OBJECTIVES: To analyze patterns of antihypertensive drug utilization and to forecast pharmacy utilization and costs in hypertensive patients. METHODS: The sample consisted of 23,272 patients who were continuously eligible for drug benefits for at least 14 quarters. Utilization was recorded and summarized for every quarter. The first two quarters represented the baseline utilization, and the following 12 quarters were used to determine individual utilization trajectories. Patients were included in the sample if they used antihypertensive drugs during the baseline period. The sample was split into analysis and test sub-samples. Model parameters developed in the analysis sample were used to forecast utilization in the test sample. Utilization for the fifth and twelfth quarters was predicted based upon information gathered from the previous quarter. The accuracy of the model was tested by comparing predicted and actual outcomes. The analysis was based on latent class models. Demographic characteristics, drug benefit details, and concurrent drugs served as covariates. Two outcome variables were computed for each patient: the number of prescriptions per quarter and the probability of exceeding a certain cost threshold. RESULTS: For number of prescriptions, the quarter-ahead forecasts were within 0.4 prescriptions of the actual figures. For prescription expenditures greater than $150, the difference between actual and estimated probabilities was 0.9% for the twelfth-quarter forecast and 4.4% for the fifth-quarter. Latent class models also accurately separated patients into low/high cost groups and increasing/decreasing cost groups by defining cost and utilization trajectories for individual patients. CONCLUSION: Latent class models produce accurate forecasts that can be used to improve the management of hypertensive patients. Compared to other forecasting techniques, these models produce results that can be more easily understood by a wide range of readers, a critical issue in outcomes research.