Show simple item record

dc.contributor.advisorBasu Roy, Senjuti
dc.contributor.authorHassan, Aftab
dc.date.accessioned2016-09-22T15:40:32Z
dc.date.available2016-09-22T15:40:32Z
dc.date.submitted2016-08
dc.identifier.otherHassan_washington_0250O_16303.pdf
dc.identifier.urihttp://hdl.handle.net/1773/36975
dc.descriptionThesis (Master's)--University of Washington, 2016-08
dc.description.abstractIn the last few years, legislations such as the Patient Protection and A ordable Care Act, also known as Obamacare, have emphasised the need for improving the quality of health care. Part of the programs introduced by this Act is the Hospital Readmissions Reduction Program (HRRP) which reduces payments to hospitals with excess readmissions. Hospitals are, therefore, constantly looking for ways to help reduce their readmission rate, and an idea of patients that are at a higher risk of getting readmitted is extremely bene cial. In this thesis, we rst look at ways to predict the chance of a patient getting readmitted, and investigate predicting future healthcare costs, to understand how much money the patient would have to spend on hospital expenses during the readmission visit. We, then design a multi objective medication recommendation framework. We would like to echo the sentiments of the medical community here, that, while reducing readmissions is important, there are also other factors that go into ensuring a successful patient discharge experience, such as reducing mortality rate and the patient's length of stay at the hospital during the subsequent visit. We propose a novel framework that recommends personalized medications to patients by analyzing the complex interplay among a multitude of factors, such as, demographic factors, medical diagnoses, clinical factors, and how they contribute to the three objectives - thirty day readmission, subsequent length of stay and mortality rate. We then nd the best medication combination to simultaneously reduce all three objectives by performing Multiobjective optimization. Our proposed framework is exible enough to include or exclude additional factors, as well as layers, and can even obey constraints provided by the domain experts (i.e., doctors) in the design of this hierarchical network. We present a case study validated by a domain expert as well as comprehensive experimental results based on the proposed approach to demonstrate the e ectiveness of our proposed framework. Our work is validated on the largest (1000+ bed) South Puget Sound healthcare system: MultiCare Health on their EMR data including medication transactions. While our primary e ort is to design medication recommendations for heart failure patients, the proposed framework could be adapted for other diseases as well. We also develop a web based service called Pathway Finder with the objective of visually exploring and discovering clinical pathways that reduce the patients chance of readmission. We, therefore, try and play our small part in improving the quality of health care in the United States.
dc.format.mimetypeapplication/pdf
dc.language.isoen_US
dc.subjectBayesian Networks
dc.subjectCost Prediction
dc.subjectHealthcare Analytics
dc.subjectMedication Recommendation
dc.subjectMulti Objective Optimization
dc.subjectPredictive Analytics
dc.subject.otherComputer science
dc.subject.otherArtificial intelligence
dc.subject.otherHealth care management
dc.subject.othercomputing and software systems
dc.titlePredictive Analytics and Decision Support for Heart Failure patients
dc.typeThesis
dc.embargo.termsOpen Access


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record