Reuters Events | Pharma

Thought leadership and innovation for the Pharmaceutical Industry

Artificial Intelligence has only just begun to unlock high value innovation. The latest predictive tools will break new ground, says Concerto HealthAI’s Jeff Elton

By Andrew Stone on Apr 10, 2020

In his 2016 book Healthcare Disrupted, Dr Jeff Elton sketched out how healthcare would move from being reactive to proactive thanks to the ubiquity of electronic health records (EHRs) and other datasets. 

 

That future is now upon us, says Elton, CEO of precision oncology business Concerto HealthAI. whose mission is to combine RWE and AI to drive better insights and patient outcomes. The firm, which recently raised $150m to invest in growing ‘next-generation’ clinical research, is now busy putting technologies and tools into hundreds of life science companies and clinical settings to ask questions of real-world data retrospectively and prospectively. 

 

The insights flowing from this near real-time clinical data have great potential to accelerate meaningful clinical innovations, complementing the more traditional and slower approaches. “We believe that research and the conduct of care, are integrally linked,” says Elton. “And we believe that faster cycles of research questions being asked leads to faster cycles, where the insights from those can be put into practice.

 

“Instead of doing surrogate measures, we're very big believers in having direct clinical measures. You can do things now with the tools in a fraction of the time that you would have done before, they will increase in utility much more rapidly. It is a frontier where we will see tremendous progress.

 

Regulatory approvals are getting faster too, thanks to the application of RWD components, says Elton. “For rarer cancers and other devastating diseases we may find an electronic medical record derived external control arm replacing what would have been historically a randomized control arm.

 

"These are now being thought of as having high validity and utility within the framework of an RCT study. There have been approvals such as in male breast cancer, where a label modification can be made based on RWD only and more will be taking place.”

 

Better outcomes, improved safety

Safety and efficacy are also beneficiaries of Concerto’s focus, says Elton, for example using data and tools to identify non-responsive patients seemingly eligible to be treated in a particular way. 

 

Other examples are identifying where patients may “hyper progress” in undesirable ways and instead identify alternative synergistic combinations of therapies. Other applications include finding strategies to mitigate a particular path of metastasis common to a disease or occurring as part of an intervention. 

 

“This ability to identify adverse events and or put mitigating strategies in place is a wonderful, brilliant application for RWD,” says Elton. “All of that can be done a fraction of the time, compared with traditional approaches.

 

“Our intent is to codesign new research models with both biopharma innovators and HCPs who don't have the time for traditional research approaches because of the incremental burden it brings to their standard-of-care workflows, but who believe in the value it might bring to their patients,” Elton adds.

 

But there is a big question mark about this AI-powered path to rapid progress and that is the as yet undefined role and reach of regulation. 

 

The speed of current innovation is beginning to exceed the ability of regulators to keep up, says Elton. “The cycles of innovation are now super rapid” and there is clear engagement from regulators Elton stresses.

 

“I give the FDA very high credit. They are open to the application of RWD to regulatory submission or using AI and data science approaches to designing clinical studies. They want people to bring those innovations forward, and they’re [deferring] guidance until they see what’s working.

 

Long-term, a lighter regulatory hand that encourages self-enforcement and transparency, will sustain a beneficial pace and range of innovations he says. “it is important that these new approaches emerge as ‘trusted’ and as having the veracity of the traditional, albeit slower, ones.”

 

Regulating the black box

Another worry is that as the potential of AI to improve outcomes increases, so does its complexity, and its workings become more opaque. Will pharma and regulators be content with ‘black box’ functionality if humans don't understand how exactly it works?

 

It is a question Elton has thought a lot about. “The question is: can you use AI to assess AI? Can you say if the training set is accurate? Does it have a bias in it? They are all super valid questions.” 

 

One way to look at it is to compare the risks involved in using AI to the risks inherent in taking a medicine owing to the fact that different people may respond to different features of a molecule in different ways, says Elton. “We do lots of things in life where there's an opaqueness on one level and transparency on another level and I think the same thing will be true of AI. 

 

“I think we’ll have published mechanisms and say: ‘This is what this model does. This is how this model was created. Here's the framework that is used. Here's where the original master data set came from. This is where the subset was taken from for doing the training, this is how that subset validated, this is where it was deployed with ongoing monitoring to assure its validity and so on.’”

 

Any regulatory approach to AI will need to evolve as the applications of it evolve, he adds. “You don't build a model and you're done. It's all based on a view of data at a point in time. The model is subject to the features of healthcare today. As the standard of care evolves, these things shift.”

 

Elton is confident of Concerto’s ability to create value here. “I'm bullish. We’re super-transparent, we have complex rules, we have predictive models, we always share and publish in the peer-reviewed forums. Our IP is in how we make the models, not the actual model itself.”

 

Concerto’s secret sauce

Its IP and its approach seems to be working for the business so far. In his 18 months as CEO at Concerto revenues have grown several-fold. What is driving this? There are a few factors that define success to date, claims Elton.

 

“We're an open architecture in all aspects of our business model. We are open in our data, we’re open about how our data are used, we're open about collaborating with other data partners, even our ostensible competitors, because our guiding North Star is doing whatever is best for the patient. Our data are designed to bring benefit to patients. It’s not our top line or bottom line that is the driver, it’s the outcomes we can realize.”

 

Technology is a crucial differentiator, he adds. “Our tech is a state-of-the-art architecture. We deploy in AWS and Azure, we can deploy other components, we can use other people's data.” 

 

Part of Concerto’s claim to have the most advanced technology, says Elton, is because it borrows from and is informed by other industries, such as telecoms or financial services, that are used to moving massive amounts of data and which routinely combine data assets from different sources to drive significant change.

 

Predicting the patient journey

Another of Concerto’s strengths is an absence of commercial conflicts, which gives it the freedom to focus on big-picture problem solving without interference, he adds. “We don't have life sciences funding. We don't have an individual provider entity or a payer entity. We’ve worked very hard to be neutral stewards that target use cases and enable things that were not possible before.

 

“We enrich, we do design, we can federate. So we design to the goal. And that's something that no one in the industry has done. For example, when EMRs on oncology patients come together, tumor staging and line-of-therapies selection is done at point of initial diagnosis. 

 

“We build AI models that read the record, and can tell the stage at any point in a particular patient journey and get an accuracy score based on that patient's record. No one did that before. We can predict features, we can predict treatment durability with alarming accuracy, we can predict mortality status, and that informs clinical study design. 

 

We are doing things on demand, on the fly, that would have taken months to execute before. We are bringing more meaningful innovations forward faster, eliminating waste, moving to value.”