Patterns of hospital utilization are multifaceted with causal mechanisms as diverse as patient populations. There is a need to identify patients with preventable high hospital utilization. Most research employs a direct modeling of utilization counts. This strategy largely identifies immutable factors (i.e., race and age) and provides little guidance on policy changes. We propose an alternate framework – segmenting the utilization population into groups and establishing that one group “uses” hospital services at higher rates than others. Using administrative data, we employ predictive modeling to determine whether individuals are likely to fall into a particular segment. We apply the framework to data from a rural hospital, examining transitions from home to long-term care among older at-risk patients. This segment represented approximately 20 percent of inpatient high utilization, 15 percent of ER utilization and 11 percent of 30-day readmissions. Predictive modeling then identified unique ICD-9 code groups with increased likelihood of later transition. These groups provide opportunities to intervene, either preventing further hospital utilization or providing guidance for likely later transfer. While this technique is popular in business analytics, it is not as widespread in health care. We demonstrate that applying the framework allows for discussion of policy changes and interventions.