AstraZeneca, Tempus Collaborate on AI Project for Cost Effective Industry R&D

July 17, 2024

By Irene Yeh 

July 17, 2024 | With the rising costs of pharmaceutical industry R&D, achieving breakthroughs in some of the hardest-to-treat diseases, such as cancer, is becoming increasingly expensive. This presents a major challenge to developers in bringing efficient, affordable treatments to patients. 

AstraZeneca and Tempus teamed up to create a framework that can gather insights, discover novel drug targets, and aim to develop therapeutics for the broader oncology community. AstraZeneca explored how the combination of AI and multimodal data can increase the probability of success of their development programs in order to boost efficiency in R&D processes and bring treatments to patients faster. The jointly developed framework earned AstraZeneca the Innovative Practices Award for Clinical and Health IT, nominated by Tempus, at this year’s Bio-IT World Conference & Expo. 

A Three-Step Framework 

There are three main elements that determine if a pharmaceutical company will invest in developing a new therapeutic: the unmet need, expected lifetime global revenues of a new drug (to support a cycle of R&D reinvestment), and the likelihood that the development effort will succeed. Tempus and AstraZeneca collaborated to determine how the combination of AI and multimodal data can help, particularly with increasing the probability of success of their development programs.  

“The inspiration came from a set of proof-of-concept projects initiated in 2020, aimed to retrospectively look at clinical trial designs and retrofit real-world data (RWD) to those designs,” elaborated Sajan Khosla, executive director of Real-World Evidence (RWE) at AstraZeneca and the lead of this project. He explained how, in the proof-of-concept setting, the team looked at what RWD brought to the design assumptions and to the traditional datasets. This led the team to believe RWD enhanced certainty in specific design elements of the study design. “Once these proof-of-concept projects were delivered, there was keen interest to systematically approach study designs by utilizing RWD.” 

Their efforts resulted in a three-step framework derived from a cross-functional team from the clinical, regulatory, biostats, and RWE groups, collectively called the core Product Team, said Khosla. This framework is designed to systematically implement and quantify the benefits of deploying AI and multimodal data at scale to accelerate and de-risk key Phase 3 registrational clinical trials: 

  1. Choosing a Metric That Matters for the Business and Measuring It 
    The team chose Probability of Technical Success (PTS) as the metric to track success of the program because it is a widely adopted metric by drug developers and investors alike. PTS is the probability of choosing the right technical parameters that will provide a positive result in a clinical trial, such as inclusion and exclusion criteria, stratification factors, and endpoints. 
    Probability of Technical and Regulatory Success (PTRS) describes the probability that an investigational new drug has of successfully going through all stages of development and ultimately receiving approval by health authorities. PTRS is the product of PTS and Probability of Regulatory Success (PRS). 
  2. Starting With a Small Pilot Collaborators are encouraged to select the most suitable programs based on one of the following two use cases: 

  • Single-arm Phase 2 studies in which an external control arm is created to better contextualize the efficacy of the experimental treatment to inform a go/no-go Phase 3 decision. 

  • Registrational Phase 3 studies where PTS can be increased by better understanding and identifying the clinical and molecular determinants for selecting the optimal population. 

  1. Library of Multimodal Data  
    The biggest value and acceleration come by selecting a Phase 2 program to help make an informed go/no-go decision, and then quickly designing the subsequent Phase 3. In all cases, Tempus provided its library of multimodal data (DNA and RNA molecular data, clinical data including outcomes, and imaging data) targeted to the needs of its collaborators, as well as the necessary staff that supported the product teams with analyzing the data to make informed decisions. 
     
    Typically, the necessary multimodal data were already available in Tempus’ library of more than 6 million de-identified research records. In other scenarios, the data simply didn’t exist, and Tempus needed to leverage its unique network to source research specimens, sequence them, and pair the molecular data with the clinical data. In both scenarios, Tempus was able to use its platform and suite of products and services to provision and generate the most suitable multimodal data assets for the programs in the pilot. 

“By developing this three-step framework, the team was able to dissect how RWD could be leveraged in design assumptions and develop a shared accountability framework that ensured the implementation of RWD in all late-stage clinical trial designs,” said Khosla. Furthermore, the framework was “an overwhelming success” in terms of the mind-set and delivery of the RWD capability to clinical teams. There is no longer a question of how the data can support decision-making. Rather, it is more of a question of extracting more value from the data, added Khosla. 

Promising Results 

After implementing a small pilot on five studies that were being designed and still undergoing government approvals, the project produced valuable information on what was going on with the current standard of care. During the pilot, there was an average increase in PTS of 5% per study. This was enough for AstraZeneca to make this approach part of its governance process. Currently, the company’s oncology Phase 3 study designs are informed by RWE dataset analyses combined with Tempus analysis and are now an integral part of AstraZeneca’s clinical design process.  

Khosla also mentioned two unexpected outcomes. First, when it comes to designing large clinical studies for next-generation oncology therapies, the data is sparse. The second outcome was how closely RWD could emulate clinical trial populations when the right balance and weights were applied, indication dependent. The “secret sauce” is in the methodology used to weight, balance, and assess any quantitative bias. 

The Future for Clinical Design Processes 

With the framework, AstraZeneca and Tempus have provided researchers with a more efficient tool for clinical trial design. RWD can provide a more holistic view of a patient—how they are tested, treated, and what are their outcomes—and increased precision and even inclusivity in approaching clinical research, which can result in more confidence in trials and drugs to help patients live longer and healthier.  

“The team has been working tirelessly to ensure all oncology studies under design consideration have been able to leverage these dynamic data sources,” said Khosla. “Moving forward, we are looking at ways these data can be shared with decision makers and regulatory authorities to highlight the evidence used to formulate specific study design choices. This could dramatically affect the way that our study data is used in those decision-making environments.”