(BSPOT46) Tree-based Scan Statistics to Identify Drug Repurposing Signals: a Test Case using Sodium-Glucose Cotransporter-2 Inhibitors

Background: Most existing drug repurposing studies using real-world data focused on validating, instead of generating, hypotheses. Tree-based scan statistics, enabled by TreeScan, is a data mining tool that conducts scan statistics across a hierarchical tree of structured data and may be used to identify repurposing signals.

Objectives: Our objective was to identify repurposing signals for sodium-glucose cotransporter-2 inhibitors (SGLT2i) by scanning across a wide range of clinical outcomes using tree-based scan statistics.

Methods: We used an active-comparator, new-user design to create a 1:1 propensity-score matched cohort of SGLT2i and dipeptidyl peptidase-4 inhibitor (DPP4i) initiators between 10/1/2015 and 10/31/2019 in the Merative™ MarketScan® Research Databases, capturing commercially-insured individuals in the US. Incident outcomes were hospital and ambulatory diagnoses during the up to 2-year follow-up while on treatment. Tree-based scan statistics were estimated across a hierarchical tree based on ICD-10-CM diagnosis codes for inverse associations between SGLT2i exposure and outcomes. We used an adjusted p ≤ 0.01 as the threshold for statistical alerting to prioritize associations for evaluation as repurposing signals. We varied the analyses by the size of the outcome tree, scanning level, and clinical setting for incident outcomes to observe the impact on signals identified and prioritized.

Results: There were 80,510 matched SGLT2i-DPP4i initiator pairs with 215,333 outcomes among 45,444 SGLT2i initiators and 223,428 outcomes among 45,931 DPP4i initiators. In the main analysis, there were 18 prioritized associations (p ≤ 0.01), which include chronic kidney disease (p=0.0001), an expected signal, and anemia (p=0.0001). Heart failure (p=0.0167), another expected signal, was identified slightly beyond the statistical alert threshold. Further narrowing the outcome tree, scanning at different levels of the tree, and including outcomes from both hospital and ambulatory settings influenced the ability to identify repurposing signals. For example, scanning at a higher level of the outcome tree increased the statistical significance of the scan statistic for liver diseases (from p=0.183 to p=0.0452).

Conclusions: We identified signals that align with recently approved indications of SGLT2i, as well as potential repurposing signals supported by existing evidence but requiring future validation. Tree-based scan statistics can be valuable to generate repurposing hypotheses for existing drugs, with important considerations, such as selection of the p-value threshold for signal prioritization, specifications of the outcome tree, and clinical settings to capture outcomes.