Do Ensembling and Meta-Learning Improve Outlier Detection in Randomized Controlled Trials?
9 November 2023
Tested outlier detection algorithms on 800+ datasets from 7 major clinical trials
Algorithms often succeeded at finding irregular data without supervision
But no single algorithm performed best across all datasets
Proposed ensemble method combining multiple algorithms improved performance
Provides guidance on using ML for data quality in clinical trials
Detecting Irregular Data in Clinical Trials with Machine Learning
This paper evaluates machine learning algorithms for detecting irregular data in large clinical trials. The algorithms were tested on over 800 datasets from 7 major trials with 77,000 patients. The algorithms often succeeded in finding irregular data without supervision. However, performance varied across datasets, so no single algorithm worked best. The paper proposes an ensemble method to combine multiple algorithms, which improved performance. Overall, the study provides guidance on applying machine learning for data quality in clinical trials.
No comments yet, be the first to start the conversation...
Sign up to comment on this paper