Protocol Development #75
"Define the need before the AI tool that you want to use" - this is the main message from the predominantly AI theme of this issue. The recent explosion of AI-augmented capabilities offers many solutions to protocol-related clinical trial problems - such as how to elect the "right" population (one that is representative and accessible), how to select the "right" outcomes (that fit the regulatory requirements and are meaningful to patients), and how to get the design and conduct "right" so that the trial is successful. In this issue there's three AI-related articles - two looking at AI application (Yingngam, Azenkot et al), and one that looks at clinical trial document searching and retrieval (Das et al). In addition, we've got barriers to trial access (Ebrahimi et al), public feedback on ICH E6(R3) Annex 2 (EMA), sustainable clinical trials (Singh et al), and participant experiences with DHTs (Reale et al).
Patient Engagement
[Education] Ebrahimi et al published Practical Guide to Clinical Trial Accessibility: Making Trial Participation a Standard of Care - an educational overview of how to make trials more accessible. Having recently spent time with academic scholars discussing the approach to protocols, this article nicely summarizes key considerations from oncology but realistically ones that could be expanded to most therapeutic areas. In the barriers to access figure (above), each stakeholder group has at least one barrier related to the protocol. Representation gap (patient advocates), misaligned eligibility (community oncologists), overly conservative criteria (industry partners), and eligibility rigidity, complex protocols, and limited generalizability (academic oncologists). Patient input into protocol development, broadening eligibility criteria, adopting inclusive design practices including ASCO/FCR recommendations, and simplified designs are proposed solutions - and the pillars of ICH E8(R1).
Regulations & Guidance
[Report] The EMA released the Overview of comments received on ICH E6 (R3) Guideline for good clinical practice – Annex 2 (EMA/CHMP/ICH/495903/2024). At 47 pages, there are more themes that can be covered here. In general, feedback requested more hands-on implementation instruction rather than high-level guidance that "may" or "should" be used. There were a few interesting points on more guidance on how to implement QbD, engage with patients, and navigate a strategy in the absence of clear regulatory guidance. Overall, it's an interesting, and revealing, series of responses covering a range of perspectives.
Clinical Trial Design
[Article - Research] Singh et al published Integrating Sustainability Pillars Into Trial Design Decision- Making: Results of an International Survey. The authors surveyed 447 participants to identify sustainability priorities over three pillars - environmental impact, economic considerations, and social criteria. Greenhouse gasses (environmental), probability of technical success (economic), and patient convenience (social) were the main priorities, with DCTs being perceived as being more sustainable in all pillars than traditional DCTs. The output of this research came in 3 recommendations: 1) perform simulation studies (to understand the impact on sustainability), 2) protect the environment, and 3) protect the supply chain (as Sponsors transition to DCTs they need to protect the complex supply chains that govern intervention access).
Digital
[Article - Case Study] Reale et al published Participant Experiences in a Decentralized Clinical Trial Using Digital Health Technologies: the ACTIV-6 Study, reporting on 28 individuals who participated in ACTIV-6 on their experiences with DHTs. The experiences were largely positive and highlight the use of DHTs/DCTs as positive - nothing we haven't heard before. What is of particular value for this article are the participant quotes littered throughout. Interpretation of PROs, the relation of pre-existing symptoms with ongoing COVID and issues of trust provide valuable insight that is relevant for anyone working in protocol development as it anchors trial design discussion to the context of the patient.
Protocol Tech & Artificial Intelligence (AI)
[Book] Bancha Yingngam published AI in Clinical Trial Design and Patient Recruitment, a book chapter that discusses the use of AI to "make protocols more specific and execute patient recruitment more efficiently". In the chapter, the author outlines 6 steps for AI-augmented design - from data collection and integrations, through protocol optimization and feasibility, to submission. The type of technology can be applied to more (or less) optimal use cases, highlighting the need to correctly identify the need before the tool (rather than trying to find a use for a tool). In the examples provided, predictive analytics are best used for e.g., recruitment, or dose optimization, whereas NLP for eligibility screening of EHR. Lastly, machine learning can be used to improve decision making processes - e.g., learning from data of prior trial outcomes to refine protocols.
[Education] Azenkot et al published Artificial Intelligence and Machine Learning Innovations to Improve Design and Representativeness in Oncology Clinical Trials as an educational tool. The scope and capabilities presented are far reaching and - in general - lacking pragmatism around implementation. For example, early detection of study endpoint data may support the use of adaptive trial design methodology, sounds nice in theory but is difficult to map out the steps needed to yield efficiency. In my opinion, what is valuable are the sections on bias and misrepresentation (i.e., AI is not objective without bias, bias can be introduced in a variety of ways such as by the way data is collected and used), ethical principles, and regulatory considerations. he FDA's risk-based credibility assessment (figure above) also reinforces the message from Bancha Yingngam's book chapter above - identify the question of interest (i.e., need) before the tool.
[Article - Research] Das et al published SECRET: Semi-supervised Clinical Trial Document Similarity Search, a detailed behind-the-scenes for what goes into searching for comparable clinical trials. In summary, to efficiently retrieve similar clinical trials - you need a methodology that can accommodate the need - simply copying-pasting large pieces of data is inefficient and time-consuming. The authors developed SECRET and compared it to existing baselines including Trial2Vec. Using a case study involved looking for a trial that most-resembled NCT00061594 using title, intervention, disease, and age variables. Trial2Vec returned NCT03470103 whereas SECRET returned NCT00433017. From a subjective review, SECRET was significantly closer - hopefully this means that searching provides more accurate results. With such a technical success, all that remains is to help teams develop strategies for using such tools to answer relevant questions.