Tech for Clinical Trials

The pharma industry has long been keyed into the excitement about AI. And the predictions for its potential impact are staggering. 

AI tools could generate $60 billion to $110 billion in value for pharma per year thanks to boosts in operating efficiencies, R&D speed and data insights, according to an analysis published by The McKinsey Global Institute in January.  

PwC also illustrated the potential returns from AI adoption on a per-company basis in a report this year.

“Overall, pharma companies that industrialize AI use cases completely across their organizations have the potential to double today’s operating profits by boosting revenues and reducing costs. We expect that this industrialization process will begin to be fully realized by companies prioritizing AI by 2030,” the PwC analysts wrote.

Although AI adoption has historically been slower in pharma than in other industries such as retail, PwC noted that “pharma [is] closing the AI value gap in an accelerated way.”

As more pharma companies leverage AI tools, what unique insights are being revealed? Here’s what several industry leaders using and developing AI tools told us. 

The factors that truly impact trial outcomes 

Dave Latshaw is CEO of Biophy, an AI company with solutions designed to predict clinical trial outcomes and speed drug development tasks.  

“For any given trial phase, [when] you can look at all [what’s] impactful to the outcome, if [there were] just a few smoking guns that would always dictate whether or not a trial succeeds, they would be very overweight factors in the model and the rest wouldn’t matter. But the truth — and why it is complex — is that when you look at a massive number of features and the magnitude of the impact over all of them, it doesn’t reveal any one or two factors that are absolutely critical. They are all interdependent and all have the same magnitude of impact. That means you have to have to do a lot of things right to have a successful trial. 

Here’s an example of what could be more important: Endpoint selection. A lot of times there will be sets of endpoints that are typically used, and some have to be for regulators, but sometimes the selection is a company’s decision. Sometimes people are surprised to understand the variability of performance for certain endpoints and that it’s very different from what they expected. The common wisdom is to use what everyone else uses. 

But there might be a better endpoint that could be used [to] capture the information at the same level of detail to provide the evidence you need but is maybe more consistent and doesn’t have the risk of some inherent mechanism for variability that could throw off results. For example, some therapeutic areas, like psychiatric indications, often have a really enhanced placebo effect that can undersell the efficacy of an individual drug. But that variability with the placebo — not the drug — is the factor that sinks the trial. And when you choose a placebo you’re subjecting yourself to more randomness than you may want.”

Predicting patient outcomes

Miruna Sasu is CEO of COTA, which specializes in solutions that leverage real-world data to support oncology research. COTA also recently introduced a suite of AI-powered solutions called CAILIN designed to generate insights from RWD and datasets for research. 

“You treat [patients but] you want to understand what’s going to happen. Is the treatment going to stop the disease? Or just lessen the progression? We’re finding that we have the ability to predict that, which is further analysis than we thought we would get. With clinical trial data, you only have a point in time. Here you get more of the patient journey. 

We also enhanced some of our data points with genomics, and [then] you have the ability to see which genes are turned on and off before and after the treatment. And if you do that again and again you understand that patient’s pattern. Doing this manually would consume a whole person’s career, but with an algorithm you can tease apart which genes are turned on and off and which cancer, for example, could be optimally treated with which biomarker inhibitor. You have the ability to almost look into the future.

In the long run, a drug candidate would come out of discovery and go right into a real-world model and be tested against patient populations to see if it’s going to work and by how much. The technology is there to allow for that.”

Deeper insights into when genes are expressed

Craig Thompson is CEO of Cerevance, a biotech leveraging a large collection of brain tissue samples along with AI/ML tools to develop drug candidates for the CNS space. 

“What we have been more surprised by is looking at certain genes by cell type and seeing when some of them are intrinsically upregulated or downregulated. And we always like to test what we’re seeing with our machine learning capabilities.

Some companies are looking at genes that we see using our machine learning are only transiently upregulated or downregulated at a certain point in disease progression, so you scratch your head about why they’re going after that target. In certain cases our technology has shown some pharma companies going after targets that aren’t expressed in the human brain and they’ve gone into phase 2 or 3 trials and then failed. At the other end, in the [neurodegenerative] space you see some targets that companies are pursuing that you don’t [understand] because the gene is extremely broadly expressed. Our anticipation is given the broad expression you would see high rates of adverse events and some of those compounds are being advanced in the clinic.”

