Gender-specific nomogram models to predict the prognosis of male and female lung adenocarcinoma patients: a population-based analysis

Background: Epidemiological and clinical prognosis differences between male and female lung adenocarcinoma (LUAD) patients have been frequently reported. To improve prognosis determinations, gender-specific nomogram models should be developed and validated to predict the prognosis of patients with LUAD. Methods: Using the Surveillance, Epidemiology, and End Results (SEER) database, LUAD patients diagnosed between 2010 and 2015 were used as SEER training and internal validation testing sets. Patients in Tianjin Chest Hospital with postoperative pathological diagnosis of LUAD from January 1, 2015 to October 1, 2016 were considered as Chinese external testing sets. Using the Kaplan-Meier method and log-rank tests, we compared all the included male and female LUAD patients' overall survival (OS) and lung cancer-specific survival (LCSS) rates. The female and male patients from SEER database were randomly divided into training and internal validation groups at a 7:3 ratio. Variables (P<0.05) in the multivariable LCSS Cox regression analysis were independent prognostic predictors of the nomogram models. Harrell's concordance index (C-index), calibration curves, decision curve analysis (DCA) curves, receiver operating characteristic (ROC) curves, and the area under the curves (AUCs) were used to test the calibration and accuracy of the gender-specific nomogram models. Results: A total of 32,654 LUAD patients (17,372 females and 15,282 males) were identified. Ten variables [age, marital status, tumor site, differentiation grade, derived American Joint Committee on Cancer (AJCC) stage, tumor size, historic stage, surgery, derived AJCC N stage and chemotherapy] were statistically significant in the multivariate LCSS Cox regression analysis, and visualized through the nomogram models. The female and male training nomogram C-indexes were 0.827 and 0.811, respectively. The 3-and 5-year AUCs of the LCSS were 0.881 and 0.872 in the female training set, respectively, and 0.879 and 0.881 in the male training set, respectively. The DCA results indicated that these nomogram models were excellent predictors of LUAD prognosis and can be used to supplement the prognostication of tumor, node, and metastasis (TNM) stage. Conclusions: Given the different incidence and prognosis of LUAD between men and women, we developed gender-specific nomogram models with good discrimination and calibration capacity to predict 3-and 5-year LUAD-specific survival.