The AI Advantage: Reimagining the Future of Drug Development 

Since 2011, when Watson won Jeopardy, there has been a growing crescendo of the promise of data, analytics, and AI. While it is impossible to watch a sporting event without a major tech company informing us how AI is improving performance athlete’s safety and how to beat the house with a betting app, the question that Pharma needs to answer is: what is hope and what is hype when it comes to improving patient outcomes and R&D productivity?  

To better understand the challenge, one needs to appreciate that based on functionality, there are some very different attributes, from practical to theoretical. For example:

• Reactive machine AI can synthesize and analyze large sets of data to make an assessment or recommendation. Think of search engines and viewer recommendations from streaming services. However, it has no memory.
• Limited memory AI can search past events and assess outcomes to make predictions.
• Theoretical AI encompasses advanced concepts such as Theory of the Mind and Self Aware AI.

From a capabilities perspective, AI, which performs very specific tasks within a subset of cognitive abilities, is sometimes referred to as Artificial Narrow AI. ChatGPT would fall into this category based on its reliance on a single task, that being text-based chat.

Sponsor companies and others are already starting to use AI to supplement their approach to data quality monitoring. Additional potential applications include:  

• Reducing the time it takes to identify targets in preclinical drug discovery—something that otherwise takes many months. 
• Analytical tools for site selection for clinical trials.
• ML, AI, and augmented intelligence are being used to garner insights from the volumes of data collected for commercialization and marketing.

While clinical development lags slightly in adopting new technology, the industry is reaching an inflection point. In healthcare, it is important to evaluate tools that could improve our ability to deliver medicines to patients in need. Therefore, it is not surprising to see the considerable investment and excitement that comes with the evolution of this AI. Paradoxically, since the health and well-being of patients is at stake, and industry research is highly regulated for the same reason, it is understandable to see both confusion and concern over the ability to use AI ethically and appropriately. Driving that concern is the inability to see “under the hood,” so to speak, to understand the accuracy of the predictions and the details related to data and data quality that support it.

Can the humans and the machines team up to give us more hope and less hype?

One of the stubborn challenges that has made it difficult to improve cycle times is recruitment.   While AI has been successfully used to identify more patients and more sites post protocol design, it has done little to reduce screening failure rates, which continue to hover at unacceptable rates from 40-90% depending on the published series.  Rather than utilizing AI to find more patients for a protocol who may not be representative of the population, companies could pivot to guide the design of a protocol that reflects the disease areas of interest, thereby reducing screening failure rates and accelerating throughput. More importantly, the results may be more broadly applicable to the relevant population.  

Some companies are using AI tools to automate data aggregation and leverage analytic capabilities. To improve the quality of the data used to train AI platforms, improve accuracy and reduce hallucinations, the pharma industry may need to commit to the adoption of data standards and common data models to make this affordable, reliable and scalable. This could support earlier and better signal detection of operational or clinical risks.

In the next several years, if companies pivot from using AI to deploy the trial and find patients to support better trial design, clinical trial protocols will look very different because their eligibility will improve — and, as a result, patient recruitment and representativeness (i.e., diversity, equity) will improve as well. 

Paving the path forward

While AI has many potential benefits, several things must happen before they can be fully harnessed for drug development: 

• Strengthened data stewardship – Over the past few years, the industry has recognized the need for good data stewardship and management — for both clinical and operational data. Steps already being taken to ensure data is in order will need to continue. This will reduce the effort and cost of data acquisition and shift the focus on the insights which is the real value.

• Balanced benefits and risks – Do AI’s benefits outweigh its risks? Broadly speaking, companies are still in the evaluation phase, but a company can look at the benefit/risk balance on a case-by-case basis. Companies can question the risks of adopting specific AI tools in specific drug development processes. An instance where the risk is low — like using generative AI to create a basic consumer leaflet, for example — might make a good AI “learning opportunity.” The risk may be too high in other instances, such as when making patient facing recommendations.  This requires a partnership with experienced, trained human experts.

• Commitment to sharing and transparency – Processes must be in place to assess AI technologies, prove their accuracy and monitor their performance. In addition, the technologies themselves can’t be black boxes. AI technology solution providers must create some transparency around how they work. Similarly, there must be a willingness among AI solution providers, users, and others to share what’s working and what’s not working. The saying goes, “Success has many fathers; failure is an orphan,” but companies will not get very far unless they are willing to try new things and sometimes fail. 

Biopharmaceutical organizations employ significant safeguards whenever they find new ways to do things — and AI is simply a new tool. Like any other tool, understanding the right problem to apply it to is the difference between success and failure. Addressing a problem is never just about technology; it’s always about people, processes, and technology. 

Leaning in to change 

Every individual across the biopharmaceutical ecosystem plays a role in the evolving use of AI to modernize R&D. Each one of us can participate by:

• Educating ourselves. Valuable starting points include discussion papers from the U.S. Food and Drug Administration (FDA) focused on AI in drug development and medical products, as well as a reflection paper published by the European Medicines Agency (EMA).
• Engaging with health authorities to find a way forward that improves outcomes, success rates and mitigates risks
• Understanding the right questions to ask.  
• Identifying the appropriate risk/benefit approach for our areas of product development and expertise. 
• Using low-risk opportunities as a “learning lab” from which to build.

Companies are appropriately cautious because public welfare is at risk, but companies may find ways to balance that risk with AI’s potential benefits. If biotech, regulators, policy makers, health care practitioners and tech companies can align on better patient outcomes as the primary objective, we will be writing about the development programs improved with AI in five years.

Photo: metamorworks, Getty Images

This post appears through the MedCity Influencers program. Anyone can publish their perspective on business and innovation in healthcare on MedCity News through MedCity Influencers. Click here to find out how.