OBJECTIVES: To predict high-risk patients with cardiovascular disease (CVD) over a 1, 2, and 3 year time period. Good prediction models will enable health care providers to target high risk patients who would most benefit from intervention programs designed to improve CVD patient outcomes. METHODS: Using classification and regression tree (C&RT) analysis from AnswerTree (SPSS 3.0), risk models were developed using California Medicaid (Medi-Cal) medical and pharmaceutical claims data for 62,154 patients with a diagnosis of CVD. Variables defined for the 6-month pre-period were used to predict year 1, year 2, and year 3 total costs. To determine the predictive ability of the model, we designated high cost patients as those with total costs of greater than $10,000, and low cost patients as those with less than $10,000. RESULTS: Outpatient cost (of approximately $3600, >1 SD above the median) in the six months prior to diagnosis was the most common split. Other contributing factors were patient comorbidities, including Other Neurological Disorders (p <0.01), Deficiency Anemias (p <0.01), and Hypertension (p <0.01). Results for years 2 and 3 were similar to year 1 findings. With further examination of the data, we found that the small group of high cost patients at Year 1 continue to be high cost patients in the subsequent years, although nearly 14.5% drop out at year 2 and 14.5% dropout from year 2 to year 3. 35% of the sample was correctly grouped into the high-cost branch, while 98% of the low-cost subjects were correctly grouped into the low-cost branch. CONCLUSIONS: C&RT is a useful method in predicting high risk patients. As demonstrated in this sample, patients incurring high costs were signaled through outpatient utilization, and were correctly identified with a sensitivity of 35% and a specificity of 98%.