Paving the Way for AI’s Potential in Clinical Trials: Q&A With Rust Felix, CEO and Co-founder of Slope

Rust Felix, CEO and Co-founder of Slope

While the potential of artificial intelligence (AI) has become more popular in the clinical trials space, compliance and quality challenges still need to be addressed. In this exclusive Q&A with Applied Clinical Trials, Rust Felix, CEO and Co-founder of Slope, sheds light on some of these concerns and discusses some of the potential uses of AI he is seeing in the industry.

ACT: When it comes to AI in clinical trials, what current challenges do you think industry needs to address before it is able to fully utilize AI?

Felix: One of the primary challenges the industry must confront is the issue of data quality and lack of standardization, especially in the context of biospecimen lifecycle management. Currently, the process is exceedingly manual, defined by paper-centric processes, complex sample management schemes, and an expanding volume of sample collections. The escalation in study complexity has precipitated a staggering increase in the number of procedures, endpoints, and data points collected, as evidenced by an analysis of nearly 10,000 clinical trial protocols.

Regulatory compliance and data quality, both essential for demonstrating protocol endpoints, are increasingly compromised due to the complicated nature of today’s studies, combined with manual, paper-centric solutions. The industry is witnessing a troubling trend, with 95% of respondents in a Fierce Biotech survey reporting study delays or quality issues due to clinical inventory or bio-sampling problems. Nearly 90% have noted instances where research sites miss required sampling timepoints or misplace patient samples.

Furthermore, an overwhelming majority of sponsors and contract research organizations (CROs) still rely on antiquated paper instructions to ensure compliance with site protocol amendments. This not only introduces errors and inefficiencies, but also compounds the challenges associated with protocol amendments. The reliance on paper requisition forms for entering sample metadata into Laboratory Information Management Systems (LIMS) exacerbates these issues, leading to significant downstream problems in managing sample data and an increased burden on lab personnel.

Before the industry can fully leverage AI in biospecimen management, there is a critical need to improve the collection and structuring of data that would be used to train and inform these models. The current reliance on siloed datasets for biospecimen analysis results in poor-quality training data, which in turn leads to ineffective AI implementations. Addressing this fundamental challenge is essential for unlocking the full potential of AI in streamlining and enhancing biospecimen lifecycle management in clinical trials.

ACT: What are some specific areas within data management that AI can streamline?

Felix: It would depend on how the AI is implemented, but the clear frontrunners would be database builds, data reconciliation, and data analytics. For example, the industry could leverage AI to expedite lab database builds that support various clinical trial logistics, including sample testing, processing, storing, and shipping. This process is very manual, takes weeks to months, and can be prone to error. In addition, it requires several key stakeholders across the sponsor, CRO, and lab to provide input and come to a consensus. AI could make this process more efficient and accurate, linking the database builds to other processes required to produce lab kits, financial reports, and more.

Data reconciliation is an area where AI would likely have a significant impact. Data reconciliation and sample tracking involves many different data sources with different formats, and these processes also fall victim to data quality issues. With data reconciliation, sponsors must identify discrepancies across various data sources, as well as protocol deviations. These queries must be investigated and resolved, as data integrity is paramount in every trial. For a steady-enrolling trial with both pharmacokinetic and pharmacodynamic sample collections that move across many sites and labs, it can take months to identify and resolve discrepancies. The traditional SAS and R programmatic data listings that facilitate this comparison are often not written correctly and are not able to handle variant data sources to identify every discrepancy. The ability to apply AI to assist in code generation and data mapping would be a game changer.

Even if a clinical study team is able to pull off a flawless study by collecting and testing all samples, with no protocol deviations, the resulting throughput of data and analysis is difficult for sponsors. In the case of precision-based medicine trials that look at genomic sequencing or other high-throughput datasets, the analysis can be especially challenging. The ability to QC the assay data, create data models that lead to more intelligent conclusions, and correlate these findings to other trials or pre-clinical data would expedite clinical trial endpoints. This could also lead to more accurate application of study drugs to disease types with particular molecular mechanisms of action that the drugs could impact.

ACT: While AI has great potential, how important is it to still keep a level of human intervention when it comes to making final decisions?

Felix: The AI movement is exciting and appears to have potential we still do not fully understand. However, there have been several instances of AI coming to incorrect conclusions and summaries, in addition to security concerns over the data that AI models collect. It is paramount that there are clear levels of oversight and validation. It’s also critical to have cross-functional committees review AI models to assess how they are functioning, the decisions they influence, and their potential biases and gaps.

ACT: As we move forward, where do you think the real potential of AI in clinical trials could end up?

Felix: I believe there is a lot of potential to leverage AI to identify novel biomarkers and mechanisms of action, as well as to enhance the field of computational biology. Removing human limitations and augmenting current tools and software that aid these areas could theoretically lead to an explosion of new drugs that not only stop disease, but cure hard-to-treat disease types. Not only could we see new drug discovery as a direct result of utilizing AI, but there is also the potential that we could identify treatments already on the market that can impact disease states they are not currently used for.

When it comes to sample and data management on clinical trials, the ability to leverage AI to manage large, siloed, and differential datasets would ensure data integrity, governance, and quality. This would give sponsors the ability to quickly identify any errors that occurred so that those discrepancies are corrected and the impact can be assessed. It’s critical to ensure that the proper samples are tested and that scientists have access to accurate sample metadata and pre-analytical variables that impact interpretation of bioanalytical and biomarker assays.

As AI becomes more prominent, better governed, and better understood by both regulators and the general public, the benefits of this exciting new technology would allow for an objective tool that could be applied to drug development, data reconciliation, and analytics approaches, thereby eliminating current gaps and expediting study timelines.