Problem: Construct data in a way that uses two years of data to predict high utilization the third year. In the data include all components that we used: demographics, diagnoses, procedures, and drug use. Does the model work better than one that uses only previous year?