Construction and evaluation of a nomogram for predicting survival in patients with lung cancer.

ABSTRACT

BACKGROUND: Lung cancer is a heterogeneous disease with a severe disease burden. Because the prognosis of patients with lung cancer varies, it is critical to identify effective biomarkers for prognosis prediction. METHODS: A total of 2325 lung cancer patients were integrated into four independent sets (training set, validation set I, II and III) after removing batch effects in our study. We applied the microarray data algorithm to screen the differentially expressed genes in the training set. The most robust markers for prognosis were identified using the LASSO-Cox regression model, which was then used to create a Cox model and nomogram. RESULTS: Through LASSO and multivariate Cox regression analysis, eight genes were identified as prognosis-associated hub genes, followed by the creation of prognosis-associated risk scores (PRS). The results of the Kaplan-Meier analysis in the three validation sets demonstrate the good predictive performance of PRS, with hazard ratios of 2.38 (95% confidence interval (CI), 1.61-3.53) in the validation set I, 1.35 (95% CI, 1.06-1.71) in the validation set II, and 2.71 (95% CI, 1.77-4.18) in the validation set III. Additionally, the PRS demonstrated superior survival prediction in subgroups by age, gender, p-stage, and histologic type (p < 0.0001). The complex model integrating PRS and clinical risk factors also have a good predictive performance for 3-year overall survival. CONCLUSIONS: In this study, we developed a PRS signature to help predict the survival of lung cancer. By combining it with clinical risk factors, a nomogram was established to quantify the individual risk assessments.