1434PA pilot study to implement an artificial intelligence (AI) system for gastrointestinal cancer clinical trial matching

Abstract Background Clinical trials improve our knowledge of diseases and treatments while providing patients access to investigational agents. However, approximately 3% of newly diagnosed cancer patients are enrolled in clinical trials in the United States. Identifying an appropriate trial for a patient is time-consuming and cumbersome in the busy clinical practice. The Watson for Clinical Trial Matching (CTM) cognitive system uses AI to derive patient and cancer-related attributes from structured and unstructured text found in the electronic health record. These attributes are matched to complex eligibility criteria in clinical trial protocols. Methods In April 2019, a pilot study was launched to test the feasibility of implementing CTM in Gastrointestinal (GI) oncology at Mayo Clinic in Rochester, MN. Two clinical research coordinators (CRCs) screened patients for potential clinical trials prior to their clinic visits using both CTM and the traditional manual screening method. To avoid bias, each CRC screened a separate set of patients by both methods alternating which methodology was used first. The clinical trial match results were blinded to both CRCs. For each method, time to complete the screen and number of potential clinical trial matches were recorded. Results A total of 35 GI cancer patients with new diagnosis, recent resection or restaging scans were analyzed. Patients were evaluated against 50 GI-specific drug therapy and multi-disease phase I clinical trials. Clinical trial matching using CTM took an average of 10.1 minutes (Range: 4 to 20 min) per patient compared to an average of 30.5 minutes (Range: 5 to 75 min) per patient (p