Twenty-gene-based prognostic model predicts lung adenocarcinoma survival

Abstract

IntroductionLung adenocarcinoma (LAC) accounts for more than a half of non-small cell lung cancer with high morbidity and mortality. Progression of treatment has not accelerated the improvement of its prognosis. Hence, it is an urgent need to develop novel biomarkers for its early diagnosis and treatment.Materials and methodsIn this study, we proposed to identify LAC survival-related genes through comprehensive analysis of large-scale gene expression profiles. LAC gene expression data sets were obtained from The Cancer Genome Atlas (TCGA). Identification of differentially expressed genes (DEGs) in LAC compared with adjacent normal lung tissues was first performed followed by univariate Cox regression analysis to obtain genes that are significantly associated with LAC survival (SurGenes). Then, we conducted sure independence screening (SIS) for SurGenes to identify more reliable genes and the prognostic signature for LAC survival prediction. Another two lung cancer data sets from TCGA and Gene Expression Omnibus (GEO) were used for the validation of prognostic signature.ResultsA total of 20 genes were obtained, which were significantly associated with the overall survival (OS) of LAC patients. The prognostic signature, a weighted linear combination of the 20 genes, could successfully separate LAC samples with high OS from those with low OS and had robust predictive performance for survival (training set: p-value <2.2×10−16; testing set: p-value =2.04×10−5, area under the curve (AUC) =0.615). Combined with GEO data set, we obtained four genes, that is, FUT4, SLC25A42, IGFBP1, and KLHDC8B that are found in both the prognostic signature and DEGs of LAC in GEO data set.DiscussionThe prognostic signature combined with multi-gene expression profiles provides a moderate OS prediction for LAC and should be helpful for appropriate treatment method selection.

Integrating genetic mutations and expression profiles for survival prediction of lung adenocarcinoma

“…Zhao et al developed a TCGA-based model using 20 filtered genes that showed lower accuracy (61.5% vs. 67%) for predicting OS compared to ours. 43 Gao et al established a TCGA-based survival model using methylation-driven genes that was also inferior to ours (66% vs. 67%). 44 There were several limitations to our study.…”

“…Our study exhibits superiority over previous studies of survival prediction. Zhao et al developed a TCGA‐based model using 20 filtered genes that showed lower accuracy (61.5% vs. 67%) for predicting OS compared to ours . Gao et al established a TCGA‐based survival model using methylation‐driven genes that was also inferior to ours (66% vs. 67%) …”

Novel prognostic molecular markers in lung cancer (Review)

“…Genetic expression analysis strongly correlates with prognosis, especially the 20-gene set (66), therefore, it may become the main prognostic asset in the nearest future requiring well equipped genetic laboratories and cost-effective procedures so that it can become a part of routine testing. If the cost of the procedure was reduced, for instance by devising a standardized kit, it could be successfully used to predict the clinical course of NSCLC.…”

“…On the basis of the prognostic score, the authors compared the 20-gene set to a 4-gene set (FUT4, SLC25A42, IGFBP1 and KLHDC8B) and the compared 4-gene set turned out to be effective, yet, its capacity to predict the clinical course was much lower than in the 20-gene set. The 5-year survival in the high and low group was ~20 and 50%, respectively (66).…”

Marginal variable screening for survival endpoints

“…However, for practitioners faced with the problem of analyzing high‐dimensional data, practical guidance is missing, which of these many methods to apply. Consequently, in most applications (Zhao, Li, & Tian, ; Wang, Teng, Li, Hu, Liu, & Zheng, ; Xia et al., ), simple Cox model–based strategies are used, possibly leading to suboptimal results.…”