Open AccessPrognostic value of cancer antigen -125 for lung adenocarcinoma patients with brain metastasis: A random survival forest prognostic model
Author(s) -
Hao Wang,
Liuhai Shen,
Jianhua Geng,
Yitian Wu,
Huan Xiao,
Fan Zhang,
Hongwei Si
Publication year - 2018
Publication title -
scientific reports
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.24
H-Index - 213
ISSN - 2045-2322
DOI - 10.1038/s41598-018-23946-7
Subject(s) - akaike information criterion , medicine , nomogram , proportional hazards model , brain metastasis , adenocarcinoma , oncology , overfitting , lung cancer , survival analysis , metastasis , cancer , statistics , artificial intelligence , computer science , mathematics , artificial neural network
Using random survival forest, this study was intended to evaluate the prognostic value of serum markers for lung adenocarcinoma patients with brain metastasis (BM), and tried to integrate them into a prognostic model. During 2010 to 2015, the patients were retrieved from two medical centers. Besides the Cox proportional hazards regression, the random survival forest (RSF) were also used to develop prognostic model from the group A (n = 142). In RSF of the group A, the factors, whose minimal depth were greater than the depth threshold or had a negative variable importance (VIMP), were firstly excluded. Subsequently, C-index and Akaike information criterion (AIC) were used to guide us finding models with higher prognostic ability and lower overfitting possibility. These RSF models, together with the Cox, modified-RPA and lung-GPA index were validated and compared, especially in the group B (CAMS, n = 53). Our data indicated that the KSE125 model (KPS, smoking, EGFR-20 (exon 18, 19 and 21) and Ca125) was the best in survival prediction, and performed well in internal and external validation. In conclusions, for lung adenocarcinoma patients with brain metastasis, a validated prognostic nomogram (KPS, smoking, EGFR-20 and Ca125) can more accurately predict 1-year and 2-year survival of the patients.