By Iris Loew-Friedrich, chief medical officer and head of development solutions, UCB
Artificial intelligence (AI) is undoubtedly transforming how we treat and manage disease. The novel insights it brings are critical to progressing patient care; and when combined with data, AI is a tool that enables more meaningful insights faster.
At UCB, we can now use AI to spot the presence of “silent,” or asymptomatic, fractures in the spine and malignant tumors invisible to the naked eye. We have even discovered how to use single molecules as universal sensor elements in a circuit to create a programmable biosensor with real-time, single molecule sensitivity, which can be used for applications like translational research, precision diagnostics, and even whole-genome sequencing. And AI is also disrupting how we discover and develop new drugs.
In clinical trials, AI can identify suitable cohorts — by analyzing real-world data (RWD) and social media content to answer questions relevant to the design of clinical trials, such as the effect of inclusion/exclusion criteria on the availability of patients for recruitment or helping to define the clinically meaningful difference for outcome thresholds. AI also has the potential to speed up patient recruitment by alerting medical staff and patients about trial opportunities, as well as simplifying entry criteria to be more accessible for potential participants.
Big Data and AI technologies are complementary, as AI can help to synthesize and analyze ever-expanding data. In fact, AI-powered capabilities, including data integration and interpretation, pattern recognition and evolutionary modelling, are essential to gather, normalize, analyze, and harness the growing masses of data that fuel modern therapy development. UCB’s partnership with Microsoft achieves this by synthesizing information to help us make the best decision at a given moment, and a by-product of this is the closer collaboration between our data scientists and biologists.
Drug discovery is a game of information: If we understand disease well, we know where to intervene and how to design a molecule to target the disease. AI can be used to integrate the patient experience and voice into solutions, further enhancing their relevance.
Putting AI Into Action
UCB is actively adopting innovative AI programs that use advanced analytics and algorithms to generate actionable outcomes from patient insights, such as personalized care pathways based on both physical and mental care options, which join up care with local communities to reduce pressure on stretched healthcare systems.
The Aetion Evidence Platform (AEP) delivers value through AI-based workflows by quickly transforming real-word data into real world evidence and helping to identify actionable unmet needs — by doing so, we are maximizing sustainable access for patients. In addition, our co-creation mindset with external stakeholders allows us to generate insights collaboratively from RWD.
Creating Patient Cohorts For Faster, More Efficient Projects
As a first step in our patient cohort creation program, together with stakeholders, we defined a psoriatic arthritis cohort within the platform. With the common platform and shared cohort definition, we enable consistency and transparency. We use this platform to set the criteria for a patient’s entry into the cohort. Currently, there is no difference in “accuracy” between using the AEP versus manually coding the requirements. One instance of using AEPs to increase speed was during the cohort creation phase of project AARDVARK (Advancing Access in Rheumatology: Delivering Value from AI-based workflows), a project which aims to leverage real world data in combination with AI algorithms to better understand spondylarthritis patients. Though the AEP did not save us time in project AARDVARK, we anticipate that by continually adding more analytic cohorts to the AEP, future projects and analyses will be faster and more efficient.
Using AI To Improve Protocol Design
AEP is also helping us design more efficient study protocols. We have already developed several UCB-specific AI models that have the potential to significantly impact the protocol design for the clinical development pipeline. Using the experimental platform, we have developed the models to:
- Generate an overview of pivotal trials for a relevant indication, which is a notoriously complex activity,
- Extract and summarize the eligibility criteria and other relevant clinical, safety, and regulatory information, and
- Produce geographical mapping of sites involved in specific studies, segmenting rare disease population for focus treatment.
Thanks to this AI-based technology, we are a step closer to supporting protocol design and the identification of relevant trial-information and patient populations. After this initial release, the UCB team and the Microsoft team are exploring the latest generative and conversational AI capabilities (like OpenAI’s ChatGPT) that could further enhance the way in which our teams use and interact with these smart systems. These tools could help to keep track of treatment schedules and trial recruitment, making trials easier to access for patients, and make the data easier to access for physicians. AI tools are also removing the barriers imposed by ethnicity or geography, because we now have access to, and an understanding of, much larger data sets, so we can learn how disease affects all populations and design solutions that are global. This is only the start of our journey. User feedback will continue to be captured and incorporated, resulting in updates and continuously evolving models.
Bringing More Value To Patients With AI
We also have created a “360-degree view of patient populations” project — a data-driven, 360-degree patient view enabled by an integrated patient population data and research platform. This platform will be based on the entire spectrum of available qualitative and quantitative data to which advanced analytical approaches can be applied. We will be able to generate new insights, for example, study design considerations to maximize value to patients, by approaching each research question with all relevant information across a spectrum of datasets.
This was piloted for people living with Hidradenitis Suppurativa (HS), a chronic, recurring, painful, and debilitating inflammatory skin disease. We explored HS patient segmentation by applying advanced analytics across multimodal datasets with clinical, genomic, and patient interview data. Integration of the findings across datasets provided a more comprehensive view of HS patient subgroups and a novel application of advanced analytic approaches across heterogeneous datasets.
This allows us to interrogate multi-modal datasets (i.e., clinical trial data, transcriptomics data, and interview data) among heterogeneous HS study populations. Various data-driven approaches were used to identify HS patient subgroups in these three datasets. While there was a common theme across the three sets of results in terms of the patient subgroup characteristics, insights were also generated in the comparison of the findings and will be used to inform future analyses.
The patient is at the center of everything we do at UCB. Our motivation to invest in digital solutions is about creating value for patients. Patient- and data-powered AI opens a realm of possibilities for novel insights and enhancing the important work our research and development teams are already doing, meaning we’re getting better solutions to patients quicker. And that is priceless.
About The Author
Iris Loew-Friedrich is chief medical officer for UCB, a member of the company’s executive committee, and head of development solutions. She provides strategic global leadership for worldwide clinical development, medical affairs, regulatory affairs, quality assurance, statistical innovation, real-world evidence, and patient safety/pharmacovigilance. Her mission is to lead UCB’s Development Solutions, ensuring high quality, innovative, cost-effective development of objectively differentiated patient solutions with proven superior and sustainable value for clearly defined patient populations.