Retrospective studies on the association between metformin and clinical outcomes have mainly been performed on patients with non-resectable pancreatic ductal adenocarcinoma and may have been affected by time-related bias. To avoid this bias, recent studies have used time-varying analysis; however, they have only considered the start date of metformin use and not the stop date. We studied 283 patients with type 2 diabetes and pancreatic ductal adenocarcinoma following pancreaticoduodenectomy, and performed analysis using a Cox model with time-varying covariates, while considering both start and stop dates of metformin use. When start and stop dates were not considered, the metformin group showed significantly better survival. Compared with previous studies, adjusted analysis based on Cox models with time-varying covariates only considering the start date of postoperative metformin use showed no significant differences in survival. However, although adjusted analysis considering both start and stop dates showed no significant difference in recurrence-free survival, the overall survival was significantly better in the metformin group (Hazard ratio (HR), 0.747; 95% confidence interval (CI), 0.562–0.993; p = 0.045). Time-varying analysis incorporating both start and stop dates thus revealed that metformin use is associated with a higher overall survival following pancreaticoduodenectomy in patients with type 2 diabetes and pancreatic ductal adenocarcinoma.