((A-264)) (A-264) A Proposed Framework to Leverage TreeScan Methods in Real-World Data for Signal Detection

Background: TreeScan is a method developed by the United States (US) Food and Drug Administration (FDA) Sentinel Program for signal detection using claims and electronic health record (EHR) databases. While the TreeScan method has been available for signal detection, a recent survey showed that only 1 out of 18 companies used real-world data (RWD) for signal detection. A practical challenge of applying this method is about how well this method performs in detecting a safety signal.

Objectives: To explore the application of the TreeScan method using real-world data and evaluate its performance in signal detection

Methods: To evaluate the adaptability of the TreeScan method, we developed a process to leverage it using a generic cancer drug with a well-described safety profile as an example. The first step was to conduct the TreeScan analysis using a claims database. Using IQVIA PharMetrics Plus, patients treated with the generic cancer drug in 2020-2022, regardless of cancer type, were included (N=32,262). Two claims of an outcome of interest within 60 days were used to identify event occurrence. The potential signals were identified using the TreeScan method by comparing the incidence of each outcome before vs. after a patient was treated (self-controlled design). Then, the signals identified by the TreeScan method, but not present on the US label, were reviewed and selected for signal refinement.

Results: Among the 52 potential signals identified by TreeScan, 35 (67%) were already included in the US product label and 17 (33%) were considered new. Of the 17 new potential signals, 8 were considered potentially important based on clinical judgment and selected as candidates for further signal refinement. These included other lesions of oral mucosa, unspecified intestinal malabsorption, hypercalcemia, other specified disorders of bone, multiple sclerosis, elevated prostate specific antigen, elevated carcinoembryonic antigen, and elevated cancer antigen 125. The relative risk ranged from 2.24 for hypercalcemia to 3.52 for other specified disorders of bone.

Conclusions: The proposed process of detecting potential signals using the TreeScan method could be used for signal detection when real-world data is also available. The potential signals identified by the TreeScan method that were not present on the US label could be further evaluated by refinement analysis to better control for confounding.