treat the 65-75% as "the evidence leans clearly positive," not as a precise odds quote.
I'd genuinely welcome the board poking holes in any of this — particularly the Weibull assumption in the timing model, the crossover characterization for anyone with the actual Stifel transcript, and the BAT era-effect magnitude. Those are where the analysis is most likely to be wrong.
The DSMB context (mostly grounded):
The trial has run ~65 months through regular DSMB safety reviews without interruption, hold, or modification. Reportedly the same DSMB members as Phase 2 (secondhand, from the Stifel report). Sustained uninterrupted conduct is meaningful evidence against a hidden safety or futility problem — though it is not direct evidence of efficacy.
The disease-modification and Phase-2-to-Phase-3 translation question (genuinely contested):
The bull mechanistic argument — that earlier rux pivoting gives imetelstat a healthier marrow substrate to work on, expanding its effect — is plausible but unquantified and untested. I'll flag this as the weakest link in the strongly bullish cases circulating. It's reasonable that the era effect helps imetelstat. It's equally arguable that it lifts both arms proportionally, leaving the HR roughly unchanged, or even that earlier/milder discontinuers are less clonally driven and thus less ideal for a telomerase inhibitor. The direction is not established. The robust version of the bull case does not need imetelstat to exceed IMbark — it only needs IMbark-level efficacy against a depleted BAT arm.
Overall read (taking a position, as requested):
Weighing all of the above, an early efficacy halt at the H2 2026 interim is more likely than not. My honest estimate is roughly 65-75%, with the residual being either a continue-to-final outcome (HR positive but DSMB elects data maturity, ~15-20%) or a genuine negative/futility surprise (~10-15%, driven not by any identified red flag but by irreducible Phase 2→3 translation risk).
The single most likely specific scenario: BAT arm ~17-20 months, imetelstat ~30 months reproducing IMbark, observed HR ~0.50-0.56, clearing the boundary, halt announced late 2026.
The honest limits of this entire analysis — please read before weighting it:
The load-bearing uncertainties are (1) the actual imetelstat Phase 3 arm performance, unknowable until interim; (2) the true BAT arm OS, inferred not known; (3) the crossover characterization, secondhand and unverified; and (4) DSMB discretion to continue even if the boundary is crossed. Convergence among multiple AI analyses on similar numbers is NOT independent validation — we are all drawing from the same public well and can share the same blind spots. The probability estimate is a reasoned synthesis, not a forecast with the reliability that a single number implies. Treat the 65-75% as "the evidence leans clearly positive," not as a precise odds quote.
I'd genuinely welcome the board poking holes in any of this — particularly the Weibull assumption in the timing model, the crossover characterization for anyone with the actual Stifel transcript, and the BAT era-effect magnitude. Those are where the analysis is most likely to be wrong.
-
biopearl123
- Posts: 2591
- Joined: Fri Jul 20, 2018 5:13 pm
Re: The poster from Dr. K is published on the Geron web site
MG, as always appreciate your excellent work. Clearly there is a lot we don't know and can't know but there are some things your analysis brings up that really help get clarity with the new Dr. K et. al. poster. Here are a few additional thoughts:
1. There are four Geron co authors on this publication. A least two will be under the spotlight at Goldman.
2. The post 2019 data does include low risk and intermediate 1. Perhaps they have been censored but I don't think so since there was no footnote to that effect.
3. I have been counting "tick marks", those little parallel lines that indicate when a patient is "censored". I count 12 before reaching Mos. Don't know if any of these censored patients left the study for another line of therapy including Imet. That we don't know but the authors do.
4. The statistical weight of finding a reportable mOS of 39.8 needs a close look. To my eye when the 50% mos mark is crossed there are only 4 patients under evaluation at that time. Not sure what to make of that and it bears closer examination since it probably falls into a gray (or black) zone for what is a true statistically valid number.
5. Goldman please take note. This data is in the public domain, we have all seen it. So have you. You will be talking with Geron authors next week in a public forum. It would be nice to get some clarity. Presumably that's why these forums exist.
6. Pretty clear everyone freaked out about "new" estimated "modern" BAT from Moffett especially if its the BAT arm that is prolonging the study against the currently accepted Imetelstat arm and might derail our hopes for a positive interim analysis..
1. There are four Geron co authors on this publication. A least two will be under the spotlight at Goldman.
2. The post 2019 data does include low risk and intermediate 1. Perhaps they have been censored but I don't think so since there was no footnote to that effect.
3. I have been counting "tick marks", those little parallel lines that indicate when a patient is "censored". I count 12 before reaching Mos. Don't know if any of these censored patients left the study for another line of therapy including Imet. That we don't know but the authors do.
4. The statistical weight of finding a reportable mOS of 39.8 needs a close look. To my eye when the 50% mos mark is crossed there are only 4 patients under evaluation at that time. Not sure what to make of that and it bears closer examination since it probably falls into a gray (or black) zone for what is a true statistically valid number.
5. Goldman please take note. This data is in the public domain, we have all seen it. So have you. You will be talking with Geron authors next week in a public forum. It would be nice to get some clarity. Presumably that's why these forums exist.
6. Pretty clear everyone freaked out about "new" estimated "modern" BAT from Moffett especially if its the BAT arm that is prolonging the study against the currently accepted Imetelstat arm and might derail our hopes for a positive interim analysis..
Re: The poster from Dr. K is published on the Geron web site
I hate to admit it, but these discussions are making me dizzy. It seems the Dr. K abstract of analysis protocols and RWD results is comparing apples and oranges with the IMET MF P2 results and IMpact P3 CT design.
-
biopearl123
- Posts: 2591
- Joined: Fri Jul 20, 2018 5:13 pm
Re: The poster from Dr. K is published on the Geron web site
Lacour, what made me dizzy was seeing a piece of information that was such an outlier that I and others sought to understand what was behind it. It certainly did not comport with what Geron and the MF investigators we have come to know and trust have been telling us all along. In addition, it gave myself and a few other board members something to do rather than just accept it at face value. Whether our (semi) conclusions are correct is another story but I would be reluctant to just let something like this stand without some scrutiny even if it turns out to be way off base.