|
" Predictive Modeling of Hospital Length of Stay for Colorectal Surgery Using Hybrid Data Mining Metaheuristics "
Al Najjar, Wareef
Khasawneh, Mohammad T.
Document Type
|
:
|
Latin Dissertation
|
Language of Document
|
:
|
English
|
Record Number
|
:
|
1105159
|
Doc. No
|
:
|
TLpq2299780599
|
Main Entry
|
:
|
Al Najjar, Wareef
|
|
:
|
Khasawneh, Mohammad T.
|
Title & Author
|
:
|
Predictive Modeling of Hospital Length of Stay for Colorectal Surgery Using Hybrid Data Mining Metaheuristics\ Al Najjar, WareefKhasawneh, Mohammad T.
|
College
|
:
|
State University of New York at Binghamton
|
Date
|
:
|
2019
|
student score
|
:
|
2019
|
Degree
|
:
|
Ph.D.
|
Page No
|
:
|
164
|
Abstract
|
:
|
As health care expenditures are growing rapidly worldwide and more specifically in the United States, health care organizations are being forced to control their costs in order to be successful and be able to stay competitive in the health care field, the challenge is also to maintain a balance between providing the best quality of care while trying to control expenditures. Extended hospital length of stay (LOS) is an issue that should be given considerable attention by hospitals to avoid facing higher costs, since they are not reimbursed for the extra days a patient spends at the hospital. Moreover, the more days of hospital stay the more vulnerable a patient would be to complications such as hospital acquired infections. This research employs a data-driven framework of hybrid evolutionary techniques for feature selection, to be applied in a case study to predict length of stay for colorectal surgery patients in a hospital in middle Georgia. This framework applies a new approach that combines particle swarm optimization with genetic algorithm, to avoid premature convergence of PSO and maintain the diversity of the population, by using the parent selection methods such as tournament and roulette wheel, standard deviation of the fitness, and crowding distance of particles, to select particles for mutation. This framework uses clinical and non-clinical preoperative patient attributes, where PSO searches for the best feature subset guided by the fitness function that is composed of weighted R2 score and dimensionality reduction, its objective is to find the best subset of features that will maximize the fitness function. Selected features by PSO are used by most commonly used, state-of-the-art machine learning techniques for regression; linear regression, decision tree regressor, random forest regressor, support vector machine for regression, and multi-layer perceptron-regression to accurately predict actual LOS values. The new proposed PSO versions provided significantly better results than the standard PSO with a 50% dimensionality reduction. Using the proposed ideas for improving PSO search ability and performance, PSO-random forest regressor (ensemble) model can predict hospital LOS for colorectal surgery patients when their preoperative attributes are known, where the prediction is objectively accurate providing a tolerance of around 3 days (mean absolute error). The result of this research will provide health care organizations with a guideline to control hospital LOS for surgery patients, by providing early predictions for the patient stay at the hospital for different diagnosis related groups.
|
Subject
|
:
|
Industrial engineering
|
|
:
|
Surgery
|
| |