Investigating Potential Differences in the Safety Profiles of Biosimilars relative to Originator Products using a Tree-Based Scan Statistic

Background: Disclaimer: The contents are those of the author(s) and do not necessarily represent the official views of, nor an endorsement, by FDA/HHS, or the U.S. Government.

Biosimilars are biologic medications that are highly similar to existing approved biologics with no identified clinically meaningful differences in safety, effectiveness, and quality. However, the complex nature of biologics generally means a biosimilar is not identical to the reference product. TreeScanTM is a signal identification approach that scans thousands of health outcomes simultaneously while adjusting for multiple scenarios that can be used to monitor the underlying assumption of no clinical differences.

Objectives: Compare the safety profile of filgrastim-sndz to that of its originator.

Methods: We queried four large health plans plus Medicare Fee-for-Service in the Sentinel Distributed Database (2016-2022) to identify adult new users of filgrastim-sndz or filgrastim. “New use” was defined as no study drug use in the 365 days prior to index. New users were 1:1 propensity-score matched on 100+ measured baseline characteristics. In the 183 days following index, we monitored 83,000+ non-pregnancy and non-cancer ICD-10-CM diagnosis codes. Diagnoses were incident if there were no similar outcomes (i.e., having same first 3 ICD-10 digits) in 400 days before the first diagnosis occurrence. Incident diagnosis codes were organized into clinically similar outcome groupings and analyzed for imbalances using unconditional Bernoulli tree-based scan statistics. P≤0.05 was the threshold for a statistical alert of imbalance. Outcomes were assessed by setting—inpatient (IP), emergency department (ED) and ambulatory (AV).

Results: 43,009 1:1 matched pairs were selected from >218 million enrollees. Characteristics imbalanced at baseline included ethnicity, year of index dispensing or administration, history of cancer and recent chemotherapy, history of organ transplant, and prescription drug utilization. After propensity score adjustment, 892,259 incident outcomes were observed, with 50% in the IP or ED settings. Four unique alerts (1 in IP/ED and 3 in IP/ED/AV) were generated but did not prompt regulatory action for reasons such as likely relatedness to factors other than a drug-event relationship.

Conclusions: Tree-based scan statistics can simultaneously evaluate thousands of safety outcomes to provide additional assurance beyond routine surveillance that biosimilar and originator products indeed have no clinically significant differences in safety.