Researchers at Florida Atlantic University’s College of Engineering and Computer Science have become the first to model the probability of COVID-19 trials finishing, terminating, pausing or being withdrawn using machine learning algorithms.

Their study, published in PLOS ONE, downloaded 4,441 COVID-19 trials listed on ClinicalTrials.gov for analysis. According to the study findings, their model was accurate in predicting success or failure for COVID clinical trials, and the results showed computational methods were effective at helping to understand differences between completed and ceased trials. They also showed that such models can predict the status of a COVID trial with sufficient accuracy and help in planning and reducing costs for trials.

They used four feature types (statistical features, study keywords, drug features and embedding features) to understand trial administration, eligibility, study information, criteria, drug types, study keywords and embedding features frequently used in machine learning. Using those categories, the researchers generated 693 dimensions to represent each clinical trial pulled from ClinicalTrials.gov. The study found that while drug features and study keywords were the most informative of the feature types, all were needed to accurately predict a trial’s success or failure.

The researchers said they will conduct further evaluation of features found to be associated with trial completion vs. cessation and plan to deliver interpretable results experts can use to understand what types of trials are more and less likely to make it past the finish line.

Read the full study here: https://bit.ly/3leTqXF.