Risk-based Monitoring: Using AI and Machine Learning to Find Systematic Bias in Clinical Data
Risk-based Monitoring (RBM) is a multifaceted approach to identifying, assessing, monitoring, and mitigating risks that could compromise the quality or safety of a clinical study. Combined with the widespread adoption of artificial intelligence (AI) and machine learning (ML), new approaches to Risk-based Monitoring may be essential for enhanced quality and efficiency as the landscape of clinical trials evolves.
Growth in the clinical trials landscape is expected to eclipse USD 78 billion by 2030, according to a recent report.
In a new paper titled “Risk-based Monitoring Using AI/ML Detection of Systematic Bias in Clinical Data” presented at the PHUSE/FDA Innovation Challenge at the PHUSE US Connect 2024 conference, authors Kostiantyn Drach (University of Barcelona/Intego Group, Barcelona, Spain) and Sergey Glushakov (Intego Group, Maitland, Florida USA) spearhead a new approach to utilizing AI/ML techniques, including graph-based machine learning algorithms, to find systematic bias in clinical data.
Risk-based Monitoring is essential
Encouraged by global health authorities like the US Food and Drug Administration (FDA), pharmaceutical companies are urged to develop monitoring plans that prioritize a Risk-based Monitoring approach. The goal is to safeguard data quality and processes critical to human subject protection and clinical trial integrity.
Centralized monitoring stands out as a key component of Risk-based Monitoring, offering many advantages over traditional on-site monitoring methods. With centralized monitoring techniques, sponsors of clinical trials can use AI/ML in Risk-based Monitoring to significantly:
- reduce human error with less manual work
- lower costs by focusing on sites where problems are likely to exist, and
- enhance data analysis and cross-site comparison capabilities.
Additionally, this approach enables the identification of potentially fraudulent, inaccurate, or biased data.
Overcoming challenges with topological data analysis (TDA)
Even with detailed guidance provided by the FDA, executing Risk-based Monitoring plans tailored to specific clinical trials remains a challenge for some pharmaceutical companies and other clinical trial sponsors.
To help overcome this, Drach and Glushakov propose innovative solutions grounded in topological data analysis (TDA) and AI/ML techniques in the paper. The goal is to supplement traditional statistical by-site data processing, identify key risk indicators (KRIs), and pinpoint problematic sites, thereby facilitating targeted on-site investigations.
To demonstrate the efficacy of these solutions, Drach and Glushakov utilized a publicly-available National Institute on Drug Abuse (NIDA) dataset, investigating the effectiveness of the buprenorphine/naloxone combination tablet in treating patients with opiate dependence.
By leveraging various domains such as general health, vital signs, and psychological health, the researchers conducted two experiments to validate the proposed Risk-based Monitoring methodologies.
