LONDON, Sept 22 (Reuters) – Big pharmaceutical companies are using artificial intelligence to quickly find patients for clinical trials and reduce the number of people needed to test drugs, accelerating drug development and realizing potential This has resulted in millions of dollars in savings.
Human research is the most expensive and time-consuming part of drug development, and it can take years to recruit patients and test a new drug from discovery to final stage. , and the process can cost more than $1 billion.
Pharmaceutical companies have been experimenting with AI for years in the hopes that machines can discover the next blockbuster drug. Some of the compounds selected by AI are currently in development, but it will take years for the bet to materialize.
But Reuters interviews with more than a dozen pharmaceutical company executives, drug regulators, public health experts and AI companies show that the technology has a big and growing role to play in human drug trials. Shown.
Companies like Amgen (AMGN.O), Bayer (BAYGn.DE), and Novartis (NOVN.S) scan billions of public health records, prescription data, health insurance claims, and their internal data. We are training AI to find clinical trial patients. In some cases, the time it takes to sign up can be cut in half.
“I don’t think it’s caught on yet,” said Jeffrey Morgan, managing director at Deloitte, which advises the life sciences industry. “But I think we’re past the experimental stage.”
The U.S. Food and Drug Administration (FDA) announced that it has received approximately 300 applications incorporating AI and machine learning in drug development from 2016 to 2022. More than 90% of these applications were filed in the past two years, and most related to the use of AI. At some point during clinical development.
Atomic AI
Before AI, Amgen spent months surveying doctors from Johannesburg to Texas to find out whether patients with clinical and demographic characteristics relevant to their clinic or hospital were participating in clinical trials. I was asking you what to do.
Existing relationships with facilities and physicians often influence the decision about which trial site to select.
However, Deloitte estimates that approximately 80% of studies are due to clinics and hospitals overestimating the number of available patients, high dropout rates, and patients not adhering to trial protocols. It is estimated that the recruitment target has not been achieved.
Amgen’s AI tool, ATOMIC, scans large amounts of internal and public data to identify and rank clinics and physicians based on their past performance in recruiting patients for clinical trials.
Amgen told Reuters that depending on the disease, it can take up to 18 months to enroll patients in mid-stage trials, but ATOMIC could cut that time in half in a best-case scenario.
Amgen is using ATOMIC in a small number of clinical trials testing drugs for diseases such as cardiovascular disease and cancer, and aims to use ATOMIC in most studies by 2024.
The company said that by 2030, it expects the use of AI to shave two years off the typically more than 10 years it takes to develop a drug.
The AI tools used by Novartis have made enrolling patients in clinical trials faster, cheaper and more efficient, said Badri Srinivasan, the company’s head of global development operations. But he said the value of AI in this context depends on the data it captures.
According to Sameer Pujari, an AI expert at the World Health Organization, less than 25% of health data is generally available for research.
external control arm
German pharmaceutical company Bayer announced it has leveraged AI to reduce the number of participants required by thousands of participants in a late-stage trial of Asundexian, an experimental drug designed to reduce the long-term risk of stroke in adults. .
Using AI, we linked interim trial results to real-world data from millions of patients in Finland and the United States to predict long-term risks in populations similar to the trial.
Once Bayer had the data, it began a late-stage trial with fewer participants. Without AI, Bayer said it would have spent millions more dollars and taken up to nine months longer to recruit volunteers.
Now the company wants to go a step further.
Bayer will use real-world patient data to generate a so-called external control group for studies testing Asundexyan in children with the same symptoms, potentially eliminating the need for patients to take a placebo. He said he plans to.
Because the disease is so rare in this age group, it is difficult to recruit patients and it is unethical to administer a placebo to trial participants when there is no proven treatment available. This is because there may be concerns about whether
Instead, Bayer aims to mine anonymized real-world data from children with similar vulnerabilities.
Bayer said it hopes the data will be sufficient to determine the drug’s effectiveness. Mining electronic patient data to find real-world patients can be done manually, but using AI can significantly speed up the process.
Although unusual, external control groups have been used in the past to replace the traditional randomized control group in which half of the participants receive a placebo, primarily for rare diseases with small patient populations or for which there are no existing treatments. It has been used.
Amgen’s drug Brincyte, aimed at treating a rare leukemia, received approval in the U.S. after using this approach, but the company needs to conduct follow-up studies to confirm the drug’s effectiveness after launch. was there.
Bryce Adamson, senior principal scientist at Roche (ROG.S) subsidiary Flatiron Health, said the benefit of AI is that it allows scientists to examine real-world patient data quickly and at scale.
He said that examining data from 5,000 patients could take months using traditional methods, but “now we can learn the same thing for millions of patients in days.” It became,” he said.
Risk of overestimation
Pharmaceutical companies typically seek prior approval from regulatory authorities to test drugs using external control arms.
Bayer said it is in discussions with regulators such as the FDA about relying on AI to create an external division for pediatric trials. The company did not provide further details.
The European Medicines Agency (EMA) said it had not received any applications from companies seeking to use AI in this way.
Some scientists, including the FDA’s head of oncology, worry that drug companies will try to use AI to develop external weapons against a broader range of diseases.
“When you compare one treatment arm to another without randomization, you are assuming that both patient populations are the same, which does not account for unknowns,” FDA Oncology Center. said Richard Pazdur, director of the Institute of Excellence.
Patients in clinical trials tend to feel better than patients in the real world because they believe they are receiving effective treatment and are receiving more treatment, which may result in There is overestimate success of medicine.
This risk is one reason why regulators tend to insist on randomized trials, as they believe that all patients are receiving the drug, even though half are receiving a placebo. It is.
Gen Li, founder of clinical data analysis company Phesi, said many companies are exploring the potential of AI to reduce the need for control groups.
But regulators say that while AI may enhance the clinical trial process, the standard of evidence for drug safety and effectiveness remains the same.
“The main risk with AI is that we don’t want to come up with the wrong answer to the question of whether a drug works,” said John Concato, associate director of real-world evidence analysis at the Office for Health Policy. I am. FDA’s Center for Drug Evaluation and Research.
Natalie Glover and Martin Coulter report from London. Additional reporting by Julie Steenhuysen in Chicago.Edited by Josephine Mason and David Clarke
Our standards: Thomson Reuters Trust Principles.