Teva’s Chief Medical Officer Eric Hughes outlines the key efficiencies to be gained from artificial intelligence (AI)

When Eric was a newly graduated infectious diseases doctor, he saw firsthand the effect of drugs such as azidothymidine (AZT), the first antiretroviral medication used to prevent and treat HIV/AIDS.

“Twenty years later those patients were still alive because of the drug development process through that period and that gave me the epiphany that drug development was what I wanted to do,” he says.

But making medicines is traditionally a long, costly and risky process.

The cost of making a medicine has risen from $179 million in the 1970s to $413 million in the 1980s, 1 billion in the 2000s and up to $2.6 billion today.

These are the sobering statistics, produced by Deloitte in 2020:

• From initial discovery to the marketplace takes about 10 to 15 years for a new medicine.
• Out of every 5,000 to 10,000 compounds that enter the pipeline, only 1 receives approval.
• Only 5 in 5,000 drugs that enter preclinical testing will progress to human testing, and only 1 out of these five drugs will get approval – medicines that reach clinical trials have only a 16% chance of being approved.
• A medicine has an average cost of $2.6 billion and an average development time of 12 years.

Traditionally there have been many inefficiencies in the drug development system – and this is where AI machine learning can help, says Eric. For example, in data mining.

“The reality is that pharma companies generate way more data than we actually ever look at,” says Eric. “The power of looking at that data with machine learning or AI technology is incredibly hopeful.” Understanding patient data in such granular detail could lead to the development of personalized medicines – deciding who's the right patient for what drug at the right time.

Using AI to diagnose and manage conditions through digital biomarkers

Biomarkers are naturally occurring molecules, genes, or characteristics by which a particular disease or condition can be identified. Digital biomarkers are objective, quantifiable physiological and behavioral data sets that are collected and measured by means of digital devices such as portables, wearables, implantables, or ingestibles. The data collected is typically used to explain, influence, and/or predict health-related outcomes. 

Digital biomarkers are one of the “holy grails” of AI, according to Eric. “If you can get something that can treat or convert a drug that's successful 20% of the time to make it successful 100% of the time and find the right people to treat, that's incredibly powerful.”

He explains how Teva is using the technology for projects including:

Tardive Dyskinesia (TD)

What: Facial recognition tool for diagnosis of TD patients
How: By extracting ~500 facial landmarks
Value: Improving diagnosis of TD patients and treatment options

Schizophrenia

What: Automated speech-based tool for prediction of clinical symptoms
How: Identification of correlation between the Positive and Negative Syndrome Scale (PANSS) and voice parameters
Value: Enhance the efficiency and success rates of clinical trials and aid in monitoring patients, leading to more effective treatments

Multiple system atrophy

What: Tool for reducing the number of patients in clinical trials by improving measurement accuracy
How: Collecting data from commercial wearable movement sensors and correlating it to clinical outcomes
Value: Reducing sample size, study timeline and time getting medicines to patients

AI technology can also be used to accurately select clinical sites

What: Better selection of sites in clinical trials
How: Combining data from multiple internal and external sources for prediction of site performance, prioritization and visualization of key performance metrics
Value: Shortening study timeline and so decreasing time getting medicines to patients, as well as improving study quality and probability of success

It is also possible to use the technology to collect patient information more efficiently

What: AI-operated avatar for clinical interview
How: Implementing virtual humans and multimodal interaction for objective assessment
Value: Consistent and objective evaluations and improving quality of the data collected in clinical studies

At the end of the day, this is extremely valuable technology, Eric concludes. “Our job here is to make medicines for patients. If we can use tools to make that happen faster and better, that’s what we’ll do”.

NPS-ALL-NP-01324-JULY-2024

Find out more

• Can AI-driven digital biomarkers redefine diagnostics and clinical trials? Meet the Teva teams working on the technology
• “The three women who inspired me”: Eric Hughes shares stories of the women who shaped his life and career
• Want to get involved in the inspiring work we do? Check out our Careers pages

You might also be interested