Prediction models in health-care systems : applications and insights
Abstract/Contents
- Abstract
- One of the canonical problems in health-care is to find a methodology for making decisions in the absence of adequate data and in the presence of uncertainty, considering the operational and non-operational costs and benefits of such decisions. Patients, physicians, and policy makers face this problem, sometimes on a daily basis, when making decisions about recommending one form of treatment over others, choosing treatment vs. palliative care, choosing the population to receive preventive diagnostic tests, and so on. The lack of such methodology is a critical problem for the health-care industry, with real and sometimes severe consequences for individuals or the industry as a whole. For example, a patient might choose to receive chemotherapy with the hope of having her cancer controlled when in fact she has only a few weeks left. A policy maker might recommend a screening policy for men older than 50 to detect early stages of prostate cancer without fully taking into account the relation between aggressive and costly interventions in the early stages of the disease and the outcome. An accessible and reliable methodology for making health-related decisions can benefit every constituency in the health-care system, however, finding such methodology is by no means an easy task for multiple reasons. First, data collection, storage, and access to the data has not been a simple and economical task until recently. As a result, the research into the relation between data and outcomes has mainly focused on small data-sets with the disadvantage that the insights from each study have been specific to the data-set used in that study. Therefore, generalizability has been an issue. Second, the uncertainty in medical outcomes is not well understood, i.e., even in the presence of patient data, two patients with seemingly equivalent characteristics can have different outcomes. There are two reasons for such paradoxical observation. First, the data-set might not be rich enough, either because of missing values or missing important variables. Second, the models chosen to characterize the outcome using the data might not be the right models for explaining the relation between the data and the outcome. This dissertation contributes to the literature on medical decision-making by proposing a framework that can enable patients, physicians, and policy makers to make more informed decisions in the absence of sufficient data and the presence of uncertainty. As examples, two health-care applications, namely, predicting non-attendance at medical appointments and colorectal cancer mortality are studied. With the first application, the underlying challenges are to find patients at risk of non-attendance at their next medical appointment and apply an optimal intervention to increase their chance of attending their appointment. While the application area is specific, the methodology developed to find optimal intervention policies is applicable to other health-care settings. Moreover, the approach to building prediction models for non-attendance to medical appointments used in this chapter enhances our understanding of the factors that can predict patient non-attendance. The second application was chosen with the goal of increasing our understanding of one of the major causes of death from cancer, namely, colorectal cancer, the factors that can predict it, and how these factors can be used by patients, physicians, and policy makers in making informed decisions. The methodology and approaches this dissertation uses and proposes can also be used to address other major problems in health-care such as predicting rehospitalization (30-day hospital readmission), chronic disorder variations, and hospital acquired infections.
Description
| Type of resource | text |
|---|---|
| Form | electronic; electronic resource; remote |
| Extent | 1 online resource. |
| Publication date | 2013 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Associated with | Zia, Leila |
|---|---|
| Associated with | Stanford University, Department of Management Science and Engineering. |
| Primary advisor | Bjarnadottir, Margret V |
| Primary advisor | Weyant, John P. (John Peter) |
| Thesis advisor | Bjarnadottir, Margret V |
| Thesis advisor | Weyant, John P. (John Peter) |
| Thesis advisor | Goel, Ashish |
| Advisor | Goel, Ashish |
Subjects
| Genre | Theses |
|---|
Bibliographic information
| Statement of responsibility | Leila Zia. |
|---|---|
| Note | Submitted to the Department of Management Science and Engineering. |
| Thesis | Thesis (Ph.D.)--Stanford University, 2013. |
| Location | electronic resource |
Access conditions
- Copyright
- © 2013 by Leila Zia
- License
- This work is licensed under a Creative Commons Attribution Non Commercial No Derivatives 3.0 Unported license (CC BY-NC-ND).
Also listed in
Loading usage metrics...