REGAL Scenario Explorer

The blinded milestones (60/72/78 deaths) stay fixed, so the pooled survival is always re-calibrated and only the split between arms moves. The headline is the plateau (GPS-cure) probability of success. The second panel is a null test, not a co-equal probability: it holds the BAT arm bit-for-bit identical and swaps only the GPS responder component — durable cure versus a fitted heavy-tailed Weibull with no cure — to ask whether the milestone plateau requires a GPS-specific durable benefit. It returns a three-state verdict: null rejected (GPS cure required), rejected (inconsistent), or not excluded (a no-cure GPS heavy tail also fits, given this BAT). Research tool, not investment advice.

Control (BAT) arm composition

Weights auto-normalize. Venetoclax cure is the key bear/bull knob (plateau panel).
ComponentMedian (mo)Cure %Shape k
Per-component survival inputs. Non-responders use the Observation row. Shape k: <1 heavy tail (more long-term survivors), 1 = exponential, >1 accelerating. Applies to the plateau panel's cure-mixture components.
The component medians above are for all CR2 transplant-ineligible patients, but the trial's eligibility bar (incl. REGAL's "life expectancy > 6 months" criterion) enrols a healthier subset. This left-truncates the pooled pool — discard the earliest-dying X%, keep the longest-surviving (100−X)% — via the conditional survival min(1, S/(1−X)); the near-flat early segment is real guarantee time, not an artifact. Applied identically to both panels before the arm split, so it cannot bias the within-trial comparison (HR stays ≈ invariant to X). The cured fraction is raised to c/(1−X). 0% = component medians taken at face value.

No-GPS-cure test — GPS tail shape

In auto mode the GPS responder tail shape sG is a fitted parameter of the no-GPS-cure null (alongside the GPS responder median mG) — the slider just displays the fitted value. Tick override to pin it and explore: sG < 1 = heavier tail; sG → 1 is exponential. sG is free to go genuinely heavy — that is what lets a no-cure Weibull try to mimic a plateau. If the fit lands on a boundary (mG at its cap, or sG pinned at an edge) the null is rejected — GPS-specific cure required.
Note (plateau panel): BAT longevity comes from the component medians and the enrollment-selection slider — the milestone deceleration is explained by a healthier enrolled cohort. The fit's one free parameter is the GPS responder cure.

GPS arm & trial

~20% anchored to the interim's ~80% WT1 response; they track the Observation row.
Slides accrual earlier (flat) ↔ later (back-loaded).
Anchored to SELLAS PRs (~20 by Apr 2022, 104 by Nov 2023, 126 by Apr 2024); those anchors pin the median to ~Q1–Q2 2023.
All-cause background mortality for a population mostly in their sixties, applied equally to both arms as a competing risk. It thins survival (incl. the plateau), brings the trigger forward, and — being non-differential — dilutes the treatment effect. ~2%/yr ≈ US ages 60–69; raise to stress-test.
Administrative censoring / dropout: patients who leave the study before dying are censored (no event), which slows event accrual and can stall the 80th-event trigger. 0 = complete follow-up; trials of this type run ~3–10%.
Event data & design
MilestoneDateDeaths
Enrolled (N)
Final-analysis events
Significance HR threshold
Interim-analysis events
Interim futility HR
Interim futility HR: the trial cleared the IDMC futility look at the 60-event interim, so a scenario whose implied HR at that point exceeds this threshold is inconsistent with reality (it would have stopped). Flagged below.

Modeled probability of a positive readout

–
Plateau (GPS cure) — headline
HR –
–
No-GPS-cure null 
HR –
Both conditional on the 80th event occurring · threshold HR ≤ 0.636 (one-sided 0.025).

Survival curves

BAT (control) GPS Pooled (blinded) event-fraction levels

Trial dynamics

Event accrual — when does the 80th event fire? modeled cumulative deaths vs calendar; dots are the blinded 60/72/78 milestones
Plateau accrual No-GPS-cure accrual blinded milestones 80-event trigger & projected dates
Distribution of simulated final hazard ratios each trial is one draw; P(success) is the mass left of the threshold
Plateau No-GPS-cure significance threshold
Where GPS-cure and no-GPS-cure disagree both pooled curves are pinned to the same milestones, then fan apart in the tail
Plateau (GPS-cure) pooled No-GPS-cure pooled the gap the data can't resolve event-fraction levels
Enrollment validation modeled cumulative enrollment vs the sourced public PR counts (~20/104/126)
modeled enrollment sourced PR anchors
P(success) vs treatment effect how P(success) moves with the true GPS-vs-BAT effect; the shaded band is the effect the blinded data allow, ★ the current fit
Plateau No-GPS-cure success threshold (HR≤0.636) effect the data allow
Enrollment selection lifts the BAT arm BAT median OS and cure fraction as the drop-weakest slider q sweeps 0→50%; dot = current q
BAT median OS (mo) BAT cure fraction (%) defensible band (~20–35%)