AI solutions need to be supported by digital infrastructure

Dr. Mert Aral is the chief medical officer of Huma Therapeutics, which develops remote patient monitoring solutions used in decentralized clinical trials that is also aiming to create a platform any company can use to build their own healthcare apps. 

“We’re looking at how we can enhance engagement and compliance and how we can provide patient education and awareness. We know these AI solutions work. We know they accelerate time to treatment change. We know we’re delivering data and real-world evidence to pharma. But how can we increase scale?  

Right now, [solutions such as digital health monitoring] still rely on clinical teams. You still have a nurse that monitors the data and calls or works with patients. The generative AI is allowing that nurse to more efficiently monitor data and trends. But our thinking is: How can we replace that nurse with a GenAI nurse so we don’t need to deploy so much staff? 

Of course, you would have to train these models. This is a high-risk environment and you can’t have a GenAI nurse giving the wrong advice to the patient. We still haven’t fully solved for that. But from a technology perspective, I don't think that’s the limiting step. We’ll get there and I don’t think it’s far off. 

So when we look at GenAI it’s not just the many applications of it. We need to also have the right digital infrastructure to launch and train and deploy these solutions at scale, and our health systems are not quite there.”

AI platforms are revealing deeper levels of biology and chemistry, while creating long-term value

Ayman AlAbdallah is a partner with Mubadala Capital, a venture firm that manages over $24 billion in assets, including about 45 life sciences companies. 

“We’ve seen the application of AI in drug discovery and development over the past 10 years and different waves of the technology have built on top of each other. One observation is that the value of AI companies will continue to accrue in the pipelines they generate. 

The intention is not to replace the scientist but to supercharge the scientists with a better understanding of biology and chemistry to develop drugs in a much more efficient and effective manner. We’re also generating insights that were otherwise not possible using conventional assays or analytics. The examples are many. 

One of our investments — Iambic Therapeutics — is leveraging physics-informed AI to propel the field forward in how we use computational biology and chemistry to discover and develop drugs. Another company Vevo Therapeutics is developing large language models for biology and moving in vivo drug discovery from a downstream step that is limited in terms of its representation of tumor heterogeneity into early-stage development to uncover misunderstood or novel mechanisms of action and new drug combinations that can result in high efficacy.”

When you’re looking to make changes to biopharma clinical trials, one of the top minds in the industry could help you get there — by welcoming Robert Langer to its advisory board, that’s exactly what Lindus Health aims to do.

Langer, an MIT professor and co-founder of Moderna (in addition to a long list of successful biotechs), is the author of more than 1,500 scientific articles, holds more than 1,400 patents issued and pending, and is widely considered the most cited engineer in history.

Bringing Langer to the advisory board of a contract research organization like Lindus, which calls itself “the anti-CRO,” sets the company up to address some of the most pressing concerns in clinical research. Langer’s mind for biotech business also contributes to the ways he can help the company innovate to mitigate some of the risks associated with the CRO model.

Since the COVID-19 pandemic, traditional clinical trials have shifted to a more flexible approach, according to industry research from WCG. With clinical trial starts expected to rise in 2024 compared to the year before and new technologies emerging at a rapid clip, CROs are exploring ways to stay ahead of the curve.

A change is gonna come

Overall, Langer sees where CROs often fall short and where Lindus could make an impact in the industry.

“I think CROs haven’t changed much [in the last decade],” Langer said in an email interview. “There are some recent attempts to incorporate AI and virtual clinical trial techniques to increase efficiencies and expand reach of studies [but] in general CRO processes have remained unchanged.”

With massive success in the science and business of biopharma, Langer is notoriously modest and focused on the work. At Lindus, that work centers around removing bottlenecks in the clinical trial process with an end-to-end business model.

Langer isn’t the only big name with a hand in Lindus’ path forward — the company has raised more than $24 million from Peter Thiel and other tech and biopharma investors.

And one of the changes Lindus is trying to bring to the CRO table is a new financial model that positions drugmakers to share the risk of the clinical process.

