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

Data-Driven Management of Intensive Care Units

Ulukus, Mehmet Yasin / MYU (2017) Data-Driven Management of Intensive Care Units. Doctoral Dissertation, University of Pittsburgh. (Unpublished)

This is the latest version of this item.

Download (3MB) | Preview


An Intensive Care Unit (ICU) is a specialized section of a hospital that provides comprehensive and continuous care for patients in critical conditions. Around 20% of hospital operating costs are due to ICUs, and this percentage has been increasing. Modeling patient flow through an ICU is challenging due to significant heterogeneity of patient cases and high variability of evolving patient conditions. Using a highly detailed data set of ICU patients from a single health system, we build a stochastic and dynamic model of patient physiology that can significantly improve ICU operations and predictions.
Scoring systems that assess the severity at admission or progression of severity during the stay have been used to predict the outcome (mortality or readmission). Existing scores are not sufficient to predict readmissions or mortalities after transferring to a lower level care unit. We present ICU outcome prediction models that perform better than existing models and could be used to benchmark ICU discharge policies and guide post-ICU resource needs.
We consider the transfer operations of patients to a downstream unit. In current practice, downstream beds are requested once a patient is clinically ready to be transferred. We investigate anticipative bed requests that can be made before a patient is ready for transfer. Patient health is described via a novel transfer readiness score created using our readmission prediction model that we incorporate into a Markov decision process model. Our numerical results indicate that an anticipative transfer request policy can significantly improve the system performance. We investigate the sensitivity of policy change upon cost parameter estimation errors by using robust models, and demonstrate that proactive strategies are more beneficial than reactive current policy in most scenarios.
We present an explicit stochastic length of stay model considering patient physiology modeled by the transfer readiness score as well as transfer delays. We characterize the stochastic process under certain assumptions. We show that the model demonstrates a moderate performance in fitting the underlying distribution of the length of stay, and improvements on the score will improve the predictive power of the model.


Social Networking:
Share |


Item Type: University of Pittsburgh ETD
Status: Unpublished
CreatorsEmailPitt UsernameORCID
Ulukus, Mehmet Yasin / MYUmyu1@pitt.edumyu1
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Committee ChairMaillart,
Committee MemberSchaefer,
Committee MemberClermont,
Committee MemberPang,
Committee MemberRajgopal,
Date: 27 September 2017
Date Type: Publication
Defense Date: 21 July 2017
Approval Date: 27 September 2017
Submission Date: 25 July 2017
Access Restriction: 5 year -- Restrict access to University of Pittsburgh for a period of 5 years.
Number of Pages: 104
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: Operations research, Markov decision processes, simulation, statistical data analysis, classification models, intensive care unit, hospital operations, medical decision making
Date Deposited: 27 Sep 2017 19:08
Last Modified: 27 Sep 2022 05:15

Available Versions of this Item


Monthly Views for the past 3 years

Plum Analytics

Actions (login required)

View Item View Item