Discovering medical resource utilization in total knee arthroplasty (TKA) using rule-based method

Abstract

TKA is a highly effective means of treating (advanced knee arthritis) degenerative joint disease. Previous studies have demonstrated that a high surgical volume for total joint arthroplasty reduces morbidity and improved economic outcome, these methods for themselves are fraught with complexity, uncertainty and non-linear problem in terms of medical datasets may be unable to more accurately finding important information. As medical datasets often include a large number of features (attributes), some of which are irrelevant, and therefore it cannot intuitively understand the corresponding to main factors which affecting the resource utilizations of healthcare. In order to solve the problems mentioned above, this study employs specialist advice to filter relevant cases (records) and proposed an integrated five features selection methods to select the important features. Based on rough set theory (RST), the rules are extracted and compared with other methods in terms of accuracy. The contributions contain: (1) data screening based on specialist opinions, (2) two stage feature selection by analysis of variance (ANOVA) and proposed an integrated feature selection approach (IFSA), and (3) data discretization and rule generation by RST. The proposed model is verified by using three datasets for comparison accuracy. The results can provide a valuable reference for National Health Insurance Bureau (NHI) in establishing the TKA standard.

Introduction

TKA is one of the most successful surgical procedures in orthopedic surgery. A variety of factors made this possible. More than 10,000 cases of TKA have occurred in Taiwan annually, warranting its further exploration. Advances in medical knowledge systems have made it increasingly difficult to determine useful knowledge. In particular, conventional data analysis methods are ineffective in diagnosing diseases, explaining the strong potential of computer-based analyses in medicine. The knowledge discovery in database (KDD) concept has received considerable attention recently in medicine (McLeish et al., 1991). As a form of data mining, KDD identifies implicit, previously unknown, and potentially useful information extracted from data. Data mining automatically searches for large volumes of digital data through the use of patterns by tools such as classification, feature selection (attribute selection), cluster analysis, association rules, and anomaly detection.

TKA treatment procedures are increasingly available in Taiwan, nearly doubling from 6500 in 1998 to 12,016 by 2006. The use of computerized administrative data sets has facilitated outcome studies of total joint arthroplasty. They have demonstrated that a high provider volume proves to be of benefit to reduce morbidity and mortality and to improve economic outcome (Carlos and Jose, 1995, Hervey et al., 2003, Kreder et al., 2003, Katz et al., 2004, Kurtz et al., 2005, Nelson et al., 2006). Most of the above-mentioned studies just used only statistical method, which cannot accurately reflect the inherently nonlinear, ambiguous, and complex characteristic of medical data. However, RST does not require a preliminary or additional parameter regarding data, while capable of modeling highly nonlinear or discontinuous functional relationships to provide a highly effective means of characterizing complex and multidimensional patterns (Pawlak, 1991, Hashemi et al., 1998). Additionally, feature selection is a unique optimization method, helping to remove noisy features and reduce the dimensionality of data sets by deleting unsuitable features and, ultimately, improving the performance of data mining algorithms (Hall and Holmes, 2003). Due to increasing annual TKA procedures, health management organizations are increasingly pressured to reduce health-care costs while maintaining quality of care. Therefore, our motivation is analyzing medical resource utilization in TKA using the proposed model.