Feasibility Medium Impact High Time horizon Long (5+ years) Barriers Lack of infrastructure and standards for data sharing and data quality. Lack of incentives for data sharing and ecosystem collaboration. Bias towards “tried-and-true” methods over untested approaches. Biggest unlock Collaboration throughout the healthcare ecosystem to align on AI’s role in clinical trial protocol design and change behaviours. Trial design today Trial protocols outline a study’s objectives, design, methodology, inclusion and exclusion criteria, clinical end points, statistical considerations, execution plan and more. They do more than anything else to dictate a trial’s success or failure. Today, the design process is complex, slowed by manual workflows and reliant on historical data and subjectivity. The result: suboptimal trial designs that drive long cycle times, high development costs and increased probability of failure. Roughly 60% of trial protocols require at least one amendment,7 nearly half of which are considered “avoidable”. Avoidable amendments cost pharmaceutical companies $2 billion per year8 in direct costs, with amendments adding an average of 260 days to development timelines.9 These costs are passed on to patients and payers, while patients wait longer for innovative therapies. Meanwhile, clinical research teams spend as much as 30% of their time10 on conducting trials, with some labour-intensive tasks accounting for 25% of clinical trial budgets.11 Trial design enhanced by Gen AI In the future, teams will use Gen AI to mine unstructured data from prior protocols and previous trial results, real-world data (RWD), regulatory precedence and guidance, patient and site feedback and more. They will use this to develop trial concepts and plans, design key statistical elements, optimize protocols and simulate scenarios to aid design decisions. In time, trial teams will use Gen AI to create digital and surrogate end points, synthetic control arms and in silico trials (which will be conducted through computer simulation only). Gen AI will unlock greater predictive power using unstructured data and will streamline trial design and protocol drafting by automating traditionally manual processes. Doing so will reduce errors, eliminate redundant work, relieve the administrative burden and accelerate trial initiation.The Holy Grail: Clinical trial design 1 US-based biotech company Insilico Medicine has developed inClinico,12 an AI platform designed to predict clinical trial outcomes, and has partnered with multiple pharmaceutical companies to enhance drug development efficiency. inClinico uses multimodal data – including omics, text, trial design parameters and small-molecule properties – to predict the success or failure of phase 2 clinical trials. To validate its model, inClinico predicted the outcomes of phase 2 clinical assets with roughly 80% accuracy. Forecasting clinical trial outcomes with this degree of accuracy can help drug companies prioritize the therapeutic programmes most likely to succeed and optimize investment decisions, thereby reducing the costs associated with failed trials and improving the efficiency of pharmaceutical research and development (R&D).USE CASE Insilico MedicineOn clinical trial design, AI could help by analysing inclusion and exclusion criteria from similar studies, predicting patient profiles and ensuring that the criteria are practical. This could prevent issues where the designed trial doesn’t match the available patient population, thus requiring amendments and wasting time. Gavin Corcoran, Chief Development Officer, Formation Bio Intelligent Clinical Trials: Using Generative AI to Fast-Track Therapeutic Innovations 7

Intelligent Clinical Trials 2024