Novel artificial intelligence (AI)-based technology to improve oncology clinical trial fulfillment.

Abstract

2052 Background: Less than 5% of US adult cancer pts are enrolled on clinical trials. Challenges in clinical trial fulfillment limit available treatment options, slow enrollment and ultimately delay new therapies from reaching market. Pt screening requires multiple clinical team members to find pts that meet strict inclusion/exclusion criteria. We evaluated the impact of new technology, Deep Lens VIPER, in identifying more qualified pts for clinical studies, and reduction of staff burden. Methods: We implemented Deep Lens VIPER at Hoag Hospital (Newport Beach, California), accessing the electronic medical records and pathology systems (EMR/LIS) to effectively identify pts who are candidates for 20 ongoing recruiting clinical studies. VIPER was fed pt data from 5,706 surgical pathology pts over a 4-month period (October 1, 2019 - January 31, 2020). Proprietary AI identification and matching technology was configured to align cancer pts with those 20 clinical studies, each with unique study criteria. Following an initial machine-assisted triage step, a research coordinator was alerted when pts who met protocol criteria were ready for final approval steps. Results were analyzed and a qualitative assessment of usability was also performed. Results: VIPER was able to triage all 5,706 surgical pathology cases (100%), identifying 1,045 pts (18.3%) with malignant neoplasms that would qualify for further analysis for clinical trials enrollment. Further triage based on inclusion and exclusion criteria led to the identification of 150 previously unidentified pts for 16 of the 20 studies. The 16 different studies for which potential pts were identified, included 11 tumor types, 12 biomarkers and 3 basket studies. Working with the VIPER system, 1 novice care team member performed initial identification of all 150 previously unidentified pts. The VIPER system increased monthly candidate pt catchment for 16 of the 20 studies under investigation, which is approximately 600 patients annually added for final triage for studies being conducted. Conclusions: We demonstrate the use of an AI-based platform to identify pts for clinical trial enrollment who would be missed using traditional recruiting methods. One staff member effectively triaged participants from 20 different studies with unique inclusion/exclusion criteria. These studies were previously managed by 6 different care team members with limited time for recruitment. Scaling this platform to additional institutions and more studies is ongoing to validate these findings.