Episode-Based Conceptual Mining of Large Health Collections

Abstract

In many countries, health care undustry is challenged by growth of costs associated with the use of new treatments or diagnostic techniques; wasteful or inefficient health care practices where funds are unnecessarily spent with no additional benefits to patients; health fraud where those who either provide or receive health services misrepresent those services to attract higher benefits. It has become very important for health service administrators to better understand current health care trends and patterns and associated costs to estimate health costs into the future. The key characteristics of a health system are hospital care, visits to medical practitioners, the consumption of pharmaceuticals calculated with regards to the particular cohorts of patients. One of the measure units for such calculations is episode of care, which has a variety of definitions. Episodes take into account various indices of patient care, for instance, a patient’s age, ethnical background, gender, location, medical services provided, information about participating physicians, fees and some other. Aggregating these attributes is important for Medicare (Australia’s universal health scheme) administrators because they can then produce extensive reports on utilisation. From a data mining point of view, applying some definition of episode is a way to preprocess data according to some temporal principle that is also clinically meaningful. Besides, it is an opportunity to filter out those irrelevant attributes that will not be included in data analyses. Episodic mining of health data is also a method to compress transactional dataset into a collection of health care episodes, that are not so diverse due to the nature of services and standardised medical practice.