Comparison of Machine Learning Methods with Traditional Models for Use of Public Trial Registry Data to Predict Sites Needed and Time from Study Start to Primary Completion Date

• Activity Number: 205 - Applications of Machine Learning
• Type: Contributed
• Date/Time: Tuesday, August 4, 2020 : 10:00 AM to 2:00 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Linghui Li* and Gabriela Feldberg and Faisal Khan and Sandra Smyth and Karin Schiene
• Companies: AstraZeneca and AstraZeneca and AstraZeneca and AstraZeneca and AstraZeneca
• Keywords: Trial Planning; ClinicalTrial.gov; Machine Learning; Traditional Models

Abstract

Estimation of clinical sites needed and time from study start date to study primary completion date during trial planning is crucial to be successful in the conduct of a clinical trial in the pharmaceutical industry. A precise estimation enables resources to be allocated to a study as needed and a determination of the time it takes for patients to complete primary endpoint from study start which allows a study team to run an efficient trial. Through utilizing machine learning methods and traditional models on ClinicalTrial.gov data, machine learning methods offer an improvement over traditional linear regression in predicting sites needed and the duration regarding indication, number of patients, study phase, and geolocation, etc. This presentation will address the comparison results in detail and the advantages of applying machine learning on the public trial registry data to predict a study’s operational metrics during the trial or a program’s planning.