Background: Biologic medications (biologics) are highly effective for diseases of the immune system, cancers, and other conditions; however, their high expense is a barrier to care and a burden to the healthcare system. Biologics cannot be exactly copied as “generic” medications. Biosimilars- similar, but not identical versions of biologic medications- are approved with large potential cost savings. However, VA providers and patients have concerns regarding biosimilar switching safety and effectiveness as disease-specific randomized controlled trials are not required for approval. Significance/Impact: Antagonists to tumor necrosis factor-α (Anti-TNFs) are the largest class of biologics with biosimilars where switching may be feasible to reduce costs; however how to safely and effectively integrate their use in a manner acceptable to patients is unknown. This proposal addresses the VA HSR priority of veteran safety, the ORD-wide research priority of increasing substantial real- world impact of VA research, and uses cross-cutting HSR methods of health systems engineering through a learning healthcare system. Innovation: Crohn’s disease (CD) and ulcerative colitis (UC) are the 1st and 2nd most common indications for Anti-TNFs in the VA and can serve as a model for a learning healthcare system approach for mitigation of adverse events related to biosimilar switching. Specific Aims: Aim 1a: To compare rates of adverse events in CD and UC patients continued on Anti-TNF originator to those switched to biosimilar. Aim 1b: To compare rates of CD or UC exacerbation in patients continued on Anti-TNF originator to those switched to biosimilar. Aim 2: To compare the accuracy and calibration of 2a) traditional regression models vs. 2b) machine learning models for predicting medication related adverse event related to Anti-TNF in VA users with CD and UC. Aim 3: To use deliberative democracy methods to engage Veterans, to elicit their preference regarding “like" medication switch programs with and without their knowledge and to develop consensus around treatment approaches. Methodology: Aim 1 will be achieved through a retrospective cohort study of CD and UC patients who received Anti-TNF from the national VA datasets from 2017-2019. Adverse events and exacerbations will be determined using a combination of administrative data and manual chart review. Analyses for Aim 1 will proceed by Poisson regression using GEE. Adjusted event rate ratios of patients switched to biosimilar compared to those who continued on originator biosimilar will be calculated with 95% confidence intervals and Wald p-values will be derived from the regression model estimates. Prediction of patients who have adverse events to Anti-TNF will inform selection of appropriate therapy, and guidance of patients for biosimilar switching. For Aim 2, both traditional regression models and machine learning models will be constructed to identify which model will be better for predicting Anti-TNF relat...