“The incentives of CROs are currently totally misaligned with sponsors and with patients — CROs typically earn more money from projects that are delayed, due to their billable hour business model,” said Meri Beckwith, co-founder of Lindus, in an email.

Founded in 2021, Lindus gained insight from Langer on what would become the company’s driving business model. Beckwith said Langer was critical in shaping a model based on hitting clinical milestones so the CRO and the drugmaker are both locked in on the same goal. Milestones are also a way to share the risk since the contractor and the client are both in the same boat, Beckwith said.

“Sponsors are understandably risk-averse,” Beckwith said. “No one gets fired for hiring [CRO giant] IQVIA.”

Under one roof

Beyond a milestone-based financial model, Langer also sees the end-to-end nature of the clinical trial model as something that sets Lindus apart.

“By building their own robust tech platform and internal site capabilities, they have been able to streamline clinical trial execution,” Langer said.

For example, software that allows for quicker study configuration can help get a trial started. And since recruitment is often a narrow bottleneck for most studies, with most not recruiting on time, according to Langer, expanding the reach through virtual assistance and access to electronic medical records can grow the patient pool faster.

Lindus has also developed software that uses machine learning to write study protocols and capture data with AI monitors from beginning to end.

Langer’s role will be that of a biotech industry guru who can speak to many different facets and disease areas where Lindus is looking to innovate.

“[Langer]’s involvement will help us better design clinical trials, drawing on his wealth of experience from having been involved with dozens of pioneering life science companies,” Beckwith said, noting that Lindus is identifying market segments that work best with its model. “He has consistently been at the forefront of new modalities like gene therapy, tissue engineering and transdermal drug delivery.”

Diseases areas like diabetes where pharma companies have had a particularly difficult time developing novel drugs is where Langer could bring his expertise to find new solutions.

“I take a lot of what I’ve learned in the biotech and drug delivery space about the needs of these companies when looking to establish clinical evidence,” Langer said. “There are a lot of common threads with the types of issues that come up during clinical development that I am able to share.”

Sponsored content

By Premier Inc

As the pace of medical discovery accelerates, so has the number of clinical studies. Yet the sector faces significant recruitment, enrollment and diversity challenges.

The good news: AI-enabled technologies coupled with real-world data (RWD) and real-world evidence (RWE) are driving innovation in clinical trials and research.

Here’s how.

Recruitment: Identifying Patients Efficiently

About 80 percent of trials miss enrollment targets and timelines, making recruitment the top cause of delays.

Precise inclusion/exclusion criteria and personalized medicines mean that trials require specific participants to prove efficacy and meet enrollment and eligibility requirements. However, medical records may be in non-digitized, narrative form and with little to no connectivity, making it difficult to find candidates.

Data and technology innovations support recruitment with rapid, accurate, at-scale screening. For example, an AI algorithm that processes medical and demographic information and pairs patients with trials expedited matchmaking by more than 40 percent.

Furthermore, natural language processing (NLP) can mine millions of medical records for unstructured data that is not visible in the electronic medical record. For instance, a trial may be limited to cancer patients in stages 3 or 4, yet the structured record only indicates a cancer diagnosis – not the disease stage included in physician notes. Using NLP, organizations can scour both sources and identify candidates – gaining efficiencies and automating a manual process.

Sponsors leveraging these capabilities have improved recruitment by 1,800 percent and tripled enrollment over baseline projections.

Enrollment: Eliminating Participation Barriers

Site selection generally favors experience and principal investigator (PI) relationships over metrics including volume, eligibility, speed, proximity and diversity. But site-driven design intensifies participation barriers. Because it funnels limited candidates to one location, it reduces data applicability, as participants are often not representative of the targeted population. It also generalizes data due to a lack of representation among racial/ethnic minorities, rural residents, women and/or patients with comorbidities.

Decentralized clinical trials (DCTs) improve accessibility by using RWD/RWE to apply proven historical performance to a location-agnostic context, bringing the trial to patients via telehealth, remote monitoring and/or local settings. Leveraging predictive analytics of past site performance, DCTs recruit based on patients’ location instead of PI relationships.

