Summary:
- Amgen has developed ATOMIC, a machine learning (ML) model that analyzes large amounts of data and predicts which clinical trial sites are likely to enroll patients more quickly and effectively.
- ATOMIC leverages many forms of data, including demographic, geographic and real-world data to ensure faster and more representative participant enrollment.
- An internal analysis of 13 Amgen-sponsored studies showed enrollment was on average three times faster at highly ranked ATOMIC sites than lower ranked sites.
About 80% of clinical trials fail to enroll enough participants to move forward. And those that complete enrollment may take years to do so. With almost half the time to bring a medicine through clinical trials spent on enrollment, this causes serious delays in getting potential new drugs to patients who need them now.
Adding to this challenge is the underrepresentation of historically marginalized populations in clinical research. Ensuring that enrolled patients reflect the demographics of those affected by the disease being studied is essential to generating results that are applicable to a broader range of people in real-world settings.
So what is Amgen's solution to these challenges? Follow the data.
A team of data scientists, engineers and analysts at Amgen created the Analytical Trial Optimization Module (ATOMIC), a machine learning (ML) model to analyze extensive data and identify potential clinical trial sites around the world that can enroll patients quickly. ATOMIC generates ranked lists of sites, predicted enrollment rates and relevant country and investigator data.
Collectively, this information can be leveraged to determine where populations with certain clinical and demographic characteristics seek care. This can help identify high potential sites that are likely to enroll eligible patients most quickly.
While Amgen's clinical study teams have been able to sort through parts of this information successfully, site selection has traditionally been a largely manual and inefficient process with no centralized source. There hasn't been a consistent method to follow all the data available.
"Current trial enrollment is often limited by our human ability to integrate all the information needed to select the best sites for a successful trial," said Matt Austin, executive director of Data Sciences at Amgen. "That's why we developed ATOMIC, an innovative machine learning approach that allows us to make informed predictions to identify clinical trial sites with a higher potential for success."
Improving, accelerating and diversifying trials
ATOMIC analyzes hundreds of factors to predict clinical trial enrollment by learning local and global trial recruitment patterns. It incorporates diverse data such as eligibility criteria, treatment duration and historical enrollment at sites. By integrating all of the available information, ATOMIC serves as a single automated source to ensure the collection of comprehensive multi-dimensional data.
The ATOMIC project team and Amgen's Center for Observational Research (CfOR) used appropriately anonymized or aggregate real-world data (RWD) such as electronic health records and medical claims to locate trial sites near patients with high lipoprotein(a), or Lp(a) levels. Elevated levels of this form of cholesterol are linked to an increased risk of cardiovascular disease, and levels differ among different populations. Using these data can help site investigators, who are testing an investigational Lp(a)-lowering drug, screen 50% fewer patients to find one with elevated Lp(a) for the trial.
Along with CfOR, the ATOMIC project team works with groups across Amgen, including Global Development and Amgen's Representation in Clinical Research (RISE) team, which aims to improve the representation of clinical trial participants to better reflect communities impacted by the conditions and diseases we aim to treat.
Together with the RISE team, the ATOMIC team collected anonymized and/or aggregated data on geography, demographics and provider and patient information. This data is used to create a scorecard that identifies high potential sites with higher concentrations of populations often underrepresented in clinical research, supporting more inclusive and targeted enrollment strategies.
Osa Eisele, executive director and head of RISE, emphasized the importance of collecting and utilizing this data when considering site selection. "This data helps identify opportunities to increase clinical trial access for historically underrepresented populations, enabling us to meet patients where they are and drive greater participation in clinical trials."
ATOMIC delivers streamlined results that clinical study teams can use to support decision-making during patient enrollment and site selection. "ATOMIC leverages powerful machine learning technology," said Sean Bruich, senior vice president of Artificial Intelligence and Data. "But it certainly doesn't remove the role and expertise of the study team. It's simply a tool that helps them more efficiently evaluate large amounts of data quickly, augmenting their decision-making abilities."
A collaborative, flexible model tailored to patients' needs
When the ATOMIC team begins assisting with the site selection process for a clinical trial, they work closely with the clinical study teams to ensure the data are curated for their trial's specific protocol. ATOMIC is flexible enough to incorporate new and different data as they come up.
"A lot of cross functional planning and partnering goes into making a trial successful," said Sheryl Jacobs, vice president of Global Development Operations. "ATOMIC allows us to streamline the process by finding those 'hidden gems' that may not have been on anyone's radar but are likely to enroll participants quickly and efficiently."
"ATOMIC represents a major innovation in how we approach clinical trial enrollment and trial site selection," said Narimon Honarpour, senior vice president of Global Development. "The commitment to enroll patients that are representative of who will receive therapies we develop, is essential."
ATOMIC was first tested as a pilot for an ulcerative colitis trial and has since been used for site selection to test potential therapeutics, including those for cardiometabolic diseases, atopic dermatitis and multiple types of cancer. A 2025 analysis of 13 Amgen-sponsored studies revealed the sites at the top of ATOMIC's ranked lists enrolled participants up to three times faster, on average, than sites lower down on the lists, demonstrating the model's effectiveness at speeding up clinical trial enrollment. Amgen is also working on additional ML models that monitor clinical trial enrollment in greater detail.
