Show simple item record

dc.contributor.authorPious, Noveenaa
dc.date.accessioned2021-04-16T12:59:29Z
dc.date.available2021-04-16T12:59:29Z
dc.date.issued2021-04-16T12:59:29Z
dc.identifier.urihttp://hdl.handle.net/10222/80383
dc.description.abstractPathology laboratory testing is central to medical practice as most diagnostic and therapeutic decisions are guided by the patient’s bloodwork results. Pathology laboratory tests are ordered by clinicians, and it has been observed that a significant number of tests ordered by physicians are inappropriate—i.e. the test is redundant, not clinically relevant, or not compliant with clinical guidelines. Inappropriate pathology test ordering not only affects laboratory utilization, but it also compromises patient safety by producing falsely abnormal results which may require unnecessary interventions. Recent laboratory utilization studies point to a discretionary behaviour in ordering tests which can be modified by providing physicians with peer comparisons, targeted education and an audit of physician’s test ordering profile. In this thesis, we aim to stratify physicians based on their patient case-mix as opposed to their order type and volume (which is circumstantial and inconclusive as comparator variables). The ensuing physician stratification will be used to generate physician phenotypes to both understand the physician’s ordering behaviour and to perform peer comparisons with a similar patient case-mix. Using pathology test ordering data spanning 6 years (2012-2017), we developed physician clusters for three temporal cohorts—i.e. 1-year, 2-year and 6-year—to track variations in the test ordering over time. We pursued a machine learning approach to investigate the phenotypical factors of physician ordering. We applied an ensemble clustering approach using three centroid models k-means, k-medoids and affinity propagation. We found the best physician clusters at k= 3 for 1st cohort, k= 4 for the 2nd cohort and k=3 for the 3rd cohort. We observed that ensemble clustering approach achieved the best results, compared to individual clustering algorithms in terms of cluster stability. We identified physician phenotypes, which interestingly change over time, which provides clear indications of underlying factors contributing to physicians test ordering pattern.en_US
dc.language.isoenen_US
dc.subjectMachine Learningen_US
dc.subjectPathology Test Orderingen_US
dc.subjectEnsemble Clusteringen_US
dc.titleINVESTIGATING CLUSTER ENSEMBLE METHODS TO DEVELOP PHYSICIAN PHENOTYPES BASED ON PATHOLOGY TEST ORDERING PATTERNSen_US
dc.date.defence2021-03-31
dc.contributor.departmentFaculty of Computer Scienceen_US
dc.contributor.degreeMaster of Computer Scienceen_US
dc.contributor.external-examinern/aen_US
dc.contributor.graduate-coordinatorDR EVANGELOS MILIOSen_US
dc.contributor.thesis-readerDR SAMUEL STEWARTen_US
dc.contributor.thesis-readerDR RIYAD ALSHAMMARIen_US
dc.contributor.thesis-supervisorDR SYED SIBTE RAZA ABIDIen_US
dc.contributor.ethics-approvalNot Applicableen_US
dc.contributor.manuscriptsNot Applicableen_US
dc.contributor.copyright-releaseNot Applicableen_US
 Find Full text

Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record