Novel Method Forecasts if COVID-19 Clinical Tests Will Fail or be Successful

Studies to develop drugs, vaccines, devices, and repurposed drugs are urgently needed to win the battle against COVID-19. Randomized clinical trials are used to supply evidence of safety and efficacy and better understand this new and evolving virus. Since July 15, over 6,180 COVID-19 clinical trials have been registered through ClinicalTrials.gov, the United States national registry and database for privately and publicly funded clinical studies conducted worldwide. Learning which ones are most likely to succeed is imperative.

The first to ever model COVID-19 completion versus cessation in clinical trials were the researchers from the Florida Atlantic University’s College of Engineering and Computer Science through machine learning algorithms and ensemble learning. The study, published in PLOS ONE, gives the most extensive features for clinical trial reports, including model trial administration, study information and design, keywords, eligibility, drugs, and other features.

This research indicates that computational methods can deliver effective models to comprehend the difference between completed vs. terminated COVID-19 trials. Furthermore, these models can also predict COVID-19 trial status with satisfactory accuracy.

Because COVID-19 is a relatively new disease, very few trials have been formally terminated. Thus, researchers regarded three types of tests for the study as cessation trials: terminated, withdrawn, and suspended. These trials stand for research efforts that have been stopped/halted for specific reasons and represent unsuccessful research efforts and resources.

“Our research’s main purpose was to predict whether a COVID-19 clinical trial will be terminated or completed, suspended or withdrawn. Clinical trials involve plenty of resources and time that includes recruiting human subjects and planning,” said Xingquan “Hill” Zhu, Ph.D., senior author and a professor in the Department of Computer and Electrical Engineering and Computer Science, who conducted the research alongside the first author Magdalyn “Maggie” Elkin, a second-year Ph.D. student in computer science who works full-time as well. “If we can predict the probability of whether a trial might be terminated or not in the future, it will assist stakeholders in better managing their resources and procedures. Someday, these computational approaches may assist our society in saving time and resources to combat the COVID-19 global pandemic.”

Zhu and Elkin collected 4,441 COVID-19 tests from ClinicalTrials.gov to produce a testbed for the study. They designed four types of features (keyword features, statistics features, embedding features, and drug features) to define clinical trial administration, eligibility, study information, criteria, drug types, study keywords, as well as embedding features commonly used in state-of-the-art machines learning. Overall, 693 dimensional features were developed to represent every single clinical trial. Researchers utilized four models, Neural Network, Random Forest, XGBoost, and Logistic Regression, for comparison purposes.

Feature selection and ranking revealed that keyword features stemmed from the MeSH (medical subject headings) terms of the clinical trial reports were the most descriptive for COVID-19 trial prediction, followed by drug features, statistics features, and embedding features. Albeit drug features and study keywords were the most informative features, all four features are vital for accurate trial prediction.

The model utilized in this study achieved more than 0.87 balanced accuracies for prediction by using machine learning and sampling. These results indicate high efficacy using computational methods for COVID-19 trial prediction. The results also showed single models with a balanced accuracy of 70 percent along with an F1-score of 50.49 percent, indicating that modeling clinical trials are ideal when defining research areas or diseases.

“The clinical trials that stopped for several reasons are expensive and often represent an enormous loss of resources. Considering the likelihood of future outbreaks of COVID-19, even after the decline of the current pandemic, it is fundamental to optimize efficient research efforts,” said Stella Batalama, Ph.D. Dean, College of Engineering and Computer Science. “Machine learning and AI-driven computational approaches have been developed for COVID-19 health care applications, and in-depth learning techniques were applied to medical imaging processing to predict outbreak, track virus spread, and for COVID-19 diagnosis and treatment. The new approach developed by professors Zhu and Maggie will be helpful to design computational approaches to foresee whether a COVID-19 clinical trial will be concluded so that stakeholders can minimize the time of the clinical study, leverage the predictions to plan resources, and reduce costs.”

The National Science Foundation-funded this study, awarded to Zhu.

Originally published on Sciencedaily.com. Read the original article.

Reference: Magdalyn E. Elkin, Xingquan Zhu. Understanding and predicting COVID-19 clinical trial completion vs. cessation. PLOS ONE, 2021; 16 (7): e0253789 DOI: 10.1371/journal.pone.0253789