A new promising way for tackling the ‘Pharma Dilemma’: artificial intelligence for clinical trials

Artificial intelligence (AI) is certainly not a panacea for solving the ‘Pharma Dilemma’, in which the cost of producing new drugs continues to spiral. However, AI can be used to fundamentally change the way we perform essential steps in clinical trial design and execution, from cohort selection to patient monitoring. Merging AI and clinical expertise across engineering and medical disciplines to explore the impact of these changes on trial performance and success rates is one of the most promising leads we have for restoring efficiency and sustainability to the drug development cycle.

Artificial intelligence (AI) is certainly not a panacea for solving the 'Pharma Dilemma' , in which the cost of producing new drugs continues to spiral. However, AI can be used to fundamentally change the way we perform essential steps in clinical trial design and execution, from cohort selection to patient monitoring. Merging AI and clinical expertise across engineering and medical disciplines to explore the impact of these changes on trial performance and success rates is one of the most promising leads we have for restoring efficiency and sustainability to the drug development cycle.

Artificial Intelligence
In light of these numbers, it comes as no surprise that exploring ways to overhaul and optimize various clinical trial design steps towards increasing trial success rates has moved into the spotlight. Recent advancements in AI, i.e. technology capable of learning (machine learning/deep learning -ML, DL), reasoning and interacting with humans like humans interact with each other (human-machine interfaces -HMI), have acted as an enabler of novel clinical trial design techniques with great potential to improve trial performance.

AI for clinical trial design
Clinical trial design can be broken down into three main steps: (1) patient selection and (2) patient recruitment before the trial starts, and (3) patient monitoring during the trial. The quality and speed with which these steps are executed directly impacts the success rates and costs of trials. Composing a cohort of suitable patients maximizes but does not guarantee the chances of a successful trial outcome. Confirming eligibility of patients to participate in the trial, then empowering and motivating them to navigate complex enrolment processes allows effective and fast trial planning. Once the trial has been launched, patients need to be monitored as reliably as possible to ensure they adhere to trial protocols, to assess clinical endpoints and prevent drop-out. All these trial design features can be supported by AI technology (Figure 2). For example, ML, DL and reasoning techniques can be used during the trial design phase to identify clinical biomarkers as well as clinical endpoints and to define the patient cohort to be included. They can also be used during the trial execution phase to assess the suitability of patients through clinical trial matching and biomarker verification. HMI in combination with ML and DL techniques can further empower doctors to perform clinical trial matching more seamlessly. When it comes to patient monitoring, state-of-the-art techniques largely rely on manual self-or third-party reporting. This is inherently unreliable in general but particularly error-prone for neurology trials where the condition of patients often renders them outright incapable of detecting and logging disease episodes or medication intake. DL techniques and HMIs in combination with wearable sensors are uniquely suited to continuously monitor patients and automatically detect and report disease episodes and events of relevance to ensure clinical endpoints are detected as fast as possible and trial protocols are adhered to. These systems must be designed such that their use does not adversely interfere with patients' normal routines in order to maximize patient engagement with HMIs and wearable devices.
Of course, integrating AI techniques into clinical trial design processes does not come without hurdles: some of the biggest challenges are centred around data security, access and interoperability, as well as around building ethical AI models. On one hand, electronic health records (EHR) come in a broad variety of types, reside in highly decentralized and incompatible environments with widely differing data ownership and access rights, and are often not digitized to begin with. This is the so-called 'EHR Interoperability Dilemma' . It makes data preparation and curation a highly complex and time-consuming endeavour and thus impacts the quality of AI model outputs. Additionally, medical data is highly sensitive, and data privacy and security need to be ensured at all times. On the other hand, it is not enough for AI algorithms to produce the desired outputs-in order to gain regulatory approval and the trust of users, AI models must be transparent, explainable and fair. These features are often referred to as the 'ethics of AI' , a field of intense ongoing research which is just as important as developing the algorithms themselves. But these hurdles are not insurmountable. As a matter of fact, regulatory bodies, pharma, biotech and AI companies, start-ups, as well as medical and clinical research organizations have jointly started to tackle these challenges and to develop frameworks for exploring a variety of AI and blockchain technologies in the drug development space. What we see at this point are predominantly early-stage, proof-ofconcept and feasibility pilot studies demonstrating the high potential of numerous AI techniques for improving the performance of clinical trials. Because AI methods have only begun to be applied to clinical trials over the past 5-8 years, it will most likely be another several years in a typical 10-to 15-year drug development cycle before AI's impact can be accurately assessed. In the meantime, rigorous research and development is necessary to ensure the viability of these innovations before the AI demonstrated in pilot studies can be fully integrated in clinical trial design. Any breach of research protocol or premature setting of unreasonable expectations may lead to an undermining of trust, and ultimately the success, of AI in the clinical sector.

