To win the fight against COVID-19, there is an urgent need for studies to develop vaccines, drugs, devices and reused drugs. Randomized clinical trials will be used to provide evidence of safety, effectiveness and a better understanding of this novel and evolving virus. As of July 15, more than 6,180 clinical trials of COVID-19 have been registered through ClinicalTrials.gov. Knowing which ones are likely to be successful is essential.
Researchers from Florida Atlantic University’s College of Engineering and Computer Science are first to model completion of COVID-19 versus completion in clinical trials using machine learning algorithms and ensemble learning. The study, published in PLUS ONE, provides the most complete clinical trial reporting capabilities, including trial management modeling, trial information and design, eligibility, keywords, medication, and other functions.
It is hoped that the new approach will be helpful in developing computational approaches to predict whether a clinical trial of COVID-19 will be completed so that stakeholders can use the predictions to plan resources, reduce costs and to minimize the time of the clinical study. The study was funded by the National Science Foundation.
This research shows that computational methods can provide effective models to understand the difference between completed and abandoned COVID-19 studies. In addition, these models can predict the status of the COVID-19 study with satisfactory accuracy.
Since COVID-19 is a relatively new disease, very few studies have officially ended. Therefore, for the study, the researchers considered three types of studies to be termination studies: terminated, withdrawn, and suspended. These studies represent research efforts that have been stopped / paused for certain reasons and research efforts and resources that have not been successful.
“The main purpose of our research was to predict whether a clinical trial on COVID-19 will be completed or terminated, withdrawn or suspended. Clinical trials require a lot of resources and time, including planning and recruiting subjects, ”said Xingquan“ Hill ”Zhu, PhD, senior author and professor at the Institute of Computer and Electrical Engineering and Computer Science. Zhu conducted the research with lead author Magdalyn “Maggie” Elkin, a sophomore in computer science who also works full-time. “Being able to predict the likelihood of whether or not a study will be abandoned at a later date helps those involved to better plan their resources and procedures. Ultimately, such computer approaches can help our society save time and resources to fight the global COVID-19 pandemic. “
For the study, Zhu and Elkin collected 4,441 COVID-19 studies from ClinicalTrials.gov to build a test environment. They designed four categories of traits (statistical traits, keyword traits, drug traits, and embedding traits) to characterize the studies, resulting in 693 dimensional traits that represent each study. The feature selection and ranking indicated that keyword features derived from medical catchphrase terms in the clinical trial reports were the most informative for predicting COVID-19 studies, followed by drug features, statistical features, and embedding features.
Edited by Gary Cramer