Rewiring Clinical Trials: Predictive, Collaborative, and Federated
The Landscape
Clinical trials are desperately in need of some rewiring. Despite advances in technologies and greater access to data, clinical trials consistently struggle to meet timelines and endpoints.
Many of the challenges trial teams faced decades ago remain the same today, not because those teams lack expertise or resist innovation, but because they continue to operate in a fragmented, non-learning system—rightfully controlled by strict security and privacy guardrails. That persistent system is a constant roadblock to enabling advancements—especially with AI—in achieving their full potential in improving trial success. The FDA recognized as much this spring, launching a major, two-pronged initiative to use AI for real-time trial monitoring and for accelerated drug review timelines.
The Problem Hasn't Changed–But the Stakes Have
Teams struggle to match protocol design with real-world patient populations, largely due to limited visibility across datasets. At the same time, protocols have become more complex over the past decade (more endpoints, procedures, eligibility criteria).
The increased complexity is directly correlated with higher dropout rates, longer cycle times, and more protocol amendments. A significant portion of amendments are due to feasibility miscalculations, enrollment challenges, and safety adjustments. With every amendment, teams are forced into reactive mode because they can’t predict issues upfront. Ultimately, many trial failures are not purely scientific, but tied to poor patient selection, suboptimal trial design, and incomplete safety/efficacy signals.
Utilizing AI to Move from Discovery to Approval, Faster
Similar to how the explosion of electronic medical record (EMR) data nearly two decades ago transformed research with more comprehensive insights into patient journeys, AI has the potential to rapidly accelerate new, more targeted treatments from discovery through approval. Early signals show that AI-derived molecules have an 80-90% success rate in Phase 1 compared to the historical industry average of 50%. At the same time, computational platforms are compressing drug discovery-to-clinic timelines, with reports of some candidates advancing to clinical trials within 18 months compared to traditional timelines of six years.
.png)
The evolution of large language models and the advent of generative AI are also driving transformations within key aspects of trial operations, especially around predicting key success drivers, such as feasibility and patient enrollment. Researchers, model developers, and biopharma are racing to operationalize these capabilities. Take TrialBench: an open-source, multimodal dataset initiative designed to tackle “AI-solvable” trial challenges like duration, dropout, adverse events, approval likelihood, and even dose selection. Importantly, this highlights just how much of the trial lifecycle is now within reach of AI.
As powerful models are developed and methods evolve, many AI initiatives will stall as soon as they hit the same persistent operational roadblocks, restricting essential collaboration, deployment, and data access due to privacy, regulatory, and compliance concerns. Federated computing is what makes AI deployable in the real-world constraints of clinical trial operations. While TrialBench shows what’s possible with the right data, it can’t drive change when that data is isolated and “untouchable.” Federated computing is what makes that data actually accessible in practice.
Within traditional computing constructs, data centralization precedes data analysis, modeling, and other computational processes that create value from the data. In contrast, federated computing is a privacy-native approach that sends the models or code to the data while the data itself remain decentralized. Instead, the compute happens where the data resides (securely behind firewalls) and only aggregated model parameters (i.e. learnings) or privacy-preserving results (e.g. insights) are shared back to a central ‘control plane.’
With federated computing, AI can securely learn from deeper and more diverse representative datasets (i.e., sensitive patient data across disparate systems [e.g., EHR + DICOM] and distinct geographies) whether the data exists within internal silos or with external partners or sites. As regulatory bodies from the FDA to NIST move quickly to define what trustworthy AI looks like in critical infrastructure environments—such as the hospital networks, EHR systems, and research institutions that clinical trials depend on—federated computing is emerging as the natural architecture for deploying AI securely, collaboratively, and at scale. Federated computing is designed to operate within regulated, distributed systems—ultimately creating a powerful, collaborative AI approach that removes the need to centralize sensitive data. Organizations no longer have to choose between insight and compliance, and they can deliver results faster because data movement restrictions no longer slow down organizational momentum.
Federated AI in Action: Pre-Screening Without Compromising Privacy
Clinical trial pre-screening is a high-value application of federated AI. Traditionally, identifying the right patients requires manual, labor-intensive chart reviews or the centralized pooling of sensitive patient data—a process that often hits a brick wall due to privacy regulations (HIPAA, GDPR) and institutional silos. By deploying prescreening algorithms via federated architecture, sponsors can securely screen distributed patient populations without ever exposing patient-level data, resulting in faster recruitment and more representative trials.
In practice, federated AI changes how sensitive data is used. Instead of a pharmaceutical sponsor asking a hospital to send their patient records, the process follows:
- Local Screening: A prescreening algorithm runs in parallel at multiple hospital sites, where it identifies eligible patients locally. This happens behind the hospital’s own firewall, thus avoiding traditional compliance risks.
- Insight Sharing: Trial sites only see patients flagged as likely eligible at their own sites, while sponsors see aggregated statistics to understand how the algorithm is performing across sites. Shared insights enable cross-network learning and optimization without sharing underlying patient data.
- Cross-Site Trial Intelligence: Scientists at the sponsor can analyze and interrogate eligibility counts, performing data exploration and statistical tests to understand the shape of the data across all of the hospital sites. Sponsors can identify the highest-yielding sites across the network for their studies, flag to trial site staff the patients deemed likely eligible for their trial who should pursue consenting and traditional screening.
- Continuous Network Learning: Insights from patient screening outcomes across the global trial network are continuously aggregated to refine the prescreening algorithm. This enables near-real-time optimization of enrollment strategies, improved identification of likely eligible patients, and progressively better performance across all participating sites.
Building Better Trials From the Ground Up
Federated AI goes beyond patient identification; it enables sponsors to design better trials from the start by learning from distributed, real-world and partner data:
- Stress-test protocol design: Run “what-if” scenarios on inclusion/exclusion criteria (e.g., adjusting thresholds such as required glucose level, from <120 to <130mg/dL) to quantify impact on eligible populations across global sites.
- Assess feasibility & diversity: Model patient populations across geographies to ensure representativeness and identify enrollment risks early, reducing costly mid-trial amendments.
- Optimize site selection: Identify high-yield sites based on actual patient demographics and availability, not assumptions, ensuring sites can enroll against protocol requirements.
As AI expands industry-wide possibilities, how exactly to embed it in clinical trial operations remains the limiting factor. While AI may get us closer to fixing what we test, federated AI is getting us closer to fixing how we test. Ultimately, federated AI enables a critical shift: rewiring siloed, reactive clinical trials to become collaborative, predictive systems designed to deliver better outcomes and more effective treatments to patients faster. The FDA’s commitment to real-time monitoring pilots makes it clear this shift is already underway. Federated AI is the architecture poised to meet it.
Explore some of our landmark federated AI programs:
Ready to see federated AI in action? Schedule a demo with Rhino Federated Computing today.
