OBJECTIVE: While SGA therapy has been associated with new-onset diabetes, the relative risk of the agents and the impact of possible confounding variables have been questioned. This study evaluated the impact of dose and treatment indication on the relative risk of new-onset diabetes associated with SGA therapy while controlling for demographic, clinical and medication variables. METHODS: A retrospective database analysis capturing electronic medical records for adult Texas Medicaid enrollees taking antipsychotic monotherapy from January 1997 to December 2001, with a maximum follow-up of 12 months was used. Patients were stratified according to treatment dose (low, medium, high) and a hierarchy of mutually exclusive diagnostic categories: schizophrenia, bipolar disorder, dementia, psychotic disorder, non-psychotic disorder and no mental health indication. The incidence of diabetes was examined using multivariate logistic regression analysis. RESULTS: Data were available for 13,731 patients. At treatment-onset, the prevalence of diabetes was 16.9 %. The mean (SD) dose by indication for the most prevalent conditions (schizophrenia, bipolar disorder and dementia, respectively) were as follows: olanzapine (12.04mg (6.73); 8.91mg (5.78); 4.87mg (3.00)); quetiapine (273.16mg (203.86); 146.33mg (142.29); 79.59 (82.57)); and risperidone (3.55mg (2.37); 2.05 (1.76); 1.12 (0.85)). The overall incidence of diabetes was 2.6%. Multivariate logistic regression analysis showed no difference in the incidence of diabetes according to the agent used (p=0.281). Compared to risperidone, the odds of new-onset diabetes were 0.879 (95% CI: 0.653 to 1.184) and 0.683 (95% CI: 0.414 to 1.126) for olanzapine and quetiapine, respectively. Neither treatment indication (p=0.876) nor antipsychotic dose (p=0.274) were associated with the development of diabetes. CONCLUSIONS: Results indicate that the risk of new-onset diabetes does not differ among SGA agents (i.e., olanzapine, quetiapine and risperidone). While the dose of antipsychotic prescribed varied significantly according to treatment indication and patient age, neither dose nor indication were associated with the development of diabetes.