Link to the University of Pittsburgh Homepage
Link to the University Library System Homepage Link to the Contact Us Form

DEVELOPMENT AND UTILIZATION OF BAYESIAN PROGNOSTIC MODELS IN A LEFT VENTRICULAR ASSIST DEVICE DECISION SUPPORT TOOL

Lohmueller, Lisa C (2018) DEVELOPMENT AND UTILIZATION OF BAYESIAN PROGNOSTIC MODELS IN A LEFT VENTRICULAR ASSIST DEVICE DECISION SUPPORT TOOL. Doctoral Dissertation, University of Pittsburgh. (Unpublished)

[img]
Preview
PDF (Thesis update 7/26/18 at 1153)
Primary Text

Download (4MB) | Preview

Abstract

Heart failure is a chronic, progressive condition that affects over 6 million Americans. The gold standard treatment for advanced heart failure is heart transplant. However, when a donor heart is not available, or the patient is not eligible, patients may receive a mechanical circulatory support device such as a left ventricular assist device (LVAD).

LVADs can improve patient survival and increase patient quality of life but they also require significant changes in lifestyle and carry with them risks of adverse events, such as re-hospitalization, gastrointestinal bleeding (GI), stroke, or right heart failure. LVAD decision making for physicians and patients requires extensive discussion of the trade-off between benefits, risks, and associated lifestyle changes. Decision support tools for patients and their caregivers are in development but are not personalized and are limited to general educational information.
Using Bayesian modeling, a machine learning method of data analysis, I developed novel predictive models for three sets of LVAD outcomes: all-cause mortality, recurrent gastrointestinal (GI) bleeding, and pump-dependent ischemic stroke. The mortality models performed better than current risk scores with receiver operating characteristic area under the curve (ROC AUC) of 70-71% in a multi-center validation cohort and 76-79% in a contemporary single-center study. The recurrent GI bleeding models performed with ROC AUCs of 68% and 60%, revealed the importance of hemoglobin/hematocrit levels and inflammation in driving risk, and are the first models for this outcome. The ischemic stroke models out-performed the current ischemic risk score with ROC AUCs of 64-66%.

In addition to model development, I explored how to present prognostic information to decision making stakeholders: physicians, patients, and caregivers. I accomplished this with three studies: pilot testing the usability of an online application for physicians, surveying potential LVAD patients’ interest in healthcare engagement, and comparing the interpretation of prognostic information in different visual formats between patients and the general population. The results of these studies indicated that survival predictions are the most important outcome in decision making; patient numeracy is a key determinant of decision making engagement; and use of line graphs to present prognostic information is well-suited to all stakeholders.


Share

Citation/Export:
Social Networking:
Share |

Details

Item Type: University of Pittsburgh ETD
Status: Unpublished
Creators/Authors:
CreatorsEmailPitt UsernameORCID
Lohmueller, Lisa Clec77@pitt.edulec770000-0002-3031-396X
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Committee ChairShroff, Sanjeevsshroff@pitt.edu
Thesis AdvisorAntaki, Jamesantaki@cornell.edu
Committee MemberHirschman, Alanalh138@pitt.edu
Committee MemberPadman, Remarpadman@cmu.edu
Date: 25 September 2018
Date Type: Publication
Defense Date: 5 December 2017
Approval Date: 25 September 2018
Submission Date: 19 July 2018
Access Restriction: No restriction; Release the ETD for access worldwide immediately.
Number of Pages: 191
Institution: University of Pittsburgh
Schools and Programs: Swanson School of Engineering > Bioengineering
Degree: PhD - Doctor of Philosophy
Thesis Type: Doctoral Dissertation
Refereed: Yes
Uncontrolled Keywords: Machine learning, LVAD, Shared decision making, User interface design
Related URLs:
Date Deposited: 25 Sep 2018 14:47
Last Modified: 25 Sep 2018 14:47
URI: http://d-scholarship.pitt.edu/id/eprint/34925

Metrics

Monthly Views for the past 3 years

Plum Analytics


Actions (login required)

View Item View Item