Mining frequent order sets

Update item information
Publication Type thesis
School or College School of Medicine
Department Biomedical Informatics
Author Che, Chengjian
Title Mining frequent order sets
Date 2007-05
Description Order sets, when used in conjunction with a Computerized Physician Order Entry (CPOE) system, has been recognized as effective methods for improving the efficiency and accuracy of the ordering process, leading to improved quality of care. However, lack of specificity in order sets has been recognized as a common problem, decreasing the efficiency of the ordering process, or even cause the failure to acceptance of a CPOE system. Patient-specific, disease-specific, or diagnosis-specific order sets have been recommended in CPOE systems. In this project we explored a methodology of discovering order sets from an existing orders database. The goal was to establish a framework to apply data mining technologies, build order sets and use them to critique existing physician-made order sets. We stated that an order set is a super set of many maximal frequent itemsets discovered from the orders database. We adopted the Maximal Frequent Itemset Algorithm (MAFIA) to discover maximal frequent itemsets. Fourteen mining-discovered order sets were created by postprocessing these maximal frequent itemsets. We employed the error rate metrics from the Fifth Message Understanding Conference (MUC-5) to critique the specificity of physician-made order sets. In addition, we use undergeneration, overgeneration, precision, and recall from MUC-5 to analyze the source of the discrepancies. The average error rate was 75.9% with a standard deviation of 0.03 among all 14 pairs of order sets. It meant that the physician-made order sets were discovered with a 75% difference when compared to the mining-discovered order sets. The differences were categorized in spurious and missing orders. The spurious orders were the orders that appeared in the physician-made order sets, but not appear in the mining-discovered order sets. The missing orders were the orders that were found in the mining-discovered order sets, but not in the physician-made order sets. By manually reviewing these differences, proper actions could be take to increase the number of disease-specific orders in the physician-made order sets, so that the specificity could be improved. Besides critiquing the specificity of existing order sets, this approach can be used to propose preliminary order sets instead of creating them from scratch, generate order sets for different time ranges, or monitor ordering pattern changes.
Type Text
Publisher University of Utah
Subject Order Sets; Intermount Health Care
Subject MESH Medical Records; Medical Records Systems, Computerized; Data Collection; Medical Errors
Dissertation Institution University of Utah
Dissertation Name MS
Language eng
Relation is Version of Digital reproduction of "Mining frequent order set". Spencer S. Eccles Health Sciences Library. Print version of "Mining frequent order set". available at J. Willard Marriott Library Special Collection. RA4.5 2007 .C48.
Rights Management © Chengjian Che.
Format Medium application/pdf
Format Extent 1,655,299 bytes
Identifier undthes,4483
Source Original: University of Utah Spencer S. Eccles Health Sciences Library (no longer available).
Master File Extent 1,655,387 bytes
ARK ark:/87278/s6fn1817
Setname ir_etd
Date Created 2012-04-24
Date Modified 2012-04-24
ID 191122
Reference URL