OBJECTIVES: Real-world health care data are increasingly used to study the effectiveness of interventions in large, heterogeneous patient populationsIt is important to understand what biases could arise in the analysis of real-world data collected shortly after a new treatment is introduced, especially in the design of comparative effectiveness research. METHODS: This retrospective cohort study used pharmacy and medical claims from the HealthCore Integrated Research Database. Eligible patients were age 18 years or older with a diagnosis of gastric or gastroesophageal junction cancer (GC) (ICD-9-CM code 151.xx) who received trastuzumab between January 1, 2010 and March 31, 2014. Patients were followed until censoring, death, or end of the data stream (July 31, 2014). Descriptive statistics were used to summarize trastuzumab use. Differences by year were evaluated using F-tests for continuous variables and χ tests for categorical variables. RESULTS: The 188 eligible patients had a mean age of 60 years; 81% were male; 52% received radiation therapy, and 34% had a resection or gastrectomy. These factors were not significantly different by year of trastuzumab initiation. However, there were statistically significant differences (p<0.05) in time between GC diagnosis and initiation of trastuzumab [e.g. mean (SD) 2010: 174 (128.2) days vs. 2013: 91 (96.4)] and trastuzumab line of treatment [e.g. n (%) first-line 2010: 10 (44%) vs. 2013: 45 (83%)]. CONCLUSIONS: Over time, trastuzumab, which was approved in 2010 for previously untreated HER2+ GC, migrated from later to earlier lines of therapy despite consistent patient demographic characteristics. Since treatment effectiveness may change across lines of treatment, bias may arise when comparing treatments used in a specific line of therapy. This potential bias associated with introductory clinical use in GC should be addressed via appropriate study design, statistical adjustments and careful interpretation of results.