Synthetic control arms (SCAs) can also streamline studies and ease patient burden. While placebo-controlled trials are accepted as the optimal approach, this overlooks the ethical considerations of giving sick patients an inactive placebo. SCAs compare de-identified patient data from pre-existing datasets to a group receiving the treatment to evaluate effectiveness.

SCAs can reduce patients needed by up to 50 percent. They also encourage enrollment by enabling all participants to receive the treatment.  

Diversity: Securing Representation Across Demographics and Conditions

Clinical trial diversity is a public health imperative. Driven by limited access, outdated enrollment practices and medical mistrust, women remain underrepresented as do patients in rural areas, various racial/ethnic groups and patients with certain medical conditions.

Underrepresentation hurts marginalized populations. Fewer than 5 percent of adult cancer patients enroll in cancer clinical trials, and limited diversity can create care inequities. A medical breakthrough unstudied in a population may become a breakthrough denied based on social determinants, geographic location or other disparities.

In 2022, Congress required that sponsors submit trial diversity action plans (DAPs) to the U.S. Food and Drug Administration (FDA). Plans “… must lay out trial enrollment goals organized by race, ethnicity, sex and age group,” and “… consider how the medical product’s intended use population is distributed according to these demographics.”

RWD/RWE and innovative technologies allow diverse patient identification and enrollment, broader network engagement and patient discovery in complete therapeutic areas. AI-driven NLP:

• Accelerates medical records analyses in underserved communities;
• Pinpoints enrollment obstacles;
• Reveals strategies to reach/engage populations;
• Identifies the right patients for the right trials; and
• Helps patients apply and participate.

Best-in-class datasets mobilize RWD for evidence-based and population-based analyses. Patient-level data can be tokenized and linked with other sources including Social Determinants of Health (SDoH) data. This uncovers insights into demographics and disease prevalence across populations, informing enrollment goals and rationales required in DAPs.

Final Thoughts

The potential to transform clinical trials is within our grasp – and patients counting on access to innovations don’t have time to waste. RWD/RWE powered by AI in the service of passion for discovery holds a bright future for us all.

The clinical trials industry is up against howling headwinds.

Novel therapies are more complicated to work with, staffing has been strained for years and the need to move efficiently and quickly without impacting quality is ceaseless.

Even positive developments can make work more complex. The regulatory push for DEI, for example, has the industry rethinking trial protocols and design. And new tech aimed at boosting efficiency brings the added challenge of training and preparing staff. 

On top of that, geopolitical issues in the Middle East and Russia’s war with Ukraine pose an ongoing threat to trials in those regions.  

“A lot of people assumed those [challenges] would stop, but they’ve gone on,” said Peyton Howell, CEO of global contract research organization Parexel, which runs trials in Israel and Ukraine. “It is concerning.”

Howell took the helm of Parexel in May and is eager to boast about its recent accomplishments, including DEI initiatives and industry awards. But Howell is also forthcoming about the challenges Parexel and other research organizations are up against. 

Tech’s promise and perils

The No. 1 question Parexel receives from pharma partners is about opportunities to use AI, Howell said.

“These tools might sound mundane, but they are the most exciting because they affect how a person spends time in their career,” Howell said, adding she’s particularly excited about tech “focused on study startup, patient recruitment and trial design. 

Still, the new platforms have to be tested and reviewed, and when they’re applied to operations, staff members have to be trained. AI platforms add another layer of tech to new demands already established by wearable devices for remote patient monitoring. 

In a study Parexel conducted with the Tufts Center for the Study of Drug Development, Howell said sites “expressed frustration with the number of different systems imposed on them and the complexity that tech burden creates.”

And those criticisms are being taken seriously, down to the letter.

“We are asking sites to do more. Some staff at sites say they feel like they’re in tech support right now,” Howell said, adding, “We’ve heard loud and clear from sites that we as an industry need to do a better job reducing the tech burden.”

The staffing crunch

The tech burden has also compounded ongoing staffing issues in clinical research. 

Although turnover has dropped significantly since the peak of the Great Resignation, and so far, the staffing crunch has yet to delay any trial starts at Parexel, recruiting new talent remains a challenge, Howell said. Research professionals for oncology, ophthalmology and neuroscience therapeutics are often in higher demand but have a smaller, more specialized talent pool.

