Phase 3 clinical trials are lengthy and expensive processes that could result in hundreds of millions of dollars in losses if they fail, enough to tank a mid-sized or small company. Direct costs alone tend to be in the tens of millions of dollars, yet failures are common. A study by researchers at MIT found that 41% of Phase 3 trials failed between 2000 and 2015. However, AI and big data could help save trials from failure.
Credit: Phesi
A report from Phesi, a clinical data science company, found that many phase 3 trials faced avoidable failure due to two factors: overly complicated trial design and issues with patient recruitment, according to Phesi founder and president Gen Li, Ph.D. “Both of these can be avoided by leveraging big data and artificial intelligence,” said Li.
Clinical trials powered by AI
Phesi’s clinical trials database contains contextualized data from over 200 million patients worldwide and could enable more efficient and wider-reaching patient recruiting. Enrolling patients from Phesi’s database could shorten the timeline and increase the precision of the recruitment process.
Once patients are recruited, Phesi can build digital patient profiles, which are used to design the trial. Phesi’s generative AI, which is based on proprietary statistical models trained on data from about 132 million patients, can use the patient database to model outcomes and enrollment cycle timelines as well as rank KOLs and investigator performance. This is all part of Phesi’s Trial Accelerator software.
“This alignment between the two [patient data and trial design] will give us the opportunity to improve the design of the trial and avoid the amendments and other problems down the road, including the failures,” said Li.
Although Phesi uses AI to power their services, they never use generative AI to interpret the patient data, because of the “tremendous risk,” said Li. However, he indicated that this might be possible in the future as AI technologies continue to evolve.
Phesi also takes measures to prevent bias in the data. “[The data] are coming from multiple institutions. They are coming from multiple investigators. They are being collected over time from different places, so that in itself carries sufficient statistical power,” said Li.
Replacing the control arm with digital twins
Phesi, in addition to other AI companies like Unlearn, can utilize patient profiles to create digital twins, which they aim to use to partially or completely replace the control arm in a clinical trial. Phesi pulls data from similar studies testing the same treatment to create digital twins of real patients who received placebo treatments. This could accelerate the clinical trial process by reducing the number of participants required, as well as solving ethical issues such as giving critically ill patients a placebo treatment.
Li emphasized that data from real patients is the foundation of each digital twin. Li said, “This is a digital world, but that digital world completely and absolutely mimics the real world.”
Digital twins can also support clinical trials in rare diseases, which often struggle to recruit enough patients. With the Phesi database, investigators can use the digital twins of deceased patients alongside their living participants. “For example, if we are dealing with a particular disease that has not been progressed or impacted by any medical innovation, then we can use very old data,” said Li.
This approach could also impact oncology. “In the cancer area, for example, lung cancer. You probably have cancers being driven by over 20 different kinds of biomarkers. So despite the fact that lung cancer is not a rare disease, that particular patient population becomes a rare disease,” said Li.
Li added that he believes Phesi is “very close” to being able to replace the control arm with digital twins, but he also emphasized the importance of building a strong foundation, “we are more aiming for a pragmatic, widely usable solution, [rather] than rushing our time to get there.”
