mesothelioma machine learning algorithms were the most important variables selected by the algorithms for the prediction of prognosis in patients with asbestos exposure. Machine learning--based machine algorithms have been well-accepted, because they are a powerful and intuitive way to identify the relationships between variables and clinical outcomes \[[@B39-cancers-11-00842],[@B40-cancers-11-00842],[@B41-cancers-11-00842]\]. For example, one-dimensional hierarchical cluster analysis was recently utilized to identify risk factors for the development of asbestosis and non-malignant pleural abnormalities \[[@B42-cancers-11-00842]\]. Moreover, the machine learning--based algorithms can improve the accuracy of the prediction model by considering interaction between variables, as was previously mentioned in the section of logistic regression modeling. Hence, to identify the relationships between variables and survival outcomes of patients with pleural mesothelioma, machine learning--based algorithms such as *random forest*, *gradient boosting*, and *deep learning* are needed.
Furthermore, we identified several factors that were commonly used to predict the prognosis of patients with pleural mesothelioma, such as histology, smoking status, asbestos exposure, and tumor stage. However, our current results showed that there was a little or no evidence to support the usefulness of these factors to predict the survival outcomes of patients with pleural mesothelioma, except the tumor stage. Notably, the tumor stage was shown to be the most important prognostic factor in the advanced stage and poor performance group in [Table 2](#cancers-11-00842-t002){ref-type="table"}. In the advanced stage and poor performance group, median survivals of the patients without tumor stages IV and IVS were 14 months and 12 months, respectively, while those of patients with tumor stages IV and IVS were 7 months and 5 months, respectively (*p* \< 0.001). Based on these findings, the survival curve of patients with tumor stages IV or IVS ([Figure 3](#cancers-11-00842-f003){ref-type="fig"}) reflected the survival rates of all patients who were diagnosed with pleural mesothelioma. Therefore, the tumor stage can predict the survival outcomes of patients with pleural mesothelioma.
In this study, we identified factors that can predict the survival outcomes of patients with pleural mesothelioma. To the best of our knowledge, this is the first study to investigate the predictors of survival outcomes by focusing on machine learning algorithms in patients with pleural mesothelioma. However, our study has several limitations. First, the sample size of our study was small. Second, we performed the analyses using statistical software programs, which are also not well-known in the medical field, such as R software and SAS/STAT software. As such, a more detailed machine learning algorithm study will be needed to evaluate survival outcomes and predictors in pleural mesothelioma.
5. Conclusions {#sec5-cancers-11-00842}
==============
We found that age, smoking history, histological type, asbestos exposure, tumor stage, and lymph node status were useful to predict the survival outcomes of patients with pleural mesothelioma. We also identified several machine learning algorithms for predicting the survival outcomes of patients with pleural mesothelioma. As a result, these machine learning algorithms are expected to predict the survival outcomes of patients with pleural mesothelioma and lead to more accurate predictions of survival outcomes.
The following are available online at , Supplementary Table S1: Univariate logistic regression modeling of variables. Supplementary Table S2: Clinical factors associated with 5-year survival outcomes by Cox regression model in the patients with pleural mesothelioma (*n* = 167).
######
Click here for additional data file.
Conceptualization, C.-Y.H. and Y.-K.C.; methodology, C.-Y.H. and Y.-K.C.; validation, Y.-K.C.; formal analysis, Y.-K.C.; investigation, Y.-K.C.; resources, J.-W.C.; data curation, S.P.; writing---original draft preparation, S.P.; writing---review and editing, C.-Y.H.; visualization, S.P.; supervision, C.-Y.H. and J.-W.C.; project administration, J.-W.C.; funding acquisition, C.-Y.H., S.P., and J.-W.C.
This study was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science, ICT & Future Planning (NRF-2014R1A2A2A01002586), the Bio & Medical Technology Development Program of the NRF funded by the Korean government (MSIT) (NRF-2015M3A9E3052338), and the Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (NRF-2017R1D1A1B03027939).
The authors declare no conflicts of interest.
{#cancers-11-00842-f001}
{#cancers-11-00842-f002}
{#cancers-11-00842-f003}
cancers-11-00842-t001_Table 1
######
Patient characteristics of the study population.
Characteristics No. of Patients Percentage
-------------------------------- ----------------- ------------
Age
≤59 78 15.6
60--69 145 28.9
70--79 152 30.4
≥80 143 28.3
Unknown 2 0.4
Gender
Female 171 33.8
Male 335 66.2
Histologic Subtype
Epithelioid 226 44.6
Biphasic 152 30.4
Pleomorphic 117 23.0
Unknown 18 3.6
Smoking History
Yes 232 46.0
No 242 48.0
Unknown 32 6.4
Asbestos Exposure History
Yes 205 41.0
No 267 53.0
Unknown 44 8.8
Tumor Stage
I 36 7.1
II 59 11.7
III 86 17.0
IV 168 33.4
Unknown 165 32.9
Lymph Node Involvement
Yes 146 29.0
No 348 68.6
Unknown 22 4.4
Grade
Well-Differentiated 25 5.0
Moderately Differentiated 107 21.2
Poorly Differentiated 134 26.5
Unknown 248 49.3
Racial Origin
White 443 88.2
Black 24 4.8
Other 40 7.9
Unknown 9 1.8
Radiation Therapy
Yes 146 29.0
No 348 69.0
cancers-11-00842-t002_Table 2
######
Factors associated with 5-year survival outcomes by univariate logistic regression modeling (*n* = 167).
Variables Odds Ratio (95% CI) *p*-Value
------------------------------------- --------------------- -----------
Age 1.054 (