 “The market remains competitive … meaning most new hires have more than one job offer,” Howell said.

To more quickly onboard and prepare research assistants, Parexel joined fellow CROs by launching a training program in 2022 for new hires. Called APEX, the program teaches the ins and outs of clinical work, such as navigating the required systems and patient monitoring. 

The Walgreens factor

Patient-guided trial design is a “biggie” in the industry, Howell said, but implementing it can be tough. And although many sponsors worry that gathering patient input will slow the trial process, Howell said it can speed the whole thing up. 

“Do it early and you’ll be more successful with recruitment and retention,” she said. “And it can reduce the cost burden because you are pressure testing the data you’ll need and the number of site visits you’ll need … which is a win-win.”

Going into the recruitment phase, Howell said patients need to remain at the center of protocol design to help retention.

“Clinical research has a burden on patients we need to understand,” Howell said.

Clinical research needs to meet patients where they are, Howell noted. 

“They want access to healthcare providers, but travel adds another burden,” she pointed out.

To bridge the access gap while improving diversity in clinical trials, pharmacy retail giants moved into the research arena — with mixed results.

Two years after launching a clinical trials division, CVS Health walked away in 2023.

CVS’ departure left more room in the market for Walgreens, which launched its clinical trials arm in 2022. Now, Walgreens said it’s reached 5 million patients to “potentially recruit into clinical trials.” 

Partnerships have also been key to the Walgreens strategy. Most recently, it struck a $100 million pact with the Biomedical Advanced Research and Development Authority to “increase innovation in decentralized clinical trials.” And earlier this year, Walgreens announced it’s teaming up with Boehringer Ingelheim to connect patients to phase 3 trials for the drugmaker’s GLP-1 candidate in obesity. 

Because of Walgreens’ capacity to recruit patients and provide trial sites, its foray into the space is a potential boon for research organizations like Parexel, Howell said.

“Seeing big companies like Walgreens in the arena is fantastic,” she said. 

Although Parexel doesn’t have an official partnership with Walgreens, Howell said the company is an “active supporter” of the pharmacy giant’s efforts.

Earlier this week, Walgreens announced it’s shuttering 1,200 pharmacies over the next three years in the wake of a $14 billion operating loss in fiscal year 2024. In an earnings report, the company didn’t indicate it plans to scale back the clinical trials business.  

Going forward with clinical trials, Howell said her advice to the pharmacy behemoth would be to “keep it simple.” 

“Focus on recruiting and bringing patients in,” she said. “I want patients to know about trials in a community setting, and I think that’s going to bring in a more diverse population — and that’s where we have strong alignment with Walgreens.”

While uses for AI in pharma have ramped up, the FDA has been working to develop standards to inform the industry on future regulatory frameworks for the technology. Now, the FDA’s Center for Drug Evaluation and Research is launching a new solution to may help address this gap.

In April, the agency announced the creation of the Center for Clinical Trial Innovation (C3TI), which aims to boost innovative approaches to clinical trials. The center will act as a hub within CDER to implement efficiencies by allowing trial sponsors to work with CDER staff. Initial demonstrations of the program can be used as case examples for future implementation.

“We have championed innovation for a long time,” Kevin Bugin, lead for C3TI and deputy director of operations in the office of new drugs, told PharmaVoice in an emailed statement. “In fact, CDER has been a leader in navigating the complexities of a rapidly changing drug development ecosystem. However, as drug development continues to evolve, clinical trials must evolve accordingly.”

The center’s launch includes three project areas that may open the door to more AI uses in clinical trial and drug development programs, including point-of-care or pragmatic trials, Bayesian analyses — which uses “multi-level statistical models” to describe data elements — and trials leveraging selective safety data collection.

“Through its engagement and coordination efforts, C3TI will also work to further enhance the adoption of innovations involving AI/[machine learning] and support the development of new CDER AI initiatives as related to clinical trials,” Bugin said. “We will also facilitate information sharing with external parties about AI/ML as appropriate, and will support knowledge sharing internally through various mechanisms, such as discussion forums and communications.”

