A large cohort study identifying a novel prognosis prediction model for lung adenocarcinoma through machine learning strategies.

Background: Predicting lung adenocarcinoma (LUAD) risk is crucial in determining further treatment strategies. Molecular biomarkers may improve risk stratification for LUAD.

Methods: We analyzed the gene expression profiles of LUAD patients from The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO). We initially used three distinct algorithms (sigFeature, random forest, and univariate Cox regression) to evaluate each gene's prognostic relevance. Survival related genes were then fitted into the least absolute shrinkage and selection operator (LASSO) model to build a risk prediction model for LUAD. After 100,000 times of calculation and model construction, a 16-gene-based prediction model capable of classifying LUAD patients into high-risk and low-risk groups was successfully built.

Results: Using a combined strategy, we initially identified 2472 significant survival-related genes. Functional enrichment analysis demonstrated these genes' relevance to tumor initiation and progression. Using the LASSO method, we successfully built a reliable risk prediction model. The risk model was validated in two external sets and an independent set. The expression of these 16 genes was highly correlated with patients' risk. High-risk group patients witnessed poorer recurrence-free survival (RFS) and overall survival (OS) compared to low-risk group patients. Moreover, stratification analysis and decision curve analysis (DCA) confirmed the independence and potential translational value of this predictive tool. We also built a nomogram comprising risk model and stage to predict OS for LUAD patients.

Conclusions: Our risk model may serve as a practical and reliable prognosis predictive tool for LUAD and could provide novel insights into the understanding of the molecular mechanism of this disease.