2. Build data infrastructure: Governments should establish centralized or federated hubs that aggregate data for improved accessibility and use. For example, the Indian state of Telangana has established itself as a leader in promoting data sharing throughout India’s fragmented healthcare system with its Citizen Health Profile initiative to collect biochemical and phenotypic data from 40 million citizens and make it available to healthcare stakeholders. In many cases, public–private partnership will be needed to fund infrastructure and ensure that sharing initiatives maximize the value of data while safeguarding privacy and security. Establishing regulatory sandboxes, where new technologies can be tested in a controlled environment, can encourage innovation while ensuring that new use cases are safe and ethical. 3. Create incentives for data sharing: Public– private partnerships must be formed to drive policies that create incentives for networked data sharing and reduce fragmentation by aligning interests throughout healthcare. The UK Biobank – a government initiative underwritten largely by the pharmaceutical industry – is a prime example.22 Monetization models can create incentives for data sharing – for instance, by monetizing aggregated data that can then be shared among contributors. Quasi-private initiatives, such as Epic’s Cosmos23 – which aggregates data from community health systems and then sells it to pharmaceutical companies for R&D – is another strong example. Clinical development leaders stressed in interviews that governments need to strike a balance between privacy, safety and innovation. The establishment of the US Food and Drug Association’s AI Council,24 which oversees AI, including its use in regulatory decision-making, is encouraging. The private sector should propose adaptive regulatory frameworks that provide guidance without stifling innovation. Barriers Dramatic changes to the clinical development status quo are inherently discomfiting. Inertia, skill deficits and lack of trust must be overcome for humans to fully embrace progress. –Overcome inertia: Although companies have invested in using AI and Gen AI to enable digital end points, adaptive trial designs and synthetic control arms, these efforts have yet to consistently demonstrate their full value. That is starting to change. The Tufts Center for the Study of Drug Development and the Digital Medicine Society (DiMe) united with industry leaders to measure potential ROI from using digital end points in clinical trials.25 They found that doing so shortened trial phases, allowed for smaller enrolment sizes, increased expected net present value (eNPV) by as much as $40 million per indication and offered returns of between four and six times investment. These results notwithstanding, leaders interviewed by the World Economic Forum and ZS said they were hesitant to fully embrace new trial methods, given the highly regulated nature of life sciences and the high stakes of each individual trial. –Build AI skills: With AI’s advance, there is a growing need for expertise at the crossroads of technology, healthcare and data science. While AI has made inroads in biostatistics for analysis and drug discovery, its integration often sits outside core development teams. AI models can produce varying outputs depending on the specific context, underscoring the importance of teams with expertise in both clinical development and AI. Additionally, AI systems are often deployed within an ensemble of specialized models, requiring teams to coordinate systems and integrate outputs into clinical and operational workflows. –Overcome trust barriers: Operationalizing Gen AI in clinical development depends on trust – from the public, who must consent to their data being used; from research entities, who must believe the benefits of data sharing outweigh the risk of intellectual property leakage; among individual scientists sceptical of AI’s value compared with traditional techniques; and among regulators, who must believe that AI’s outputs are safe, ethical and reliable. Recommendations Trust must be promoted in the ecosystem. 1. Develop smart AI policies: Governments must establish guidelines that ensure the safe and effective use of AI, while also promoting workforce upskilling. This could take the form of certifications or training programmes that build confidence in trial teams’ skills in using AI in clinical development. 2. Enforce data transparency: While there is a push for greater transparency in AI models, companies are hesitant to reveal the data used to train models. Regulators should provide 2.2 Innovation culture, trust and workforce considerations Intelligent Clinical Trials: Using Generative AI to Fast-Track Therapeutic Innovations 13

Intelligent Clinical Trials 2024