Clinical trials lack innovation

In general, clinical trials are “not very innovative,” according to Charles Fisher, CEO of Unlearn.AI, an AI company that creates digital twins and offers a web-based app for clinical trial planning. 

“By and large, clinical trials are run by really large pharmaceutical companies,” Fisher said. “The big pharma companies don't move very fast [so] it always takes them a long time to adopt new things.”

However, Big Pharma argues that the regulatory environment around clinical trials has been one of the biggest impediments to innovation. 

“One reason for the lack of consistent, widespread adoption of innovative approaches is that efforts to revise FDA’s regulations have not kept up with the breadth and pace of change and advancement in the industry,” said Eli Lilly’s senior vice president of clinical capabilities Janelle Sabo in a public comment to the FDA. 

C3TI isn’t the agency's first foray into attempts at innovation. In the last few years, the FDA has launched a string of programs aiming to improve clinical trials, including the Complex Innovative Trial Design Paired Meeting Program, which was introduced in 2022 and aimed to advance “the use of complex adaptive, Bayesian and other novel clinical trial designs;” the Innovative Science and Technology Approaches for New Drugs Pilot Program, which supports “novel approaches to drug development” such as AI-based algorithms; and the Fit-for-Purpose Initiative, which creates a regulatory pathway for drug development programs that use “dynamic tools.”

C3TI is specific to innovation in clinical trials, but the program may have “similar objectives” to past programs, the FDA said. 

“We see C3TI as a continuing evolution of our longstanding clinical trial innovation efforts,” Bugin said. “By acting as a hub, C3TI plans to strengthen the implementation of current innovation efforts and encourage the development of new ones through enhanced communication and collaboration.”

Despite this history, some companies have found these initiatives lacking in real-world guidance.

“Currently, most guidances on innovative approaches provide an outline of FDA expectations and concerns about the uncertainties of the approach, but lack practical instruction on implementation,” Dr. Mwango Kashoki, senior vice president, global head of regulatory strategy at Parexel, wrote in the company’s comment to the FDA.

And the FDA’s top-down approach to innovation may not be impacting how trials are run.

“The agility of the FDA regulatory review process impacts innovation,” Kashoki’s comment continued. “Although we are seeing a pragmatic, patient-focused approach from FDA leadership, this is not filtering down to the division and individual FDA staff member levels.”

Informing regulations

As part of the launch, CDER has said C3TI will include additional touchpoints with CDER staff, with expanded opportunities for “enhanced communication and collaboration.”

This belief that the FDA isn’t willing to accept innovation may be a false perception held by the pharma industry, according to Fisher. 

“My experience, working with both the FDA and the [European Medicines Agency], has been that they're quite open to innovation, often encouraging innovation, but that's not the perception in the industry,” Fisher said. “Any programs that the regulators create to try to combat that perception and encourage the development of new innovative approaches is very welcome.”

With the addition of C3TI comes the question of how this initiative is different from the FDA’s past forays in innovation, and how the center will add clarity to the regulatory framework — something pharma companies say is impacting their decisions to get involved in pilot programs and demonstration projects. 

“It is worth noting that there appears to be a disconnect between the public statements made by senior FDA leaders advocating for innovative approaches, and the feedback and views expressed within the review divisions and medical product centers,” Genentech said in its comment to the FDA. “Ultimately, such inconsistent feedback creates uncertainty and regulatory risk, which can discourage investment and participation in pilot programs.”

Without guidance from the FDA or CDER, the future of AI in clinical trials can’t be hammered out completely, Fisher said.

“There's really broad confusion within the whole industry as to how AI can be used in clinical trials, and even what it is, how it's going to be regulated,” Fisher said.

According to CDER, C3TI and its goals could help inform how the agency ultimately regulates AI.

“Initially, C3TI is considering how our regulatory framework could be optimized to support the rapid evolution in clinical trial design and conduct and the use of technological innovations,” Bugin explained. “We will consider whether targeted changes could help further accommodate and advance modern operational approaches to clinical trials; facilitate the incorporation of technological innovations that promote efficiency and enable diversity; and emphasize the importance of good risk assessment and management principles to the quality of clinical investigations.”