Where to go from here: an interdisciplinary call to action
Using AI for clinical trial design is a high-potential, high-risk early-stage technological opportunity-the risk and cost burden of pursuing it could be attenuated by accompanying it with other gradual strategic shifts in the pharma business model. For example, instead of focusing on very few diseases which cater to large patient populations with maximum market potential, the development portfolio should be more diversified and include diseases which are usually neglected by big pharma. This removes fierce competition and particularly the need to salvage one's investment in a drug that ends up performing similarly to a competitor's product, requiring costly additional studies Artificial Intelligence

Human-machine interfaces
Artificial Intelligence https://pharmaphorum.com/views-and-analysis/how-blockchain-will-revolutionise-clinical-trials-clinical-trials/ to demonstrate marginal differentiation (Scannel and colleagues have called this strategy flaw the 'better than the Beatles' problem). Smaller biotech companies, startups as well as joint programmes between non-profit and academic institutions have successfully demonstrated this approach yielding substantially decreased drug development costs. What's more, these players are in an advantageous position with respect to big pharma as they can establish new processes from scratch and have the flexibility for rapid adjustments to existing processes. This highly competitive landscape adds a forcing function to the transformation process of big pharma which will only intensify as the AI toolbox expands and IP protection of blockbuster drugs expires.
In many ways the situation is similar to one faced previously by the semiconductor industry: as Moore's law, an unquestioned model, accepted and followed for decades by stakeholders in a billion-dollar business loses validity, a paradigm shifting reorientation of the technical innovation roadmap is necessary to return to sustainable growth and productivity. One may liken the unfolding scenery to an ocean steamer heading down a canal that gets narrower every mile: it is foreseeable that ploughing ahead full steam in disregard of the changing surroundings will inevitably bring the entire vessel to a standstill, no matter how big it is. What is needed instead to keep the momentum going is an adaption of the mode of transportation and a clear understanding of the evolving new environment that needs to be navigated. It will mean slowing down to make time and free resources for exploring alternative scenarios. New ideas will be needed, and true innovation will be crucial: just painting the ship in a different colour will not solve the issue. New skills and new techniques will have to be brought on board just as redundant ones will have to be thrown overboard. There will be fundamental changes and the need to head into uncharted territory. The business will look different after the transition and risk will no doubt be involved to get there. The fact that others are steaming alongside in the same direction does not change the fact that shorelines are closing in left and right-onto everyone. Neither does going faster than others nor sinking competitors change the outlook. Real, sustainable solutions will carry risk, require strong executive sponsorship, rely on global collaboration, utilize deep scientific skills in healthcare analytics and AI, and follow visionary change management with a transition plan in place. Such a plan has to take the timing factor into account-innovation will be integrated gradually alongside existing processes -and should carefully balance change with stability: on one hand existing expertise, product offerings and market advantages need to be leveraged to the fullest to minimize risk for the business. On the other hand, pharma needs to collaborate with AI experts and double down on R&D commitment to data privacy, data security and explainable AI as success in these domains will be key to overcoming resistance amongst adopters, regulators and decision makers and eventually be the yardstick for establishing leadership positions in the evolving post-blockbuster drug market. ■