Closing the Dead Time Gap in Clinical Data: Q&A with Raviv Pryluk, PhaseV

The promise of real-time clinical trial monitoring runs headlong into a problem most discussions skip over: the gap between when data is measured and when it is actually clean enough to analyze. Closing that gap requires more than streaming infrastructure—it requires rethinking how data is standardized, validated, and made analysis-ready at speed.

To explore this further, Applied Clinical Trials recently caught up with with Raviv Pryluk, PhD, CEO and co-founder of PhaseV, about what has to happen on the back end before real-time monitoring becomes meaningful, how artificial intelligence (AI) is beginning to compress the standardization bottleneck, and what the industry needs to get right in responding to the FDA's push toward continuous data review.

ACT: What is the "dead time" problem in clinical data, and why does it undermine the promise of real-time trial monitoring?

Pryluk: There is a delay. We're collecting data in a clinical trial—trying to show that one drug is better than standard of care or placebo—patients are being recruited, data is being measured, but it takes time before that data can be used for analysis. We need to clean the data, validate the data, and there are so many processes before we can refer to it as ready for analysis. This delay prevents us from analyzing the data in real time unless we have certain technologies and solutions in place. Without them, there is a gap between when the data was actually measured and when it can be analyzed.

ACT: Real-time data streaming gets a lot of attention, but what has to happen on the back end in mapping and standardization before that streaming actually becomes useful?

Pryluk: A lot of data is being collected in clinical trials—blood tests, biopsies, many other measures, collected across several sources: central labs, CRA queries logged into computers, and eventually flowing into systems like EDC and CTMS, arriving at CROs and sponsors in various shapes and formats. In order for that data to be ready for analysis, it needs to follow standards. One of the most common in the industry comes from CDISC—ADAM and STDM—and these formats require quite a lot of work. It takes many weeks to translate data from the way it was collected into those standards, and until recently that required a lot of manual work from statistical programmers.

Today, technology—and specifically machine learning and AI—allows us to streamline and accelerate that process for the first time, and this is what opens the door to real-time monitoring at scale. It closes the dead time problem, to the point of potentially eliminating it.

One thing I'm hearing from sponsors following the FDA announcement about the two pilots is: this is all great, but we need to be very careful reviewing data that hasn't been cleaned, integrated, and standardized. There's a risk of surfacing issues simply because the data hasn't been meticulously curated before analysis. So on one hand, we want real-time monitoring to surface issues as they happen. On the other hand, we want to do it in a way that doesn't just surface noise that wasn't cleaned. That's the balance and the challenge—but I do believe the technology many companies are working on, including us, will allow us to bridge that gap and make real-time monitoring the future for all of us.

ACT: The FDA real-time clinical trials pilot involves two large pharma companies. What needs to be true for that model to work for sponsors without that type of scale and infrastructure?

Pryluk: Even for big pharma, the specific pilot that took place was a pilot—it wasn't yet scale ready. There are a lot of concerns and things to be resolved before it scales. It should scale. This is the future, and it's great for patients, sponsors, and regulators. But in order for that to happen, we need the right tools to standardize, clean, and have data readily available for analysis before doing that analysis in real time.

If you think about the LLM revolution, that's exactly the enabler that allows us today to create ADAM, STDM, and TLF tables—table listings and figures—almost at the click of a button. Once you have them, you can start analyzing and asking questions. That's the enabler that will allow this to happen at scale, and not only in a bespoke pilot here and there.

ACT: What should the industry get right in responding to the FDA's May RFI to ensure the framework that emerges is practical and broadly adoptable?

Pryluk: First and foremost, for something like this to be successful, we need the right blend of big pharma, regulators, technology vendors, and CROs that are executing a lot of these trials—all forming the right ecosystem. It's a big problem with huge rewards at the end, and it can't be resolved without that ecosystem coming together.

It's about bringing stakeholders and technology companies together to facilitate cleaning and standardization at scale, creating the infrastructure that allows regulators to connect and sponsors to learn, and making it a transparent and learning environment. When things are not perfectly operating, we shouldn't shut it down—we should keep learning and allow that learning to be shared across sponsors and CROs. We really want to avoid the siloed approach, where one company makes it work but the learnings aren't shared. If ten pharma companies are developing drugs for the same target and the same indication, and there are safety events, we want those learnings to be shared. We have the right mechanisms to manage security, compliance, and privacy—it's about building the ecosystems that bring stakeholders together despite those challenges.

ACT: How do you prevent the push toward continuous data review from simply shifting the manual burden rather than eliminating it?

Pryluk: Quick wins—that's the first thing. There is naturally a lot of skepticism. People will say it was a pilot, it's going to disappear. So we need to create more wins and pursue a gradual shift, starting in early phases, focusing on one therapeutic area, then another, and gradually moving into late-phase and pivotal trials. We have to see a gradual plan, because otherwise it will disappear.

We also have to involve the right technology players. If this is tried only within individual companies, we're going to lack the network effect. There is a massive opportunity for cross learnings, shared learnings, and technology that can grow at scale. Those are the things I would highlight.