Utilization of Data-Mining Techniques for Evaluation of Patterns of Asthma Drugs Use by Ambulatory Patients in a Large Health Maintenance Organization

A major problem of drugs utilization is to identify outlier patients who are using large quantities of drugs over extended periods of time. Today, healthcare and health insurance systems have to deal with an increased number of patients suffering from chronic diseases, such as asthma, who are continuously using a combination of several medications. This has caused a substantial increase in the cost of providing healthcare for such patients. In Israel, 11% of the national health care budget is spent on medications. However, healthcare management operations do not have the information that can assist in determining whether extensive multi-year drug utilization by a chronic patient is an outlier or misuse of resources. In this work, we construct a prediction model for asthma drug utilization by applying novel methods of knowledge discovery in time-series databases to a multi-year asthma drug utilization data set. Methods of mining utilization patterns combine clustering algorithms, clustering validity measures, and decision-tree classification algorithms. This methodology is applied to a regional patients' database maintained in 'Clalit Health Services' HMO, Beer-Sheva, Israel between January 2000 and November 2002. The clustering results reveal that 274 asthma patients who received 9,319 prescriptions during that period can be partitioned into three groups of utilization patterns, where ten patients (3.6%) who used 1,333 prescriptions (14.3%) are classified as outliers. The classification results show that the use of corticosteroids medications (oral or by inhalation) and the age of a patient can be considered as the main predictive factors in the induced models.