OBJECTIVES: The main objective is to define clusters of patient diagnoses and to use them to analyze breast cancer health outcomes. METHODS: The NIS records of 2005 were used. Patients diagnosed with breast cancer were extracted, and then clusters of strings of all diagnoses per hospitalization were defined. Logistic regression models were used to determine the risk of dying from hospitalization associated with each cluster and ANOVA models were used to evaluate the effect of diagnosis clusters on length of stay. Time series models were used to fit the data and predict one month of total charges and the number of hospitalizations for each cluster. The analysis was performed with SAS, SAS Enterprise Guide (EG), SAS Enterprise Miner (EM), and the SAS Time Series Forecasting System. RESULTS: Four clusters were found. These clusters had a significant effect on length of stay and in-hospital death. The best time series models were found to be the mean, linear trend and log linear trend. The cluster defined by breast cancer with internal body organ failure was found to be the worst condition with a longer in-hospital stay and a higher risk of in-hospital death. The one month predicted values for this cluster were found to be 942 hospitalizations and about $26 million in total charges. CONCLUSIONS: Cluster analysis is a useful method to study health outcomes. Enterprise Miner is an effective software for cluster analysis and Data Mining in general.