Fig. 3.

The impact of an unmeasured confounder on the current use of metformin examined by using the rule-out approach. We assumed that the prevalence of an unmeasured confounder was 60% in a-1 and b-1, as well as 10% in a-2 and b-2, respectively. The prevalence of current use of metformin was estimated to be approximately 50% in the study cohort. In all panels, the combinations of RR_CD and OR_EC that lie on the line or in the lower right zone of each line indicate an unmeasured confounder could account for the reduced risk of severe COPD exacerbation from current use of metformin among diabetic patients with COPD (Model 1: adjusted RR = 0.81, the upper 95% confidence limit = 0.91; Model 2: adjusted RR = 0.85, the upper 95% confidence limit = 0.95). OR, odds ratio; RR_CD, risk of severe COPD exacerbation from an unmeasured confounder; OR_EC, odds ratio between current metformin use and an unmeasured confounder; P_c, the prevalence of an unmeasured confounder in this study. *In model 1, we adjusted for all covariates with standardized difference > 0.1 in Table 1. ^†In model 2, we adjusted for all covariates with standardized difference > 0.1 in Table 1 and DM therapy regimen in the 180 days prior to the index date