Reimagining Drug Development

1. Why is it critical to rethink traditional drug development approaches, particularly for cancer therapies?

BUDHWANI: In her 1983 book Sudden Death, Rita Mae Brown defined insanity as doing the same things over and over but expecting different results. Most of us would agree, and yet, we continue to invest almost U$100 billion a year in cancer drug development, only to face a staggering 96 per cent failure rate in clinical trials. Setting financial losses aside, consider the human toll. Now, imagine if patients were matched to trials that were right for them. Imagine if preclinical drugs were prioritised based on likelihood of clinical success. Think about what this could mean for trial participants, sites, sponsors, and future patients—people who would never receive the treatment that is best for them, if it fails in trials. This is why it is critical to break out of the insanity loop and rethink our approach.

CLARK: In addition to clinical trial failure rates, cancer is fundamentally a personalised medical threat that is heavily dependent on the variable of time. These points effectively neutralise many, even most, traditional development approaches involving a groups-based design through typical clinical development methods for regulators globally. Whether tackling tumor heterogeneity, single or multidrug resistant mechanisms, or off-target toxicities, each of these challenges alone further highlight the problems with identifying critical targets for drug development. Cancer represents not only a multivariate temporal challenge, but it represents a wicked problem for providers as they attempt to isolate the variables that produce each type of cancer. These are only a few of the many reasons we need to rethink drug development in oncology.

2. How are emerging technologies, like personalised medicine and AI/ML, different from traditional approaches, and why do you think these could be a game-changer in cancer drug development?

BUDHWANI: For the better part of a century, cancer research has relied on generalised cell-line and animal models which, while technologically sound, fail to represent the incredible complexity and uniquely personal nature of patient tumors. Consequently, we have become remarkably adept at curing cancer in Petri dishes and in mice! Emerging precision personalised medicine technologies can finally enable us to defeat the “emperor of all maladies” in humans. For example, patientderived 3D tissue models can help optimise treatment strategies on an individualised basis. If approved therapies are not suitable, these models can guide trial enrollment. Further upstream, they can be used to prioritise licensing and development of preclinical drugs. Data from these models can, in turn, power AI/ML systems to accelerate progress by identifying patterns and generating hypotheses for next-generation drug discovery. Bottom line: by responsibly developing and adopting these tools, we will not just optimise drug development – we will redefine what’s possible.

CLARK: Perhaps the most obvious role of personalised medicine and AI/ML tools is that they are useful for accelerating discovery and design (often via structural analysis), optimising manufacturing, and identifying biomarkers. These tools are also useful for tailoring treatments to host response, which increases the space for therapeutic decision making and may have broad spectrum applications across the immune system. Likewise, these activities could predict treatment response for different types of patients, thereby helping with development of personalised therapies and companion diagnostics. National Cancer Institute scientists have recently demonstrated that deep reinforcement models, a mathematical modeling technique, combined with machine learning that optimises learning through trial and error, punishments and rewards, leads to clear, more effective treatments, with fewer side effects.

3. And where are the traps? What might prevent these technologies from delivering on their promise?

BUDHWANI: Nothing kills a much-needed innovation more rapidly than hype and FOMO. Ironically, it is because these emerging technologies hold incredible promise that they are at an even greater risk of never arriving – or worse – getting discredited by a frenzy to “move fast, break things.” Perhaps the easiest way to turn promise into pitfall here is by running fast and loose with data. Data is the lifeblood of this wave of innovation, yet it can also become its Achilles’ heel due to insufficiency, skew, or a frantic tendency to force-fit data simply because it is available—after all, doesn’t everything look like a nail when you have a shiny new hammer? On the flip side, there’s the very real danger of inertia: a pathologically risk-averse mindset rooted in the mantra, “this is how we’ve always done it,” so why change? Finally, the fear of collaboration—given the sheer size of the problem space, even the most promising technologies will fail to see the light of day if we try to go at it alone.

CLARK: On their face, Karim’s comments about “hype” and “hope” resonate for change management, particularly in digital health. They represent the ability to fund the future and sink the success of AI, from predictive to generative. Data is the lifeblood and Achilles’ heel, and yet since the 1950s when data and telehealth started us on the journey, growth in the area of digital health has largely stagnated due to emotions and the opposite side of hope. Access to deidentified electronic health records represent some of the greatest data that will help accelerate from concept to solutions, but their availability is limited. The limitation simultaneously engenders confidence and limits the potential of these tools. Beyond the variance of emotional confidence in the capability, there are tangible risks related to intellectual property and attribution that risk undermining the burgeoning technologies before they are fully able to “learn” necessary capacities to generate solutions, maximise quality, and express the qualities of good character for the development of morally and ethically appropriate pharmaceutical solutions. Furthermore, a lack of their access diminishes the post marketing research that maximizes and sustains the quality products needed for success.

4. How do you address resistance to change and what strategies can convince stakeholders of the urgency?

CLARK: As the generative AI revolution emerges, the strongest barriers to the capability revolve around its inherent likelihood of “hallucinating” producing novel, albeit false, misleading, or even nonsensical “solutions.” These contributions arise from bias or error in the training data. Therefore, convincing stakeholders that there is value in rapid adoption of these tools necessarily must focus on the quality of the results obtained leading to objective outcomes. Tangibly, this means a rapid increase in the use of digital twins with a tangible outcome of reducing the oncology clinical trial failure rate and costs (both for patients and companies), even a small amount could have a significant impact on adoption of the capabilities while also creating the opportunities to generate more data that would further creating training for greater expansion. These digital twins have the potential to transform clinical development to forecast outcomes for individuals, which has statistical potential to represent a within-subjects design and greater statistical power of trials. Both scientifically and practically, the outcome would be reduced risks for patients, faster clinical trials, and less costly clinical trials.

