- Pharmaceutical giant Bristol-Myers Squibb is quietly testing a range of artificial intelligence ideas to improve its research.
- In an interview with Insider, one of its executives identified three promising AI programs.
- These programs are used in clinical trials to make them more efficient and powerful.
In the past several years, there has been an explosion of biotech companies specializing in artificial intelligence that have raised hundreds of millions, such as InstroAnd the pharmacy recursiveAnd the deep genome The big drug companies are now up and running. Bristol-Myers Squibb, the New Jersey pharmaceutical giant, is quietly testing several AI programs, hoping that the ideas will make pharmaceutical research more efficient and robust.
Venkat Sithuraman, senior vice president of biometrics and data science at Bristol-Myers, spoke to Insider about three artificial intelligence and machine learning programs being used in clinical trials at the $161 billion company. While artificial intelligence and machine learning have become ubiquitous buzzwords in the pharmaceutical industry, these programs show how the technologies can impact drug research in the near-term, said Sethuraman, a drug industry veteran who previously worked at GSK and Novartis.
If the ideas work, they could boost clinical trials by reducing the use of placebos, creating better metrics for measuring drug efficacy, and improving the experience of trial volunteers.
Reducing dependence on placebos
The long-term gold standard for whether a drug is effective is a randomized controlled trial. Volunteers either get the experimental treatment, a placebo, or the current standard of care.
The design creates tension for the participants. Researchers need data from a control group, but patients do not want to take a placebo. Bristol Myers hopes to use artificial intelligence and massive amounts of historical data to reduce the use of placebos.
Over the past few years, the pharmaceutical industry has toyed with the idea of synthetic control weapons, where data from the real world replaces the need for a placebo. This historical data can come from a variety of places, Including electronic health records, insurance claims and disease records. If these data are of high quality, they could replace the need for a placebo group. These are the dummy control groups It has now supported some drug approvalsbut regulators usually still want to see appropriate trial data.
But Sitthurman said the real-world data is often of lower quality. It usually provides much less information than what comes from participants who are closely followed in a clinical study.
“It’s moving in the right direction, but there are still roads to go,” Sithuraman said.
Bristol-Myers is taking a middle ground, hoping to reduce placebos rather than replace them. The idea is to supplement the study’s control arm with historical data.
Sethuraman outlined a study of 1,000 patients as an example. Typically, 500 volunteers will receive the experimental drug and 500 will receive a placebo.
With Bristol Myers’ new concept, the same study could end with the same amount of data but recruit fewer volunteers to get the placebo. The trial aimed to enroll 250 people in the placebo arm. The experimental researchers hope that the historical data set for the disease will mimic the rest of the placebo group.
As the study recruits patients, the researchers were getting an early look to see how the real-world data stacked up against the placebo group. Ideally, the placebo groups and the real-world groups would look similar in what the study measures. If the study is testing a cancer drug, for example, the focus will likely be on how quickly the disease gets worse and the cancer grows.
If the two data sets were not the same, the researchers would scrap the historical data and go back to the model method, expanding the placebo group to 500 volunteers.
But if the real-world data is consistent with the placebo group, researchers can rely more on historical data. They could adapt the study to use 350 patients from real-world data and only 150 in the control group. In general, studies can be done faster, recruit fewer patients, and reduce the odds of patients getting a placebo.
The Sethuraman team has already applied this approach retrospectively, analyzing previous studies to see if it could work. He said the research was a promising proof of concept, and the results will be published soon. Sithuraman said Bristol is now using this approach in a mid-stage clinical study, with results expected next year.
Finding better ways to measure drug efficacy
Another challenge for clinical trials is deciding what to measure. Sethuraman’s vision is to use artificial intelligence to turn mountains of images, such as PET and CT scans, into valuable data. This data can then be used to create endpoints that can make clinical trials better and faster.
For example, his team Presenting results earlier this month at a cancer conference which defined a new scale called the “g-score”, which uses artificial intelligence to convert images of tumors into data. This data is then used to predict whether a patient’s tumor is likely to grow or shrink.
Sithuraman said predictive measures such as a g-score can support faster decisions about whether a drug is helping patients. Cancer studies usually take years to collect survival data, to see if more volunteers died in the control arm than those who got the experimental treatment. Experiments can run faster if measurements such as the g-score are adopted.
Conversational AI improves patient experience
The Seturaman team is also running a pilot program to make participating in clinical trials an even better experience.
Working with a startup who declined to be named, Bristol Myers introduced an element of conversational and artificial intelligence into some clinical trials earlier this year. Study volunteers can ask questions about AI, which can avoid the usual pain of trying to call a doctor’s office or connect with a researcher.
Bristol-Myers hopes that AI will keep patients engaged and automatically flag chats in which a patient reports side effects.
Sethuraman said he expects to have enough data next year to assess the impact of the program.