The future of clinical development is set for significant change, driven by the integration of new digital data sources, advanced computing power for detecting meaningful patterns using artificial intelligence (AI) and machine learning (ML), and increasing regulatory support through new collaborations.[1] The United States Food and Drug Administration (FDA) has noted a significant increase in the use of AI/ML in drug development, with over 100 submissions in 2021 alone.[2] These submissions cover the entire drug development process, from drug discovery and clinical research to post market safety and advanced pharmaceutical manufacturing. Here, however, we concern ourselves with how AI and ML approaches can impact the way clinical trials are designed, conducted, and analyzed, offering potential enhancements in efficiency, reduced costs, and improved patient outcomes.

In this article, I explore the opportunities and challenges of AI/ML in clinical development, particularly in the context of Cytel’s operations.

Opportunities

Deeper Data Insights

The growing data volume and diversity often demand innovative, sophisticated analysis methods to draw insight and understand relationships between factors influencing disease prognosis and treatment benefits. AI, and more specifically ML, offers an effective solution by analyzing complex datasets to uncover patterns and predict outcomes. These methods use algorithms derived from statistics and computer science to develop models with high predictive accuracy, if not always good explanatory properties. However, in many cases, those are the methods of choice when integrating data from various sources like clinical trials, real-world evidence, and genomic databases to make better-than-human predictions that provide deep insights. Ultimately, these predictive analytics enhance treatment effectiveness and patient care by identifying trends and making forecasts on outcomes, disease progression, and patient responses.

Improved Patient Recruitment and Retention

Clinical trial failures often arise from ineffective patient cohort selection and recruitment methods that fail to enroll suitable participants quickly. Selecting the right patient population for your study design is a critical step to realizing a successful clinical development plan and commercial strategy. Running classification algorithms, using semi-supervised learning, for example, can help the team refine study inclusion/exclusion criteria and update recruitment strategies during the trial. Additionally, AI has the potential to predict patient dropout risks, enabling targeted retention strategies.

Personalized Medicine

AI can identify subgroups of patients likely to benefit from specific treatments based on their genetic, demographic, and clinical profiles, thereby advancing personalized medicine. This approach not only enhances patient outcomes but also optimizes resource use in clinical development by focusing efforts on patient groups most likely to respond to the therapies being tested. Interpretable ML methods are particularly useful to overcome the black-box nature of most AI algorithms and validate the explanatory aspects of any modeling exercise with clinical input and review.

Enhanced Safety Monitoring

ML algorithms can identify adverse events early, facilitating timely interventions and enhancing patient safety. By continuously monitoring patient data, AI ensures real-time detection of safety signals and allows for prompt corrective actions. Naturally, this can be applied in post-approval surveillance studies, and there is also an opportunity to utilize such algorithms during clinical development and benefit-risk assessment.

Pitfalls

Data Safety and Privacy

Handling sensitive patient data in AI systems requires rigorous privacy protections to prevent unauthorized access and data breaches. It involves encrypting data, implementing strict access controls, and adhering to regulatory frameworks like GDPR and HIPAA. Advances in data technology, such as tokenization and AI-driven extraction from unstructured physician notes, may enable the creation of detailed datasets to identify narrow patient cohorts that match clinical trial participants.

Regulatory Challenges

Developing and adhering to regulatory standards for AI in clinical settings is essential to ensure these systems are safe and effective. This includes setting standards for validation, reliability, and ethical use. Navigating the complex regulatory landscape — which is just emerging or has yet to fully develop — to gain approval for AI-driven solutions requires careful planning and extensive documentation. Close exchanges and collaborations with health authorities are needed to progress thinking and practices in the industry.

Algorithm Transparency and Complexity

A significant challenge in integrating AI and ML in clinical development is thoroughly understanding the algorithms, especially their design and potential biases. Biases can stem from the data used to train these models, leading to skewed or inaccurate outcomes that may negatively impact patient care and treatment efficacy. Furthermore, the risk of technical failures in AI systems is a substantial concern. In the case of large-language models, hallucinations are a well-documented issue that arises from the stochastic nature of the underlying algorithms. Other ML models are not immune to similar problems, and processes need to be established to ensure quality, reproducibility, and generally speaking, trust in the resulting outcomes.

Skill Gaps and Training

In clinical development, integrating AI/ML requires comprehensive knowledge of both technology and clinical research. The effective implementation of these technologies relies on skilled professionals who can manage these complex systems and accurately interpret their outputs. Moreover, continuous learning and investment in training are necessary to keep up with the rapid advancements.

Final Takeaway

AI/ML is reshaping clinical development with its ability to analyze complex data, predict which patient groups will respond to therapy, and therefore personalize treatments. Although the technology promises significant improvements over the next few years, it is easy to overpromise and set unrealistic expectations. To fully capture the opportunities and realize the benefits, the industry must address challenges such as data privacy, algorithm transparency, and integration complexities. By proactively managing these pitfalls and leveraging AI/ML effectively, stakeholders can drive innovation, optimize resource use, and accelerate the development of therapies that better meet patient needs.

At Cytel, we are adapting to this transformation by integrating advanced statistical thinking into our approach to AI/ML technology. This enables us to deliver tailored solutions that not only optimize product development pathways but also accelerate time to market.

References:

[1] Shah, P., Kendall, F., Khozin, S. et al. Artificial intelligence and machine learning in clinical development: a translational perspective. npj Digit. Med. 2, 69 (2019). https://doi.org/10.1038/s41746-019-0148-3

[2] https://www.fda.gov/science-research/science-and-research-special-topics/artificial-intelligence-and-machine-learning-aiml-drug-development

Interested in learning more? Contact our expert team for unparalleled guidance in trial design, deployment, real-world solutions, and regulatory strategies.

Share Via