[Adobe Stock]
More automated data-collection processes
Already, clinical trials are pulling considerable data from external sources, whether that be outside labs providing supplemental test results or real-world data generated through biomarkers, genomics sequencing, wearables, and other sources circumventing conventional data capture processes.
“There’s more and more data that we’re collecting directly from patients or other sources,” Indupuri said. “And the amount of data that we’re collecting through EDC — or sites entering into EDC — is reducing, which is a good sign because you’re eliminating this duplicative data effort that leads to a lot of inefficiencies.” The growing amount of data streams also provides new opportunities for cross-referencing, limiting the need for manual data-proofing. While the frequency of manual data entry in EDC systems varies widely, corrections are often common to address erroneous data, especially in cases when data sources are limited.
EDCs: Not a complete cure for data headaches
On the one hand, EDCs helped clinical trials transition away from the paper-based systems in use in the 1990s and early 2000s, which involved an “army of data entry people and data managers,” Indupuri recalled. “It was completely labor intensive.” But on the other, EDCs did not completely remove data-handling and data-entry headaches. For instance, EDCs addressed transcription errors, which were “a big issue” with paper-based systems, but the web forms that replaced them were something of a mixed bag. While it is true that the software could programmatically check data values and prompt a physician or site coordinator to proof potentially problematic data, it also increased the oversight burden at sites and for sponsors in overseeing and monitoring of clinical trials.
As Gartner has noted, adaptive clinical trial designs can pose challenges to existing EDC tools, which were designed for more static trial designs. Similarly, traditional EDC tools often lack the necessary infrastructure to support AI algorithms, which can provide advanced analytics and insights for more dynamic trial management.
Raj Indupuri
To address these limitations, modern data infrastructure aims to ingest and standardize data from a variety of sources, connecting the dots between data acquisition, data infrastructure and analytics.
Modern infrastructure offers new data integration possibilities
Indupuri highlights the power of a data cloud to synthesize data from a range of different sources, structures, or formats into a modern infrastructure. That spans ingestion, mapping, computing, and integrating other data products into a cloud environment.
From there, machine learning models can automate repetitive tasks for stakeholders like data managers and programmers while also spotting outliers to report to data managers. “The time it takes to clean and produce high-quality data is reduced,” Indupuri said. “Machine learning models analyze the data, identify outliers, and make recommendations to data managers.”
For programmers, these models can standardize diverse data formats for regulatory submission. “I believe it’s quite exciting times in terms of technology — whether it’s cloud or gen-AI, it is going to impact every stakeholder within clinical development,” Indupuri said.
Here, generative AI offers new possibilities. “Our data science teams have built a machine learning model that examines Open FDA and other public datasets. It can predict whether a medication indication is accurate,” Indupuri said.
Indupuri also shared about experimenting with large-language models (LLMs) for tasks such as conversation analytics that can help with programming tasks for regulatory submissions. “We’re now working on a use case where you can ask questions and have a conversation with the system,” Indupuri explained. “Additionally, we’re looking into feeding data into LLMs to automatically standardize them.”
As a wave of cloud and AI technologies emerge, the speed and efficiency with which complex clinical trial challenges can now be tackled have changed significantly. “These are still complex problems that we can solve at a speed that I would not have envisioned even four or five months back,” he concluded.
Filed Under: clinical trials, Drug Discovery, machine learning and AI
