Using cluster analysis for medical resource decision making.

Escalating costs of health care delivery have in the recent past often made the health care industry investigate, adapt, and apply those management techniques relating to budgeting, resource control, and forecasting that have long been used in the manufacturing sector. A strategy that has contributed much in this direction is the definition and classification of a hospital's output into "products" or groups of patients that impose similar resource or cost demands on the hospital. Existing classification schemes have frequently employed cluster analysis in generating these groupings. Unfortunately, the myriad articles and books on clustering and classification contain few formalized selection methodologies for choosing a technique for solving a particular problem, hence they often leave the novice investigator at a loss. This paper reviews the literature on clustering, particularly as it has been applied in the medical resource-utilization domain, addresses the critical choices facing an investigator in the medical field using cluster analysis, and offers suggestions (using the example of clustering low-vision patients) for how such choices can be made.

Clinical low vision resource usage prediction.

In an era of increased demands and constrained budgets, it is necessary to make the best use of all available resources. This is difficult when specialized vision care, such as low vision clinical assessment, is involved because of the heterogeneity of the patient populations seen by such clinics. PURPOSE. This research attempts to discover if these diverse patient populations can be identified and clustered into groups based upon similarity of clinical resources use. Specifically, the inquiry examines the potential for a low vision patient resource utilization classification scheme at the Low Vision Clinic (LVC) in the Centre for Sight Enhancement (CSE), University of Waterloo. METHODS. From a sample of 99 patients consulting the LVC in a 3-month period, retrospective data collection involved abstracting and coding medical records containing information detailing each patient's demographic, diagnostic, therapeutic, and resource utilization characteristics. Cluster analysis using Hartigan's block clustering algorithm was then applied to the data. A replication study was completed using a sample of 99 patients visiting the LVC 1 year later. RESULTS. Patients can be classified into five iso-resource groups, hereby termed low vision patient resource groups (LVPRGs). The clusters represent a resource consistent and clinically coherent scheme for classifying low vision patients based upon resource requirements. As a measure of repeatability, the groups reemerged in the replication study. CONCLUSIONS. If the groupings demonstrate robustness in a field test, clustering algorithms in general, and LVPRGs in specific, may offer useful tools to enhance resource utilization in the LVC setting.