Drug-device combinations containing active pharmaceutical ingredients marketed together with their delivery devices are among the costliest medications in the US. Generic competition is vital for lowering prices in the US and improving access, but generic entry can be slow for drug-device combinations. The Food and Drug Administration (FDA) applies a necessarily rigorous set of requirements before approving these products, given the complex interactions between ingredients and devices and the need for proper self-administration. Despite the many regulatory challenges required for evaluation of drug-device combinations, the FDA does not actively surveil adverse effects after approval, relying instead on spontaneous reports by patients and clinicians. These may be subject to reporting bias and often lack the granularity needed for regulators to distinguish among specific generic and brand-name products. Therefore, the FDA is increasingly considering real-world evidence (RWE) generated from large longitudinal healthcare database studies as a tool for informing regulatory decisions. Routine use of RWE to monitor the safety and effectiveness of generic drug- device combinations can help ensure the reliability of pharmaceuticals, instill confidence in patients and prescribers, facilitate timely regulatory action, and promote the public health goals of access and affordability. The proposed research will develop a principled, scalable, and reproducible approach for rigorous and timely evaluation of the safety and effectiveness of generic drug-device combinations compared to brand-name reference products using healthcare databases. We will demonstrate this approach for generating multi-phase actionable RWE in 2-3 use cases, including description of utilization patterns, evaluation of comparative effects, and systematic screening for unsuspected safety issues. The work will build on state-of-the-art methods in pharmacoepidemiology to conduct research in a transparent and reproducible fashion. We will first rigorously describe the uptake and clinical characteristics of patients using generic drug- device combinations compared to those taking branded versions (Aim 1). We will then examine several prespecified safety and effectiveness outcomes for selected products, adjusting for confounding using traditional propensity score methods with investigator selected covariates and high-dimensional propensity score approaches, which semi-automatically adjust for hundreds of covariates selected through machine- learning algorithms (Aim 2). We will supplement these analyses with studies to detect unsuspected safety signals using tree-based scan statistic methods, which simultaneously evaluate thousands of correlated safety outcomes while formally controlling type 1 error (Aim 3). We will then refine our understanding of potential safety signals in fully specified protocols with confounding adjustment tailored to specific outcomes (Aim 4). The end result will be a r...