Research Article
Identifying the Prognosis Factors in Death after Liver
Transplantation via Adaptive LASSO in Iran
Hadi Raeisi Shahraki, Saeedeh Pourahmad, and Seyyed Mohammad Taghi Ayatollahi
Department of Biostatistics, School of Medicine, Shiraz University of Medical Sciences, Shiraz, Iran
Correspondence should be addressed to Seyyed Mohammad Taghi Ayatollahi; ayatolahim@sums.ac.ir
Received 28 April 2016; Accepted 7 August 2016
Academic Editor: Pam R. Factor-Litvak
Copyright © 2016 Hadi Raeisi Shahraki et al. Tis is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
Despite the widespread use of liver transplantation as a routine therapy in liver diseases, the efective factors on its outcomes are
still controversial. Tis study attempted to identify the most efective factors on death afer liver transplantation. For this purpose,
modifed least absolute shrinkage and selection operator (LASSO), called Adaptive LASSO, was utilized. One of the best advantages
of this method is considering high number of factors. Terefore, in a historical cohort study from 2008 to 2013, the clinical fndings
of 680 patients undergoing liver transplant surgery were considered. Ridge and Adaptive LASSO regression methods were then
implemented to identify the most efective factors on death. To compare the performance of these two models, receiver operating
characteristic (ROC) curve was used. According to the results, 12 factors in Ridge regression and 9 ones in Adaptive LASSO
regression were signifcant. Te area under the ROC curve (AUC) of Adaptive LASSO was equal to 89% (95% CI: 86%–91%),
which was signifcantly greater than Ridge regression (64%, 95% CI: 61%–68%) ( < 0.001). As a conclusion, the signifcant factors
and the performance criteria revealed the superiority of Adaptive LASSO method as a penalized model versus traditional regression
model in the present study.
1. Introduction
Liver transplantation is recognized as a well-established
therapy for patients with acute liver failure [1–3]. Despite the
fact that it has become widespread and recently the number
of liver transplants throughout the world has exceeded 15000
cases in a year, the clinical efective risk factors on liver
transplantation outcome are still controversial [4].
Logistic regression is the most common method for
assessing the efects of various factors on the binary outcome
[5]. Usually, in order to avoid modeling bias, at the initial stage
of modeling, a high number of variables are candidates [6].
But logistic regression may encounter with multicollinearity
problem (strong correlation between two or more than two
independent variables in regression models) in modeling the
relation among a high number of variables [7, 8]. In these
settings, Ridge regression is a traditional remedial method
which can control multicollinearity by imposing a slight bias
in the estimation of coefcients. Penalized regressions have
recently developed models in facing high dimensional data.
Imposing a penalty on the coefcients in penalized methods,
besides controlling the multicollinearity, represents a sparse
and interpretable model [9]. For instance, least absolute
shrinkage and selection operator (LASSO), as one of the
most famous penalized models, is applicable regardless of the
number of variables and sample size [10].
In some of the previous researches in liver disease, penal-
ized methods were applied and superiority of them versus
conventional statistical methods was confrmed by some
authors [11, 12]. On the other hand, although the risk of death
and their associated efective factors afer liver transplanta-
tion was investigated in some studies, due to the limitations of
conventional statistical methods, a few potential factors were
considered. Recently, penalized regression had been widely
used in medical sciences for modeling and identifying the
most important factors. However, artifcial neural networks
(ANNs) are frequently used as the nonparametric substituted
modeling methods in the issue of large sample size and a high
number of variables [2].
Hindawi Publishing Corporation
Journal of Environmental and Public Health
Volume 2016, Article ID 7620157, 6 pages
http://dx.doi.org/10.1155/2016/7620157