Eskandari, Hanie
(2023)
Advancing Data-Driven Healthcare Models with Optimization-Based Calibration Methods.
Doctoral Dissertation, University of Pittsburgh.
(Unpublished)
Abstract
A number of health studies consider Markov models of individual-level disease progression. Typically these studies use simulation-based methods (SBMs) to calibrate the parameters of the models so that model outputs agree with target data related to the disease/condition under consideration. SBMs, however, require a large amount of computation time, ultimately limiting the complexity of models that these studies can actually calibrate
in practice. The black-box nature of these methods also has left open a number of fundamental questions (e.g., understanding conditions under which the parameters of Markov models of disease progression are identifiable). This thesis develops new optimization-based calibration methods that require less computation time than SBMs, explores fundamental questions left open by SBMs, and constructs Markov models for Opioid Use Disorder (OUD) progression and Total Joint Replacement (TJR) recovery.
In Chapter 2 of this thesis, we consider using disease prevalence target data to calibrate a class of discrete-time Markov chain (DTMC) models that have covariate-dependent transition probabilities. We formulate the calibration problem as a (deterministic) non-convex optimization problem and consider solving it with first order methods that just require relatively inexpensive matrix-vector multiplications (instead of simulations). We investigate the
performance of our methods through computational experiments and apply them in a case study on Opioid Use Disorder.
Chapter 3 of this thesis considers the problem of identifying the transition probabilities
of time-homogeneous DTMC models of natural history of disease from target mortality data.
We establish mathematical conditions under which the transition probabilities are identifiable, and we present a polynomial-time algorithm for computing the values of the transition probabilities when they are identifiable. Our approach is premised on an interesting connection to the theory of homogeneous symmetric polynomials.
In Chapter 4, we propose a Markov Decision Process (MDP) model that concurrently uses multiple patient-reported and performance-based measurements as variables that define
the state of the patients to dynamically assess the recovery progress of TJR patients. The model can be used as a tool to devise personalized post-discharge intervention plans for TJR patients.
Share
| Citation/Export: |
|
| Social Networking: |
|
Details
| Item Type: |
University of Pittsburgh ETD
|
| Status: |
Unpublished |
| Creators/Authors: |
|
| ETD Committee: |
|
| Date: |
14 September 2023 |
| Date Type: |
Publication |
| Defense Date: |
3 July 2023 |
| Approval Date: |
14 September 2023 |
| Submission Date: |
11 July 2023 |
| Access Restriction: |
2 year -- Restrict access to University of Pittsburgh for a period of 2 years. |
| Number of Pages: |
141 |
| Institution: |
University of Pittsburgh |
| Schools and Programs: |
Swanson School of Engineering > Industrial Engineering |
| Degree: |
PhD - Doctor of Philosophy |
| Thesis Type: |
Doctoral Dissertation |
| Refereed: |
Yes |
| Uncontrolled Keywords: |
Markov Models, Calibration, Identifiability, Opioid Use Disorder, Total Joint Replacement |
| Date Deposited: |
14 Sep 2023 13:42 |
| Last Modified: |
14 Sep 2023 13:42 |
| URI: |
http://d-scholarship.pitt.edu/id/eprint/45079 |
Metrics
Monthly Views for the past 3 years
Plum Analytics
Actions (login required)
 |
View Item |
![[img]](https://d-scholarship.pitt.edu/style/images/fileicons/application_pdf.png)
![[feed]](/images/feed-icon-32x32.png){kind=icon}