What a New 2,610-Patient AML Registry Study Does and Does Not Tell Us About SELLAS’s REGAL Control Arm

A retrospective, non-randomized study published in Leukemia on October 6, 2026 analyzed 2,610 patients from the PETHEMA registry. All had newly diagnosed acute myeloid leukemia (AML) and were judged unfit for intensive chemotherapy. Patients treated with venetoclax-based regimens lived longer than those given a hypomethylating agent (HMA) alone: median overall survival was 11.4 months versus 8.2 months (HR 0.66). Among patients who achieved a composite complete remission (CR, CRi or MLFS), the study found no statistically significant difference in disease-free survival: medians were 10.3 versus 11.5 months, HR 0.88, 95% CI 0.74 to 1.05 (Leukemia, 2026).

The study is relevant to SELLAS Life Sciences (SLS) because these drugs are among the options in the control arm of REGAL, the Phase 3 trial of galinpepimut-S (GPS). How long that control arm survives is one of the main open questions in the trial.

The biggest caveat is that the two studies use the drugs in different settings. The registry looked at first-line treatment of newly diagnosed patients. REGAL looks at patients who relapsed, were brought back into a second remission, and then received either GPS or best available therapy afterwards. The PETHEMA study provides context for drugs permitted in REGAL’s control arm, but it neither estimates that arm’s survival nor resolves whether modern therapy explains the trial’s slow event accrual.

The study at a glance

The patients came from 188 centers in Spain, Portugal, Argentina, Colombia and Chile, and patients treated in front-line clinical trials were excluded. Follow-up was updated to November 13, 2025, with a median of 25.3 months. Of 622 responders with transplant data, 53 (8.5%) went on to an allogeneic stem cell transplant in first remission (Leukemia, 2026).

The authors also note that real-world results trail trials. The registry’s venetoclax median of 11.4 months is shorter than the 14.7 months in the VIALE-A trial. They cite a meta-analysis of 19 real-world cohorts with a pooled median of 9.37 months (11.5 months for venetoclax plus azacitidine). They also cite the prospective REVIVE study, in which patients who met VIALE-A’s eligibility criteria had a median overall survival of 17.8 months, versus 10.7 months for those who did not (Leukemia, 2026). That gap reflects how patient selection and prognosis are linked. It does not isolate an effect of trial eligibility itself.

Why this study comes up in the REGAL debate

REGAL is an event-driven trial: its final analysis is triggered by the 80th death across both arms. SELLAS reported that 78 events had occurred as of May 11, 2026 (SELLAS 8-K, May 12, 2026). In August it said it was “approaching the pre-specified 80th event” and would announce when it occurs (SELLAS 8-K, August 11, 2026). As of October 8, 2026, no such announcement has been filed.

Events have come in more slowly than SELLAS once forecast. This suggests that patients overall are living longer than originally assumed, but the blinded count does not say in which arm. Better survival on GPS is one possible explanation. A control arm doing better than assumed is another. On a November 2022 call, SELLAS said statistical significance would correspond to an estimated hazard ratio of about 0.636, “for example” 12.6 months on GPS versus 8 months on best available therapy (BAT). On the same call it said the original statistical plan had assumed about 8 months for the whole study population, and a BAT median of 5.4 months in that plan’s example (SELLAS 8-K, November 15, 2022, call transcript). These are illustrative design figures, and these public examples do not establish the control-arm assumption used in the final statistical analysis plan.

The registry’s headline overall survival numbers are not the closest comparison for REGAL. Every REGAL patient is in remission and alive at randomization, while the registry’s overall survival also counts patients who never responded and those who died early. Disease-free survival among responders is closer, because those patients were in remission. Among responders, the study did not demonstrate a statistically significant difference in disease-free survival, and that does not establish equivalent remission duration. The authors interpret venetoclax’s overall survival benefit as driven mainly by the higher remission rate and lower early mortality.

The registry does not directly test the strong-control-arm explanation for REGAL. It measured how long first-line responses lasted. It did not measure how well the same drugs maintain a second remission, which is the question REGAL’s control arm answers.

Where the comparison breaks down

Using overall survival rather than disease-free survival includes survival after relapse, but because REGAL also differs in patients and starting point, this does not establish how REGAL survival compares with the registry. SELLAS has not published REGAL’s baseline characteristics or how control patients were split among the BAT options, so it is unknown how many received the drugs the registry studied. And the registry’s disease-free survival comparison covers only responders, who were a far smaller and differently selected group on HMA (20%) than on venetoclax-based therapy (64.4%).

