Author: Erica Jonas, VP, Services Delivery
Snapshot:
Precision oncology has changed what a clinical trial looks like. Where trials once focused on cancer by location — lung, breast, colon — the science now tells us that patients sharing a genetic mutation have more in common with each other than with patients who share a tumor type but differ at the molecular level. Basket trials were designed to follow that logic: test a single targeted therapy across multiple tumor types and indications simultaneously, grouped by the mutation rather than the anatomy.
The promise is real. Basket trials can reach efficacy signals with fewer patients and shorter timelines than conventional designs. But they introduce statistical complexity that conventional trial management approaches aren’t built for — and as the field has matured, so have the methods sponsors use to navigate it.
In a basket trial, not every cohort performs equally. Some tumor types respond strongly to the investigational therapy. Others show early signs of futility. The question sponsors face at every interim analysis is: which cohorts should stay open, and which should close?
Bayesian Futility Analysis has become the methodology of choice for answering that question rigorously. Rather than waiting for a fixed checkpoint to evaluate a cohort, Bayesian methods allow sponsors to assess the Probability of Success continuously — updating the analysis as new patients enroll and new effectiveness data comes in. When a cohort’s probability of success drops below a defined threshold, enrollment into that tumor-type cohort closes. New patients with that indication stop being enrolled, protecting them from a treatment that the evidence suggests won’t work for their cancer type, while the trial continues in the cohorts where it is showing promise.
The statistical methods continue to evolve. Researchers are moving beyond evaluating each cohort in isolation — Bayesian hierarchical models can allow information borrowing across baskets, where response patterns in one tumor-type cohort provide weak but meaningful signal about others sharing the same mutation and therapy. This approach has the potential to improve statistical power for smaller or rarer indications that struggle to reach the sample sizes needed for a confident interim look. Interim analysis frequency is also increasing — novel Bayesian methodology can support continual assessments after each patient observed, rather than at fixed checkpoints, a level of granularity that was computationally impractical not long ago.
The operational question these advances raise is the same one Suvoda has already been solving for: how do you execute on sophisticated statistical logic in real time, at the pace of patient enrollment, without creating a manual burden that slows the trial down?
This is where Suvoda RTSM, the next evolution of Suvoda IRT, comes in. We built Suvoda RTSM to automate the complexities of adaptive oncology trial design — and Bayesian Futility Analysis is a direct application of that capability.
Working closely with a sponsor, we developed an automated two-way integration between Suvoda RTSM and the sponsor’s statistical algorithm. The result: real-time data flow between the RTSM / IRT and the Bayesian model, with cohort enrollment decisions driven by live data rather than manual calculation.
Here’s how it works:
Integrating the statistical algorithm into Suvoda RTSM. The integration required both Suvoda RTSM and the statistical algorithm to exchange data in real time — determining which cohorts should remain open or closed for enrollment, and which cohort each incoming patient should be assigned to.
Automatic enrollment using real-time data. Suvoda RTSM and the statistical algorithm operate as a continuous loop, not a one-time transaction. As patients progress through the trial, site users enter two data points for each enrolled patient:
Suvoda RTSM uses those inputs to automatically maintain each patient’s evaluability status — tracking, at any given moment, how many patients in each cohort are evaluable and how many are responding. That running picture is what feeds the algorithm.
When a new patient is ready to enroll, Suvoda RTSM automatically sends the most current cohort-level data to the statistical algorithm — not the new patient’s data, which doesn’t exist yet, but the aggregate status of every patient already in the study. The algorithm performs the Bayesian calculations and returns the current open/closed status for each cohort. If the incoming patient’s tumor-type cohort is open, they enroll. If it has closed due to futility, they don’t. No statistician intervention required.
The result is that every enrollment decision is grounded in the most current picture of the trial — and that picture updates continuously as existing patients hit new milestones.
Built-in flexibility for mid-study changes. Oncology trials change. Suvoda RTSM was built with that as a given. Study teams can update data points without triggering system change orders, manually override both the integration and individual patient evaluability or responder status, and correct site data errors quickly — all without programming changes or delays.
The efficiency gains are direct:
The broader implication is that adaptive basket trial designs become operationally viable at a scale they weren’t before. The statistical sophistication the field has developed over the past several years — continuous interim analyses, information borrowing across cohorts, real-time Probability of Success assessment — is only as useful as the operational infrastructure that can execute it. Suvoda RTSM provides that infrastructure.
Ultimately, this is about patients. Every patient enrolled in a cohort trending toward futility is a patient who could have been in a different trial. Automating the enrollment decision — grounding it in live data and rigorous statistical logic — is one of the most concrete ways clinical trial technology can reduce the human cost of drug development.