To make an informed decision to invest in a research stage biotech company one has to face two analytical tasks. The first is to evaluate the net present value (NPV) of the therapy under development, which is typical done either through a discounted cash flow (DCF) model or as a multiple of peak sales. The other task is to assess the probability of success (PoS) that is, the likelihood that the therapy will ultimately achieve regulatory approval. In this article we concentrate on the second task. Using SELLAS Life Sciences’ REGAL study as an example we illustrate how Monte Carlo simulations can provide insights into the probability of success of a clinical trial.
Background
Sellas Live Sciences has two clinical assets under development. SLS009 (tambiciclib) a selective CDK9 inhibitor for relapsed/refractory acute myeloid leukemia (AML) and Galinpepimut-S (GPS), an immunotherapeutic agent targeting the WT1 antigen, currently in the pivotal Phase 3 REGAL trial also in AML. Although both programs are initially focused on AML, their potential therapeutic applications extend well beyond this indication. For the purposes of this analysis, however, our primary interest lies in the REGAL study.
REGAL is a pivotal Phase 3 study designed to demonstrate the effectiveness of GPS as a maintenance therapy for patients with AML in second complete remission (CR2). REGAL is a two-arm trial comparing GPS with best available therapy. The primary endpoint is overall survival. Enrollment started in November 2020 and, largely due to the impact of COVID-19 and other factors, was only completed in April 2024. The original design targeted approximately 116 patients, with interim and final analyses planned at 80 and 105 events. However, due to longer than expected overall survival in the pooled, blinded population, the statistical analysis plan was amended. Ultimately, 126 patients were enrolled, the interim analysis was triggered at 60 events, and the final analysis is planned at 80 events.
According to SELLAS, a statistically significant advantage over the standard of care is achieved if the hazard ratio (HR) falls below 0.636. This would also be clinically meaningful. Given the favorable safety profile and AML CR2's orphan disease designation, regulatory approval would be highly probable should the trial meet this endpoint.
Method
Over the last twelve months, we have received a substantial amount of information from SELLAS that helps us assess the probability of success for REGAL. The 60 event occurred in early December or late November 2024 and the Independent Data Monitoring Committee (IDMC) recommended that the trial continue without modification. No unexpected safety signals were identified. The study was not halted for futility nor for efficacy. A time-scheduled interim analysis was conducted presumably in August 2025, and again the IDMC recommended that the trial continue without modification.
To incorporate this information into an estimate of the PoS, a common approach is to begin with the overall historical success rates of oncological trials and adjust them based on study-specific factors. A more rigorous and quantitatively transparent alternative is the use of Monte Carlo simulations. How this is done is demonstrated in the following. Skip to results if you are not interested in the details.
We are interested in the PoS conditional on the observed data. By Bayes’ theorem, this requires computing the posterior probability that the trial succeeds given specific assumed OS values in the BAT and GPS arms, weighted by the likelihood that these OS values are true (i.e., the posterior probability of those OS values given the data).
We accomplish this by generating a large number of Kaplan–Meier curves for each pair of OS values (e.g., 7 months for OS BAT and 20 months for OS GPS). We then eliminate all curves that do not fit the data and count the results (whether the trial was successful or not). OS value pairs that are more likely to produce Kaplan–Meier curves fitting the data will be less likely to be eliminated. By normalizing with the number of surviving curves and multiplying by the prior probability, we obtain the posterior probability of each OS value pair. Summing these and applying the corresponding PoS for each pair yields the overall PoS for REGAL.
The a priori probability of the OS values follows a log-normal distribution with a median and standard deviation that depend on chosen parameters. The Kaplan–Meier curve is generated by drawing individual survival times for each patient in each arm from a Weibull distribution. The Weibull distribution has an advantage over the exponential distribution because it allows for variable hazard rates. The shape parameter k of the Weibull distribution also depends on the chosen settings.
The initial choice of the parameters is mOS of 10 month for BAT, the higher end of the historical standard of care. MOS of 18 month for GPS based on the Phase 2 data but with a small correction to adjust for the usual biases which occur in clinical trials. k of 0.5 for BAT and 0.6 for GPS. This parameter is the hardest to estimate and the choice is somewhat arbitrary, but should be conservative.
The filters used to determine whether a curve fits the observed data are as follows: The trial must not have been stopped for futility at the 60th event. We calculate the conditional power and exclude all curves with a conditional power below 10%. Similarly, the trial must not have been stopped for efficacy. We calculate the p-value at the same event and exclude all curves with a p-value below 0.005, which corresponds to the alpha spending assumed for the interim analysis. Additionally, we exclude all curves for which the 80th event has already occurred. Finally, we reject all curves that would have led to futility at the interim analysis in summer 2025, that are those with a conditional power below 10% at the 21-month mark.
One limitation of this approach is that we assume recruitment occurred at a single point in time, whereas in reality, recruitment was spread over several years. This is not expected to change the overall trends observed in the results, as discussed below.
The source code will be provided on git hub in near future.
Results
With the assumptions outlined above, we find that the probability of success (PoS) is around 77%. This is considerably better than the baseline for oncological phase 3 trials, however, prudent risk management remains advisable for those investing in Sellas Life Sciences.
The calculated PoS may appear low for some investors. A key reason is the decreasing hazard rate in this setting, which reduces the ability of the rejection criteria to filter out ultimately unsuccessful curves. If constant hazards are assumed instead (shape parameter k equal to 1 in both arms), the PoS typically exceeds 90 %, even with overly pessimistic effect priors. Thus, in this particular setting, the PoS is driven more by the assumed shape of the survival distribution than by the location (median OS) of the prior.
Discussion
Despite the simplifying assumption of instantaneous recruitment, extensive sensitivity checks (varying accrual patterns, shape parameters, and filter stringency) show that the directional conclusion and the order of magnitude of the PoS remain robust. In summary, 77% represents a realistic, biology-informed estimate of success probability for REGAL that appropriately penalizes the limited informativeness of interim survival data in the presence of declining hazard rates. Realistic choices for shape will ultimatly depend on the biological dynamics at work. This supports cautious continued investment in SELLAS while underscoring that, in this specific setting, even favorable interim signals do not translate into near-certain success.
Disclaimer
This article is provided for informational and educational purposes only. It is not intended to be, and should not be construed as, financial, investment, legal, or professional advice of any kind. The analyses, calculations, and opinions expressed herein represent the personal views and research of the author and are based on publicly available information as well as certain assumptions and methodological choices that are explicitly described.Nothing in this article constitutes a recommendation to buy, hold, or sell shares of SELLAS Life Sciences Group, Inc. (NASDAQ: SLS) or any other security, nor does it represent investment advice tailored to the specific situation of any individual reader. Past performance and interim clinical data are not indicative of future results.Investments in clinical-stage biotechnology companies are inherently speculative and carry a high risk of total loss. Readers are strongly encouraged to perform their own due diligence and to consult qualified financial, tax, and medical advisors before making any investment decision.The Monte Carlo simulations and source code described are research tools developed independently by the author for educational illustration; they have not been reviewed or endorsed by SELLAS Life Sciences or any regulatory authority. The author has no financial relationship with SELLAS Life Sciences Group, Inc. or any related entities. He has not received compensation, consulting fees, equity, or any other form of payment or benefit directly or indirectly from the company, its management, investors, or any third parties associated with SELLAS. The author currently holds a long position in SLS shares and options that profit from a rising share price. This analysis was conducted independently and solely for educational purposes.