Using Joint Variable Importance Plots to Prioritize Variables in Assessing the Impact of Glyburide on Adverse Birth Outcomes

The only pharmacologic treatment for gestational diabetes (GDM) approved by U.S. Food and Drug Administration is insulin. However, due to improved ease of use and lower cost, oral antidiabetic medications, such as glyburide, are prescribed more commonly than insulin. We investigate glyburide's impact on two adverse perinatal outcomes compared to medical nutritional therapy, the universal first-line therapy, in a large, population-based cohort. At the design stage, we employ matching to select comparable treated subjects(received glyburide) and controls (received medical nutritional therapy). Multiple background variables were associated with GDM treatment modality and perinatal outcomes; however, there is ambiguity about which of the many potential confounding variables should be prioritized in matching. Standard selection methods based on treatment imbalance alone neglect variables' relationships with the outcome. Thus, we propose the joint variable importance plot (jointVIP) to guide variable prioritization for this study. This plot adds outcome associations on a second dimension to better contextualize standard imbalance measures, further enhances variable comparisons using unadjusted bias curves derived under the omitted variable bias framework, and can produce recommended values for tuning parameters in existing methods. After forming matched pairs, we conduct inference for adverse effects of glyburide and perform sensitivity analyses to assess the potential role of unmeasured confounding. Our findings of no reliable adverse effect of glyburide inform future pharmacologic treatment strategies to manage GDM.