Hemoglobin A1c (HbA1c) has been widely used as a measure of glycemic control in patients with diabetes mellitus. However, the role of HbA1c in the prediabetic stage has received little attention, although the effectiveness of early intensive intervention for the prediabetic patients defined by plasma glucose is well documented. In this study, routine clinical data in a Japanese university hospital were used to address whether the level of HbA1c predicted drug treatment for diabetes mellitus. The study population consisted of 38,628 adolescent and adult patients (>or=15 years old) who used routine medical care services provided by the hospital and who were prescribed no anti-diabetic drugs in the 12 months before the 3-year study period. The rates of starting anti-diabetic drugs during the study period were calculated for subgroups divided by the level of HbA1c followed by Kaplan-Meier survival analysis and Cox proportional-hazard regression. In addition, the medical records of middle-aged patients with borderline (5.6-6.4%) and high (>or=6.5%) levels of HbA1c were examined to estimate the rates of initial drug treatment for type 2 diabetes mellitus. The rate (95% confidence interval; CI) of beginning anti-diabetic drug therapy was 1.5 (0.9-2.5)/1000 patient-years in patients with a normal level of HbA1c (<5.6%), which is similar to 2.1 (1.8-2.5)/1000 patient-years in patients with no HbA1c data, but lower than the value of 23.0 (18.6-28.6) and 161.8 (144.8-180.7)/1000 patient-years in those with borderline and high levels of HbA1c, respectively (P < 0.001). In a small group of 513 middle-aged patients, the rates of initial drug treatment for type 2 diabetes mellitus were 39.4 (28.1-55.1)/1000 patient-years and 270.4 (209.4-349.0)/1000 patient-years in those with borderline and high levels of HbA1c, respectively. A borderline (5.6-6.4%) or high (>or=6.5%) level of HbA1c was found to strongly predict future drug treatment for diabetes mellitus.