As pharma continues to leverage AI, some companies are seeking more information from the FDA on how to move forward.

“An important challenge is that although industry is moving (and needs to move) forward quickly with [AI], it will be many years before FDA issues guidance,” Kashoki wrote. “The FDA needs to share more quickly with industry and other stakeholders what the agency is learning, what are their areas of greatest focus and concern, and for what issues do they want more detailed information from industry.”

The promise of artificial intelligence in the biotech and pharma sector is vast, from drug discovery to patient enrollment to clinical trial design. But the industry is also approaching the future with caution, and addressing the technology’s shortcomings is essential to making it a useful tool in the long run.

“Hallucinations” represent one of the challenges to widespread use of AI and machine learning. In a healthcare setting, the introduction of false information into the algorithms can be especially harmful, said Wael Salloum, co-founder and chief science officer at natural language processing company Mendel AI.

Currently, large language models are used to decipher patient records and guide physicians and other caretakers to the correct interventions, including drugs. If a patient is taking one drug, for example, the LLM will not push for a second treatment that can lead to complications. But when that output is designed to be trustworthy, false information is dangerous, Salloum said.

“What these LLMs are designed to do is produce really good English and convince you that’s the truth — when you’re summarizing a medical record for a doctor to make a decision, any potential mistake can be a matter of life and death,” Salloum said.

Hallucinations are defined by the more generalist LLM company ChatGPT as the “generation of content that is not based on real or existing data but is instead produced by a machine learning model's extrapolation or creative interpretation of its training data.” Many of these can be small in nature but represent bigger problems down the road, according to IBM — from the spread of misinformation to enabling bad actors to formulate a cyber attack.

Mendel’s platform, called Hypercube, combines an LLM with a logic-based AI to cut the rate of hallucinations in medical research even at large scale. The company this month joined a Google Cloud partnership giving pharmas and other healthcare companies access to the platform via the tech giant’s marketplace.

And while AI has become a massive game-changer in the life sciences, Salloum hopes platforms like Hypercube can improve the trust between AI and the researchers, scientists and doctors who use it.

“It’s very important that everything a system produces be explainable and traceable — and so many of the doctors we speak to say they spend more time verifying a summary from an original record than reading the record themselves,” Salloum said. “Any overhype of AI might cause damage to the whole industry, because once you lose trust, you don’t get it back.”

Building a trustworthy AI

Joining the Google Cloud marketplace is no accident — in truth, Hypercube works much like the Google search engine that has come to be so ubiquitous in many people’s online lives. Every time a user searches for something, the engine has already created an internal referential index rather than combing through the whole internet, Salloum said.

Similarly, Hypercube builds a knowledge base of millions of patient records that traces back to the original record, cutting the risk of hallucination. In this sense, Hypercube is more of a “research engine” than a search engine, Salloum said.

Like many AI applications, Hypercube works better when used as a tool.

“We are there to retrieve literature for review, but at the end of the day, you need human intelligence to lead, and you can use technology and AI as a support for humans, but not as a replacement,” Salloum said. “We’re nowhere close to that, and it’s important not to overhype.”

Many of Mendel’s customers are companies with data that’s ready to be used for scientific research but still too raw, Salloum said. Mendel has also helped pharma companies make connections between the data and the patient population.

Hypercube is also infused with a “world model,” or fundamental definitions within medicine that are almost philosophical in nature, Salloum said.

“What is medicine? What is a treatment? What is cancer? What is cell differentiation? These get distilled into the LLM until it’s forced to memorize that there is a certain structure that’s then fine-tuned with tasks,” Salloum said. “So the system itself is producing it and also justifying it.”

The larger purpose of AI in a healthcare setting is to provide access to information that might otherwise be out of reach, Salloum said. And making that information more trustworthy by mitigating the presence of hallucination is at the heart of that access.

“Our mission is to democratize healthcare by creating a centralized medical knowledge where everything we learn from a patient journey can be understood, from every success to every failure of a treatment,” Salloum said. “And the applications, ultimately, are infinite.”

The FDA has issued a final rule that pulls lab developed tests under the same review process as medical devices and could slow work in clinical trials, according to critics. 

