Mining clinical data for trauma patients
- Degree Grantor:
- University of California, Santa Barbara. Computer Science
- Degree Supervisor:
- Linda R. Petzold
- Place of Publication:
- [Santa Barbara, Calif.]
- Publisher:
- University of California, Santa Barbara
- Creation Date:
- 2016
- Issued Date:
- 2016
- Topics:
- Computer science
- Keywords:
- Clinical Data,
Data Mining,
Text Mining, and
Survival Analysis - Genres:
- Online resources and Dissertations, Academic
- Dissertation:
- Ph.D.--University of California, Santa Barbara, 2016
- Description:
Trauma is the leading cause of death from age 15 to 49 worldwide. After admitted to intensive care units (ICUs), the majority of these deaths take place in the first hours or days, and much of the initial treatments and decision-making actions occur in the first minutes or hours after injury. Not surprisingly, the ICU has been found to be one of the sites where medical errors are most likely to occur. Tools that can provide quick and accurate assessments of a patient's condition can provide immense value in helping physicians to make well-informed critical decisions, and ultimately to curb the mortality rate for trauma patients.
In this thesis, we demonstrate our uses of the data-driven approach to study trauma patients' clinical data. Specifically, we first show our use of the hidden Markov model to identify blood states for trauma patients. We apply a hidden Markov model to patients' time-series multivariate blood factor-related measurements. Missing data in the time-series dataset is considered in the hidden Markov model. The hidden Markov model identifies 6 disease states and 3 stages. We analyze their relationships to the blood composition data and the coagulation cascade.
Next we demonstrate the cure time model, which predicts the mortality/survival and time to mortality/survival of trauma patients. It models the static data for dying patients, surviving patients, and their death/cure times jointly. We pro- pose a joint log-odds ratio, which can predict the mortality of patients using the information from both the logistic regression and Cox models. We show the performance of the cure time model.
Lastly we propose the survival topic model, which models patients' measurements, notes and mortality/discharge jointly, and predicts the probability of mortality/discharge as functions of time. It models each patient as a latent distribution of disease conditions, which we call topics. These conditions generate the measurements and notes and determine the patients' mortality. We derive a mean- field variational inference algorithm for this model. The empirical results of the survival topic model are presented in the thesis.
- Physical Description:
- 1 online resource (124 pages)
- Format:
- Text
- Collection(s):
- UCSB electronic theses and dissertations
- Other Versions:
- http://gateway.proquest.com/openurl?url_ver=Z39.88-2004&rft_val_fmt=info:ofi/fmt:kev:mtx:dissertation&res_dat=xri:pqm&rft_dat=xri:pqdiss:10194196
- ARK:
- ark:/48907/f3sq90jr
- ISBN:
- 9781369341218
- Catalog System Number:
- 990047190240203776
- Copyright:
- Yuanyang Zhang, 2016
- Rights:
In Copyright
- Copyright Holder:
- Yuanyang Zhang
Access: This item is restricted to on-campus access only. Please check our FAQs or contact UCSB Library staff if you need additional assistance. |