Machine learning for detecting centre-level irregularities in randomized controlled trials: A pilot study.

Area of Science:

• Clinical Trials
• Data Science
• Machine Learning

Background:

• Centralized statistical monitoring is an alternative to onsite monitoring for randomized control trials (RCTs).
• Current methods are resource-intensive and struggle with novel irregularity patterns.
• Machine learning (ML) has not been applied to detect irregularities in clinical trials.

Purpose of the Study:

• To pilot the use of unsupervised machine learning for detecting center-level irregularities in multicenter clinical trial data.
• To develop a flexible ML approach capable of discovering novel irregularity patterns without prior labeling.
• To evaluate the performance of ML-based irregularity detection against current centralized monitoring methods.

Main Methods:

• Employed unsupervised ML to compute distance matrices between centers, generating continuous features.
• Utilized a one-class support vector machine to model data distributions and identify outliers.
• Evaluated the ML approach on two trials with known irregularities, comparing against existing automated methods.

Main Results:

• The ML approach demonstrated superior performance in detecting irregularities on one trial (AUROC 0.728 vs. 0.140).
• Current methods performed better on the other trial (AUROC 0.752 vs. 0.584 for ML).
• Results indicate ML's feasibility and potential value for detecting novel irregularities in RCTs.

Conclusions:

• Unsupervised machine learning shows promise for identifying center-level irregularities in clinical trials.
• This approach offers automated discovery of previously unseen patterns, enhancing flexibility.
• Further research is warranted to explore the full potential of ML in clinical trial monitoring.