LDTs, or in vitro diagnostic products (IVDs), are developed by a single laboratory and used in the clinical setting and for a range of assays — from genetic to COVID-19 tests. 

After the rule was initially proposed, it received more than 6,700 comments during the public period last fall, with many arguing it would limit access to testing and negatively impact clinical trials. In response to the comments, the FDA updated the phaseout policy, with a four-year plan for IVDs to meet the new requirements, and included additional exceptions. While labs that create LDTs are already regulated under the Clinical Laboratory Improvement Amendments of 1988 to meet those requirements, the FDA said the new rule will solidify the regulatory framework.

“By applying the same general oversight approach to laboratories and non-laboratories that manufacture IVDs, the FDA will reduce regulatory uncertainty, which will give stakeholders more stability, clarity and confidence, and facilitate investment in the development of innovative IVDs,” the FDA said in a release.

Final rule highlights

Despite their wide use, LDTs are not without limitations, and regulators have long discussed oversight of the tests, which have become more complex and ubiquitous over time, according to the FDA. 

The agency in 2015 highlighted 20 case studies of tests that were not up to par. The most well-known case involved the now-defunct blood testing company Theranos, which produced inaccurate LDTs that were used in clinical settings before the Centers for Medicare and Medicaid Services revoked the lab’s CLIA certification in 2016. The company subsequently fell apart and its top leaders were charged with fraud.

The oversight issue also came to light early in the pandemic when diagnostic tests weren’t available and labs couldn’t distribute LDTs due to emergency use authorization demands at the time. 

The final rule now defines LDTs similarly to medical devices when it comes to regulation, allowing the FDA a more active and traditional oversight process. By May 2028, IVDs manufactured by a lab will be required to fully comply with the new regulations in the FDA’s will phaseout plan over the four-year period. The “benefits of the phaseout policy outweigh the costs,” according to the regulator. The agency also said the policy benefits will become clearer over time as more LDTs hit the market.

One loophole the FDA included was that LDTs already available under the old policies and those designed to meet “unmet needs” will not have to comply with the new market standards. 

Clinical trial concerns

Pharmas placed emphasis on the potential impact to clinical trials last year when the rule was first proposed. Eli Lilly, which leveraged LDTs by academic medical centers — which are different than those created by a commercial lab — in a trial for the cancer drug Retevmo, noted that LDTs manufactured by academic medical centers can accelerate enrollment in clinical trials by defining the patient population sooner. Conversely, centralized testing “can lead to significant delays,” the company said of its own experiences with precision oncology clinical trials.

“Lilly emphasizes the importance of ensuring patients have uninterrupted access to diagnostic tests accessed through AMC laboratories resulting in greater and more efficient access to clinical trials,” said Dr. Anthony Sireci, head of diagnostics, development and commercialization at Eli Lilly, in a comment. “We are concerned that if the proposed rule is implemented as proposed, it will significantly reduce the amount of testing performed by AMCs thereby limiting access to clinical trials for the communities they serve.”

Regeneron Pharmaceuticals generally supported the proposed rule in its comment to the FDA, but also noted potential barriers to implementation, including delays in clinical trials. The company criticized the FDA’s lack of definition between assays in the clinical setting versus clinical trials.

“Overly broad application of IVD regulation that does not allow for sufficient flexibility for use of well-characterized assays in clinical research could delay initiation of clinical trials and subsequently the availability of new treatments for patients in need,” Dr. Edward Cox, vice president of regulatory affairs, and Cuancha-Maria Manning, senior director of regulatory intelligence at Regeneron, wrote in the company’s comment.

Now that the rule is finalized, the labtech industry could launch a legal challenge. According to the American Clinical Laboratory Association, the rule is also outside the FDA’s purview.

“ACLA has grave concerns about this rule as a matter of both policy and law,” the association, which represents diagnostic labs, said in a statement immediately following the final rule. “The rule will limit access to scores of critical tests, increase health care costs, and undermine innovation in new diagnostics. The rule also exceeds FDA’s statutory authority, as Congress has never granted the agency authority to regulate laboratory developed testing services offered by laboratory professionals.”