Comparison of four variable selection methods to determine the important variables in predicting the prognosis of traumatic brain injury patients by support vector machine.

Variable selection filter prediction prognosis support vector machine traumatic brain injury wrapper

Journal

Journal of research in medical sciences : the official journal of Isfahan University of Medical Sciences
ISSN: 1735-1995
Titre abrégé: J Res Med Sci
Pays: India
ID NLM: 101235599

Informations de publication

Date de publication:
2019
Historique:
received: 17 02 2018
revised: 22 03 2019
accepted: 13 08 2019
entrez: 19 12 2019
pubmed: 19 12 2019
medline: 19 12 2019
Statut: epublish

Résumé

Large amounts of information have called for increased computational complexity. Data dimension reduction is therefore critical to preliminary analysis. In this research, four variable selection (VS) methods are compared to obtain the important variables in predicting the prognosis of traumatic brain injury (TBI) patients. In a retrospective follow-up study, 741 TBI patients who were hospitalized for at least 2 days and had a Glasgow Coma Scale score of at least one were followed. Their clinical data recorded during intensive care unit (ICU) admission and eight-category extended GOS conditions 6 months after discharge were utilized here. Two filter- and two wrapper-based VS methods were applied for comparison. A support vector machine (SVM) classifier was then used, and the sensitivity, specificity, accuracy, and the area under the receiver characteristic curve (AUC) values were calculated. Theoretically, the variables selected by sequential forward selection (SFS) method would better predict the prognosis (AUC = 0.737, 95% confidence interval [0.701, 0.772], specificity = 89.2%, sensitivity = 58.9% and accuracy = 79.1%) than the others. Genetic algorithm (GA), minimum redundancy maximum relevance (MRMR), and mutual information method were in the next orders, respectively. The use of an SVM classifier on optimal subsets given by GA and SFS reveals that wrapper-based methods perform better than filter-based methods in our data set, although all selected subsets, except for the MRMR, were clinically accepted. In addition, for prognosis prediction of TBI patients, a small subset of clinical records during ICU admission is enough to achieve an accepted accuracy.

Sections du résumé

BACKGROUND BACKGROUND
Large amounts of information have called for increased computational complexity. Data dimension reduction is therefore critical to preliminary analysis. In this research, four variable selection (VS) methods are compared to obtain the important variables in predicting the prognosis of traumatic brain injury (TBI) patients.
MATERIALS AND METHODS METHODS
In a retrospective follow-up study, 741 TBI patients who were hospitalized for at least 2 days and had a Glasgow Coma Scale score of at least one were followed. Their clinical data recorded during intensive care unit (ICU) admission and eight-category extended GOS conditions 6 months after discharge were utilized here. Two filter- and two wrapper-based VS methods were applied for comparison. A support vector machine (SVM) classifier was then used, and the sensitivity, specificity, accuracy, and the area under the receiver characteristic curve (AUC) values were calculated.
RESULTS RESULTS
Theoretically, the variables selected by sequential forward selection (SFS) method would better predict the prognosis (AUC = 0.737, 95% confidence interval [0.701, 0.772], specificity = 89.2%, sensitivity = 58.9% and accuracy = 79.1%) than the others. Genetic algorithm (GA), minimum redundancy maximum relevance (MRMR), and mutual information method were in the next orders, respectively.
CONCLUSION CONCLUSIONS
The use of an SVM classifier on optimal subsets given by GA and SFS reveals that wrapper-based methods perform better than filter-based methods in our data set, although all selected subsets, except for the MRMR, were clinically accepted. In addition, for prognosis prediction of TBI patients, a small subset of clinical records during ICU admission is enough to achieve an accepted accuracy.

Identifiants

pubmed: 31850086
doi: 10.4103/jrms.JRMS_89_18
pii: JRMS-24-97
pmc: PMC6906917
doi:

Types de publication

Journal Article

Langues

eng

Pagination

97

Informations de copyright

Copyright: © 2019 Journal of Research in Medical Sciences.

Déclaration de conflit d'intérêts

There are no conflicts of interest.

Références

Crit Care Nurse. 2009 Aug;29(4):12-22; quiz following 22
pubmed: 19648595
J Oral Maxillofac Surg. 2007 Sep;65(9):1693-9
pubmed: 17719385
J Neurotrauma. 2013 Jan 1;30(1):17-22
pubmed: 22931390
Comput Biol Med. 2015 Nov 1;66:1-10
pubmed: 26327447
Methods Inf Med. 2016 Oct 17;55(5):440-449
pubmed: 27492342
S Afr Med J. 2014 Jun 17;104(7):492-4
pubmed: 25214051
Lancet Neurol. 2010 May;9(5):543-54
pubmed: 20398861
IEEE Trans Pattern Anal Mach Intell. 2005 Aug;27(8):1226-38
pubmed: 16119262
BMC Emerg Med. 2012 Nov 19;12:17
pubmed: 23157693
Theor Biol Med Model. 2014 May 7;11 Suppl 1:S7
pubmed: 25077572
J Res Med Sci. 2015 Dec;20(12):1166-71
pubmed: 26958051

Auteurs

Saeedeh Pourahmad (S)

Bioinformatics and Computational Biology Research Center, Shiraz University of Medical Sciences, Shiraz, Iran.
Department of Biostatistics, Medical School, Shiraz University of Medical Sciences, Shiraz, Iran.

Soheila Rasouli-Emadi (S)

Department of Biostatistics, Medical School, Shiraz University of Medical Sciences, Shiraz, Iran.

Fatemeh Moayyedi (F)

Department of Computer Engineering, Larestan University, Lar, Iran.

Hosseinali Khalili (H)

Trauma Research Center, Shahid Rajaee (Emtiaz) Hospital, Shiraz University of Medical Sciences, Shiraz, Iran.

Classifications MeSH