BUDHWANI: Drawing on Doc. Feller’s physics classes at Coe College, I would like to think that resistance here is really disconnected capacitance. To unleash this stored energy, we must complete the circuit by addressing disconnections like uncertainty, risk aversion, administrative burden, or a “this is how we’ve always done it” mindset. Ironically, hype and FOMO might actually make matters worse—forcing broccoli down someone’s throat won’t make them love broccoli. Instead, we must repair the circuit with empathy and strategy. First, messages must address concerns so that conversations can flow from resistance to potential. Second, data. We must generate data that demonstrates measurable benefits to build credibility. Third, collaboration. As I often say, this is not a solo sport. Finally, emphasise the cost of inaction: misdirected investments, growing inefficiencies, falling behind competitors, and missing blockbuster opportunities for income and impact. Echoing Matt, the approach must be rooted in de-risking adoption through respect and results.

5. What regulatory challenges do you foresee and how should regulatory frameworks evolve?

CLARK: A key challenge for regulatory bodies is that their processes are inherently slow, necessarily slowing the development of innovations. There are real implications in oncology unless action is taken to address these problems. When paired with data sharing challenges and privacy or legal concerns, the slow pace of confirmatory regulatory actions that protect the drug supply along with a likely lack of transparency from AI models, the interaction of regulatory processes and the developer using these novel tools will ultimately serve to create a logjam. This probable bottleneck could propagate across the clinical trial enterprise and impair drug development across disease conditions. Clinical trial resources are already finite. If global regulatory bodies cannot develop frameworks for a solution, the oncology space alone could generate significant problems for clinical trials at large. Fortunately, the benefit-risk equation in historical drug development for oncology may contribute to solutions at the speed of relevance. Initiatives like Project Optimus and the recent draft guidance for improved oncology clinical trials for accelerated approval by the U.S. Food and Drug Administration are clear examples showing that while there is a risk, regulators may already be leaning into the opportunities.

BUDHWANI: Matt is the expert in this area and has deftly covered most of the ground, so all I will add is that the key to a productive regulatory framework lies in a kind of Goldilocks zone for innovation. Think of the regulatory framework as the sun: too much regulatory gravity, and all the planets collapse under its weight; too little, and we are flung into the cold, dark space of unbridled greed and fear. Neither is sustainable. We must strive for a framework that ensures patient safety while fostering an environment that drives the transformative potential of emerging technologies.

6. What steps can industry, regulators, and investors take to strike the right balance in bridging the drug development “valley of death” instead of tripping into it?

BUDHWANI: We can start by weighing costs of continuing “what we’ve always done” against those of breaking out of the insanity loop. Emerging precision personalised medicine technologies have the potential to redefine what’s possible, but Rome wasn’t built in a day, and certainly not by one person. The same applies to bridging the drug development valley of death. Pharma and biotech should prioritise collaboration, pooling resources and expertise to de-risk development and adoption. Investors can navigate with a steady hand, steering clear of traps that prioritise a quick buck at the expense of their pipeline’s transformative potential—both income and impact. Industry groups and regulators can work together to establish frameworks that provide clarity and foster innovation. Ultimately, it comes down to aligning on common goals: improving patient outcomes, reducing inefficiencies and missed opportunities, and accelerating the path to market.

CLARK: The costs and risks associated with drug development are immense and inhibit meaningful development of safe and effective drugs, especially in oncology. Costing US$1.2 billion to upwards of US$4.4 billion and over 18 years to market per successful oncology drug according to a recent 2024 analysis in JAMA and in 2023 by Queen’s College Belfast, bridging the valley of death must involve innovation that accelerates development while exploiting newly developed advantages of global data sources and AI. The Queen’s College study specifically indicated that a precision medicine approach for drug development with a companion diagnostic could save upwards of a billion dollars per approved drug. This situation indicates a clearer path to return on investment, thus, providing hope that the valley of death can be largely avoided. Predictability for drug development investments along with companion diagnostics motivates the innovation to market pipeline while also providing an opportunity for even safer drugs that are targeted for greatest effect.

7. Looking ahead, how do you see cancer drug development evolving over the next decade?

CLARK: It is hard to not be excited about the future of oncology drug development over the next decade, particularly when considered with AI-informed companion diagnostics. Add to this that the current overall probability of success is approximately 4.1 per cent, the emerging tools that exploit AI/ML and companion diagnostics allows informed pharmaceutical companies to dramatically improve the ability to tackle past problems from nonclinical through post-marketing studies for exciting technologies that are badly needed. Demonstrated returns on investment generates the energy to fuel the urgency and the action to drive the machine whereas a service-oriented regulatory environment is properly motivated by powerful benefit-risk frameworks to make the market move with alacrity. Perhaps most importantly, these tools and the current regulatory moment contain the ultimate potential to significantly improve the lives of people living with cancer.

BUDHWANI: I agree. I believe we are on the brink of a transformative shift in cancer drug development, rooted in innovation, collaboration, and an unyielding focus on patients. Precision personalised medicine, synergising synthetic intelligence with human wisdom, will take center stage—enabling us to better match treatments to tumors and reduce the staggering failure rates we face today. This will create a dynamic feedback loop in the cancer care continuum where patient tumors, not cell lines or animal models, will refine and prioritise drug development at the bench, which in turn will reduce failure rates in clinical trials and at the bedside. Rinse, repeat. A collaborative ecosystem will evolve across industry, academia, regulators, investors, and other stakeholders—with patients at the core. Cost savings from reduced failure rates and improved clinical success will translate into broader access—across geographies—not just quarterly profits, but generational impact. Together, we will redefine what’s possible and finally crush the emperor of all maladies.

--Issue 58--