013: Use of Artificial Intelligence in Drug Development

Statement of Purpose, Innovation or Hypothesis: The purpose of this study was to investigate the potential of artificial intelligence (AI) in streamlining pharmacokinetic non-compartmental analysis and data cleaning tasks. The goal was to assess whether AI could enhance efficiency and accuracy in these complex tasks, which traditionally require significant manual effort and expertise.

Description of Methods & Materials: The study employed an AI system designed to review and rectify errors in R code, a common language for statistical computing and graphics. The AI system was tested on a variety of tasks, including data cleaning and pharmacokinetic non-compartmental analysis. The system’s adaptability was also assessed by introducing unexpected scenarios and changes in the data structure. The aim was to evaluate how well the AI system could adapt to these changes and continue to perform its tasks effectively.

Data & Results: While the current AI system demonstrated proficiency in several tasks, it also revealed significant limitations. When the system encountered complex coding errors, it was unable to rectify them and instead entered a loop of repeated attempts to fix the same error. This not only failed to resolve the issue but also led to a waste of computational resources and time. Furthermore, the AI system’s inability to understand the context or the bigger picture of the task at hand was another significant limitation. Unlike human analysts who can leverage their domain knowledge and experience, the AI system was confined to its pre-programmed instructions and could not adapt to unexpected scenarios or changes in the data structure. Additionally, the AI system lacked the ability to explain its decision-making process, a limitation often referred to as the ‘black box’ problem in AI.

Interpretation, Conclusion or Significance: These findings highlight the critical role of human intervention in using AI for complex tasks. While AI can enhance the efficiency and accuracy of these processes, currently human oversight is essential to address AI limitations and ensure the validity of the results. The study underscores the importance of a mixed-method approach that combines the strengths of AI and human expertise for optimal results. Future research should aim at improving the error-handling capabilities of the AI system to reduce the need for human intervention and maximize the benefits of AI in pharmacokinetic non-compartmental analysis and data cleaning. Furthermore, the advancement of AI should prioritize training using real world datasets that encompass a broad spectrum of pharmacokinetic data sourced from various clinical pharmacology studies. This will contribute to more robust, efficient, and reliable pharmacokinetic studies in the future, to set clear and realistic algorithms that will enable paving the way for significant advancements in the field.

Disclosures: Authors have no disclosure to report.