Diabetes patient heterogeneity is well recognized. Statistical clustering methods are gaining popularity as a precision medicine tool for identifying diabetes subgroups. We evaluated data-driven clusters among individuals with newly diagnosed diabetes and examined associations with glycemic response to initial treatment in real-world care. We studied individuals in the US Veterans Affairs Health Care System with newly diagnosed diabetes based on diagnosis codes and medications who were initially treated with a single diabetes medication from 2005-2013. We performed k -means clustering as previously described based on age, body mass index (BMI), and hemoglobin A1c (HbA1c) at the time of diabetes diagnosis. We compared 5-year glycemic response by initial medication class (metformin, sulfonylurea, or insulin) in each cluster using linear mixed effects models, adjusted for age, race, sex, BMI, glomerular filtration rate, and HbA1c, and including a medication*time term to assess medication-specific HbA1c trends. Four clusters best fit the data for the 325,250 individuals studied (mean age 63.5 years, BMI 32.7 kg/m 2 , HbA1c 7.5%). Clusters corresponded to those identified previously in other studies with three characterized by extreme values of clustering variables: Cluster 1 with mean HbA1c 11.3% (N=35,071), Cluster 2 with mean BMI 41.1 kg/m 2 (N=70,909), Cluster 3 with mean age 76.8 years (N=92,363), and Cluster 4 with intermediate values for all the three variables (N=126,907). HbA1c trends differed by medication class in all four clusters (p<0.0001 for all). However, differences in HbA1c trends by medication class were clinically small in all 4 clusters (Figure). Mean HbA1c trend was lowest among metformin users, intermediate among sulfonylurea users, and highest among insulin users in all clusters. Conclusion: Data-driven clustering of individuals with newly diagnosed diabetes using routinely collected clinical variables may not help guide initial diabetes treatment choice with regard to glycemic control.