What changes in our earlier analysis, and what does not

In July we published two pieces on REGAL. Why SELLAS Might Fail argued that the control arm may be stronger than old AML CR2 assumptions. What the Public Data Can and Cannot Tell Us modeled the pooled event counts (60, 72 and 78 deaths). Under its assumptions, the best-fitting model put the pooled median overall survival (both arms combined) at about 15.5 months, with point estimates of roughly 6.5 to 17.7 months across models. It also estimated that at 80 events an observed hazard ratio near 0.64 is roughly the significance boundary, and that 80% power requires a true hazard ratio around 0.53. These boundary figures are standard approximations that assume two-sided 5% significance and roughly equal arms; the final analysis plan and the interim analysis affect the exact boundary.

The registry adds outside context: in a large real-world population, the first-line drugs permitted in REGAL’s control arm produce modest results. It does not directly revise our previous model estimates, which remain conditional on the enrollment, censoring and survival assumptions described in the July analysis. Nor does it tell us which row of that analysis’s conditional scenario tables, covering BAT medians from about 7 to 15.5 months, matches reality.

What will settle it

• The 80th event announcement. SELLAS has said it will announce it, followed by database lock, blinded data review, statistical analysis, unblinding and topline results (SELLAS 8-K, August 11, 2026).
• Arm-level results. The control arm’s median overall survival, the hazard ratio with its confidence interval, the Kaplan-Meier curves and the censoring pattern. See also PFS vs OS in cancer trials.
• The BAT mix and baseline characteristics. These will show how relevant a first-line registry is to REGAL’s control patients.

BioRadar’s SLS page lists SELLAS’s new SEC filings as they arrive.

Frequently asked questions

Does the PETHEMA study show that REGAL’s control arm will live about 10 to 11 months?

No. The 10.3 and 11.5 month figures are median disease-free survival among newly diagnosed patients who responded to first-line treatment, counted from first response to relapse or death. REGAL measures overall survival from randomization in patients in second remission. The patients, the setting, the endpoint and the starting point all differ, so the registry cannot be converted into a REGAL estimate.

What drugs are in REGAL’s control arm?

Investigator’s choice of best available therapy: observation with palliative hydroxyurea allowed, an HMA (azacitidine or decitabine), venetoclax, low-dose cytarabine, or combinations of these, per ClinicalTrials.gov (NCT04229979). SELLAS has not disclosed how control patients were distributed among these options.

Has REGAL reached 80 events?

No announcement has been filed. SELLAS reported 78 events as of May 11, 2026 (8-K filed May 12, 2026), and said on August 11, 2026 that it was approaching the 80th event and would announce it. As of October 8, 2026, no such announcement had appeared in its SEC filings.

Why does the control arm matter so much for REGAL?

With 80 events, the trial has adequate statistical power only for a large treatment effect under the assumptions used in our July analysis. That analysis estimated the significance boundary at an observed hazard ratio near 0.64, and SELLAS has cited about 0.636. The longer the control arm lives, the longer GPS patients must live to produce a hazard ratio that low.

Sources

• Aspas Requena G, et al., Montesinos P. Leukemia (2026), published online October 6, 2026: Venetoclax-based therapy versus hypomethylating agents alone in unfit patients with newly diagnosed acute myeloid leukemia: a real-world study from the PETHEMA registry
• Wolach O, et al. Blood Advances (2025): Trial eligibility, treatment patterns, and outcome for venetoclax-based therapy in AML: a prospective cohort study (REVIVE)
• ClinicalTrials.gov: NCT04229979, REGAL (arms, eligibility, primary endpoint, sites)
• SELLAS 8-K, Exhibit 99.1, May 12, 2026: first quarter 2026 results, 78 events as of May 11, 2026
• SELLAS 8-K, Exhibit 99.1, August 11, 2026: second quarter 2026 results, approaching the 80th event
• SELLAS 8-K, Exhibit 99.1, November 15, 2022: third quarter 2022 call transcript, significance threshold and survival assumptions
• SEC EDGAR: SELLAS 8-K filings (checked October 8, 2026)
• BioRadar: What the Public Data Can and Cannot Tell Us About SELLAS’s Galinpepimut-S in AML (pooled-event model, power at 80 events)
• BioRadar: Why SELLAS Might Fail: The Hidden Risks Behind GPS, REGAL, and the AML CR2 Bull Case

Data as of October 8, 2026. This article is for information only and is not investment advice.

BioRadar analyses take no bull or bear side, make no buy or sell calls and are never paid for. Read our editorial policy.